Neural Computing and Applications (2023) 35:23711–23724

https://doi.org/10.1007/s00521-020-05362-z
(0123456789().,-volV)(0123456789().,-volV)

S.I.: DEEP NEURO-FUZZY ANALYTICS IN SMART ECOSYSTEMS

Hybrid bio-inspired algorithm and convolutional neural network

for automatic lung tumor detection
Surbhi Vijh1 • Prashant Gaurav2 • Hari Mohan Pandey3

Received: 24 July 2020 / Accepted: 9 September 2020 / Published online: 19 September 2020
 The Author(s) 2020

Abstract
In this paper, we have proposed a hybrid bio-inspired algorithm which takes the merits of whale optimization algorithm
(WOA) and adaptive particle swarm optimization (APSO). The proposed algorithm is referred as the hybrid WOA_APSO
algorithm. We utilize a convolutional neural network (CNN) for classification purposes. Extensive experiments are per-
formed to evaluate the performance of the proposed model. Here, pre-processing and segmentation are performed on 120
lung CT images for obtaining the segmented tumored and non-tumored region nodule. The statistical, texture, geometrical
and structural features are extracted from the processed image using different techniques. The optimized feature selection
plays a crucial role in determining the accuracy of the classification algorithm. The novel variant of whale optimization
algorithm and adaptive particle swarm optimization, hybrid bio-inspired WOA_APSO, is proposed for selecting optimized
features. The feature selection grouping is applied by embedding linear discriminant analysis which helps in determining
the reduced dimensions of subsets. Twofold performance comparisons are done. First, we compare the performance against
the different classification techniques such as support vector machine, artificial neural network (ANN) and CNN. Second,
the computational cost of the hybrid WOA_APSO is compared with the standard WOA and APSO algorithms. The
experimental result reveals that the proposed algorithm is capable of automatic lung tumor detection and it outperforms the
other state-of-the-art methods on standard quality measures such as accuracy (97.18%), sensitivity (97%) and specificity
(98.66%). The results reported in this paper are encouraging; hence, these results will motivate other researchers to explore
more in this direction.

Keywords Medical imaging  Artificial intelligence  Feature extraction  Hybrid WOA_APSO  Convolutional neural
network

1 Introduction

According to WHO statistics, the estimated worldwide

death of people from cancer is 9.6 million in 2018. How-
ever, it is observed that 30–50% of people can be prevented
& Hari Mohan Pandey
Pandeyh@edgehill.ac.uk from cancer by providing preventive measures and treat-
ment at an early stage of cancer. Lung cancer is also
Surbhi Vijh
surbhivijh428@gmail.com referred to as lung carcinoma characterized by uncontrol-
lable cell growth in tissues which generally have been
Prashant Gaurav
prashantgaurav36@gmail.com categorized as small cell and non-small cell carcinoma on
the basis of cellular structure [1]. On the basis of tumor
1
Department of Information Technology, KIET Group of lymph node location and tumor size, there are four stages
Institution, Ghaziabad, India of lung cancer from I to IV [2, 3].
2
Department of Computer Science and Engineering, Amity Computed tomography (CT) imaging is considered the
University, Sector 125, Noida, Uttar Pradesh, India finest way of analyzing and visualizing the abnormalities
3
Department of Computer Science, Edge Hill University, present due to less distortion [4]. Image pre-processing
Ormskirk, Lancashire, England, UK

123
23712 Neural Computing and Applications (2023) 35:23711–23724

consisting of normalization and enhancement is usually prognosis of lung cancer done by the experts. The com-
performed to improve the quality of the image and reduce parative classification analysis is performed by considering
the distortion. There are different types of filters applied for techniques such as artificial neural network, support vector
image enhancement techniques in spatial and frequency machine and convolutional neural network [19, 20]. Sup-
domain [5]. The wiener filter minimizes the noise using a port vector machine uses the multiclass model learning
low pass filter and performs deconvolution using high pass technique to determine the prediction of medical imaging.
filter (Inverse filtering) on lung CT image obtained from The support vector machine (SVM) considers a linear
the Lung Image Database Consortium [6]. The segmenta- function hypothetical space in higher dimension features
tion is the most challenging task in medical imaging for which are instructed using a learning bias originated from
appropriately extracting the features from the segmented statistical learning theory [21]. The artificial neural net-
tumor nodule region. There are various segmentation work represents a multilayer network consisting of three
techniques [7] such as Watershed transform edge-based different layers with the backpropagation algorithm
segmentation, Region-based segmentation [8] and [22, 23]. Deep learning [24] is a promising field and pro-
Thresholding [9]. The global thresholding and morpho- vides enhanced performance in various medical imaging
logical post-processing operation are performed for applications [24, 25]. The fully connected convolutional
detecting the tumored region from surrounding and seg- neural network consisting of a mesh connection of neurons
menting the lung region nodule. The different statistical comprises activation function and backpropagation for
features, textural features, shape and geometrical-based adjustment of weights [26, 27]. The presented state-of-art
features are extracted from the segmented region [10]. The outperforms for the fully connected convolutional neural
number of features is prominently more in neuroimaging, network in computer-aided diagnosis of lung cancer
so the technique used for feature extraction is Gray-Level imaging by showing better performance.
Co-occurrence Matrix (GLCM), Gray-Level Run-Length
Matrix (GLRLM), Histogram features, Gray-Level 1.1 Motivation and contributions
Dependence Matrix (GLDM) and Local Binary Pattern
(LBP) [11, 12]. These sets of extracted features provide Metaheuristic hybridization is the recent trend in the field
significant information from medical imaging which helps of optimization, achieved by combining one metaheuristic
in evaluating the pattern and decision-making process. To algorithm with a complementing metaheuristic algorithm.
reduce the dimensional features space of different modal- The novel variant of whale optimization algorithm and
ities of the image, the hybrid WOA_APSO feature selec- adaptive particle swarm optimization, hybrid bio-inspired
tion algorithm is proposed, which removes the redundant WOA_APSO, is proposed for selecting optimized features.
features [13, 14]. The feature selection technique optimizes Hybrid WOA_APSO is implemented for the selection of
the selection of extracted features and provides more the optimized features subset. The whale optimization
dominant information. The extracted feature selection is algorithm is a recently developed metaheuristic algorithm
grouped using a linear discriminant algorithm [15–18] for with many good properties like few parameters, covers a
selecting the more dependent and relevant features on class broader area in the search space, and exploration of the
values to improve the performance of classification. The search space is made efficient by using randomly selected
subset of the best feature is selected to reduce the dimen- search agents in place of the best search agent so far.
sionality of problem space and to maximize the perfor- Further, APSO uses additional adaptive parameters—
mance of the learning algorithm (Fig. 1). adaptive factor fa and perturbation factor Ga to improve the
Classification is the most critical aspect in determining convergence speed of the algorithm and to achieve a bal-
the performance parameters, i.e., accuracy, sensitivity and ance between exploitation and exploration of the search
specificity of the model that helps in the process of space [28–32]. Thus, WOA_APSO performs better than
other common algorithms in terms of convergence speed
and balancing between exploration and exploitation
[28, 29, 33]. So, the hybridization of WOA and APSO is
proposed and is explained in detail in the paper.
The key contribution of this paper is highlighted as
follows:
• Firstly, we propose an intelligent lung tumor segmen-
tation algorithm for the detection of tumor and treat-
(i) Tumored (ii) Non-tumored ment of cancer patients.
Fig. 1 i Tumored lung CT image ii Non-tumored lung CT image

