Intelligent Medicine 2 (2022) 69–81

Contents lists available at ScienceDirect

Intelligent Medicine
journal homepage: www.elsevier.com/locate/imed

Review

Comparative analysis of breast cancer detection using machine learning

and biosensors
Yash Amethiya 1, Prince Pipariya 1, Shlok Patel 1, Manan Shah 2,∗
1
Dharmsinh Desai University, Nadiad, Gujarat, India
2
Department of Chemical Engineering School of Technology, Pandit Deendayal Energy University, Gandhinagar, Gujarat, India

a r t i c l e i n f o a b s t r a c t

Keywords: Breast cancer is a widely occurring cancer in women worldwide and is related to high mortality. The objective
Breast cancer of this review was to present several approaches to investigate the application of multiple algorithms based on
Detection machine learning (ML) approach and biosensors for early breast cancer detection. Automation is needed because
Machine learning
biosensors and ML are needed to identify cancers based on microscopic images. ML aims to facilitate self-learning
Biosensors
in computers. Rather than relying on explicit pre-programmed rules and models, it is based on identifying patterns
in observed data and building models to predict outcomes. We have compared and analysed various types of
algorithms such as fuzzy extreme learning machine – radial basis function (ELM-RBF), support vector machine
(SVM), support vector regression (SVR), relevance vector machine (RVM), naive bayes, k-nearest neighbours
algorithm (K-NN), decision tree (DT), artificial neural network (ANN), back-propagation neural network (BPNN),
and random forest across different databases including images digitized from fine needle aspirations of breast
masses, scanned film mammography, breast infrared images, MR images, data collected by using blood analyses,
and histopathology image samples. The results were compared on performance metric elements like accuracy,
precision, and recall. Further, we used biosensors to determine the presence of a specific biological analyte by
transforming the cellular constituents of proteins, DNA, or RNA into electrical signals that can be detected and
analysed. Here, we have compared the detection of different types of analytes such as HER2, miRNA 21, miRNA
155, MCF-7 cells, DNA, BRCA1, BRCA2, human tears, and saliva by using different types of biosensors including
FET, electrochemical, and sandwich electrochemical, among others. Several biosensors use a different type of
specification which is also discussed. The result of which is analysed on the basis of detection limit, linear ranges,
and response time. Different studies and related articles were reviewed and analysed systematically, and those
published from 2010 to 2021 were considered. Biosensors and ML both have the potential to detect breast cancer
quickly and effectively.

1. Introduction lives could be saved through innovations in intelligent decision-making

strategies [2] and technologies [3–4].
Technologies in healthcare include maintenance and retrieval of Cancer is characterized by the rampant and aberrant growth of cells
electronic medical records of patients and devices involved. Cancer de- due to a combination of characteristic genetic and epigenetic defects.
tection has always been a challenge in the diagnosis and treatment plan This uncontrolled growth of the cells contributes to tumour develop-
for haematological diseases. Currently, an overwhelming percentage of ment. If the tumour begins to rapidly metastasize to other organs and
the population is affected by one or more diseases. Recent years have systems of the body as the cancer progresses, the disease may already be
seen tremendous advances in medical science. Despite these advance- incurable when discovered [5]. Breast cancer primarily affects women
ments, there is still a huge lack of information among the public regard- (with < 1% of cases affecting non-females); roughly one in eight women
ing health and disease. A large proportion of the population likely suffer develop breast cancer in their lifetime [6]. Roughly 2.1 million women
from health issues, some of which may even be fatal [1]. In addition to are diagnosed with breast cancer annually, and the most severely af-
improving the accuracy of the rapid detection of fatal conditions, adopt- fected are those between the ages of 40 and 70 years [7]. Therefore, the
ing safe, realistic techniques and using modern technology can reduce early diagnosis of breast cancer is paramount to good prognosis. Despite
the need for caregivers and reduce overall health care costs. Several the fact that the symptoms may be weak in the early stages, chances of

∗
Corresponding author: Manan Shah, Department of Chemical Engineering School of Technology, Pandit Deendayal Energy University, Gandhinagar, Gujarat,
India (Email: manan.shah@spt.pdpu.ac.in).

https://doi.org/10.1016/j.imed.2021.08.004
Received 22 June 2021; Received in revised form 24 August 2021; Accepted 24 August 2021
2667-1026/© 2021 The Authors. Published by Elsevier B.V. on behalf of Chinese Medical Association. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

survival dramatically increase if detected early [1]. The various screen-

ing methods used to diagnose breast cancer include fine needle aspira-
tion cytology (FNAC), ultrasound-guided surgical biopsy, and mammog-
raphy. In dense breasts, the rate of cancer detection using mammogra-
phy is very poor and about 10%–30% of cases go undetected [8–9].
It is important to identify cancer cells accurately to decrease mortality
rates, and this involves effective early cancer diagnosis and treatment
to increase the survival rate of cancer patients [10–13].
Machine learning (ML) is one of the most popular models to eas-
ily train machines and create predictive models for successful decision-
making. Machine learning helps with early diagnosis of breast can-
cer and determines the nature of the cancer by analysing the tumour
size. Machine learning methods are the leading approaches to obtain
favourable outcomes among classification and prediction problems.
Breast cancer research could benefit from ML techniques used to iden-
tify cancer and predict the presence or absence of tumours. Machine
learning techniques can also be used to predict tumour malignancy [14–
15]. To diagnose and monitor the diseases, conventional methods used
are highly based on the detection of the presence of specific signal fea-
tures by a human observer. In the past decade, the development of sev-
eral computer-aided diagnosis (CAD) approaches have been prompted
by many patients in intensive care units requiring constant monitor-
ing. The predominantly qualitative diagnosis criteria are turned into a
more concrete quantitative feature to improve upon the issue of clas-
sification in these techniques [16–17]. Multiple classifiers were tested
on three separate datasets, from the point of view of classification ac-
curacy. Classifiers like sequential minimal optimization and multilayer
perceptron (MLP) were also included in the study apart from classifiers
such as Naive Bayesian, IBK, and Decision Trees. The above were tested Figure 1. Block diagram of the breast cancer recognition system.
on various datasets which mainly consisted of the Wisconsin Breast Can-
cer (Original) Dataset (WBCD) and other Wisconsin Datasets such as the
Breast Cancer Wisconsin (Diagnostic) Dataset(WDBC) and Breast Can- fluid-based biomarkers are taken into account for the production of
cer Wisconsin (Prognostic) Dataset(WPBC). For the WDBC dataset, the electrochemical biosensors. Gene- or immune-based sensors are some
MLP and J48 classifiers were fused and assisted by the feature selection examples of modern electrochemical bio-affinity sensors that provide
method that exhibited a higher accuracy rate than non-fused versions. exceptional sensitivity, which is an essential tool for the detection of
For the WPBC dataset, the fusion of all four above-mentioned ML algo- early-stage cancer. Electrochemical devices are highly efficient, and
rithms exhibited a greater accuracy in finding tumour than conventional offer clinical advantages in identifying cancer and monitoring treat-
methods, while the suggested SMO proved to be a better approach and ment as well as a more precise method of detection. Advancement of
more precise for the WDBC dataset. Further, Salama et al. suggested a these portable devices will be crucial to quicken the process of can-
multi-classifier by figuring out which one of the fused classifiers would cer diagnosis and rapidly obtain analytical results [26]. Electrochemical
provide the best performance for any particular dataset [18–19]. The biosensors are among the best options for biochemical analysis given
Breast Cancer Model with ML algorithms is represented in Figure 1. The their added advantages of ultra-sensitivity, cost efficiency, simplicity
model can be used for the prediction of benign and malignant cancer of preparation and background as well as several unlisted advantages
cells. First, the Breast Image data is loaded, then Feature Extraction [13, 27–30].
takes place, and then the final classification model can be trained to
fulfil the task stated above. Malignant cancer begins with irregular cell
growth and can quickly spread to or infiltrate the surrounding tissues, 2. Search methodology and article selection
which makes it a life-threatening condition; by contrast, benign tumours
are considered non-cancerous and usually, non-lethal [15,20]. In this literature review, relevant articles were retrieved by querying
The International Union of Pure and Applied Chemistry (IUPAC), terms like [“Breast Cancer” or “Cancer Detection”] + “Machine Learn-
the world authority on chemical nomenclature and terminology, did ing” + [“SVM or ANN”] + [Biosensors or FET or Electrochemical”] in
not provide a formal definition for biosensors until 1997. A biosensor the following databases: Google Scholar, Research Gate, PubMed, Sci-
works in a way that can detect abnormalities in the human body by ence Direct, IEEE, and Springer.
monitoring biomarkers within the human body. Mammogram screen- All the applications where breast cancer detection can be used have
ing is recommended for women at normal risk of breast cancer and, by been discussed in this review. Studies conducted prior to 2010 are con-
combining this with certain biomarkers, we can increase the chances sidered to be less relevant as the techniques introduced had less accu-
of early detection. The main objective for the designing of a biosen- racy, were costly, and the scope of implementation was comparatively
sor was to detect bio-analytical molecules from various samples. It was less.
integrated in a way to fuse three different areas of science-chemistry,
biology, and engineering-into a single device [21-22]. The presence of
a tumour biomarker could indicate cancer in the human body [23]. 3. Advanced technology in breast cancer detection
Biomarkers for breast cancer can be categorised into one of the three
types: prognostic, diagnostic, and therapeutic; based on the severity Technological advancement in the medical sector has continuously
of the cancer, their presence can be detected [24-25]. Clinical and increased over the years. There are various technologies deployed for
medical teams have recognized several tumour biomarkers for deter- breast cancer detection. The major ones among these are presented in
mining the presence of cancer in an individual, and several specific Figure 2.

70
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

Figure 2. Technologies used for breast cancer detection.

3.1. Breast screening techniques 3.4. Big data analysis

Breast Screening refers to the early detection of diseases by testing Big data methods allow experts to exploit data sources, including
before symptoms appear. Genetic tests to identify inherited diseases and unstructured ones such as textual patient reports or images; thereby
physical or imaging-related examinations are all examples of testing that influencing medical research and ultimately patient care [33]. Cross-
can be done as part of a screening test. Breast cancer screening involves Database, Operational Big Data, and Analytical Big Data are some of
modalities such as mammography, magnetic resonance imaging (MRI), the methods used in Big Data analysis.
whole-slide imaging (WSI), and ultrasound. Usually hailed as the gold standard for breast imaging and diagnosis,
screen-film mammography is widely used for breast imaging [34]. Mam-
mography is used as a breast cancer detection, diagnostic, and screen-
3.2. Biosensors
ing tool and involves examining the human breast using low-level X-
rays [35-36]. Mammography is considered a good method for the de-
Biosensors are analytical devices used to quantify the biological char-
tection of ductal carcinoma in situ (DCIS) and calcifications. Before le-
acteristics of tissues and body fluids [31-32]. Some examples of biosen-
sion manifestation, breast cancer screening is best used for detecting
sors include optical biosensors, piezoelectric biosensors, and electro-
early-stage cancer. Since the advent of mammography, the mortality
chemical biosensors.
rates due to breast cancer have seen a dip. Mammography has a sen-
sitivity of 67% for true positives. Although widely accepted through
3.3. Machine learning the years, the detection of breast cancer by mammography is presently
considered quite weak when compared to other more advanced tech-
ML techniques are widely recognized as the method of choice in niques. Some of the disadvantages of mammography include radiation
breast cancer pattern classification and decision forecast modeling as exposure and a high false-positive rate. It is also possible that the ra-
it provides certain unique advantages in the critical feature detection diologist in charge of the diagnosis might occasionally miss a detail
from the overall complex datasets. KNN, SVM, DT methods have shown and generate a false-negative report. Last, this method is not very sensi-
promising results in the classification of data. Moreover, these methods tive for detecting early-stage cancer in women with dense breast tissue
greatly help in clinical decision-making and diagnosis. [37].

