Hacettepe University

Department of Artificial Intelligence Engineering

Report
Artificial Intelligence Applications in Medicine

Authors:
Talha KABA 2210765037
E-mail: kababey111@gmail.com

Atalay ÖNAL 2200765019

AIN101 Freshman Seminar Fall-2022

Senior: Suat ÖZDEMİR

Date: 31.12.2022
 Abstract
Artificial intelligence applications are mostly used effectively in the field of diagnosis of
diseases and the most progress has been made in this field. They usually do this by processing
images of patients. With machine learning, deep learning decision tree, artificial neural
networks methods; high-precision results are obtained. Applications that have been done were
mentioned: Detection of upper gastrointestinal cancer [3], lung cancer detection [6], heart
failure detection [9], breast cancer detection [2] & [4], nuclear medicine and temporal
subtraction [1], digital pathology [4], cataract detection via near infrared eye images [8]. By
utilizing the system in which patients are registered, deep learning-based natural language
processing systems are used to diagnose diseases [5]. The relationship between humans and
artificial intelligence in medicine has been studied from many angles [10].

1
Table of Contents

1. Introduction…………………………………………………….……..3

2. Clinical Decision Support System (CDSS)……………………….….4

2.1 D Design and Performance Considerations for AI-CDSS………………….4

3. Deep Learning and Machine Learning………………………………5

3.1. The difference between machine learning and deep learning……………......6

4. Methodologic Guide for Evaluating the Clinical Performance of

AI Technology………………………..…………………………………………….7

5. Use of AI in Field………………………………………………………7
5.1. Breast Cancer…………………………………………………….8
5.2. Upper gastrointestinal cancer…………………………………...8
5.3. Cataract Detection via Near Infrared Eye Images……………...9
5.4. Health Failure…………………………………………………….9
6. Discussion………………………………………………………….….....10
7. Conclusion……………………………………………………………….10
8. References……………………………………………………………….11

2
1. Introduction
The size of the data we obtain in healthcare field is increasing day by day with technology,
and the interpretation of this data has become difficult to process and interpret a large amount
of data. That makes things slow; thus, time -the most remarkable issue of healthcare- will loss
if processes would not be faster that's why we certainly need the use of artificial intelligence
while maintaining the diagnosis and treatment processes of diseases. Cancer is an aggressive
disease with a low average survival rate. It is costly due to high recurrence and mortality
rates. Accurate early diagnosis and prognosis of cancer is essential to increase the patient's
survival rate. Developments in statistics and computer engineering over the years have
encouraged many scientists to apply computational methods such as multivariate statistical
analysis to analyze the prognosis of the disease, such as multivariate statistical analysis and
the accuracy of these analyses is significantly higher than that of empirical predictions.
Machine learning and deep learning methods are often used to achieve these goals. Patient
diagnostic applications in medicine are typically performed through image, video analysis and
natural language processing (NLP), . When we talk about a lot of developments in the field of
artificial intelligence, there is some limits and obstacles: Ethical concerns, mistrust, the need
for large amounts of meaningful data to train AI are some of these barriers. Despite all that
total number of articles per year related to AI in diagnostics is continuously increasing. (As
can be seen in figure:1)

Total number of articles per year related to AI in diagnostics

Figure:1

3
2. Clinical Decision Support System (CDSS)
CDSS or Computer Aided Diagnosis (CAD) [4] is a healthcare IT system that is designed to
take inputs from various clinical sources to provide the clinician support with their decision
making. It has been defined as an active knowledge system that uses multiple items of patient
data to generate case-specific advice
Artificial intelligence (AI) is poised to make a veritable impact in medicine. Clinical decision
support (CDS) is an important area where AI can augment the clinician's capability to collect,
understand and make inferences on an overwhelming volume of patient data to reach the
optimal clinical decision. Advancements in medical image analysis, such as Radiomics, and
data computation, such as machine learning, have expanded our understanding of disease
processes and their management.
2.1. Design and Performance Considerations for AI-CDSS

Transparency: An important pitfall of an AI-based CDSS is the proverbial “black box,”