123
Neural Computing and Applications (2023) 35:23711–23724 23713

• Secondly, we show the hybridization of two meta- performance is computed by evaluating the correlation
heuristic algorithms, namely, whale optimization algo- values. Joon et al. [44] showed the segmentation of
rithm (WOA) and adaptive particle swarm optimization cancerous and non-cancerous lung region nodule using
(APSO). The proposed algorithm is referred as the k-mean clustering and fuzzy c-mean technique. The
hybrid WOA_APSO algorithm. Hybrid WOA_APSO is structural and texture extracted features are used in the
implemented for the selection of the optimized features classification performed by the support vector machine.
subset. Here, feature selection grouping is performed by Prabukumar et al. [45] proposed a hybrid segmentation
embedding linear discriminant analysis. Our proposed technique comprising Fuzzy C-means (FCM) and region
model utilizes a CNN for classification purposes. growing algorithm to segment the nodule. The statistical,
• Thirdly, extensive computer simulations are performed texture and geometrical features are extracted from the
to determine the effectiveness of the proposed model. segmented nodule, and the optimized features are selected
We reported a twofold performance comparison. First, using a cuckoo search algorithm. The optimal features are
the performance of the proposed model is tested against considered for evaluating the classification using a support
different state-of-the-art classification techniques and vector machine while obtaining the overall accuracy of
evaluated accuracy, sensitivity and specificity. Sec- 98.5%. Mittal et al. [46] proposed a methodology for
ondly, the performance of the proposed algorithm is multilevel image thresholding for segmentation and intro-
compared with the standard WOA and APSO algo- duced the exponential kbest gravitational search technique.
rithms based on the computational cost of convergence Shankar et al. [47] stated a methodology for Alzheimer
to the optimal results. detection in which various features are extracted using
gray-level run-length matrix, histogram features, gray-level
occurrence matrix, the local binary pattern features and
1.2 Organization
scale-invariant transform. The feature selection is per-
formed using the gray wolf optimization algorithm to attain
The remainder of the paper is organized as follows: Sect. 2
the optimized features for performing the classification.
presents the related work; the proposed model is compre-
The convolutional neural network classification technique
hensively discussed in Sect. 3; Sect. 4 represents the
is used to achieve the accuracy, sensitivity and specificity
experimental results and computational performance mea-
as 96.23%, 94% and 96%. Vijh et al. [48] proposed an
sures; the conclusion and future avenue of the proposed
approach for developing the computer-aided lung tumor
work is given in Sect. 5.
segmentation system. The whale optimization algorithm is
used for feature selection and support vector machine for
classification. The performance of methodology is com-
2 Related work
pared using different SVM kernels. However, the RBF
support vector kernel provided the accuracy, sensitivity and
Cancer is the deadly disease worldwide whose detection at
specificity of 95%, 100% and 92%.
an early stage provides preventive measures and treatment
Reddy et al. [49] presented algorithm consists of parallel
to increase the survival rate of patients. The research in
thresholding, feature extraction and fuzzy neural network
medical imaging is growing rapidly to recognize the pat-
for identifying the lung tumor on CT imaging evaluating
tern of disease showing the development of computation-
the accuracy of 96.5%. Zhang et al. [50] stated a novel
ally intelligent systems. Data visualization for medical
approach for computer-aided diagnosis for lung tumor
image analysis is one of the most promising fields in the
detection by implementing a multiscale mask region-based
development of a robust expert system using artificial
convolutional neural network on PET imaging. The per-
intelligence [34, 35], computer vision [36] and pattern
formance computed by estimating recall, precision and F-
recognition application [37–39]. The computer-aided
value as 1, 0.90 and 0.95. Uçar et al. [51] presented the
medical diagnosis is the most challenging task in the
methodology for automatic detection of lung nodules using
interpretation of the decision-making process by a radiol-
deep learning convolutional neural network architecture
ogist for extracting the abnormalities in the image [40, 41].
with Laplacian of gaussian filter model obtaining the
The related research work and findings are presented in this
accuracy of 72.97%. Naqi et al. [52] proposed a novel
section.
hybrid approach consisting of 3D neighborhood connec-
Uzelaltinbulat et al. [42] presented a lung tumor seg-
tivity, active contour model (ACM) and geometric prop-
mentation technique using Otsu thresholding and mor-
erties for 3D nodule candidate detection. The comparative
phological operations. Kumar et al. [43] proposed a hybrid
analysis of classification is performed using Naı̈ve Bayes,
of the 2D Otsu method and modified artificial bee colony
KNN, SVM and AdaBoost to evaluate the model effec-
method for the segmentation of the lung CT image. The
tiveness. Table 1 presents a summary of the existing works.