71
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

With time, biosensors are being widely accepted, as they have 20 years have passed since the introduction of the SVM based models as
demonstrated excellent selectivity in recent times to isolate and distin- a prognostic model for cancer [6]. Several optimized systems have been
guish the complementary sequences from the non-complementary ones, introduced for image processing in the medical sector such as CADe
showing promising outcomes for breast cancer diagnosis [38]. DNA- and CADx systems with the help of several algorithms based on ML.
based biosensors and electrochemical DNA biosensors are very attrac- CADe systems help detect objects that hold a high significance in clinical
tive compared to other various types of biosensors due to their high sen- terms, while CADx systems help quantify the malignancy of the clinical
sitivity, simplicity, and miniaturization [39-41]. There are many ways objects that are detected either manually or automatically [50-51]. Var-
in which the biosensors interface with each other on a molecular level; ious data mining algorithms and techniques have been applied in the
these include methods where the electric signal is obtained from hy- several studies conducted on a multitude of datasets to classify breast
bridizing with the target DNA and from the DNA probe itself [39, 41]. cancer. The primary advantage of such techniques and algorithms is
Given the simplicity of the process and application for nucleic acid their superior classification results. This has led to many researchers in-
hybridizations that have been useful in the detection of several dis- corporating techniques like data mining and the use of optimized ML
eases, some of which include tuberculosis [42], meningitis [43], and models into their studies that revolve around solving challenging and
lung cancer [44], DNA-biosensor-based diagnostics have attracted con- convoluted tasks [52].
siderable attention in the field of breast cancer detection too [41]. In Researchers have also worked on increasing the accuracy for the sur-
contrast to detection of biological cells and imaging of biomolecules, vival time prediction for someone diagnosed with breast cancer, by us-
electrochemical nano biosensors have many advantages, including low ing ML [6]. Essentially, this study attempts to evaluate the efficiency and
cost, versatility, high sensitivity and precision, durability, and rapid accuracy of the ML algorithms already in place for predicting survival
response. Variables such as capture efficiency, nanomaterials, and the time. The authors proposed an approach that used ML and added a new
size of the sensors considerably affect the sensitivity of electrochemical feature based on the concatenation of three features, namely tumour
nanobiosensors [34]. Even though biosensor and biomarker-based tech- stage, tumour size, and age at the time of diagnosis, into a single new
niques are on the rise, it comes with its own set of limitations. These feature. Then, they applied ML models using Support Vector Machine
techniques, although sensitive and selective, are associated with certain - Regression (SVR). The methods used by them provided optimistic re-
problems. Biomarker-based techniques are quite expensive and also re- sults. The authors also showed that more accurate predictions could be
quire trained personnel for the labelling process. They are also highly made using both linear and decision tree-based regression models of SVR
time-consuming, which discourages its wide adoption [24, 34]. and confirmed the same using cross-validation. Last, they concluded that
Medical diagnosis using ML techniques has gained considerable mo- the new tumour integrated clinical feature (TCIF) outperformed the ex-
mentum in the past decade. This growth in the adoption of ML tech- isting Nottingham prognostic index (NPI) feature.
niques is partly due to the fact that it affords better diagnosis of sev- Asri et al. [53] proposed an interbred classifier based on the ap-
eral diseases, attributed to improved symptom detection. Further, ML proach of knowledge discovery to detect the presence of breast cancer.
also helps to design a more personalized therapeutic proposal, as re- The paper’s objective was to perform a thorough analysis and provide
sults of the analysis can be used to submit new diagnostic hypotheses a comparison of the various ML algorithmic programs like SVMs, Naive
[45]. Breast cancer detection methods that have been employed utiliz- Bayes, K-Nearest Neighbours, and DTs. The WBCDs were used to imple-
ing ML algorithms include SVM, decision trees, and neural networks. ment the algorithms. Their main aim was to assess the performance of
The most widely adopted and the most useful ones are neural networks the algorithms on multiple parameters to develop a novel fusion algo-
(NNs) and DTs. However, their working mechanisms are quite differ- rithm that would display optimal execution. Based on their experiments,
ent. The main benefit of using a decision tree-based ML model is that the authors found that a classifier that fused three model types of SVM,
the algorithm used is easy to understand and has also been proven to be NB, and C4.5 could achieve 97.31% accuracy, which was the best per-
quite efficient. Moreover, the order of the training instances shown does former. They showed that this novel approach of using multi-classifiers
not affect the training efficiency of the model. Many ML models suffer was quite efficient and reliable in the prediction of breast cancer and its
from overfitting; however, pruning in the decision tree addressed this diagnosis. They also concluded that building an accurate and computa-
problem of overfitting. Nonetheless, the decision tree-based ML model tionally efficient classifier is a real challenge given that patients’ lives
also has some limitations. They come with the prerequisite that the data depended on it and should therefore be carefully monitored.
classes must be mutually exclusive. The end result decision tree is de- Sadhukhan et al. [54] employed a model, based on the variables
pendent upon the order in which the attributes were selected. Overly from the cell nucleus, that had been trained using ML. The algorithms
complex and difficult-to-interpret decision trees can be obtained owing at work in this study were K-NN and SVM. The performance of their
to certain errors while training. Furthermore, the missing values of cer- classifiers was determined and analyzed. With the help of a Bayesian
tain attributes can lead to confusion in the decision tree [46]. Neural Network, they aimed to compare and contrast the task using a dataset
networks can be used to address the issue of classification or regres- containing the feature values that were collected from the images of
sion. Representation of Boolean functions is also unlocked using neural cell slides using FNAC. Their research showed that all attempts were
networks, and they can handle noisy inputs with ease. However, neural directed towards developing an algorithm that enables one to predict
networks are more complex and difficult to understand than the simple whether the tumour is malignant or benign. The image provided had
working of decision tree algorithms. Moreover, the use of too many at- varying illumination at various locations that would differ depending on
tributes can lead to overfitting-a problem that does not exist when using the intensity of illumination. Three levels of classification were applied
decision trees [47]. to the image: black, white, and grey. The model showed a significantly
high level of efficiency (97.49%). Therefore, the authors concluded that
4. Machine learning in breast cancer detection faster diagnosis of breast cancer could be achieved in the future using a
process whereby the slide images from the cells obtained by FNAC could
ML is a subfield of artificial intelligence (AI) that uses soft-coding in- be pre-processed using automation to extract the relevant features and
stead of the conventional hardcoded approach. ML refers to the mecha- immediately be fed directly into the ML-based model.
nism wherein despite an absence of explicitly programmed instructions, Based on the Wisconsin Breast Cancer dataset and by measuring the
the machine can continue to learn from experience [48-49]. In cancer performance of the 11 ML algorithms used for the classification task, a
research, ML models have a long history of being helpful with not just comparative analysis between them was presented by Benbrahim et al.
the research but also with the practical implementation in cancer detec- [7]. To differentiate between benign and malignant breast lumps, the
tion [46]. For almost 30 years now, decision trees and artificial neural authors proposed a method for developing two classifiers using the fea-
networks have helped in the detection and diagnosis of cancer. Almost tures derived from the post-diagnosis images from FNAC. They aimed

72
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

to examine and analyze the accuracy of the given 11 different ML al- They observed a large number of incorrect classifications with the C4.5
gorithms and conclude which one would provide the best result. The algorithm and for the k-NN algorithm (34 and 33 incorrect instances, re-
results of the experiments showed that the neural network proved to spectively). Lucas Borges reported a comparison of two ML techniques-
be the most accurate among all the other 11 with an accuracy index of Bayesian network algorithm and J48. They mainly worked with the
96.49%, followed by linear discriminant analysis (LDA) and logistic re- WBCD dataset and developed two classifiers to distinguish benign from
gression (LR), and then random forest (RF), decision tree (DT), and Sup- malignant lesions [58]. The pre-processing of data is an important step
port Vector Classifier (SVC) (linear). They concluded that in the future, before running the algorithm, as it would not be able to deal with miss-
the neural net algorithm could be deployed and implemented into a Big ing values otherwise. Moreover, the performance of the model would
Data test that might perform better over present modules like Hadoop be enhanced by supervised learning. Further, by decreasing the num-
and Apache Spark. ber of values inside the continuous data using label encoding, the ef-
Osmanović et al. [55] collected data from 699 samples at the UCI ficiency of the model would also be increased. Weka 3.6 software was
ML repository. The artificial neural network (ANN) used for this study employed for all the tests that were conducted. The optimal performance
was configured such that it could make use of nine neurons (num- was achieved when eight attributes were used in addition to the class
ber of attributes) and a single output neuron (the nature of the lump containing missing values being removed. The Bayesian networks algo-
i.e., benign or malignant). A backpropagation network-specific confu- rithm, which achieved optimal performance among all the algorithms
sion matrix was also presented. The results indicated that there was a tested in this paper, was able to achieve an accuracy of 97.80% in its
99% chance of correct diagnosis if the ANN was functionally deployed. best configuration.
This also indicated that there was a 97.6% chance of them being neg- Al-Hadidi et al. [36] proposed a novel kNN ML approach to more ac-
atively classified. The ANN still has scope for further improvements in curately detect breast cancer. They designed a method consisting mainly
the future. It could help doctors with a faster method of diagnosis and of two parts: the initial part of the approach processed the images re-
also monitor the patient’s condition. The authors proposed to develop a ceived as input for feature extraction, while the second part consisted
graphical user interface (GUI)-based tool for ease of use by doctors who of the extracted features that were processed using the two models, one
are not particularly adept in using such tools without a proper interface. of which was a neural network and the other model employed logis-
Negi et al. [56] presented four ML model-based algorithms for breast tic regression. Further, they compared both models and analyzed them
cancer detection: Bayesian networks, kNN, SVM, and random forest. The using the Matlab software. During the study, the reported error value
authors aimed to develop a meticulous approach for diagnosing and was < 0.07, and the neural network-based model made use of a fewer
classifying breast cancer. In contrast to the random forest method, the number of features compared to the one that used LR. Although the
Support Vector Machine (SVM) method achieved optimal accuracy and number of features consumed by the BPNN model was 24 less than the
distinctiveness, with the former coming up with the greatest probability LR-based model, they were able to obtain a success rate > 93% using that
of classifying tumours appropriately. With the help of properties such as model. Sinha et al. presented their work on employing an ML-based ap-
training time for a small and large dataset, recall, area under the receiver proach for the early diagnosis of breast cancer [59], wherein they used
operating characteristic curve, accuracy of the predictions made, and ML methods to analyse the number of cancer patients with tumours and
roll-up of the feature count, an accurate comparison can be drawn. Their present a report on the same. There were 567 rows of data representing
study showed that there is still scope for future investigation in the field. 30 distinct attributes of breast cancer traits in the Breast Cancer dataset
Al Bataineh used the WBCD dataset and presented a direct compar- used by the authors. The rows containing information about the nature
ison among five algorithms for breast cancer diagnosis: Multilayer Per- of the tumour (i.e., benign vs malignant) were taken out and set as tar-
ceptron, kNN, Classification and Regression Trees, NB, and SVM [15]. get attributes. Based on the data collected, a percentage was calculated
Throughout the study, the primary objective was to evaluate the perfor- to represent patients with tumours, and the predictions were presented
mance of the various ML methods by evaluating the reliability, accuracy, using illustrative visualizations. To obtain unbiased estimates, the re-
and precision of each algorithm individually for data classification. The searchers used supervised ML algorithms. Based on the testing results,
aim was to discover which ML method could predict the nature of the they found the K-Nearest Neighbours to be the most accurate predictor
tumour. Their results showed that the Multilayer Perceptron model of- with an accuracy of 91.6%. The least accuracy was displayed by the NB
fered optimal performance in terms of precision, recall, and accuracy. approach, which showed an accuracy of 75.6%, while the kNN-based
The accuracy shown by the MLP on the training data was 96.70%, which approach showed an accuracy of 90.9%, which was the highest among
outperformed other algorithms. These models were later tested on un- the lot.
seen data to analyze their real-world performance. Gayathri and Sumathi [60] presented a comparative analysis for the
Asri et al. [57] presented four different classification algorithms with RVM algorithm, which brought forth considerably lower computational
comparison and contrast between them: Support Vector Machine, Naive costs when pitted against other algorithms that were also utilized for
Bayes, kNN, and C4.5 decision trees. Based on the performance measures breast cancer diagnosis. The study evaluated how the RVM approach
of sensitivity, precision, specificity, and accuracy, they sought to evalu- provided an edge to correctly diagnose breast cancer even when the
ate the algorithms’ efficiency and effectiveness. The Weka ML environ- features were reduced over other ML models. The RVM was compared
ment was used to derive all the classifiers employed by them. The WBCD against ML algorithms like Naive Bayes, neural networks, DT, SVM, and
was used in their study which comprised 699 cases, 458 of which were Fuzzy inference systems for performance analysis. The study showed
benign and 241 were malignant, which are spread across two classes that RVM performed significantly better over the other models. Previous
and 11 features, with 65.5% of malignant tumours and 34.5% benign studies showed that an RVM-based approach had rarely been used for
tumours. They applied a method known as the K-Folds cross-validation. the WBCD dataset for diagnosis of breast cancer, but was more widely
The value of k was selected as 10 to achieve the least amount of bias. accepted for other types of cancers such as blood cancer and lymphatic
The results from the model were later evaluated using K-Folds cross- cancer. Hence, Gayathri utilized the original WBCD in their study to de-
validation. These researchers assessed the performance results in terms tect breast cancer, which outperformed all the other approaches of the
of five metrics: the Kappa statistics, the Mean Absolute Error, the Root time and displayed an accuracy of 97%. Even when the features (vari-
Mean Squared Error, the Relative Absolute Error, and the Root Rela- ables) were reduced, RVM still showed better performance than others.
tive Squared Error. Based on the results of their experiment, they found Still, the study stated that there was scope for improvement in the fu-
that the SVM-based model could achieve the best performance with an ture and suggested that it could also be fused with other ML algorithms
accuracy of 97.13% and the lowest rate of error was 0.02%. The per- to further increase the accuracy by fine-tuning. Karabatak presented a
formance of the ML model based on other algorithms varied between novel classifier based on Naive Bayesian for diagnosis of breast cancer.
95.12% and 95.28%, and the error rate varied between 0.03 and 0.06. Naïve Bayesian, which is based on the Bayes theorem, is an uncompli-

