IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 3, September 2024, pp. 2912~2922

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i3.pp2912-2922  2912

Optimizing pulmonary carcinoma detection through image

segmentation using evolutionary algorithms

Moulieswaran Elavarasu, Kalpana Govindaraju

Department of Computer Science, Faculty of Science and Humanities, SRM Institute of Science and Technology, Kattankulathur, India

Article Info ABSTRACT

Article history: This paper’s goal is to suggest an image segmentation technique for use with
medical images, specifically computer tomography scan images, to aid
Received Aug 19, 2023 doctors in understanding the images. To address a variety of picture
Revised Oct 8, 2023 segmentation issues, it is necessary to investigate and apply novel
Accepted Dec 14, 2023 evolutionary algorithms. The study focuses on pulmonary carcinoma, which
is the cancer that affects males the most frequently across the globe. For
proper treatment and life-saving measures, early identification of lung cancer
Keywords: is essential. To identify lung cancer, doctors frequently employ the
computed tomography imaging technique. In order to extract tumours from
Evolutionary algorithms lung scans, the study analyses the effectiveness of three optimization
Image filtering algorithms: k-means clustering, particle swarm optimization, and modified
Image segmentation guaranteed convergence particle swarm optimization. The study also
Preprocessing examines the pre-processing performance of four filters, namely the mean,
Pulmonary carcinoma bilateral, gaussian, and laplacian filters, shows that the bilateral filter is best
suited for CT scans of the body. To test the proposed technique on 30
examples of lung scans. The proposed algorithm is tested on 30 sample lung
images. The results show that the modified guaranteed convergence particle
swarm optimization algorithm has the highest accuracy of 96.01%.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Kalpana Govindaraju
Department of Computer Science, Faculty of Science and Humanities, SRM Institute of Science and Technology
Kattankulathur, Chennai-603203, India
Email: [email protected]

1. INTRODUCTION
Pulmonary carcinoma is one of the most common and deadliest types of cancer worldwide. Early
detection is crucial for successful treatment and improved patient outcomes. Medical imaging techniques,
such as computed tomography (CT) scans, are widely used for the diagnosis and monitoring of pulmonary
carcinoma. However, accurately detecting cancerous regions from medical images can be challenging due to
the complexity of lung structures and variations in image quality. In recent years, machine-learning
techniques, such as support vector machines (SVM), have been increasingly used in biomedical image
processing for the detection of pulmonary carcinoma. SVMs are a type of supervised learning algorithm that
has been widely used in various fields, including image processing, natural language processing, and
computer vision. SVMs are particularly useful in solving classification problems, which makes them suitable
for detecting cancerous regions from medical images [1]. The use of evolutionary algorithms in automated
systems for pulmonary carcinoma detection several benefits, including increased accuracy, reduced
inter-observer variability, and improved efficiency. These systems can also provide early detection of
pulmonary carcinoma, which can significantly improve patient outcomes [1]. The choice of noise removal
and filtering technique depends on the type and level of noise present in the image and the specific
application of the image. Moreover, the effectiveness of these techniques also depends on the quality of the

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  2913

image, the imaging modality used, and the parameters selected for the filter. In this paper, we propose an
automated system for pulmonary carcinoma detection using image segmentation by means of various
evolutionary algorithms. We compare the performance of different evolutionary algorithms, including
genetic algorithms, particle swarm optimization (PSO), differential evolution, and artificial bee colony
optimization, for pulmonary carcinoma detection. We also evaluate the performance of different
segmentation techniques, including thresholding, region-based segmentation, and edge-based segmentation,
with different evolutionary algorithms. The proposed system is evaluated using a dataset of CT images, and
the results are compared with the ground truth. The potential of evolutionary algorithms in improving the
accuracy and efficiency of pulmonary carcinoma detection is emphasized, along with areas for further
research and development [2].