123
23714 Neural Computing and Applications (2023) 35:23711–23724

Database Image Pre-processing Segmentation Nodule detection Features Optimization Classification Final Outcome

Norma Global GLCM

lize Thresh WOA_
Image olding APSO
Source GLDM
(hybrid) Tumore
image Tumored d
/Non- GLRL
M CNN
Enhan Tumored
ced Region
Mathem Feature
Image: atical Groupin
HOG
Morphol g with
Wiener
ogical LDA
Filterin
operatio Non-
g LBP
n tumored

Fig. 2 Flow process of proposed methodology

Table 1 Comparative study of the existing techniques

Author Dataset Segmentation methods Classification Performance

Manikandan Lung CT images Fuzzy auto seed cluster SVM kernel’s Specificity-
et al. [53] 93%
Accuracy-
94%
Kavitha et al. Lung CT images Otsu’s Thresholding and Artificial neural network Accuracy-
[54] Morphological segmentation method 92.68%
Precision-
87.50%
Recall-
100%
Kumar et al. Brain MRI T1 Gradient vector flow model Principal component analysis and Artificial Accuracy-
[55] weighted neural network (PCA-ANN) 91.17%
El Abbadi MRI brain images Morphological segmentation Probabilistic neural network Accuracy-
et al. [56] 98%
Nibali et al. LIDC/IDRI dataset- – Residual network (ResNet)—deep residual Accuracy-
[57] Lung CT images network 89.90%
Sensitivity-
91.07%
Specificity-
88.64%
Ali et al. [58] LIDC/IDRI dataset— – Reinforcement learning convolutional –
lung CT images neural network
Duarte et al. MIAS Fisher discriminant analysis and – Accuracy-
[59] texture features 0.94 ±
0 .019
Esteva et al. Clinical images Deep learning convolutional Accuracy-
[60] neural network architecture 72%
Vijh et al. Brain MRI Otsu Thresholding ? APSO, Convolutional neural network (CNN) Accuracy-
[61] morphological operation 98%
Loss
function-
4%
The comparison is done based on the dataset, segmentation methods, classification method and performance of the techniques. For performance
standard, measures such as specificity, accuracy, precision and recall have been reported

123
Neural Computing and Applications (2023) 35:23711–23724 23715

3 Proposed methodology Opening : D  F ¼ D  F  DF ð5Þ

This section sheds light on the proposed methodology. The

simulated phases implemented in the proposed methodol- Closing : D  F ¼ DFF ð6Þ
ogy of computer-aided automatic diagnosis system for the
detection of lung tumor as shown in Fig. 2 are (1) Image
Acquisition and Normalization, (2) Image pre-processing,
5. Feature extraction is the most important phase in
(3) Image segmentation, (4) Mathematical morphological
obtaining the pattern information of the segmented
operations, (5) Feature extraction, (6) Feature selection
nodule. In the proposed methodology, the total 60
with grouping and (7) Classification.
different geometrical, statistical, texture and structure
1. 120 samples of lung CT images are obtained from the features are extracted from each segmented nodule.
NCI Lung Cancer Database Consortium [62]. After the The techniques used for extraction are Gray-level co-
acquisition of lung CT images, the normalization is occurrence matrix (GLCM), Gray-level run-length
performed on tumored and non-tumored CT images matrix (GLRLM), Histogram-oriented gradient fea-
which are available in the cancer imaging archive. tures (HOG), Gray-level dependence matrix (GLDM)
2. Image pre-processing is considered an important phase and Local binary pattern (LBP). GLCM is referred to
in clinical research application of neuroimaging for as a second-order statistics method which considers the
improving the quality. Image enhancement is achieved spatial relationship between a couple of pixels.
by applying a Wiener filter for denoising the image GLRLM helps in obtaining higher-order statistical
with a minimum mean square error. The Wiener features consisting of a set of continuous pixels having
filtering is the statistical approach for reducing the similar gray levels [63]. GLDM extracts the features by
blurring and smoothing effect in the image. F ½n computing a gray-level absolute difference method
represents the outcome of the Wiener filter as depicted between two pixels separated by specific displacement
in Eq. (1). [64]. Histogram-oriented gradient extracts feature by
Xn focusing on the structure of the image and uses the
F ½ n ¼ ak w½n  i ð1Þ feature descriptor for counting the occurrence of
k¼0 gradient orientation in localized portion [65]. LBP
3. Image segmentation is the process of partitioning the considers the shaping-based LBP operator for the lung
image in multiple regions consisting of a different set CT image which threshold the neighboring pixels
of the pixel. The global thresholding technique is based on the value of current pixels [66]. Table 2
projected to partition the image depending upon the shows the name of the features extracted from the
intensity of the gray-level pixels for threshold T. The segmented tumored and non-tumored lung images for
segmented image acquired from global thresholding the analysis.
can be represented as H(x, y) and using Eq. (2). Here, 6. Nature inspired metaheuristic optimization algorithms
tðx; yÞ is the pixel value of the image. mimic a physical or biological phenomenon for solving
 real-world optimization problems. The novel hybrid
1 if tðx; yÞ [ T algorithm comprising whale optimization algorithm
H ðx; yÞ ¼ ð2Þ
0 if tðx; yÞ  T
and adaptive particle swarm optimization
4. Mathematical morphological operations are estimated (WOA_APSO) is proposed for the selection of the
by applying the certain structuring element at all optimized dimension of features and for providing
possible location for smoothing the region of interest. effective results.
The mathematical operations are performed where D is 6:1 Mathematical formulation of the hybrid WOA_APSO
binary image and F is the structuring element as optimization algorithm
depicted in Eqs. (3), (4), (5) and (6), respectively. Initially, the hybrid of WOA_APSO algorithm
  begins with a random solution. However, search agents
Erosion : D  F ¼ AjðF ÞA  D ð3Þ modify their position with respect to specific agent
behavior. The objective function Fitob used for each
  iteration is outlined in Eq. (7) and also 2D Matlab plot
Dilation : D  F ¼ AjðF ÞA \ D 6¼ ; ð4Þ is depicted in Fig. 3
Fitob ¼ E ð1 þ bÞ=RF ð7Þ
RF ¼ m=S ð8Þ