73
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

cated classifier [61]. The experiments presented in the study were dis- detecting breast cancer is used by Jeeva et al. to detect it at an earlier
cerned by using a k-fold cross-validation test. In the first step, the input stage with greater accuracy. In 2021, using deep learning, Amit et al.
data was randomly divided into five parts and for every computation, [65] presented a method to detect breast cancer by classifying thermo-
one was used as the test data set and the other four were marked as graphic images. A CAD method was proposed to identify three distinct
the training dataset (80:20 split). The most significant advantage of the case scenarios under database management: cancerous, non-cancerous,
method was that it was irrelevant to the predictions in what manner or and no cancer. Among the most well-known applications of thermogra-
order the data were divided. Several techniques for performance eval- phy are screenings for breast cancer. However, thermography is not yet
uation were also used. The outcome showed a sensitivity of 99.11%, proven to be the ideal method for CAD application. Additionally, doctors
accuracy of 98.54%, and specificity of 98.25%. As shown by the re- prefer thermography results over mammography results because ther-
sults, it was proven that the weighted NB approach outperformed the mography is safer than mammography. Thermal imaging breast cancer
regular NB and several other models as well. Owing to the grid search screening can replace other forms of breast cancer screening if it is im-
methodology used in the model, NB had certain drawbacks like being proved sufficiently. According to this study, thermographic testing and
computationally expensive. The authors suggested that these shortcom- treatment should be performed using a CNN, which addresses the dis-
ings could be further investigated and improved upon with the help of advantages mentioned above. In this study, an effective and accurate
genetic algorithms. technique for segmenting thermograms and detecting breast cancer is
Celik et al. [62] presented a paper on the early detection of breast proposed so that it can be categorized as normal or pathological. To
cancer by using four ML algorithms that were applied on data retrieved produce clear and distinct images, pre-processing is critical. First, data
from blood analysis. The study aimed to compare four algorithms and augmentation is performed. This step contributes to increasing the size
analyze the outcomes of the ML models. The methods used were k-NN, of the dataset by applying multiple conversions to the original input.
ANN, SVM, and Extreme Learning Machine. Hyperparameter tuning and Translations, symmetries, and rotations were used to repeat the input.
optimization were also used to boost the results of the classification. The Below are the steps involved in pre-processing augmentation: (1) Trans-
main advantage provided by this hyperparameter optimization was that lation – An image can be rendered in a particular direction with a cer-
it affected the accuracy of the system based on the number of hidden tain number of pixels. (2) Centring-each row and column was cut-off
layer neurons and the range of these ranges could be manually tuned at the sides. Consequently, a variety of sizes of photographs were ob-
by the user. The highest level of accuracy was shown by the ELM model tained. After cropping all rows and columns, the number of images is
that displayed an average accuracy of 80%. The optimal number of hid- counted. Before being resized, the files are levelled into a single size.
den neuron layers was found to be 1800 as a result of the various tests Based on transparency, randomized images for healthy subjects and pa-
conducted. The use of standard ELM is more advantageous in terms of tients with cancer are selected after pre-processing. The CNN algorithm
accuracy. Mojrian presented a paper on the hybrid ML model for diagno- overall performs very well in this experiment, with a 99.65% accuracy
sis of breast cancer that was based on a Multilayer Fuzzy Expert System and an overall 0.0067% loss of accuracy while training the model. On
[63]. The ELM-RBF model uses a classifier known as ELM or Extreme the same dataset, the accuracy of the random forest and SVM algorithms
Learning Machine, combined with a radial basis function kernel—an were examined and their results were 90.55% and 89.84%, respectively
RBF. These authors also used WBCD. Several evaluation metrics were (Table 1). By segmenting thermographic sinus images, a novel region
established for analysis of the said model. These metrics comprised the of interest (ROI) was introduced. By generating representative data sets
MAPE, RMSE, confusion matrix, R2, and k-fold cross-validation (k = 10 based on experts’ images and providing input to CNN, ground-truth im-
in this case) methods. After thorough testing, an average accuracy of ages were used for the ML training. Additionally, future studies should
98.05% was achieved by this hybrid ELM-RBF model with the help of a aim to provide a scientifically valid method for evaluating these find-
10-fold cross-validation technique. However, at the same time, a linear- ings.
SVM model only showed an accuracy of 90.56%. The results obtained by Among all the papers we examined, the ML models that performed
the implementation and analysis of the hybrid ELM-RBF model proved relatively the worst were models using the NB algorithm. At the same
that this hybrid model outperformed the linear-SVM in almost every time, our tests revealed that the most robust performance and the
field, and the accuracy obtained with the hybrid model was also signif- highest level of accuracy in the early diagnosis of breast cancer were
icantly higher than that of the linear-SVM one. achieved with the help of a custom ML model that used the fuzzy ELM-
Jeeva et al. [64] analyzed breast cancer detection, classification, and RBF algorithm [63]. Statistics showed that this model had an average
detection capabilities using image processing. In their study on image accuracy rate of 98.05%. A variety of algorithms and ML models are
processing, the image was pre-processed before being utilized to discard currently available that can achieve the same goal. However, it can be
the redundancies present in the input images without affecting the final observed that the best results are obtained whenever two different algo-
product images. The proposed method used the discrete wavelet trans- rithms are fused into one, highlighting the advantages of both models
form (DWT) coefficients as a feature vector and used SVM and ANN and diminishing the limitations, as seen in this model. One of the main
algorithms. Images of patients with medical conditions have been pro- applications of this model is that it helps to almost eliminate the need
cessed with the wavelet, the powerful mathematical tool for feature ex- for sampling from patients, with high accuracy, and process the results;
traction, to calculate DWT coefficients. Wavelets are especially useful this helps to reduce the cost of the medical procedure. This also helps
for classification because they provide information about the frequency avoid all major complications involved with the procedure. Thus, this
of a signal by providing localized information. The accuracy was re- model [63] not only provides the highest accuracy in breast cancer de-
ported as 99.51% for SVM and 98.54% for ANN. One of the limitations tection but also has further applications and implications for the entire
of ANNs is the difficulty in obtaining accurate results. Thus, multiple im- medical field. Moreover, this model can be modified for other types of
ages to detect breast cancer in a short time cannot be analyzed with this cancers with little-to-moderate effort.
method. In the beginning, pre-processing is done. Changing the dimen-
sions, resolution, and contrast of the training and input images helped 5. Biosensors using breast cancer detection
arrange all the images in the same proportions without affecting the
details. By using the DWT method, various parts of the image can be Biosensors have been developed as a means of testing for bioana-
identified for feature selection. A grey image is produced after feature lytical molecules in the areas of engineering, biology, and chemistry
extraction, with pixel parameters ranging from 0 to 255. To identify [21-22]. Biosensors have two main components: (1) a transducer that
differences, we used SVM to segregate the breast cancer samples into converts biochemical response into measurable output signals, and (2) a
two groups-those affected by cancer and healthy breast tissue. A fea- bioreceptor on the sensor to capture the matching sample [22]. Biosen-
ture extraction and image processing technique that aids radiologists in sors work in a way that can indicate the presence of abnormalities within

74
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

Table 1 Comprehensive study on breast cancer detection using machine learning

Machine learning algorithm Dataset Database Accuracy Limitations

(%)