which means that there are no explanations available for its results. Furthermore, radiomics
may be perceived yet another unexplained piece of the puzzle, and collectively may pose a
challenge for its full-scale acceptance as most clinicians do not feel confident about relying on
the results of black boxes that do not draw their inferences or insights from discrete
knowledge and logic rules. Having said that, more real-world evidence demonstrating high
accuracy (80%-85%) and improved clinical efficiency (in terms of patient outcomes) is
required to mitigate this conventional notion
Efficiency and Integration: Another factor that can be critical in successful implementation
and increased physician use of an AI-CDSS is the way it incorporates into the workflow.44
User experience considerations should be taken into account to ensure that the systems work
seamlessly in the background, automatically extracting and processing data, and giving results
that are timely, easy to access and apply to the next steps. Comprehensive models and
frameworks for AI-CDSS, as integrating ML with ontological reasoning, and building toward
a transparent AI, have been adopted to improve scientific robustness, patient-centric design,
workflow efficiency, transparency and physician compliance.
Relevance and Context: With the correct clinical questions being at the heart of the design,
the AI-CDSS should be programmed to extract pertinent information from multiple sources
and process them to produce results that are actionable in terms of being relevant to the given
clinical scenario. This includes the most relevant inferences from radiomics, as well as
radiopathologic correlations that improve one's understanding of the current status of the
disease process.
Robustness and Reliability: Any insights that the AI-CDSS provides, while they can't be
“explained”, the results still need to be validated through robust statistical analysis. The
robustness of these models is typically evaluated by performing studies comparing human
performance vs that of the human + AI-CDSS paradigm. In that sense, it is thought to assist or
augment the capabilities of the human to make the right clinical decision. Having said that, for
the role or impact of AI-CDSS to increase, and for it to make more critical or important
decisions, it will need to meet rigorous standards of accuracy and reproducibility.

4
3. Deep Learning and Machine Learning
Deep learning and machine learning are two related but distinct approaches to artificial
intelligence.
Machine learning is a broader term that refers to any type of artificial intelligence that can
learn and adapt based on data. Machine learning algorithms can be trained on data sets to
perform a specific task, such as predicting the likelihood of an event occurring or identifying
patterns in data. Example of the use of machine learning in medicine is in the analysis of
genetic data to identify genetic variations associated with specific medical conditions. These
algorithms can be used to identify genetic markers for conditions such as cancer and to
develop personalized treatment plans based on a patient's genetic profile.
Deep learning is a type of machine learning that uses artificial neural networks to learn and
make decisions. These neural networks are designed to mimic the way the human brain
processes information and can learn and adapt based on the data they receive. Deep learning
is often used for tasks that require the analysis of large amounts of data, such as image and
voice recognition, natural language processing, and predictive modeling. In medicine, deep
learning algorithms can be used to analyze medical images, such as X-rays, CT scans, and
MRIs, and identify abnormalities or potential diseases [1]. For example, deep learning
algorithms have been developed to detect breast cancer in mammograms and lung cancer in
CT scans. Another use of deep learning in medicine is in the analysis of electronic health
records (EHRs). These algorithms [5] can be used to extract and analyze data from EHRs to
identify patterns and trends, and to predict the likelihood of certain medical conditions. This
can help healthcare providers to make more accurate diagnoses and to develop personalized
treatment plans for their patients. (In figure 2 difference between machine learning and deep
learning them shown.)

5
3.1. The difference between machine learning and deep learning

Figure 2: Deep learning vs traditional machine learning. (A) In the traditional paradigm,
several steps requiring expert human knowledge is required to recognize cancer in images.
First, image processing such as segmentation corrects the image and breaks it into
manageable parts. Next, hand-crafted measurements, or features, are extracted from each part.
A machine learning algorithm is provided those features as a vector, which it uses to learn a
predictive model. (B) In contrast, deep learning is an end-to-end approach to learning that
takes raw images as input and directly learns a model to produce the desired output. Deep
learning uses biologically inspired networks to represent data through multiple levels of
simple but nonlinear modules that transform the previous representation into a higher, slightly
more abstract representation. The compositional nature of the architecture allows deep
networks to form highly complex and nonlinear representations as each layer forms a more
abstracted representation than the last. The result is a rich representation that provides
unprecedented discriminatory power.

6
4.Methodologic Guide for Evaluating the Clinical Performance of AI Technology
First step: Define the clinical problem: The first step in the evaluation process is to clearly
define the clinical problem that the AI technology is being used to address. This may involve
identifying specific patient populations, diagnostic or treatment goals, or other relevant
factors. Second step: Select the evaluation method: several different evaluation methods that
can be used to assess the clinical performance of AI technology, including retrospective chart
review, prospective cohort studies, and randomized controlled trials. Third step: Determine
the performance measures: Selecting performance measures that are relevant to the clinical
problem being addressed and that are sensitive to the impact of the AI technology. Final step:
Analyze and interpret the results: The final step in the evaluation process is to analyze and
interpret the results of the evaluation in a transparent and rigorous manner. This may involve
comparing the performance of the AI technology to that of existing methods, or to a
benchmark or control group.
The authors argue that a methodologic guide [6] such as this is necessary for ensuring that the
clinical performance and effect of AI technology in the field of medical diagnosis and
prediction is properly evaluated and understood. They conclude by calling for increased
collaboration and transparency in the evaluation process, and for the development of best
practices and guidelines for conducting such evaluations.