123
23716 Neural Computing and Applications (2023) 35:23711–23724

where E calculates the overall error, b is constant having The bubble net phase formulation for a spiral updating
value 0.5, m represents no of selected features, S is the no position to mimic the helix-shaped movement of humpback
swarms. The parameters and the values used in hybrid bio- whales and prey is shown using Eq. (13).
inspired algorithm are represented in Table 3.
Y ði þ 1Þ ¼ E0  eal cos ð2plÞ þ Y ðiÞ ð13Þ
The hunting functionality is performed with the help of
the best search agent chasing the position of prey to The humpback whales revolve around the prey within a
encircle. This behavior can be mathematically represented spiral-shaped and shrinking circle simultaneously. The
using Eqs. (9) and (10), respectively. prey updates their location by using Eqs. (14), (15) and
 (16).
! ! !  h  i
~
E ¼  L  Y ðiÞ  Yð iÞ ð9Þ ~
Y ði þ 1Þ ¼ ðCtji viþ1 Þ þ f1 vi1j RtBest;t  xitj
! h  i
Y ði þ 1Þ ¼ Y ðiÞ  P  ~
~ E ð10Þ þ f2 vi2j RtBest;t  xitj : ð14Þ
L and P are the coefficient vectors, i is the latest ongoing Here, Rbest represents the local best search
!
iteration, Y shows the position vector and Y represents ytj ði þ 1Þ ¼ ytj ðiÞ þ fa Ctj ði þ 1Þ ð15Þ
the position vector of the best solution acquired. However,
the coefficient vector is shown in Eqs. (11) and (12). Gbest ¼ Gbest þ Ga ¼ Gbest þ max yj  min yj rand
P ¼ 2b  r  b ð11Þ ð16Þ

L¼ 2r ð12Þ Here, fa represents Adaptive factor and Gbest global best
location
where b is linearly decreasing from 2 to 0 and r is random In the exploration phase, the coefficient vector P is used
vector [0, 1]. for searching the prey, and it can be shown through
Eq. (17).

123
Neural Computing and Applications (2023) 35:23711–23724 23717

Table 2 Extracted features

Feature Name of feature Formula Feature Name of feature Formula

GLCM Dissimilarity P
g GLCM Cluster prominence P
g1
3
ji  jjH ði; jÞ Hij i  Fi þ j  Fj
i;j i;j¼0

Energy P
g1 Inverse difference moment P
g1
Hij
logðHijÞ2 normalized (IDM) 1þðijÞ2
i;j¼0 i;j¼0

Inverse difference P
g1
Hij Sum variance P
2g
normalized (IDN) 1þjijj  ð1  lÞ2 IH ði; jÞ
i;j¼0 i;j¼1

Autocorrelation P
g LBP Local binary Pattern P
H1
ðijÞH ði; jÞ 2g yðGh  GðZa1 Za2 ÞÞ
i;j¼0 h¼0
Contrast P
g1 GLDM GLDM prob½Kaðe; r Þ
H ði; jÞði  jÞ2
i;j¼0

Maximum probability Max(GPij) HOG Skewness P

g1
ðoi  meanÞ3 H ðoi Þ
h¼0
Sum of squares P g1
g1 P Kurtosis P
g1
ði  lÞ2 H ði; jÞ ðoi  meanÞ4 H ðoi Þ
i¼0 j¼0 h¼0

Sum average P
2g2 Mean P
g1
iHiþj ðiÞ oi H ðoi Þ
i;j¼2 h¼0

IM of correlation 1 Xij Xij 1 Variance P

g1
max fXi ;Xj g ðoi  meanÞ2 H ðoi Þ
h¼0
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Sum entropy P
2g2 Stand Deviation P
g1
Hiþj ðiÞlog Hiþj ðiÞ ðoi  meanÞ2 H ðoi Þ
i;j¼0 h¼0
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
P
IM of correlation 2 1  exp ðXij2  HijÞ High gray-level Run Emphasis t i2 H ði; jÞ
i;j

Cluster shade P
g1 GLRLM Run-length non-uniformity II 2
4 1
P P
Hij i  Fi þ j  Fj t H ði; jÞ
i;j¼0 i w
P w
Homogeneity P
g1
Hij Run percentage H ði;jÞw
i;j
1þðijÞ2
i;j¼0

Correlation P
g1 Inertia P g1
g1 P
Hij ðil ÞðjlÞ
1þðijÞ2 ði  jÞ H ði; jÞ
i;j¼0 i¼0 j¼0
1
P
Difference variance P
g1 Long run emphasis I w2 H ði; jÞ
ð1  lÞ2 H ði; jÞ t
i;j
i¼1
P X ði;wÞ
Difference entropy P
g1 Low gray-level run emphasis 1
 Hiþj ðiÞlog Hiþj ðiÞ t i2
i;j
i¼0
Angular second moment g1 P
P g1 Gray-level non-uniformity 2
P P
H ði; jÞ2 1
t H ði; jÞ
i¼0 j¼0 i w
P H ði;jÞ
Energy P
g1 Short run emphasis 1
log ðHijÞ2 t
i;j
i2
i;j¼0

~ !
Y ði þ 1Þ ¼ Yrandom  ~
P~
E ð17Þ 6:2 Feature selection grouping is performed using linear
discriminant analysis for reducing the dimensions
Time Complexity The proposed hybrid bio-inspired and selecting the best-optimized subsets which
algorithm has two inner loops for population ‘n’ and one enhance the classification performance effectively.
outer loop for iteration ‘t’ where n = 20 and t = 35. The
extreme case complexity of hybrid WOA_APSO can be
O(n2). However, when the population size is large, the time 7. Convolutional neural network is a deep learning
complexity of the algorithm can be represented as follows: classification technique for training and testing the
T(n) = 2n ? s ? nlog(n)t = O(nlog (n)t). learning network [67, 68]. The neural network

123
23718 Neural Computing and Applications (2023) 35:23711–23724

Table 4 Parameter and values used in convolutional neural network

Algorithm Parameter Value

CNN Layer 3
Activation function relu, softmax
Optimizer Adam
LOSS categorical_crossentropy
Epochs 200
validation_split 0.2
Fig. 3 2D Matlab plot for F(x, y) = E * (1 ? b)/RF BATCH SIZE 2