Decision Tree J48 algorithm [66] Wisconsin Digitized images of 94.56 Due to the data mining process being used to
FNAC of a breast mass collect data, additional prognostic factors need to
be explored from a more recent timeframe and be
added to the decision tree.
Gaussian Mixture Model (GMM) [67] DDSM Scanned film 86 It provides worse results when detecting benign
mammog-raphy tumours than the SVM classifier.
k-Nearest Neighbours classifier Thermograms from Federal Breast infrared images 94.44 This study does not use a thermographic database
(k-NN) [68] Fluminense University Hospital to store and evaluate normal and abnormal
images, leading to suboptimal results.
Genetic algorithm (GA) optimized MRI from radiologists of the MR Images 89.77 The entire part that begins with MRI acquisition
artificial neural network (ANN) [69] University of Bari Aldo Moro and ends with lesion classification is not currently
automated here during the phases of MR analysis.
k-Nearest Neighbours classifier Wisconsin Digitized images of 95.90 KNN is expressed as a non-parametric ML
(k-NN) [70] FNAC of a breast mass algorithm, whereas an NB algorithm can be
expressed in either non-parametric or parametric
terms.
Sequential Minimal Wisconsin Digitized images of 96.99 –
Optimisation(SMO) [71] FNAC of a breast mass
Artificial neural network (ANN) [72] Department of Obstetrics and Data collected by using 86.95 The size of the dataset used was quite small (116
Gynaecology of the University of blood analyses. instances) so might have fewer chances of being
Coimbra (CHEA) accurate in the real-world data.
Support Vector Machine (SVM) [73] Wisconsin Digitized images of 94.3 Convolutional neural networks (CNN) provides
FNAC of a breast mass better accuracy for mammogram-based detection.
Fuzzy ELM-RBF [63] Wisconsin Digitized images of 98.05 –
FNAC of a breast mass
Deep neural network with Support M. G Cancer Hospital & Research Histopathology image 97.21 Because of the small size of the dataset, it needs to
Value (DNNS) [74] Institute, Visakhapatnam, India samples be enlarged using data augmentation.

the human body by detecting certain biomarkers within the human ing a label-free approach. An embedded surface acoustic wave (SAW)
body. By definition, a biomarker is a molecule found in blood, tissues, device was used in a flow cell to develop a novel biosensor system
or other body fluids that can be an indicator of normal or abnormal for the flow injection analysis (FIA). To develop a procedure that al-
process, a condition, or a disease. Recent research on biomarkers has lowed the directed orientation of the capture antibodies, two biotiny-
indicated their significance in the detection and treatment of a variety lated molecules—neutravidin and biotinylated protein A—were used to
of cancers. Tumour biomarkers are considered the most valuable among assemble the linker system. Extracellular domain (ECD) of HER-2/neu
other biomarkers [26]. Since the first tumour biomarker was discovered (HER-2/neu-ECD), a clinically relevant protein marker, could be de-
in 1847, more than 100 tumour markers have been discovered over the tected with this technique at a concentration of about 10 ng/mL. For
past two centuries. Many of these markers are found in the bloodstream diagnostic applications, the cut-off value was 13–20 ng/mL. Based on
and provide vital information about a person’s health. The two main published literature, the surface modification was designed for low
biomarkers relevant to breast cancer are protein-based biomarkers and concentrations of HER-2/neu-ECD. They developed an efficient cou-
gene markers [75]. Various kinds of biosensors, DNA-based sensors, and pling process that could accommodate both anti-HER-2neu and anti-
other technologies for breast cancer detection have been developed over HER-2neu-ECD, the latter of which is a potential biomarker for breast
the years. Given their advantages over others, electrochemical biosen- cancer. Because antibodies are widely used in the biotechnology in-
sors have garnered greater interest and are widely used. The highly sen- dustry as binding molecules in binding assays, the findings may be
sitive, miniaturized design, simplicity, and low cost of electrochemical useful for other principles and applications of biosensors in clinical
biosensors are their main advantages [39–41]. They offer elegant ap- diagnosis.
proaches to interfacing, including signal-transduction and DNA recog- Rafiee-Pour et al. [78] presented a novel electrochemical miRNA
nition, by which the focus is on the directionality of electrical signals biosensor that used a label-free approach that employed a redox indica-
generated by hybridization DNA probes and target DNA [39, 41]. tor, methylene blue (MB). The biosensor may be applied to the breast
Gohring et al. [76] presented a study for the diagnosis of a breast cancer biomarker miRNA-21. The main concept in this study was to de-
cancer biomarker, HER2, (in human serum samples) by applying a label- termine miRNA without excessive labelling or operational complexity.
free approach with an optofluidic ring resonator. This study was aimed In consequence, a detection limit of 84.3 fM which is considered rela-
at developing a low-cost detection system that could be used to mon- tively low, can be achieved with this strategy, thus making it possible to
itor disease progression using HER2 proteins detected in the serum detect miRNAs at concentrations from 0.1 to 500 pM. This study used
at varying concentrations. Based on the opto-fluidic ring resonator’s a comparison of the proposed miRNA biosensor on perfect complement
(OFRR) results, serum samples were detectable at medically relevant and mismatch targets to determine its selectivity. With the same concen-
concentrations of HER2 in a short span of 30 min (detection range: 13– tration of 200 pM, the electrochemical signal of the miRNA-21 (com-
100 ng/mL). HER2 biomarker measurements show that the bulk refrac- plementary target) was significantly larger than the current from the
tive index sensitivity for the OFRR was found to be 30 nm per refractive miRNA-192 (non-complementary sequence). Using MB as a hybridiza-
index unit (RIU). The OFRR was tested with ethanol, passing through it tion indicator, a novel platform for sensitive detection of miRNAs was
at increasing concentrations of a known refractive index, to determine developed in this work. MB was used to measure the miRNA-21 concen-
its sensitivity. With this device, one can assess the clinical prognosis of tration by determining the difference in oxidation peak current between
patients with improvements in accuracy and monitor the effectiveness the DNA or duplex DNA/miRNA. Therefore, these results indicated that
of treatment at a low cost. the electrochemical miRNA biosensor is a highly effective biosensor with
Gruhl et al. [77] presented an article on the use of an acoustic biosen- the following notable advantages: fast, low cost, simple preparation pro-
sor to detect the biomarker indicative of cancer, HER-2/neu, by us- cedure, wide linear range, and relatively rapid production.

75
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

It is crucial to detect breast cancer early and boost patient survival composite they used was made from reduced graphene oxide, chitosan,
rates. Arif et al. [79] reviewed a device designed for self-screening as and gold nanoparticles on the electrochemical biosensor surface to im-
well as early detection of breast cancer. Their paper describes the de- prove the detection limits and signal enhancement. Furthermore, an ap-
sign of a small device that can be placed and used at home. The process tamer for nucleolin, AS1411, was used as a biosensor recognition ele-
utilized several components like salivary autoantibodies, breast-cancer ment, which is an overexpressed protein present on the surface of cancer
specific biomarkers, and biosensors to detect the cancer at an early stage cells. The aptasensor demonstrated a wide linear range 10–1,000,000
by measuring the concentration variation of a biomarker in saliva. Au- cells/mL, a confirmed limit of detection of 4 cells/mL on the lower side,
toantibodies present in the saliva can be detected early and lead to pre- and a linear correlation with the mathematical logarithm of the cell
cise detection of different cancer types. There is evidence in the litera- concentration. The aptasensor fabrication process was studied by elec-
ture that autoantibodies against ATP6AP1 or other autoantigens begin trochemical impedance spectroscopy and cyclic voltammetry. As a result
to surface in the initial stage of a disease, and could be used as po- of the higher concentrations of cells, those cells were more likely to be
tential biomarkers to diagnose the disease early. There is an increasing captured by the aptasensor that blocked the access of ferro/ferricyanide
reliance on saliva as a detection medium of choice. Using quartz crystal to the sensor, increasing the charge transfer resistance. Thus, this ap-
microbalance biosensors to detect autoantibodies against ATP6AP1 in tasensor identified MCF-7 cancer cells with high efficiency and without
saliva, the study provides information on the sensitivity of this detec- the presence of an electrochemical or optical label. Using this simple
tion technique. A prototype could also be made of the device. However, and highly sensitive aptasensor, one can accurately distinguish healthy
concerns about the saliva containing enough ATP6AP1 autoantibodies cells from cancer cells.
need to be addressed to validate the specificity of disease detection. Mansor [83] studied an electrochemical DNA biosensor with em-
Jabin [80] presented an article on biosensors with titanium coating bedded nanowire made out of zinc oxide with the goal of detecting
for cancer cell detection based on surface plasmon resonance technol- the BRCA1 gene. Research conducted in this study focused on show-
ogy. Basically, the goal was to develop a sensor that could achieve a ing that nanowires of zinc oxide fabricated on gold surfaces can be
greater sensitivity towards amplitude, wavelength, and birefringence. an appropriate platform for the development of biosensors capable of
One of the major objectives of the proposed cancer sensor was to define detecting BRCA1. Demonstrating a 3.32 𝜇M detection limit, the pro-
the light absorption characteristics of a desired biosample by means of posed DNA biosensor was capable of detecting the target sequence at
plasmonic resonance, and ultimately assist in the detection of cancer concentrations ranging between 10.0–100.0 𝜇m. Field emission scan-
using a real-time finite element method (FEM) approach. Displaying a ning electron microscopy (FESM) images showed that the nanostruc-
coupling length all the way up to 66 m, the sensor exhibits a high bire- ture grown on the gold electrode had a mean diameter ranging from
fringence of 0.04, so it can provide information that will be valuable approximately 450 to 550 nm. Tests with the electrodes prepared with
for both science and engineering. However, with a resolution ranging ZnONWs/Au showed that the electrodes are ideal for DNA immobiliza-
between 1.5 × 10−2 RIU and 9.33 × 10−3 RIU, the optimal wavelength tion. Additionally, DPV measurements were used to determine direct
sensitivity in the proposed structure ranges between 10,000 nm/RIU electrochemistry of DNA, allowing them to identify the BRCA1 gene by
and 17,500 nm/RIU. Depending on the type of cancer cell, the transmit- detecting short DNA sequences when DNA probes were chemisorbed
tance variance of ranges between 3300 and 6100 dB/RIU, and the am- onto the ZnONWs/Au surface. As a consequence, a satisfactory surface
plitude sensitivity for major polarization mode is between −340 RIU−1 oxidation peak was observed at about 0.8 V in the presence of com-
and −420 RIU−1 , ensuring a detection limit that can reach up to 0.025 in plementary DNA sequences. It was found that the optimal pH for DNA
a state of major polarization. Additionally, the sensitivity of this biosen- biosensors was at pH 7. Therefore, this technique can be carried out
sor is measured through comparison to other structures influenced by on real breast cancer cells using more complex array formats for the
similar data, and this makes it one of the best structures proposed so far. detection of genes such as BRCA2 and p53 among many others. Hos-
The proposed model features enhancements to the previously described sain et al. [84] were the first to demonstrate a numerical model of a
D-shaped photonic crystal fiber (PCF) that allows for greater flexibility hybrid SPR biosensor utilising graphene coating and fibre-optic surface
in terms of the optical parameters and a noticeable improvement in their for the detection of BRCA2 (breast cancer gene 2) and BRCA1 (breast
sensing capability. cancer gene 1) genetic breast cancer. The objective of their study was to
Graphene’s excellent mechanical and electrical properties make it present the numerical modelling of the biosensor for detection of breast
particularly suitable for breast cancer field effect transistor (FET) biosen- cancer involving two specific mutations found in the BRCA1 and BRCA2
sors; hence, graphene-based materials have become popular for this ap- genes—916delTT and 6174delT, respectively. This study analysed per-
plication. There is a wide array of biosensor materials that are either formance in terms of sensitivity and open windows toward early detec-
made purely of graphene or have a certain concentration of graphene in tion of genetic breast cancer that was derived from BRCA1 or BRCA2.
their composition, which contains single to few-layer pristine graphene, The quantification results showed that graphene had an approximately
nanoribbons, graphene oxide (GO), reduced graphene oxide (rGO), 35% higher sensitivity than the usual surface plasmon resonance (SPR)
doped graphene, and even “damaged” graphene. Novodchuk et al. [81]. biosensor most likely owing to graphene’s absorption ability. Sensors
presented a review to highlight the progress made in graphene-based such as these, when used to detect tumour cells, provide an easy and
functional electrochemical sensors for bio-sensing of breast cancer cells convenient method for the identification of specific mutations in the
and compared them to previous graphene-based biomarkers, biosensors, genes BRCA1 and BRCA2 genes. By examining the variation in SRF and
and device configurations. It has been observed by analyzing several SPR angles, the sensor can distinguish between the matched DNA inter-
biosensors that increased electrical properties and mobility significantly actions from mismatched DNA interactions that take place between the
increase the sensitivity of the biosensor. Further, the threshold limits p-DNA, a sh-DNA mutation type, and the target DNA (a mr-DNA muta-
of current top-of-the-line devices were compared for multiple analytes, tion type). As a result of graphene’s prominent properties, it is placed
current ratios, and the mobility for each device. Various biomarkers are between the gold film to serve as a medium for the efficient absorption
often targeted by different bioreceptors to perform selective detection of light. Early breast cancer can be detected using DNA hybridization
in a bio environment. In addition, size is an important characteristic of event analysis on the basis of four key factors—reflectance, maximum
a bioreceptor material, as it is typically limited by its Debye screening transmittance, SRF angle, and SPR angle. Its improved sensitivity makes
length when it is an FET biosensor. the proposed hybrid SPR biosensor ideal for future applications in early
Shafiei et al. [82] presented a reduced graphene oxide-chitosan-gold breast cancer detection based on the genetic biomarkers 916delTT and
composite-based electrochemical aptasensor that allows for the detec- 6174delT.
tion of breast cancer cells without the use of labels. The authors aimed Li et al. [85] used an electrochemical-based approach for the de-
to detect cancer cells (MCF-7) with an ultrasensitive aptasensor. The tection of breast cancer cells. Using two sets of tumour markers to-