2. RELATED WORK
K-means clustering is a technique used for analyzing CT images and localizing lung tumors. It
involves several steps such as preprocessing the images, extracting the region of interest (ROI), extracting
meaningful features, creating a feature matrix, determining the number of clusters, applying k-means
clustering, and localizing the tumors based on the centroids of the clusters. However, it is important to note
that the accuracy of this approach depends on various factors and combining multiple techniques may be
necessary for results that are more reliable [3].
The combination of k-means clustering and Cuckoo Search optimization is proposed as an effective
method for improving lung cancer segmentation in CT scans. K-means clustering is used to separate tumor
regions from healthy lung tissue, while Cuckoo Search optimizes the clustering solution to enhance
segmentation accuracy. The algorithm iteratively updates the positions of candidate solutions based on their
fitness, aiming to find the optimal clustering configuration. This combined approach leverages the strengths
of both methods and can be followed by post-processing techniques for further refinement. However, the
performance of the method may vary depending on implementation, parameters, and dataset characteristics,
necessitating proper validation and evaluation [4].
Foggy k-means, also known as fuzzy k-means, is a clustering algorithm used to cluster data points
with uncertain or ambiguous membership. When applying foggy k-means to lung cancer data, the general
steps involve data preprocessing, initializing cluster centroids, calculating membership degrees for each data
point, updating cluster centroids based on the membership degrees, repeating these steps until convergence,
assigning data points to clusters based on their highest membership degree, and interpreting and evaluating
the resulting clusters. The algorithm allows for data points to belong to multiple clusters with varying degrees
of membership. It is important to adapt the algorithm to the specific dataset and use appropriate evaluation
measures [5].
In the optimal reactive power dispatch (ORPD) problem, PSO is a commonly used algorithm. To
enhance its performance, a modified inertia weight control strategy can be employed. The strategy involves
adjusting the inertia weight during the optimization process to balance exploration and exploitation effectively.
The steps include initialization, fitness evaluation, global best update, particle movement, inertia weight update,
and termination. The modified strategy can be linearly decreasing or nonlinear, depending on the desired decay
rate. By using this approach, PSO can find near-optimal solutions for the ORPD problem. Experimentation and
parameter tuning may be necessary for optimal performance [6].
Convergence-guaranteed PSO methods for mobile robot global path planning aim to improve the
convergence properties of PSO by incorporating various techniques. These techniques include designing an
appropriate fitness function, handling constraints, using adaptive parameters, incorporating local search
strategies, preserving diversity within the swarm, and hybridizing with other algorithms. By combining these
techniques, convergence-guaranteed PSO methods enhance the optimization process, enabling the generation
of optimal paths while considering obstacles and constraints. It is important to experiment and select the most
effective approach based on the specific path planning problem [7].
The improved convergence particle swarm optimization (CPSO) algorithm with random sampling of
control parameters enhances the traditional CPSO algorithm by incorporating random sampling. This allows
for better exploration of the search space, improving the chances of finding optimal solutions and avoiding
local optima. The algorithm initializes a swarm of particles, updates their velocities and positions based on
control parameters, evaluates fitness, and performs random sampling of control parameters. It continues this
process until a termination criterion is met. The algorithm aims to converge faster and produce better
solutions compared to standard CPSO. However, experimentation and fine-tuning may be required for
optimal performance depending on the problem at hand [8].
In summary, to use PSO for generating stable structures of carbon clusters (Cn, where n=3–6, 10),
you would initialize a population of particles representing carbon cluster structures. Each particle’s position
represents the coordinates of carbon atoms within the cluster. The velocity and position of each particle are
Optimizing pulmonary carcinoma detection through image segmentation using … (Moulieswaran Elavarasu)
2914  ISSN: 2252-8938

updated based on its current position, best personal position, and best global position found by the swarm.
The potential energy or stability of each structure is evaluated using an objective function. Constraints can be
applied to ensure structural integrity. The algorithm iteratively searches for structures with low potential
energy, updating personal and global best positions along the way. The process continues until a termination
criterion is met, such as reaching a maximum number of iterations or achieving a desired level of stability [9].

3. METHOD
Accuracy is critical in medical image segmentation since it involves human lives. It is critical to
eliminate noise content and improve image quality before an evaluation [10]. The operational approach for
the current study is shown in Figure 1. The medical images from the Kaggle dataset which contains scans of
1,098 unique individuals, selected as the source for input CT images, these are grouped into 3 clusters consist
of 120 Benign cases, 562 Malignant cases, and 416 Normal cases.
Since it involves human lives, precision is of utmost importance in medical picture segmentation. Prior
to an inspection, it is imperative to reduce the occurrence of noise content and to enhance image quality.
Preprocessing is the term for this stage of the work. Noise removal and contrast enhancement are the two main
procedures in the preprocessing stage. The effectiveness of mean, bilateral, gaussian, and laplacian filters to
isolate the existence of speckle noise has been compared in the current work. Segmentation is the second phase
of the work. Applying three techniques-k-means, PSO and modified guaranteed convergence particle swarm
optimization (MGCPSO)-is what this stage entails. The segmentation results of the five procedures mentioned
above were split up, and the tumour component was retrieved and manually excised. The outcomes demonstrate
that the segmentation based on the MGCPSO is more accurate than the others.