4 Experiment result and analysis

The 120 lung CT tumored and non-tumored images were

acquired from the lung cancer database consortium for
evaluation of experimental results. The optimized features
that are taken as input for classification are segregated in
7:3 ratio for training and testing purpose.
The optimized threshold value achieved from the
WOA_APSO algorithm is compared with the whale opti-
mization algorithm (WOA) and adaptive particle swarm
optimization (APSO). The achieved threshold value of
Fig. 4 Fully connected convolutional neural network WOA_APSO, WOA and APSO is 1.16, 2.09 and 1.9. So,
the proposed bio-inspired hybrid WOA_APSO algorithm is
Table 3 Usage of hybrid bio-inspired algorithm parameters and
giving precise informational optimized features subsets
values (Tables 5, 6).
The performance measures of intelligent lung tumor
Algorithm Parameter Values
diagnosis systems are acquired by comparing the different
WOA_APSO Lb 1 classification algorithms. The evaluation analysis parame-
Ub 160 ters used for determining the effectiveness of the model are
Dim 20 accuracy, sensitivity and specificity shown in Eqs. (18),
Max_iteration 35 (19) and (20), respectively.
Gamma 0.5 Accuracy : ½TP þ TN=Total 100 ð18Þ
SearchAgents_no 25
Sensitivity : ½TP=TP þ FN 100 ð19Þ
Specificity : ½1  FPR 100 ð20Þ
comprises three densely connected layers consisting of
where TP depicts the total number of correctly segmented
activation function connecting one neuron to another
images of true positives which are classified properly, FN
neuron as depicted in Fig. 4. The backpropagation
depicts a total number of correctly segmented images of
algorithm is used for updating the weights and deltas
true negatives which are not classified properly, FPR is the
with a learning rate of 0.001. The testing is performed
number of incorrect segmented images which are classified
on different parameters to identify the best combina-
properly.
tion for determining the robustness of the experiment.
The time complexity of the proposed algorithm (Algo-
The various parameter are as follows: Layer neurons:
rithm 1) is presented in Sect. 3. The computational time of
[5, 10] [10, 15], [15, 20]; Activation function: relu,
the proposed algorithm against the standard WOA and
softmax; Validation split: 0.1, 0.2, 0.3; Batch size: 1, 2,
APSO is reported in Table 7. Table 7 reveals that the
3; Learning rate: 0.1, 0.01, 0.001; and Epochs: 10, 20,
proposed algorithm takes less time to converge as com-
40, 60, 80, 100, 200. Table 4 represents the parameters
pared to the other two existing algorithms.
and values used for the CNN.

123
Neural Computing and Applications (2023) 35:23711–23724 23719

Table 5 Optimized selected features subset

Sum Long run Run-length non- Low gray-level High gray-level Mean Skewness Kurtosis Run-length Class
average emphasis uniformity run emphasis run emphasis non-uniformity value

2.29 19,218.84 96.4479 100.4414 5893.224 1.1447 2.01957 5.0786 96.44796 0

2.1947 28,150.36 163.6982 97.12592 5331.39 1.0971 2.7202 8.3994 163.6982 0
2.2103 25,320.4 80.3332 95.31727 5074.938 1.1049 2.57784 7.6452 80.33326 0
2.1522 47,876.4 78.6185 83.83347 3964.353 1.0759 3.20094 11.2460 78.61852 0
2.5389 12,683.62 128.558 106.8538 6759.883 1.2689 1.04237 2.0865 128.5588 0
2.5003 13,692.13 118.406 106.0758 6529.926 1.2496 1.15684 2.3382 118.4067 0
2.2186 26,600.17 85.7822 95.54607 5105.718 1.1090 2.50816 7.2908 85.78228 0
2.2494 30,015.29 79.6508 92.85763 4770.801 1.1244 2.27514 6.1762 79.65085 0
2.2017 27,257.69 92.2708 95.26133 5067.48 1.1006 2.6546 8.0468 92.27081 0
2.0691 101,773.3 232.107 49.68066 2765.702 1.0345 5.10078 27.01794 2.0691 1
2.0328 121,067.6 330.5319 39.3437 2633.991 1.01636 7.62568 59.15096 2.0328 1
2.0493 104,101.7 285.0864 50.65326 2781.266 1.02461 6.13689 38.66141 2.0493 1
2.0441 107,260.8 293.685 47.32784 2730.529 1.02203 6.51332 43.42328 2.0441 1
2.0268 141,059.5 439.4875 27.46381 2541.249 1.01336 8.4763 72.84769 2.0268 1
Run-length non-uniformity LBP 1 LBP 2 LBP 4 LBP 6 LBP 8 LBP 18 LBP 20 LBP 22 Class value

96.4479 0.6241 3.9027 5.6613 5.5266 9.4334 166.0049 178.3931 209.1352 0

163.6982 1.38394 14.4269 6.4487 14.8731 14.9557 187.4467 158.0956 183.8711 0
80.3332 0 2.6145 3.5119 2.0653 1.9645 120.6988 126.473 160.5806 0
78.6185 0 1.2176 0.5654 3.2268 2.6067 71.84998 70.3722 101.3662 0
128.5588 3.75799 8.1659 10.0845 8.1010 11.4551 203.7716 186.0544 235.3808 0
118.406 2.97823 9.1889 12.3220 8.2400 10.1207 201.6179 162.3591 225.4596 0
85.7822 0.68514 1.3914 5.8027 6.5657 4.9089 136.2018 135.387 166.0467 0
79.6508 0.97669 1.5798 4.3095 3.8455 1.6442 143.2166 137.4865 156.7091 0
92.2708 0 6.0355 3.6341 6.9010 4.4248 146.1222 153.3348 176.1286 0
2.0691 0.88508 0.9473 0.8688 2.6195 1.6197 30.97928 55.0529 32.1778 1
2.0328 0 0.8622 0 2.6543 0 52.63489 20.3920 39.4177 1
2.0493 0 1.3256 0.6096 0.6547 2.0256 14.30879 47.1313 16.4106 1
2.0441 0 0.884 1.6993 0.9474 1.4987 45.7859 38.8361 28.7902 1
2.0268 0 0.9572 0.9010 0 0 11.9131 15.9448 17.2959 1