76
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

gether on the surface of the cells, the main objective and aim of this responses were almost double those of miR-155 derived from the three-
study was to detect breast cancer cells. Observations from the exper- base mismatch miR-155. Thus, the observation of the differences be-
imental results showed that only under special conditions could both tween the two responses suggested that the probe/miR-155 had a higher
tumour markers be detected on the tumour cells’ surface, along with a chance of duplex formation than the probe/three-base mismatched miR-
satisfactory electrochemical response. MCF-7, a cell specific to breast 155. Cyclic voltammograms were generated, and the mean value (dl) for
cancer, can be differentiated from several other types of cells if this miR-155 was 0.70 ± 0.01; the three-base mismatched miR-155 showed
method is used. Further, the prepared cytosensor has good reproducibil- a mean value of 0.56 ± 0.08. The introduced biosensor used an ul-
ity and a low detection limit for MCF-7 cells at a number of concentra- trasensitive method which displayed a detection limit of 20 zmol dis-
tions between 10−4 and 10−7 cells/mL with possible medical applica- playing a linear range of 2 × 10−20 to 2 × 10−12 mol for the detec-
tions. Breast cancer cells were screened by using a method to simul- tion of miR-155. The electrochemical biosensor introduced in the report
taneously detect MCF-7 and CEA on their outer surface of cancer cells. used polyethyleneimine-silver nanoparticles as a redox-active indicator.
With the ability to measure cancer specifically, systematically, and cost- The result of electrochemical biosensor analysis of the serum samples
effectively, this method offers great promise for fast, simple, and repro- showed high sensitivity and cost-effectiveness.
ducible cancer detection. Benvidi et al. [38] designed an ultrasensitive Freitas et al. [25] presented a study on the detection of HER2-ECD
and highly selective electrochemical DNA biosensor that could detect breast cancer biomarkers using electrochemical sensing platforms. They
breast cancer in its preliminary stages. The authors’ aimed to report studied the electrochemical composition of HER2 with the objective of
a new type of electrochemical biosensor for detecting breast cancer- early detection of breast cancer by screening and electrochemical anal-
associated genes such as BRCA1, by using a gold electrode via elec- ysis. Briefly, 7.5–50 ng/mL linear calibration plots were obtained and
trochemical impedance spectroscopy (EIS), cyclic voltammetry (CV), the assay took 2 h and 20 min to be completed. During this time, the
and differential pulse voltammetry (DPV) methods. A comprehensive limit of detection (LOD) of 0.16 ng/mL (from SPCE-MWCNT/AuNP) and
electrochemical redox study was carried out with ferricyanide and fer- 8.5 ng/mL (from SPCE-AuNP) were obtained and these concentrations
rocyanide as an electrochemical redox couple after the construction were smaller than the established threshold of 15 ng/mL for this spe-
process was explored using CV and EIS. An electrochemical biosen- cific cancer biomarker. The following sensing platforms were used to de-
sor based on EIS displayed a wide dynamic range (1.0 × 10−19 to tect HER2-ECD: gold nanoparticles-modified SPCEs, graphene, SWCNT,
1.0 × 10− 7 M) under optimum conditions and a low detection limit multiwalled carbon tube, and hybrid graphene-modified SPCEs. The
(4.6 × 10−20 M) for the target DNA as opposed to the DPV technique, SWCNT or MWCNT fused with gold nanoparticles, formed hybrid nanos-
which has a detection limit of 2.5 × 10−15 M for a linear range of tructures. Based on the analysis of other biomarkers and spiked hu-
10−14 to 10−7 M. On the basis of breast cancer data acquired from man serum samples, the platform’s applicability was confirmed and its
GenBank, particular primers and probes were modelled. Because of the selectivity was determined. In this manner, the developed HER2-ECD
ability to detect BRCA1 at extremely low levels, this study proves that detection strategies would be a viable alternative to the conventional
the proposed biosensor can successfully identify BRCA1 in the earliest way of analysing the HER2 gene. Salahandish et al. [88] presented an
stages. article on nano-biosensor for ultrasensitive detection of breast cancer
Cardoso et al. [86] presented a novel design for biosensing and de- based on HER2. The objective was to design a biosensing system with
signed a highly sensitive electrical biosensor to detect the presence of a high-performance linear response and selective performance with a
miRNA-155, which had a linear response down to the attomolar range. large linear response range. They created a sandwich structure of a Ni-
Their main aim was to develop a simple and inexpensive solution with doped graphene (NFG)/AgNPs/PANI multilayer structure. This struc-
a concurrent approach design, use a rendering of the serum samples, ture showed excellent conductivity and stability and allowed increased
and minimize the complexity by optimizing the critical variables. The anchorage of HER2 antigens from cancer cells, which was made possi-
resulting biosensor showed efficiency in detecting miRNA-155 at con- ble due to the nanocomposite components being arranged in the proper
centrations as low as 5.7 aM with a sensing range of 10 aM to 1 nM. combination and order. The sensor displayed a rapid response time of
Selectivity was high towards bovine serum albumin (BSA) and breast 30 min, and an average detection limit of 2 cells/mL for detecting SK-
cancer antigen CA-15.3. The biosensor response against miRNA-155 was BR3 cells between 10 and 5 × 106 cells/mL. The nano-biosensor demon-
not affected by the raw fluid extracts obtained from melanoma(skin can- strated high sensitivity and specificity with a detection efficiency > 90%
cer cells), but a positive signal was obtained when extracts from breast for cancer cell detection by using a label-free approach in whole blood
cancer cells were at play. The biosensor was constructed by immobi- samples without sample preparation or the need for staining. The au-
lizing the anti-miRNA-155 on a gold screen-printed electrode (Au-SPE) thors also developed ultrasensitive nanocomposites for detecting breast
and using mercaptosuccinic acid to clog the non-specific sites. Surface cancer cells with the help of economical AgNP and optimised the meth-
modification of these devices was confirmed by atomic force microscopy ods for nanocomposite synthesis and functionalization. Consequently,
and electrochemical methods such as square wave voltammetry, CV, and their results showed that the new and optimised nanocomposite has
EIS, which were also able to hybridize with miRNA-155 while still main- potential applications in electrochemical biosensing and super capac-
taining their stability. This simplistic and robust strategy could prove itances.
to be an efficient method for the simultaneous quantitative analysis of In 2019, Kim et al. [89] presented a study focused on a label-free
multiple miRNAs 155 in physiologic fluids, for POC diagnosis as well as surface-enhanced Raman spectroscopy biosensor to detect breast can-
various other biomedical research. Hakimiam and Ghourchian [87] pre- cer in real-time using human tears. At that time, there was no practi-
sented a study on highly sensitive electrochemical biosensors for the de- cal sensor that could balance signal enhancement, reproducibility, and
tection of microRNA-155 as a risk factor for breast cancer. Recent studies uniformity. They demonstrated the use of a Raman spectrometer to esti-
in the field suggested that the concentration of miRNA-155 (miR-155) in mate or detect asymptomatic breast cancer by utilizing human tears and
the human body is directly related to breast cancer. The main objective combining a multivariate identification algorithm based on multivariate
was to assess miR-155 and analyse the significance of its impact on the statistics to demonstrate feasible practical applications. Mucins, glyco-
presence of breast cancer in an individual, by developing and report- proteins, non-glycosylated proteins, peptides, and lipids could be found
ing an electrochemical biosensor that can outperform others in terms in tears, making them a special source of body fluids. Tears are a good
of detection speed and sensitivity. Two sequences, namely a three-base source of biomarkers, because they can be collected noninvasively, do
mismatched miR-155 and a standard miR-155, were used for the study not contain many solid proteins such as albumin, and do not require fur-
to determine the specificity of the biosensor. When the biosensor re- ther purification. By analysing human tear fluids for breast cancer, they
sponses in the two scenarios (with and without the use of the two se- evaluated the feasibility and versatility of the Au/HCP-PS monolayer as
quences) were compared, it was found that miR-155-induced biosensor a biosensor. PC-LDA identification using leave-one-out cross-validation