Figure 1. Process flow diagram

3.1. Image acquisition

The process of collecting a picture from a source, often hardware equipment such as cameras, is
referred to as image acquisition. Images from intrinsically electronic devices such as CT and magnetic
resonance imaging (MRI) should be acquired via direct digital DICOM capture. Digital interfaces gather and
transport high-resolution image data from modality and grayscale. Meanwhile, analogue frame grabbers
convert the voltage output of a video signal, such as a scanning console display, into an image. The
frame-grabbing method, like printing an image to film, limits image quality to only 8 bits, whereas colour
data is collected at 12, 16, or even 32 bits. Capturing in 8 bits may limit the image’s ability to display all
clinical data, levels, contrast, and brightness settings. The initial stage in image processing techniques is to
read input images from a designated source for later processing. By smoothing the image, Gaussian noise and
spatial filtering are removed [11].

3.2. Pre-processing
Preprocessing the grayscale image to reduce noise is the first step. The gaussian filter, bilateral
filter, laplacian filter, and mean filter are a few examples of filtering methods. Table 1 represents the various
filtering techniques, description, advantages and its disadvantages [12]–[22].

3.3. K-Means clustering algorithm

The k-means clustering algorithm is the most basic and widely used method in cluster analysis. This
programme divides a dataset into two or more clusters. The accuracy of this procedure is entirely dependent
on the cluster centre chosen. To achieve the best results, the best cluster centre must be chosen. The general
measure used to separate the dataset is the Euclidean distance. The Euclidean distance is used to allocate
pixels to separate clusters. In this algorithm, the objective function is:

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2912-2922

Int J Artif Intell ISSN: 2252-8938  2915

𝐶 2
𝐽(𝑣)= ∑𝐶𝑖=1 ∑𝑗=1
𝑖
(‖𝑥𝑖 −𝑣𝑗 ‖) (1)

where xi represent pixels, vj represent cluster centres, ‖𝑥𝑖 −𝑣𝑗 ‖ is the Euclidean distance between xi and vj, Ci
represents the number of data points for the ith cluster, and C represents the number of cluster centres [23], [24].

Table 1. Various filtering techniques [12]–[22]

Algorithm Description Advantages Disadvantages
Gaussian Image blurring to minimise noise Removes Gaussian noise edges are blurred, which might not
effectively and retains edges be ideal for photos containing
better than the mean filter. non-Gaussian noise.
Bilateral Based on spatial and range information, it Removes noise while keeping Slower than other filters and
smooths the image while keeping the edges, with less smoothing than maybe ineffective for photographs
edges. the Gaussian filter. with intricate textures.
Laplacian Edge enhancement is achieved by Edges are preserved, thus edge Amplifies noise, may not be
subtracting the blurred picture from the detection jobs are possible. effective for low contrast images.
original image.
Mean Replaces each pixel with the mean value of Effectively reduces noise and Edges are blurred, therefore they
its neighbours. better maintains edges than the might not be ideal for high-
Gaussian filter frequency noisy photos.

3.4. Particle swarm optimization

PSO is a metaheuristic algorithm used extensively in medical image analysis. It mimics the social
behaviours of birds seeking food. The primary idea behind PSO is information sharing and communication.
In this method, each particle has an initial position and velocity. The fitness value is used to update the
velocity and position. The relevant PSO equations for updating location and velocity are [25]:

𝑣(𝑡+1)=𝑣 (𝑡) + 𝑐1 𝑟1 [𝑝𝑏𝑒𝑠𝑡(𝑡)−𝑥(𝑡)] + 𝑐2 𝑟2 [𝑔𝑏𝑒𝑠𝑡(𝑡)−𝑥(𝑡)],𝑥(𝑡+1)=𝑥(𝑡)+𝑣(𝑡+1) (2)

where r1 and r2 are the random integers and c1 and c2 are the acceleration coefficients. PSO's success is
dependent on the fitness function. The following fitness function was utilised in this study:

𝑖𝑛𝑡𝑒𝑟𝑐𝑙𝑢𝑠𝑡𝑒𝑟 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒
𝑚𝑎𝑥𝑖𝑚𝑖𝑧𝑒𝑓=∑𝑛𝑖=1 𝑖𝑛𝑡𝑟𝑎𝑐𝑙𝑢𝑠𝑡𝑒𝑟 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 (3)

3.5. Modified guaranteed convergence particle swarm optimization

The MGCPSO focuses on a new particle that deals with the region's current best location. This
particle is handled as a member of the swarm in this task, and the velocity update equation for this new
particle is as follows [26]:

𝑣𝜑(𝑡+1)=𝑥𝜑(𝑡)+𝑝𝑏𝑒𝑠𝑡(𝑡)+𝜔𝑣𝜑(𝑡)+𝜌(𝑡) (1−2𝑟) (4)

The social component improves the search ability. This will improve the random search in the area of the gbest
position.

4. RESULT AND DISCUSSION

The techniques are being practically applied in Python code with the aid of Jupyter and Google Colab,
and the outcomes are being validated. Several filters, including mean, bilateral, gaussian, and laplacian filters,
were compared in the initial step of preprocessing. Table 2 displays the streptomyces subtilisin inhibitor (SSI)
and streptomyces metallo-protease inhibitor (SMPI) values. The results show that the bilateral filter is more
accurate than the mean, bilateral, gaussian, and laplacian filters for segmenting medical images. By contrasting
the algorithm’s outputs with manual segmentation results, the true positive rate, true negative rate, false positive
rate, and false negative rate were used to gauge the accuracy of the segmentation. Figure 2 display the
preprocessed results of the images. In Table 3, numerical outcomes for k-means algorithm are displayed, Table
4 displays the numerical outcomes for PSO algorithm. Table 5 displays the numerical outcomes for MGCPSO
algorithm, Table 6 displays the statistical comparative result of accuracy, and Table 7 represents the
comparative evaluation of the predicted method’s accuracy.

Optimizing pulmonary carcinoma detection through image segmentation using … (Moulieswaran Elavarasu)
2916  ISSN: 2252-8938