LBP 24 LBP 34 LBP 36 LBP 38 LBP 40 LBP 50 LBP 52 LBP 54 LBP 56 Class value

195.0578 351.533 301.095 312.6975 319.8348 176.9504 203.6467 190.9865 161.6447 0

148.8005 302.4375 208.926 291.9901 216.2246 158.332 182.0684 151.3116 185.1942 0
155.4493 292.4331 207.4571 249.5617 208.6418 123.4223 158.4634 151.4662 119.5931 0
102.5716 181.7596 147.5484 147.1823 136.703 66.70948 101.6018 98.88786 72.21412 0
223.4218 395.9502 286.0657 329.443 339.4295 187.1072 235.5035 225.1967 199.3057 0
217.505 367.9648 290.6264 302.8536 318.8569 165.3334 224.8615 220.4193 198.9715 0
191.1737 366.9759 243.2482 293.0461 228.3787 132.0325 168.4901 186.2506 136.3982 0
161.79 332.7053 170.2044 315.1821 155.3123 136.9516 157.8566 162.9077 145.1957 0
169.3257 332.2018 213.7075 301.8197 231.957 151.542 176.2257 167.812 146.2633 0
55.4414 79.2554 54.8933 66.2689 56.7798 55.1357 31.49076 54.7443 30.1848 1
14.1459 61.3162 22.4648 82.7475 17.5012 18.6852 38.5865 12.4575 51.8421 1
58.1105 41.1567 50.6474 33.6050 52.2777 44.1412 15.8069 55.8905 12.3614 1
36.9702 52.3596 57.4833 62.0686 47.2138 38.1989 26.5194 36.2923 43.05667 1
22.5867 49.9930 18.9753 41.1125 18.3220 15.1034 16.4373 21.7003 11.0442 1

123
23720 Neural Computing and Applications (2023) 35:23711–23724

Table 6 Comparative performance analysis of classification 4.1 Result and discussion

Parameters SVM ANN CNN
Figure 1 shows the sample of considered CT tumored and
Accuracy 80 95.79 97.18 non-tumored lung images taken from the cancer imaging
Sensitivity 85 89.33 97 archives. The segmented image result analysis of the pro-
Specificity 95.12 95.45 98.66 posed algorithm at each successive for efficient detection
of the nodule is shown in Figs. 5 and 6, respectively.
Table 5 contains the optimized and discriminant set of
grouped features obtained by applying the proposed hybrid
Table 7 Comparison of efficiency in terms of time taken by different
whale optimization algorithm and adaptive particle swarm
metaheuristic technique optimization (WOA_APSO) grouped by applying linear
discriminant analysis (LDA). The comparative perfor-
Algorithm WOA APSO WOA_APSO
mance analysis of different classification techniques is
Time taken in seconds 189.9583 193.55621 184.48222 shown in Table 6. Figure 7 depicts that a convolutional
neural network provides better accuracy, sensitivity and
specificity in comparison to support vector machine and

Fig. 5 The sample result

analysis on lung tumored image
for detection of nodule

123
Neural Computing and Applications (2023) 35:23711–23724 23721

Fig. 6 The sample result analysis on lung non-tumored image for detection of nodule

artificial neural network. The achieved accuracy, sensitiv- process. The proposed state-of-art method provides a better
ity and specificity are 97.18, 97 and 98.66. consolidated optimized dimension of features selection
grouping approach by implementing a hybrid WOA_APSO
algorithm embedding LDA. The convolutional neural net-
5 Conclusion and future work work classification technique outperforms by providing an
accuracy of 97.5% in comparison to support vector
In this paper, we have presented a novel approach for early machine and artificial neural network. The methodology
detection, diagnosis and prediction to improve the treat- demonstrates the effectiveness and promising results for
ment of patients and take preventative measures. Here, we clinical application as compared to existing algorithms.
implemented a hybrid WOA_APSO algorithm (see Algo- The proposed method limits the working for 3-dimensional
rithm 1). Image pre-processing and segmentation technique medical imaging. In the future, the work can be extended
is applied for partitioning and segmenting the tumor region. by using different modalities of medical imaging, and other
The different features are extracted to gather the statistics metaheuristic techniques can be incorporated for enhancing
information analysis which assists in the decision-making the system performance.