77
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

(LOOCV) provided 96% classification accuracy. Tear-fluid-based asymp- itself to further medical research [99]. To train the model, the images
tomatic breast cancer detection achieved 92% clinical sensitivity and in the datasets are magnified, but by doing so, the histology images are
100% specificity. It is anticipated that if appropriate sensitivity is guar- altered in size that can greatly enhance quality processing during train-
anteed, the new diagnostic method will overcome the shortcomings of ing. Low magnification analysis of the histology images is challenging
conventional labelling techniques, such as high costs and low yields. In because the images contain so many tissues. Learning discrete features
these studies, the non-invasive, label-free, real-time screening technol- from an image at multiple magnification levels can be beneficial to make
ogy was found to be effective at detecting asymptomatic cancers early, a distinct and accurate diagnosis, but is quite challenging at the same
reducing the likelihood of recurrence. By the end of 2020, Abrao Ne- time [5]. Heterogeneous breast densities make mass detection and clas-
meir et al. [90] developed a new method of detecting HER2 in saliva, a sification more challenging than calcifications. Traditional ML method-
better alternative to blood, by using EIS. The current detection method ologies present confined approaches that only operate on a specified
of HER2 remains highly specialized, costly, and time-consuming, given density type or dataset [74]. In ML applications, one of the biggest bio-
the difficulty in using biosensors for low concentrations. With label- logical challenges in constructing efficient and accurate classifiers that
free electrochemical biosensors, the goal is to detect HER2 in saliva are suitable for practical use. It is predominantly the following chal-
at a low cost. The detection produced a biosensor that is highly lin- lenges in gene expression classification that are of concern: the huge
ear, sensitive towards HER2, twice as sensitive as HER3, and four times number of gene expressions, investigation of relevant features, the ex-
more sensitive than the EGFR interference molecule. A standard addi- istence of noise in the datasets, and accuracy as well as reliability of
tion method was used to detect HER2 in real saliva, which proved to classifications. Major future trends in computational biology designed to
have good linearity (R∗ 2 = 0.9904), hence proving that this technique check for breast cancer recurrence may get good results by using a com-
could be used to detect HER2 in real patients. In this method, a diazo- bined effort of different recent ML techniques and findings from cancer
nium salt was initially deposited on the SPE by means of cyclic voltam- researchers [100]. Large epigenetic datasets need to be made publicly
metry. Later, the anti-HER2 antibodies were immobilized using 1-ethyl- available and more accessible for novel ML approaches to be properly
3-(3-dimethylamino) propyl carbodiimide/N-hydroxysuccinimide. In a developed, improved, and tested. Despite improved technologies, there
microfluidic system, electrochemical impedance spectroscopy was used is room for improvement in accurate cancer detection, characterization,
to detect HER2 biomarker concentrations. In its physiological concentra- and monitoring. The ability to make clinical decisions is dependent on
tion range of 5–40 pg/mL, the biosensor showed a higher linear detec- the accurate diagnosis of the disease. There is a great need for reliable
tion of HER2. With the standard addition methods, the biosensor was tracking of neoplastic diseases over time as due to certain treatments
used to detect HER2 concentrations of up to 10 pg/mL in real saliva. certain surrounding tissues are affected. Due to this, the surrounding
Hence, this method of diagnosis has proven to be highly effective and tissue may acquire signal characteristics that are difficult to distinguish
efficient on real patients, which means it can be used instead of conven- from the tumour [101].
tional methods. Diagnosis of younger women is more difficult than that of older
During this research, we found that there are various types of biosen- women owing to the higher breast tissue density of the former group. In
sors that utilise different biomarkers to detect tumour cells. Others al- previous models, the detection of one cancer cell was based on the per-
lowed finding specific mutations in biomarkers. We discovered that the formance of the neighbouring cancer cells. A cancerous cell could only
biosensor that performed relatively worse was the ZnO nanowires-based be evaluated in relation to another cancerous cell. The use of biosen-
DNA biosensor used to identify the BRCA1 gene with the detection limit sors for cancer detection in the future shows great promise. Biosensor
of 3.32 𝜇M and a linear range between 10.0 and 100.0 𝜇M [83]. On the design poses the challenge of storing and processing biological sam-
other hand, our tests also revealed that the most robust performance, ples, as they are highly perishable and must be stored before testing.
as well as the highest sensitivity and specificity, were observed with an Furthermore, some biomarkers are less specific and are apt to serve as
electrochemical biosensor for detecting the presence of miRNA-155 at markers for more than one type of cancer. It is possible to solve this
attomolar levels, during a breast cancer diagnosis. These electrochem- problem by using a biosensor-based platform that is capable of detect-
ical biosensors have a detection limit of 5.7 aM and a linear range of ing multiple biomarkers simultaneously and exhibiting high sensitiv-
1–10 aM that was achieved by immobilization of the anti-miRNA-155 ity and specificity. miRNAs regulation is different in patients compared
in Au-SPEs [86]. Several biosensors of this type are common; however, to healthy individuals. The data suggest that miRNAs in the body can
the aforementioned device can be repeatedly used with subsequent read- serve as effective biomarkers for diagnosing and tracking cancer. How-
ings, most notably in EIS measurements, and shows high selectivity to- ever, circulating miRNAs continues to be challenging to detect with-
ward extracts from other tumours and diverse proteins from biological out prior extraction and purification [93]. Despite the low detection
fluids [86]. In addition, this device provides more realistic opportunities level, miRNA modification requires professional operation like purifica-
for analyzing miRNA cancer biomarkers in a clinical context, as well as tion and exact working conditions for real-time sample determination,
allows these markers to be reused because of its de-hybridization prop- which may impose some difficulties. In spite of the significant advance-
erties. Further, reusing it in an online sensor approach (Table 2). ment in cancer diagnosis owing to research around tumour markers,
current diagnostic procedures are invasive, uncomfortable, and incon-
6. Challenges and future scope venient, and these issues need to be addressed. Therefore, the discovery
of non-invasive biomarkers for the early detection of cancer is of great
The primary challenges around cancer detection and treatment in- importance [94]. An important challenge in electrochemistry is how to
clude designing research pipelines, identifying the growth mechanism bring the technology to the patient in an accurate and reliable manner
of the specific cancer type, designing preclinical models, managing com- without compromising accuracy or reliability. It would therefore take
plex cancers precisely, early detection, and developing innovative clini- a great deal of developmental work to realize decentralized electronic
cal trial methods for enhanced accuracy that could be useful to doctors cancer tests. Innovation in the field of sensing has made electrochemi-
and patients for early detection and as a second opinion [74]. In order cal devices widely applicable in clinical settings. These devices may be
for ML technology to be reliably used for the diagnosis and treatment of particularly suitable for providing information in a straightforward and
cancer, some key challenges must be addressed. It is crucial to extract rapid manner at a lower cost, making them uniquely suited for point-
features from medical imaging data and process them before they can of-care cancer screening [39]. Despite their promising capabilities, the
be used as input data. Therefore, medical imaging data cannot be used electrochemical biosensors that will determine circulating biomarkers
directly as input data. Along with the development and popularization of breast cancer are still in the early design stages or prototype stages at
of technology, the weights of coefficients in neural models are calcu- best, and none of them has been tested with a large enough patient pop-
lated, so a reasonable confidence interval can be drawn, which can lend ulation to demonstrate their reliability in clinical practice. This problem

78
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

Table 2 Comprehensive study on cancer detection using biosensors

Name of biosensor Biosensor specification Analyte Performance Linear range

FET biosensor [91] Transducer rGO-Encapsulated SiO2 NPs HER2 The detection limit is 1 pM 1 pM to 1 𝜇M
Mobility is 3 cm2 V−1 s−1
Electrochemical [92] Hemin/miR/DNA-Au/probe/AuNPs/Au electrode miR-21 Detection limit is 6 fM 0.01–500 pM
Electrochemical [93] Two auxiliary probes that self-assemble to form 1D miRNA 21 LOD is 100 aM 100–105 aM
DNA concatemers
Electrochemical [94] P19 captured dual aptamer complex on SPCE miR-205 Detection limit is 0.6 nM 2–10 nM
Electrochemical Aptamer on a gold electrode and iron redox probe Progesterone (PR) LOD is 0.90 ng/mL 10–60 ng/mL
(EIS) [95] readings Response time: 40 min
Electrochemi-cal [96] Anti-miR probe/GNR/GO/GCE miR-155 Detection limit is 0.6 fM 2.0 fM to 8.0 pM
Electrochemi-cal [86] Immobilization of anti-miRNA-155 miRNA LOD is 5.7 aM 10–109 aM
on Au-SPE 155
Sandwich polydA-aptamer-modified gold electrode MCF-7 cells The detection limit of 10–105 cells/mL
electrochemi-cal polydA-aptamer functionalized gold 8 cells/mL (3𝜎/slope)
biosensor [13] nanoparticles/graphene oxide hybrid
FET biosensor [97] Transducer G-graphene/PBASE DNA Detection limit is 100 fM 100 fM to 1 nM
Mobility is 64 cm2 V−1 s−1
FET biosensor [98] Transducer rGO/SA BRCA1 Detection limit is 0.2 nM 0.2–75 nM

needs to be addressed by testing the methodology on large numbers of ical performance analysis for different types of biosensors and Machine
real-life samples and comparing it with other competitive approaches learning models. Prince Pipariya investigated the data and information.
[102]. In future, increasingly sophisticated technological innovations in Further, Shlok Patel did the comparative analysis for better insights and
biosensors that incorporate biomarker patterns, software, and microflu- again validating the data. Now, the original draft was prepared by all
idics can make these devices highly effective in this application area the authors and at last, it was again reviewed by Shlok Patel. Manan
[26]. Shah has worked on writing and editing part.

7. Summary
Acknowledgements
In the past few years, various breast cancer detection methods have
been developed by applying ML algorithms and biosensors. In addition, The authors are grateful to Dharmsinh Desai University and Depart-
breast screening techniques and Big Data analysis techniques are also ment of Chemical Engineering, School of Technology, Pandit Deendayal
increasing over time. Most of the databases require pre-processing de- Energy University for the permission to publish this research.
pendent on the ML model used. However, there were some differences
in performance measuring factors, and the different models use different References
metric elements depending on the dataset. Whereas, according to its use,
biosensors require an analyte that will be used to generate an electric [1] Kamboj A, Tanay P, Sinha A, et al. Breast cancer detection using supervised ma-
chine learning: a comparative analysis. Singapore: Springer;2021:263–9.
signal to identify biomarkers. Similar to ML inputs, analytes also need to
[2] Solanki P, Baldaniya D, Jogani D, et al. Artificial intelligence: new age of transfor-
be processed prior to testing. The specification of each biosensor varies mation in petroleum upstream. Pet Res 2021. doi:10.1016/j.ptlrs.2021.07.002.
according to the analyte used. After carrying out these comparisons and [3] Fotouhi H, Čaušević A, Lundqvist K, et al. Proceedings - International Computer
analyzing all the aspects of each ML model and biosensors that are used Software and Applications Conference, Atlanta, United States. IEEE Computer So-
ciety; 2016. doi:101109/COMPSAC20168.
to detect breast cancer efficiently, we found two optimum techniques, [4] Hady AA, Ghubaish A, Salman T, et al. Intrusion detection system for health-
one of each. First, the Fuzzy ELM-RBF ML algorithm was found to be care systems using medical and network data: a comparison study. IEEE Access
the best model for breast cancer detection with 98.05% accuracy. The 2020;8:106576–84. doi:10.1109/ACCESS.2020.3000421.
[5] Singh S, Deep A, Mohanta G, et al. In next generation point-of-care biomedical
WBCD has used digitized images of FNAC analysis of a breast mass. Sec- sensors technologies for cancer diagnosis. Singapore: Springer;2017:253–78.
ond, the electrochemical biosensor with a detection limit of 5.7 aM that [6] Mihaylov I, Nisheva M, Vassilev D. Machine learning techniques for survival time
was used to detect miRNA-155 from serum samples by immobilization of prediction in breast cancer. Lecture Notes in Computer Science Springer, Cham.
2018: 186–94, doi:10.1007/978-3-319-99344-7_17.
the anti-miRNA-155 on Au-SPE was found to be the best biosensor. The [7] Benbrahim H, Hachimi H, Amine A. Comparative study of machine learning al-
biosensor could detect very low concentrations of miRNA-155, down to gorithms using the breast cancer dataset. Adv Intell Sys Comp 2020;1103:83–91.
1–10 aM in serum samples. A significant area was analyzed in this re- doi:10.1007/978-3-030-36664-3_10.
[8] Siegel RL, Miller KD, Jemal A. Cancer statistics, 2016. CA Cancer J Clin
search of breast cancer detection at an early stage. Researchers’ interest
2016;66(1):7–30. doi:10.3322/caac.21332.
in the field of breast cancer detection is evident from the exponential [9] Kaushal C, Singla A. Analysis of breast cancer for histological dataset based on
growth of published research articles on a yearly basis. Analyzing the ex- different feature extraction and classification algorithms. Adv Intelligent Sys Comp
2021;1165:821–33. doi:10.1007/978-981-15-5113-0_69.
isting research will enable clinicians to make highly accurate diagnoses,
[10] Uhr JW. Cancer diagnostics: one-stop shop. Nature 2007;450:1168–9.
ultimately leading to more effective treatment of patients. doi:10.1038/4501168a.
[11] Geiger TR, Peeper DS. Metastasis mechanisms. Biochim Biophys Acta
Conflicts of interest statement 2009;1796:293–308. doi:10.1016/j.bbcan.2009.07.006.
[12] Liberko M, Kolostova K, Bobek V. Essentials of circulating tumor cells
for clinical research and practice. Critical Rev Oncol 2013;88:338–56.
The authors declare that there are no conflicts of interest. doi:10.1016/j.critrevonc.2013.05.002.
[13] Wang K, He MQ, Zhai FH, et al. A novel electrochemical biosensor based on
Funding polyadenine modified aptamer for label-free and ultrasensitive detection of human
breast cancer cells. Talanta 2017;166:87–92. doi:10.1016/j.talanta.2017.01.052.
[14] Kononenko I. Machine learning for medical diagnosis: history,
This research did not receive any specific grant from funding agen-
state of the art and perspective. Artif Intell Med 2001;23:89–109.
cies in the public, commercial, or not-for-profit sectors. doi:10.1016/S0933-3657(01)00077-X.
[15] Al Bataineh A. A comparative analysis of nonlinear machine learning algo-
Author contributions rithms for breast cancer detection. Int J Mach Learn Comput 2019;9:248–54.
doi:10.18178/ijmlc.2019.9.3.794.
[16] Güler NF, Übeyli ED, Güler I. Recurrent neural networks employing Lyapunov
In this paper, Yash Amethiya worked on data curation from various exponents for EEG signals classification. Expert Syst Appl 2005;29:506–14.
sources. Then after Yash Amethiya and Prince Pipariya did the theoret- doi:10.1016/j.eswa.2005.04.011.