5. Use of AI in Fields
Artificial intelligence is often developed for use in the detection of pathological diseases. For
diagnoses to be made from radiological images, a large number of radiological images are
required to train the AI. These data should be accurately diagnosed with the help of experts so
that more precise data can be obtained. The consensus in many fields is that the most accurate
diagnoses are made when AI and experts work together, rather than individually [9]. (Can be
seen on the diagnostic accuracy graph on figure 3)

Figure: 3

7
5.1 Breast Cancer
Breast cancer is the most common malignant disease in women worldwide. In recent years,
early diagnosis and better adjuvant therapy have significantly improved patient outcomes.
Diagnosis by histopathology has proven effective in guiding breast cancer treatment, but
new challenges have emerged over the years as our increasing understanding of cancer has
revealed its complex nature. As patient demand for personalized breast cancer treatment
increases, there is an urgent need for more sensitive biomarker assessment and more
accurate histopathological breast cancer diagnosis to make better treatment decisions.
Digitization of pathology data has opened the door to faster, more reproducible, and more
precise diagnoses through computerized image analysis. Software that aids diagnostic breast
pathology through image processing techniques has been around for years. But recent
breakthroughs in artificial intelligence (AI) promise to fundamentally change the way we
detect and treat breast cancer soon. Machine learning, a subfield of AI that applies statistical
methods to learn from data, has seen an explosion of interest in recent years due to its
ability to recognize patterns in data with less need for human instruction [4]. In particular, a
technique known as deep learning has produced groundbreaking results in many important
problems such as image classification and speech recognition. In this review, we will discuss
the use of AI and deep learning in diagnostic breast pathology and other recent advances in
digital image analysis

5.2 Upper gastrointestinal cancer

The study [3] aimed to assess the accuracy of a real-time artificial intelligence (AI) system for
detecting upper gastrointestinal (GI) cancer during endoscopy procedures. Upper GI cancer
includes cancer of the esophagus, stomach, and duodenum (the first part of the small
intestine). The study was conducted at multiple centers in China and included a case-control
design; cases were patients diagnosed with upper GI cancer and controls were patients
without cancer. During the endoscopy procedure, the AI system was used to analyze images
of the GI tract in real time and provide a diagnosis. The AI system's diagnosis was then
compared with the final diagnosis made by an experienced gastroenterologist. A total of 1,080
patients were included in the study, 540 in the case group and 540 in the control group. The
AI system had a sensitivity of 93.3% and a specificity of 97.0% for detecting upper GI cancer.
This means that the AI system correctly detected 93.3% of patients with upper GI cancer and
97.0% of patients without cancer. The experts of the study [3] concluded that the real-time AI
system was accurate for detecting upper GI cancer during endoscopy and has the potential to
improve the efficiency and accuracy of cancer diagnosis. However, they also noted that
further research is needed to assess the long-term performance and clinical utility of the AI
system.

8
5.3 Cataract Detection via Near Infrared Eye Images
In this study [8], the experts present a new method for detecting cataracts in the eye using
near infrared (NIR) eye images. Cataracts are a common condition that results in the
clouding of the lens in the eye, causing vision loss. Traditional methods for detecting
cataracts involve examining the eye through a microscope or using specialized imaging
techniques, such as ultrasound or magnetic resonance imaging (MRI). However, these
methods can be invasive and may not be accessible in resource-limited settings. The authors
propose using NIR eye images as a non-invasive method for detecting cataracts. NIR images
are taken by shining a NIR light onto the eye and capturing the reflection. They have the
advantage of being able to penetrate deeper into the eye and provide a clearer image of the
lens. To test the effectiveness of their method, the authors collected NIR images from 50
individuals with cataracts and 50 individuals without cataracts. They used a machine learning
algorithm to analyze the images and identify the presence of cataracts. The algorithm was
able to correctly identify cataracts in 88% of the cases. The authors conclude that NIR eye
images have the potential to be a reliable and non-invasive method for detecting cataracts.
Further research is needed to confirm the accuracy and effectiveness of this method, as well
as to explore its potential for use in resource-limited settings.