123
23722 Neural Computing and Applications (2023) 35:23711–23724

6. Perumal S, Velmurugan T (2018) Preprocessing by contrast

enhancement techniques for medical images. Int J Pure Appl
Math 118(18):3681–3688
7. Srimani PK, Mahesh S (2013) A Comparative study of different
segmentation techniques for brain tumour detection. Int J Emerg
Technol Comput Appl Sci 2(4):192–197
8. Mesanovic N, Grgic M, Huseinagic H, Males M, Skejic E,
Smajlovic M (2011) Automatic CT image segmentation of the
lungs with region growing algorithm. In: 18th international
conference on systems, signals and image processing-IWSSIP,
pp 395–400
9. John J, Mini MG (2016) Multilevel thresholding based segmen-
tation and feature extraction for pulmonary nodule detection.
Procedia Technol 24:957–963
10. Novitasari DCR, Lubab A, Sawiji A, Asyhar AH (2019) Appli-
cation of feature extraction for breast cancer using one order
statistic, GLCM, GLRLM, and GLDM. Adv Sci Technol Eng
Syst J 4(4):115–120
11. Bala R (2017) Survey on texture feature extraction methods. Int J
Eng Sci 3:10375
Fig. 7 Comparative performance analysis 12. Zu C, Wang Y, Zhou L, Wang L, Zhang D (2018) Multi-modality
feature selection with adaptive similarity learning for classifica-
tion of Alzheimer’s disease. In: 2018 IEEE 15th international
symposium on biomedical imaging (ISBI 2018). IEEE,
pp 1542–1545
13. Mirjalili S, Lewis A (2016) The whale optimization algorithm.
Compliance with ethical standards Adv Eng Softw 95:51–67
14. Zhan ZH, Zhang J, Li Y, Chung HSH (2009) Adaptive particle
Conflict of interest The authors declare that they have no conflict of swarm optimization. IEEE Trans Syst Man Cybern Part B (Cy-
interest. bern) 39(6):1362–1381
15. Sharma A, Paliwal KK, Imoto S, Miyano S (2014) A feature
Open Access This article is licensed under a Creative Commons selection method using improved regularized linear discriminant
Attribution 4.0 International License, which permits use, sharing, analysis. Mach Vis Appl 25(3):775–786
adaptation, distribution and reproduction in any medium or format, as 16. Balakrishnama S, Ganapathiraju A (1998) Linear discriminant
long as you give appropriate credit to the original author(s) and the analysis-a brief tutorial. Inst Signal Inf Process 18:1–8
source, provide a link to the Creative Commons licence, and indicate 17. Anter AM, Hassenian AE (2018) Normalized multiple features
if changes were made. The images or other third party material in this fusion based on PCA and multiple classifiers voting in CT liver
article are included in the article’s Creative Commons licence, unless tumor recognition. In: Hassanien AE, Oliva DA (eds) Advances
indicated otherwise in a credit line to the material. If material is not in soft computing and machine learning in image processing.
included in the article’s Creative Commons licence and your intended Springer, Cham, pp 113–129
use is not permitted by statutory regulation or exceeds the permitted 18. Vijh S, Gaur D (2017) Secured biometric template matching by
use, you will need to obtain permission directly from the copyright using linear discriminant analysis. In: International conference on
holder. To view a copy of this licence, visit http://creativecommons. health information science. Springer, Cham, pp 194–203
org/licenses/by/4.0/. 19. De Bruijne M (2016) Machine learning approaches in medical
image analysis: from detection to diagnosis. Elsevier, Amsterdam
20. Jindal A, Aujla GS, Kumar N, Chaudhary R, Obaidat MS, You I
References (2018) SeDaTiVe: SDN-enabled deep learning architecture for
network traffic control in vehicular cyber-physical systems. IEEE
Netw 32(6):66–73
1. Sluimer I, Schilham A, Prokop M, Van Ginneken B (2006) 21. Nalepa J, Kawulok M (2019) Selecting training sets for support
Computer analysis of computed tomography scans of the lung: a vector machines: a review. Artif Intell Rev 52(2):857–900
survey. IEEE Trans Med Imaging 25(4):385–405 22. Bertolaccini L, Solli P, Pardolesi A, Pasini A (2017) An overview
2. Sivakumar S, Chandrasekar C (2013) Lung nodule detection
of the use of artificial neural networks in lung cancer research.
using fuzzy clustering and support vector machines. Int J Eng J Thorac Dis 9(4):924
Technol 5(1):179–185 23. Ganesan N, Venkatesh K, Rama MA, Palani AM (2010) Appli-
3. Dimililer K, Ever YK, Ugur B (2016) ILTDS intelligent lung cation of neural networks in diagnosing cancer disease using
tumor detection system on CT images. In: The international demographic data. Int J Comput Appl 1(26):76–85
symposium on intelligent systems technologies and applications. 24. Singh A, Aujla GS, Garg S, Kaddoum G, Singh G (2019) Deep
Springer, Cham, pp 225–235 learning-based SDN model for internet of things: an incremental
4. Makaju S, Prasad PWC, Alsadoon A, Singh AK, Elchouemi A
tensor train approach. IEEE Internet Things J 7(7):6302–6311
(2018) Lung cancer detection using CT scan images. Procedia 25. Shen D, Wu G, Suk HI (2017) Deep learning in medical image
Comput Sci 125:107–114 analysis. Annu Rev Biomed Eng 19:221–248
5. Bari M, Ahmed A, Naveed S (2019) Lungs cancer detection using 26. Suzuki K (2017) Overview of deep learning in medical imaging.
digital image processing techniques: a review. Mehran Univ Res
Radiol Phys Technol 10(3):257–273
J Eng Technol 38(2):351–360