79
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

[17] Osareh A, Shadgar B. Proceedings of the 2010 5th international symposium [46] Cruz JA, Wishart DS. Applications of machine learning in cancer prediction and
on health informatics and bioinformatics. HIBIT;2010:114–20. doi:10.1109/HI- prognosis. Canc Inform 2006;2:59–77. doi:10.1177/117693510600200030.
BIT.2010.5478895. [47] Kharya S, Dubey D, Soni S. Predictive machine learning techniques for breast cancer
[18] Salama G, Abdelhalim MB, Zeid MA. Breast cancer diagnosis on three different detection. Int J Comput Sci Inf Technol 2013;4:1023–8. doi:10.3390/s17071572.
datasets using multi-classifiers. 2012. [48] Michie D, Spiegelhalter DJ, Taylor CC. Machine learning, neural, and
[19] Ghasemzadeh A, Azad SS, Esmaeili E. Breast cancer detection based on Gabor- statistical classification. 1994; Available from: http://citeseerx.ist.psu.edu/
wavelet transform and machine learning methods. Int J Mach Learn Cybern viewdoc/summary?doi=10.1.1.27.355. (Accessed May 20, 2021).
2018;10:1603–12. doi:10.1007/S13042-018-0837-2. [49] Bazazeh D, Shubair R. Comparative study of machine learning algorithms for breast
[20] Nahid AA, Kong Y. Involvement of machine learning for breast cancer cancer detection and diagnosis. Int Conf Electron Devices Syst Appl 2017:1–5.
image classification: a survey. Comput Math Methods Med 2017:3781951. doi:10.1109/ICEDSA.2016.7818560.
doi:10.1155/2017/3781951. [50] Nemoto M. Machine learning for computer-aided diagnosis. Igaku Butsuri
[21] Muhammad TI, Mohammad AU. Biosensors, the emerging tools in the identification 2016;36:29–34. doi:10.11323/jjmp.36.1_29.
and detection of cancer markers. J Gynecol Women’s Health 2017;5(4):555667. [51] Sahran S. Machine learning methods for breast cancer diagnostic. Breast Cancer
doi:10.19080/JGWH.2017.05.555667. Surg Intech Open 2018.
[22] Sobiepanek A, Kobiela T. Application of biosensors in cancer research, Laboratory [52] Mohammed SA, Darrab S, Noaman SA, et al. Analysis of breast cancer de-
of biomolecular interactions studies (Tomasz Kobiela’s Lab), 2018. Available from tection using different machine learning techniques. Comm Comp Inform Sci
https://www.researchgate.net/publication/329310717_Application_of_biosensors_ 2020;1234:108–17. doi:10.1007/978-981-15-7205-0_10.
in_cancer_research (Accessed May 20, 2021). [53] Asri H, Mousannif H, Al Moatassim H, et al. A hybrid data mining clas-
[23] Maia M, Freitas A, Handschuh S. Proceedings of the 12th IEEE Inter- sifier for breast cancer prediction. Adv Intel Sys Comp 2020;1103:9–16.
national Conference on Semantic Computing (ICSC). IEEE;2018:318–19. doi:10.1007/978-3-030-36664-3_2.
doi:10.1109/ICSC.2018.00065. [54] Sadhukhan S, Upadhyay N, Chakraborty P. Breast cancer diagnosis using im-
[24] Mittal S, Kaur H, Gautam N, et al. Biosensors for breast cancer diagnosis: a re- age processing and machine learning. Adv Intel Sys Comp 2020;937:113–27.
view of bioreceptors, biotransducers and signal amplification strategies. Biosens doi:10.1007/978-981-13-7403-6_12.
Bioelectron 2017;88:217–31. doi:10.1016/j.bios.2016.08.028. [55] Osmanović A, Halilović S, Ilah LA, et al. Machine learning tech-
[25] Freitas M, Nouws HPA, Delerue-Matos C. Electrochemical Sensing platforms for niques for classification of breast cancer. IFMBE Proc 2019;68:197–200.
HER2-ECD breast cancer biomarker detection. Electroanalysis 2019;31:121–8. doi:10.1007/978-981-10-9035-6_35.
doi:10.1002/elan.201800537. [56] Negi R, Mathew R. Machine learning algorithms for diagnosis of breast cancer.
[26] Hasanzadeh M, Shadjou N, de la Guardia M. Early-stage screening of breast can- lecture notes on data engineering and communications technologies. Springer Sci
cer using electrochemical biomarker detection. Trends Anal Chem 2017;91:67–76. Bus Media Deutschl GmbH 2020;31:928–32.
doi:10.1016/j.trac.2017.04.006. [57] Asri H, Mousannif H, Al Moatassime H, et al. Using machine learning algorithms for
[27] Hong J, Kandasamy K, Marimuthu M, et al. Electrical cell-substrate impedance breast cancer risk prediction and diagnosis. Procedia Comp Sci 2016;83:1064–9.
sensing as a non-invasive tool for cancer cell study. Analyst 2011;136:237–45. doi:10.1016/j.procs.2016.04.224.
doi:10.1039/c0an00560f. [58] Borges L. Analysis of the wisconsin breast cancer dataset and machine learning
[28] Liu J, Zhou H, Xu JJ, et al. An effective DNA-based electrochemical switch for breast cancer detection. 2015, Available from: https://www.researchgate.net/
for reagentless detection of living cells. Chem Commun 2011;47:4388–90. publication/311950799_Analysis_of_the_Wisconsin_Breast_Cancer_Dataset_and_
doi:10.1039/c1cc10430f. Machine_Learning_for_Breast_Cancer_Detection (Accessed May 20, 2021).
[29] Seriburi P, McGuire S, Shastry A, et al. Measurement of the cell-substrate separation [59] Sinha N, Sharma P, Arora D. Prediction model for breast cancer detection
and the projected area of an individual adherent cell using electric cell-substrate using machine learning algorithms. Adv Intel Sys Comp 2021;1227:431–40.
impedance sensing. Anal Chem 2008;80:3677–83. doi:10.1021/ac800036c. doi:10.1007/978-981-15-6876-3_33.
[30] Yang J, et al. Quadruple signal amplification strategy based on hybridization chain [60] Gayathri BM, Sumathi CP. Proceedings of the IEEE International Conference on
reaction and an immunoelectrode modified with graphene sheets, a hemin/G- Computational Intelligence and Computing Research ICCIC; 2017.
quadruplex DNAzyme concatamer, and alcohol dehydrogenase: ultrasensitive de- [61] Karabatak M. A new classifier for breast cancer detection based on Naïve Bayesian.
termination of influenza virus subtype H7N9. Microchim Acta 2015;182:15–16. Meas J Int Meas Confed 2015;72:32–6. doi:10.1016/j.measurement.2015.04.028.
doi:10.1007/s00604-015-1583-8. [62] Celik Y, Sabanci K, Durdu A, et al. Breast cancer diagnosis by different machine
[31] Ikegwuonu T, Haddow G, Tait J, et al. Horizon scanning implanted biosensors in learning methods using blood analysis data. Int J Intell Syst Appl Eng 2018;6:289–
personalising breast cancer management: first pilot study of breast cancer patients 93. doi:10.18201/ijisae.2018648455.
views. Heal Sci Reports 2018;1. doi:10.1002/hsr2.30. [63] Mojrian S. 2020 RIVF International Conference on Computing and Communication
[32] Mendes IL, das Neves MF, Lopes FS. Biosensor applicability in breast can- Technologies, RIVF. IEEE; 2020.
cer diagnosis. Int J Biosens Bioelectron 2019;5(4):125–30. doi:10.15406/ijb- [64] Jeeva R, Subramaniyam D, Harshathunnisa A, et al. An accurate breast
sbe.2019.05.00165. cancer detection and classification using image processing. 2021; Available
[33] Ibnouhsein I, Jankowski S, Neuberger K, et al. The big data revolution for breast from: https://www.researchgate.net/publication/350609625_An_Accurate_Breast_
cancer patients. Eur J Breast Heal 2018;14(2):61–2. doi:10.5152/ejbh.2018.0101. Cancer_Detection_and_Classification_using_Image_Processing (Accessed August 09,
[34] Wang L. Early diagnosis of breast cancer. Sensors (Basel) 2017;17:1572. 2021).
doi:10.3390/s17071572. [65] Amit S, Thampi GT, Rao M. Inter-comparison of artificial neural network algo-
[35] Kulkarni V, Sahoo N, Chavan SD. Simulation of honeycomb-screen combinations rithms for time series forecasting: predicting Indian financial markets. Int J Comput
for turbulence management in a subsonic wind tunnel. J Wind Eng Ind Aerodyn Appl 2017;162(2):1–13. doi:10.5120/ijca2017913249.
2011;99:37–45. doi:10.1016/J.JWEIA.2010.10.006. [66] Sumbaly R, Vishnusri N, Jeyalatha S. Diagnosis of breast cancer using
[36] Al-Hadidi MR, Alarabeyyat A, Alhanahnah M. Proceedings of the 2016 9th inter- decision tree data mining technique. Int J Comput Appl 2014;98:16–24.
national conference on developments in esystems engineering. DeSE;2017:35–9. doi:10.5120/17219-7456.
doi:10.1109/DeSE.2016.8. [67] Aminikhanghahi S, Shin S, Wang W, et al. Proceedings of the symposium
[37] Sadoughi F, Kazemy Z, Hamedan F, et al. Artificial intelligence methods for on applied computing, 13. Association for Computing Machinery;2015:2252–6.
the diagnosis of breast cancer by image processing: a review. Breast cancer, doi:10.1145/2695664.2695832.
10;2018:219–30. doi:10.2147/BCTT.S175311. [68] Mejía TM, Pérez MG, Andaluz VH, et al. Proceedings of the 2015 Asia-pacific con-
[38] Benvidi A, Dehghani Firouzabadi A, Dehghan Tezerjani M, et al. A highly sen- ference on computer-aided system engineering. IEEE;2015:24–9. doi:10.1109/AP-
sitive and selective electrochemical DNA biosensor to diagnose breast cancer. J CASE.2015.12.
Electroanal Chem 2015;750:57–64. doi:10.1016/j.jelechem.2015.05.002. [69] Bevilacqua V, Brunetti A, Triggiani M, et al. Proceedings of the GECCO 2016 com-
[39] Wang J. Electrochemical biosensors: towards point-of-care cancer diagnostics. panion - genetic and evolutionary computation conference. Association for Com-
Biosens Bioelectron 2006;21:1887–92. doi:10.1016/j.bios.2005.10.027. puting Machinery, Inc;2016:1385–92. doi:10.1145/2908961.2931733.
[40] Rasheed PA, Sandhyarani N. Graphene-DNA electrochemical sensor for the sen- [70] Sharma S, Aggarwal A, Choudhury T. Proceedings of the international con-
sitive detection of BRCA1 gene. Sensors Actuators B Chem 2014;204:777–82. ference on computational techniques, electronics and mechanical systems.
doi:10.1016/j.snb.2014.08.043. CTEMS;2018:114–18. doi:10.1109/CTEMS.2018.8769187.
[41] Senel M, Dervisevic M, Kokkokoğlu F. Electrochemical DNA biosensors for label- [71] Bayrak EA, Kirci P, Ensari T. 2019 Scientific Meeting on Electrical-
free breast cancer gene marker detection. Anal Bioanal Chem 2019;411:2925–35. Electronics and Biomedical Engineering and Computer Science. EBBT 2019.
doi:10.1007/s00216-019-01739-9. doi:10.1109/EBBT.2019.8741990.
[42] Das M, Sumana G, Nagarajan R, et al. Zirconia based nucleic acid sen- [72] Saritas MM, Yasar A. International journal of intelligent systems and applications in
sor for Mycobacterium tuberculosis detection. Appl Phys Lett 2010;96:133703. engineering performance analysis of ann and naive bayes classification algorithm
doi:10.1063/1.3293447. for data classification. Orig Res Pap Int J Intell Syst Appl Eng IJISAE 2019;7:88–91.
[43] Patel MK, Solanki PR, Kumar A, et al. Electrochemical DNA sensor for doi:10.1039/b000000x.
Neisseria meningitidis detection. Biosens Bioelectron 2010;25:2586–91. [73] Rufai MA, Muhammad AS, Garba S, et al. Machine learning model for breast cancer
doi:10.1016/j.bios.2010.04.025. detection. FUDMA J Sci 2020;4(1):55–61.
[44] Liu S, Su W, Li Z, et al. Electrochemical detection of lung cancer specific microR- [74] Vaka AR, Soni B. SRK. Breast cancer detection by leveraging machine learning. ICT
NAs using 3D DNA origami nanostructures. Biosens Bioelectron 2015;71:57–61. Exp 2020;6:320–4. doi:10.1016/j.icte.2020.04.009.
doi:10.1016/j.bios.2015.04.006. [75] Gajdosova V, Lorencova L, Kasak P, et al. Electrochemical nanobiosen-
[45] Houfani D, Slatnia S, Kazar O, et al. Machine learning techniques for breast sors for detection of breast cancer biomarkers. Sensors 2020;20(14):4022.
cancer diagnosis: literature review. Adv Intel Sys Comp 2020;1103:247–54. doi:10.3390/s20144022.
doi:10.1007/978-3-030-36664-3_28.