5.4 Health Failure

Heart failure is a serious and debilitating condition that affects millions of people worldwide
and is a leading cause of hospitalization and death. The experts [9] describe the current state
of heart failure diagnosis, which is often based on subjective evaluations and manual
analyses of patient data. This approach can be prone to errors and biases and may lead to
misdiagnosis or delayed treatment. To address these issues, the authors propose the use of
AI to improve the accuracy and speed of heart failure diagnosis. They describe several AI-
based approaches that have been developed, including machine learning algorithms, natural
language processing, and image analysis. These techniques allow for the automatic analysis
of large amounts of patient data, including electrocardiograms, echocardiograms, and other
imaging modalities. The authors also discuss the potential benefits of using AI in heart failure
diagnosis, including increased accuracy, reduced workload for healthcare providers, and
improved patient outcomes. They also address potential concerns and challenges, such as
the need for high-quality data, the potential for bias in AI algorithms, and the need for
proper validation and testing.

9
6. Discussion
The use of artificial intelligence in medicine, as long as the necessary studies are carried out,
should be used together with experts because there will not be enough experts to provide
health care to people in the world equally and therefore an auxiliary system will be needed, so
that the processes will be accelerated and more diagnoses, treatments or recommendations
will be offered in the same state of capacity. The level of auto transformation in the use of AI
technology in health depends on the sophistication and reliability of the technology. Different
situations occur according to the rate at which it assists health personnel. We can illustrate
this by analogy with the automatization of a car (as indicated in figure: 4). I foresee that
technical obstacles can be easily overcome, and these problems can be overcome with a
common regulation that appeals to all people to overcome the doubts and ethical concerns
about the use of artificial intelligence in medicine.

Figure: 4

7. Conclusion
Due to the nature of science, progress is made through accumulation, and accordingly, the
applications of artificial intelligence science in different fields develop as required by science.
in other words, developments in similar fields contribute to a brand-new progress. Due to the
nature of artificial intelligence, it develops together with other disciplines and every field it
encounters gives a great impetus to the normal development process. Therefore, the
applications of artificial intelligence in medicine will be a milestone for the development of
medical science and will provide important opportunities for the provision of health services
all over the world.

10
8. References:
[1] Junji Shiraishi, Qiang Li, Daniel Appelbaum; Kunio Doi, Computer-Aided Diagnosis
and Artificial Intelligence in Clinical Imaging, Seminars in Nuclear Medicine, Volume 41,
Issue 6; 2011, Pages 449-462; ISSN 0001-2998.
[2] Shigao Huang, Jie Yang, Simon Fong, Qi Zhao; Artificial intelligence in cancer
diagnosis and prognosis: Opportunities and challenges; Cancer Letters, Volume 471; 2020;
Pages 61-71; ISSN 0304-3835.
[3] Huiyan Luo, Guoliang Xu, Chaofeng Li, Longjun He, Linna Luo, Zixian Wang,
Bingzhong Jing, Yishu Deng, Ying Jin, Yin Li, Bin Li, Wencheng Tan, Caisheng He,
Sharvesh Raj Seeruttun, Qiubao Wu, Jun Huang, De-wang Huang, Bin Chen, Shao-bin Lin,
Qin-ming Chen, Chu-ming Yuan, Hai-xin Chen, Heng-ying Pu, Feng Zhou, Yun He, Rui-hua
Xu;Real-time artificial intelligence for detection of upper gastrointestinal cancer by
endoscopy: a multicentre, case-control, diagnostic study; The Lancet Oncology; Volume 20,
Issue 12; 2019; Pages 1645-1654; ISSN 1470-2045.
[4] Stephanie Robertson, Hossein Azizpour, Kevin Smith, Johan Hartman; Digital image
analysis in breast pathology—from image processing techniques to artificial intelligence;
Translational Research; Volume 194; 2018; Pages 19-35; ISSN 1931-5244.
[5] Liang, H., Tsui, B.Y., Ni, H. et al. Evaluation and accurate diagnoses of pediatric
diseases using artificial intelligence. Nat Med; Volume 25; 2019; Pages 433–438.
[6] PARK, Seong Ho; HAN, Kyunghwa. Methodologic guide for evaluating clinical
performance and effect of artificial intelligence technology for medical diagnosis and
prediction. Radiology, Volume 286.3; 2018; Pages 800-809.
[7] Briganti, Giovanni; LE MOINE, Olivier. Artificial intelligence in medicine: today and
tomorrow. Frontiers in medicine; 2020, 7: 27.
[8] Pavani Tripathi, Yasmeena Akhter, Mahapara Khurshid, Aditya Lakra, Rohit Keshari,
Mayank Vatsa, Richa Singh, MTCD: Cataract detection via near infrared eye images;
Computer Vision and Image Understanding; Volume 214; 2022; 103303; ISSN 1077-3142.
[9] Choi, DJ., Park, J.J., Ali, T. et al. Artificial intelligence for the diagnosis of heart
failure. npj Digit. Med. 3, 54 (2020)
[10] Topol, E.J. High-performance medicine: the convergence of human and artificial
intelligence. Nat Med 25, 44–56 (2019).

11