123
Neural Computing and Applications (2023) 35:23711–23724 23723

27. Anwar SM, Majid M, Qayyum A, Awais M, Alnowami M, Khan 46. Mittal H, Saraswat M (2018) An optimum multi-level image
MK (2018) Medical image analysis using convolutional neural thresholding segmentation using non-local means 2D histogram
networks: a review. J Med Syst 42(11):226 and exponential Kbest gravitational search algorithm. Eng Appl
28. Pandey HM, Chaudhary A, Mehrotra D (2014) A comparative Artif Intell 71:226–235
review of approaches to prevent premature convergence in GA. 47. Shankar K, Lakshmanaprabu SK, Khanna A, Tanwar S, Rodri-
Appl Soft Comput 24:1047–1077 gues JJ, Roy NR (2019) Alzheimer detection using Group Grey
29. Pandey HM, Chaudhary A, Mehrotra D (2016) Grammar induc- Wolf Optimization based features with convolutional classifier.
tion using bit masking oriented genetic algorithm and compara- Comput Electr Eng 77:230–243
tive analysis. Appl Soft Comput 38:453–468 48. Vijh S, Gaur D, Kumar S (2019) An intelligent lung tumor
30. Pandey H (2017) Performance review of harmony search, dif- diagnosis system using whale optimization algorithm and support
ferential evolution and particle swarm optimization. In: IOP vector machine. Int J Syst Assur Eng Manag 11:1–11
conference series: materials science and engineering, vol 225. 49. Reddy UJ, Reddy BRVR, Reddy BE (2019) Recognition of lung
IOP Publishing cancer using machine learning mechanisms with fuzzy neural
31. Pandey HM (2016) Performance evaluation of selection methods networks. Trait Signal 36(1):87–91
of genetic algorithm and network security concerns. Phys Pro- 50. Zhang R, Cheng C, Zhao X, Li X (2019) Multiscale mask
cedia 78:13–18 R-CNN–based lung tumor detection using PET Imaging. Mol
32. Dhiman G, Soni M, Pandey HM et al (2020) A novel hybrid Imaging 18:1536012119863531
hypervolume indicator and reference vector adaptation strategies 51. Uçar M, Uçar E (2019) Computer-aided detection of lung nodules
based evolutionary algorithm for many-objective optimization. in chest X-rays using deep convolutional neural networks.
Eng Comput. https://doi.org/10.1007/s00366-020-00986-0 Sakarya Univ J Comput Inf Sci 2(1):41–52
33. Pandey HM (2020) Secure medical data transmission using a 52. Naqi SM, Sharif M, Lali IU (2019) A 3D nodule candidate
fusion of bit mask oriented genetic algorithm, encryption and detection method supported by hybrid features to reduce false
steganography. Future Gener Comput Syst 111:213–225 positives in lung nodule detection. Multimed Tools Appl
34. Jindal A, Aujla GS, Kumar N, Prodan R, Obaidat MS (2018) 78(18):26287–26311
DRUMS: demand response management in a smart city using 53. Manikandan T, Bharathi N (2016) Lung cancer detection using
deep learning and SVR. In: 2018 IEEE global communications fuzzy auto-seed cluster means morphological segmentation and
conference (GLOBECOM). IEEE, pp 1–6 SVM classifier. J Med Syst 40(7):181
35. Gulati A, Aujla GS, Chaudhary R, Kumar N, Obaidat MS (2018) 54. Kavitha P, Ayyappan G (2018) Lung cancer detection at early
Deep learning-based content centric data dissemination stage by using SVM classifier techniques. Int J Pure Appl Math
scheme for Internet of Vehicles. In: 2018 IEEE international 119(12):3171–3180
conference on communications (ICC). IEEE, pp 1–6 55. Kumar V, Sachdeva J, Gupta I, Khandelwal N, Ahuja CK (2011)
36. Aujla GS, Jindal A, Chaudhary R, Kumar N, Vashist S, Sharma Classification of brain tumors using PCA-ANN. In: 2011 world
N, Obaidat MS (2019) DLRS: deep learning-based recommender congress on information and communication technologies. IEEE,
system for smart healthcare ecosystem. In; ICC 2019–2019 IEEE pp 1079–1083
international conference on communications (ICC). IEEE, pp 1–6 56. El Abbadi NK, Kadhim NE (2017) Brain cancer classification
37. Teramoto A, Fujita H, Yamamuro O, Tamaki T (2016) Auto- based on features and artificial neural network. Brain
mated detection of pulmonary nodules in PET/CT images: 6(1):123–134
ensemble false-positive reduction using a convolutional neural 57. Nibali A, He Z, Wollersheim D (2017) Pulmonary nodule clas-
network technique. Med Phys 43(6Part1):2821–2827 sification with deep residual networks. Int J Comput Assist Radiol
38. Gore JC (2020) Artificial intelligence in medical imaging. Magn Surg 12(10):1799–1808
Reson Imaging 68:A1–A4, ISSN 0730–725X. https://doi.org/10. 58. Ali I, Hart GR, Gunabushanam G, Liang Y, Muhammad W,
1016/j.mri.2019.12.006 Nartowt B, Kane M, Ma X, Deng J (2018) Lung nodule detection
39. Jiang J, Trundle P, Ren J (2010) Medical image analysis with via deep reinforcement learning. Front Oncol 8:108
artificial neural networks. Comput Med Imaging 59. Duarte MA, Pereira WC, Alvarenga AV (2019) Calculating
Graph 34(8):617–631 texture features from mammograms and evaluating their perfor-
40. Chen CH (2015) Handbook of pattern recognition and computer mance in classifying clusters of microcalcifications. In:
vision. World Scientific, Singapore Mediterranean conference on medical and biological engineering
41. Bi WL, Hosny A, Schabath MB, Giger ML, Birkbak NJ, Mehr- and computing. Springer, Cham, pp 322–332
tash A, Allison T, Arnaout O, Abbosh C, Dunn IF, Mak RH 60. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM,
(2019) Artificial intelligence in cancer imaging: clinical chal- Thrun S (2017) Dermatologist-level classification of skin cancer
lenges and applications. CA Cancer J Clin 69(2):127–157 with deep neural networks. Nature 542(7639):115–118
42. Uzelaltinbulat S, Ugur B (2017) Lung tumor segmentation 61. Vijh S, Sharma S, Gaurav P (2020) Brain tumor segmentation
algorithm. Procedia Comput Sci 120:140–147 using OTSU embedded adaptive particle swarm optimization
43. Kumar S, Sharma TK, Pant M, Ray AK (2012) Adaptive artificial method and convolutional neural network. In: Hemanth J, Bhatia
bee colony for segmentation of CT lung images. Int J Comput M, Geman O (eds) Data visualization and knowledge engineer-
Appl iRAFIT 5:1–5 ing. Springer, Cham, pp 171–194
44. Joon P, Bajaj SB, Jatain A (2019) Segmentation and detection of 62. https://wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI
lung cancer using image processing and clustering techniques. In: 63. Preetha K, Jayanthi SK (2018) GLCM and GLRLM based feature
Saeed K, Chaki N, Pati B, Bakshi S, Mohapatra DP (eds) Pro- extraction technique in mammogram images. Int J Eng Technol
gress in advanced computing and intelligent engineering. 7(2.21):266–270
Springer, Singapore, pp 13–23 64. Ahmadi N, Akbarizadeh G (2018) Iris tissue recognition based on
45. Prabukumar M, Agilandeeswari L, Ganesan K (2019) An intel- GLDM feature extraction and hybrid MLPNN-ICA classifier.
ligent lung cancer diagnosis system using cuckoo search opti- Neural Comput Appl 32:1–15
mization and support vector machine classifier. J Ambient Intell 65. Dalal N, Triggs B (2005) Histograms of oriented gradients for
Hum Comput 10(1):267–293 human detection. In: 2005 IEEE computer society conference on

123
23724 Neural Computing and Applications (2023) 35:23711–23724

computer vision and pattern recognition (CVPR’05), vol 1. IEEE, diseases using a deep convolutional neural network. IEEE Trans
pp 886–893 Med Imaging 35(5):1207–1212
66. Dornaika F, Moujahid A, El Merabet Y, Ruichek Y (2017) A 68. Ker J, Wang L, Rao J, Lim T (2018) Deep learning applications
comparative study of image segmentation algorithms and in medical image analysis. IEEE Access 6:9375–9389
descriptors for building detection. In: Samui P, Roy SS, Balas VE
(eds) Handbook of neural computation. Academic Press, Cam- Publisher’s Note Springer Nature remains neutral with regard to
bridge, pp 591–606 jurisdictional claims in published maps and institutional affiliations.
67. Anthimopoulos M, Christodoulidis S, Ebner L, Christe A, Mou-
giakakou S (2016) Lung pattern classification for interstitial lung

123