80
Y. Amethiya, P. Pipariya, S. Patel et al. Intelligent Medicine 2 (2022) 69–81

[76] Gohring JT, Dale PS, Fan X. Detection of HER2 breast cancer biomarker using the [89] Kim S. Label-free surface-enhanced raman spectroscopy biosensor for on-site breast
opto-fluidic ring resonator biosensor. Sensors Actuators B Chem 2010;146:226–30. cancer detection using human tears. ACS Appl Mater Interfaces 2020;12:7897–904.
doi:10.1016/j.snb.2010.01.067. doi:10.1021/ACSAMI.9B19421.
[77] Gruhl FJ, Rapp M, Länge K. Label-free detection of breast cancer marker [90] Abrao Nemeir I, Mouawad L, Saab J. Electrochemical impedance spectroscopy
HER-2/neu with an acoustic biosensor. Procedia Eng 2010;5:914–17. characterization of label-free biosensors for the detection of HER2 in saliva
doi:10.1016/j.proeng.2010.09.258. 2020;60:7081. doi:10.3390/IECB2020-07081.
[78] Rafiee-Pour HA, Behpour M, Keshavarz M. A novel label-free electrochem- [91] Myung S, Solanki A, Kim C, et al. Graphene-encapsulated nanoparticle-
ical miRNA biosensor using methylene blue as redox indicator: application based biosensor for the selective detection of cancer biomarkers. Adv Mater
to breast cancer biomarker miRNA-21. Biosens Bioelectron 2016;77:202–7. 2011;23:2221–5. doi:10.1002/adma.201100014.
doi:10.1016/j.bios.2015.09.025. [92] Meng X. Electrochemical determination of microRNA-21 based on bio bar
[79] Arif S, Qudsia S, Urooj S, et al. Blueprint of quartz crystal microbalance biosen- code and hemin/G-quadruplet DNA enzyme. Analyst 2013;138:3409–15.
sor for early detection of breast cancer through salivary autoantibodies against doi:10.1039/c3an36788f.
ATP6AP1. Biosens Bioelectr 2015;65:62–70. doi:10.1016/j.bios.2014.09.088. [93] Hong CY. Ultrasensitive electrochemical detection of cancer-associated circulat-
[80] Jabin MA. Surface plasmon resonance-based titanium coated biosensor for cancer ing microRNA in serum samples based on DNA concatamers. Biosens Bioelectron
cell detection. IEEE Photonics J 2019;11(4). doi:10.1109/JPHOT.2019.2924825. 2013;50:132–6. doi:10.1016/j.bios.2013.06.040.
[81] Novodchuk I, Bajcsy M, Yavuz M. Graphene-based field effect transistor biosen- [94] Torrente-Rodríguez RM. Simultaneous detection of two breast cancer-related miR-
sors for breast cancer detection: a review on biosensing strategies. Carbon NAs in tumor tissues using p19-based disposable amperometric magnetobiosensing
2021;172:431–53 N Y. doi:10.1016/j.carbon.2020.10.048. platforms. Biosens Bioelectron 2015;66:385–91. doi:10.1016/j.bios.2014.11.047.
[82] Shafiei F, Saberi RS, Mehrgardi MA. A label-free electrochemical aptasen- [95] Contreras Jiménez G, Eissa S, Ng A, et al. Aptamer-based label-free impedi-
sor for breast cancer cell detection based on a reduced graphene oxide- metric biosensor for detection of progesterone. Anal Chem 2015;87:1075–82.
chitosan-gold nanoparticle composite. Bioelectrochem 2021;140:107807. doi:10.1021/ac503639s.
doi:10.1016/j.bioelechem.2021.107807. [96] Azimzadeh M, Rahaie M, Nasirizadeh N, et al. An electrochemical nanobiosen-
[83] Mansor NA. Detection of breast cancer 1 (BRCA1) gene using an electrochemi- sor for plasma miRNA-155, based on graphene oxide and gold nanorod,
cal DNA biosensor based on immobilized ZnO nanowires. Open J Appl Biosens for early detection of breast cancer. Biosens Bioelectron 2016;77:99–106.
2014;3:9–17. doi:10.4236/OJAB.2014.32002. doi:10.1016/j.bios.2015.09.020.
[84] Hossain MB, Akib TBA, Abdulrazak LF, et al. Numerical modeling of graphene- [97] Xu S, et al. Ultrasensitive label-free detection of DNA hybridization by sapphire-
coated fiber optic surface plasmon resonance biosensor for BRCA1 and BRCA2 based graphene field-effect transistor biosensor. Appl Surf Sci 2018;427:1114–19.
genetic breast cancer detection. Opt Eng 2019;58:1. doi:10.1117/1.oe.58.3. doi:10.1016/j.apsusc.2017.09.113.
037104. [98] Filippidou MK, et al. Detection of BRCA1 gene on partially re-
[85] Li T, Fan Q, Liu L, et al. Detection of breast cancer cells specially and ac- duced graphene oxide biosensors. Microelectron Eng 2019;216:111093.
curately by an electrochemical method. Biosens Bioelectron 2010;25:2686–9. doi:10.1016/j.mee.2019.111093.
doi:10.1016/j.bios.2010.05.004. [99] Huang S, Yang J, Fong S, et al. Artificial intelligence in cancer diagno-
[86] Cardoso AR, Moreira FTC, Fernandes R, et al. Novel and simple electrochemical sis and prognosis: opportunities and challenges. Canc Lett 2020;471:61–71.
biosensor monitoring attomolar levels of miRNA-155 in breast cancer. Biosens Bio- doi:10.1016/j.canlet.2019.12.007.
electron 2016;80:621–30. doi:10.1016/j.bios.2016.02.035. [100] Shukla R, Yadav V, Ram Pal P, et al. Machine learning techniques for detecting and
[87] Hakimian F, Ghourchian H. Ultrasensitive electrochemical biosensor for detection predicting breast cancer. Int J Innov Technol Explor Eng 2019;8(7):2658–62.
of microRNA-155 as a breast cancer risk factor. Anal Chim Acta 2020;1136:1–8. [101] Bi WL. Artificial intelligence in cancer imaging: clinical challenges and applica-
doi:10.1016/j.aca.2020.08.039. tions. Cancer J Clin 2019;69:127–57. doi:10.3322/caac.21552.
[88] Salahandish R, Ghaffarinejad A, Naghib SM, et al. Nano-biosensor for highly sensi- [102] Campuzano S, Pedrero M, Pingarrón JM. Non-invasive breast cancer diagnosis
tive detection of HER2-positive breast cancer. Biosens Bioelectron 2018;117:104– through electrochemical biosensing at different molecular levels. Sensors (Basel)
11. doi:10.1016/j.bios.2018.05.043. 2017;17(9):1993. doi:10.3390/s17091993.

81