Table 2. SMPI and SSI values of input images

Sample SMPI SSI
Images Mean Bilateral Gaussian Laplacian Mean Bilateral Gaussian Laplacian
Image 1 1.00005099 0.99883772 1.00015143 0.99983883 0.00653812 0.00207065 0.00548567 0.685815
Image 2 1.00006564 0.99885763 1.00016626 0.99982655 0.00656545 0.00202673 0.00550924 0.685577
Image 3 1.00006497 0.99880494 1.00013502 0.99980607 0.00649593 0.00200639 0.0054503 0.68777
Image 4 1.00007082 0.99880084 1.00014042 0.99979297 0.0065783 0.0020208 0.00552204 0.686029
Image 5 1.00008755 0.99886379 1.00013866 0.99975728 0.00584951 0.00150239 0.00490082 0.70714
Image 6 1.00008619 0. 9887512 1.00014111 0.99975457 0.00572866 0.00151213 0.00480674 0.708324
Image 7 1.00002471 0.99949843 0.99985296 0.99997888 0.00574208 0.00169107 0.00480342 0.719542
Image 8 0.99998011 0.99876035 0.99985778 0.99996715 0.00588469 0.00173544 0.00492531 0.714687
Image 9 1.00000285 0.99876677 0.99989671 0.99996982 0.00600199 0.00172164 0.00503817 0.710399
Image 10 1.00001908 0.99928420 1.00000917 0.99996443 0.00616713 0.00172735 0.0051611 0.702873
Image 11 1.00001004 0.99933914 0.99996487 0.99996091 0.00611228 0.00161426 0.00513139 0.702475
Image 12 1.00001499 0.99940113 1.00003242 0.99995779 0.00587849 0.00164259 0.00493284 0.706968
Image 13 1.00001268 0.99927493 1.00002306 0.99995558 0.00277204 0.00127928 0.00233801 0.776927
Image 14 1.00001508 0.99908803 1.00001072 0.99995312 0.00363379 0.00174865 0.00306998 0.739167
Image 15 1.00002226 0.99909637 1.00003674 0.99995350 0.00378776 0.00169811 0.00319446 0.734687
Image 16 1.00001885 0.99912012 1.00006004 0.99993713 0.00367232 0.00104465 0.00309586 0.728587
Image 17 1.00001577 0.99911254 1.00003237 0.99996591 0.00374972 0.00102567 0.00315888 0.727015
Image 18 1.00001296 0.99928709 1.00007336 0.99996315 0.00378583 0.00112909 0.0031973 0.723038
Image 19 1.00002629 0.99931720 1.00004326 0.99996596 0.00372441 0.00137282 0.00314496 0.72536
Image 20 0.99998131 0.99972701 0.99981695 0.99996780 0.00407759 0.00127076 0.00342971 0.714843
Image 21 1.00001090 0.99846019 0.99996623 0.99997494 0.00394663 0.00128628 0.00333205 0.717366
Image 22 1.00001165 0.99846254 1.00002183 0.99997072 0.00397981 0.00134342 0.00335919 0.716257
Image 23 1.00002200 0.99843795 1.00002915 0.99997714 0.00395729 0.00122709 0.00333939 0.721084
Image 24 1.00000463 0.99841464 1.00002746 0.99997623 0.00351772 0.00099441 0.00297505 0.731775
Image 25 1.00002220 0.99858012 1.00001403 0.99997446 0.00353507 0.00103015 0.00298315 0.736536
Image 26 1.00000053 0.99857502 1.00001434 0.99997606 0.00145602 0.00093932 0.00322325 0.828436
Image 27 1.00000855 0.99855847 1.00000520 0.99997855 0.00606584 0.00158877 0.00510341 0.677864
Image 28 1.00000874 0.99854366 0.99997554 0.99998330 0.00639493 0.00177743 0.00538764 0.669129
Image 29 1.00009861 0.99886061 1.00013501 0.99973294 0.00646719 0.00183678 0.00545048 0.665425
Image 30 1.00008776 0.99886256 1.000182787 0.99973633 0.0065535 0.00185939 0.00552473 0.662408

Figure 2. After preprocessing

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2912-2922

Int J Artif Intell ISSN: 2252-8938  2917

Table 3. Numerical outcomes for k-means algorithm

Images True positive rate True negative rate False positive rate False negative rate Accuracy
Image 1 83.0198 88.4375 16.9802 11.5625 85.7287
Image 2 86.0726 81.3622 13.9274 18.6378 83.7174
Image 3 78.4182 89.1677 21.5818 10.8323 83.793
Image 4 85.3457 82.7848 14.6543 17.2152 84.0653
Image 5 78.4954 78.8414 21.5046 21.1586 78.6684
Image 6 85.7631 85.6644 14.2369 14.3356 85.7138
Image 7 87.8826 77.4612 12.1174 22.5388 82.6719
Image 8 85.4438 89.7083 14.5562 10.2917 87.5761
Image 9 83.6784 83.5609 16.3216 16.4391 83.6197
Image 10 76.5976 80.3489 23.4024 19.6511 78.4733
Image 11 79.4071 80.2404 20.5929 19.7596 79.8238
Image 12 77.1947 87.7569 22.8053 12.2431 82.4758
Image 13 81.3153 79.0573 18.6847 20.9427 80.1863
Image 14 82.2217 77.6233 17.7783 22.3767 79.9225
Image 15 87.8811 87.4071 12.1189 12.5929 87.6441
Image 16 76.4974 89.9263 23.5026 10.0737 83.2119
Image 17 82.4508 77.1854 17.5492 22.8146 79.8181
Image 18 86.9717 88.1683 13.0283 11.8317 87.57
Image 19 80.0838 89.8913 19.9162 10.1087 84.9876
Image 20 79.6117 89.3348 20.3883 10.6652 84.4733
Image 21 82.2061 85.9213 17.7939 14.0787 84.0637
Image 22 84.5572 87.1664 15.4428 12.8336 85.8618
Image 23 79.2736 81.4655 20.7264 18.5345 80.3696
Image 24 79.0333 89.4537 20.9667 10.5463 84.2435
Image 25 82.7447 88.289 17.2553 11.711 85.5169
Image 26 87.3402 87.9803 12.6598 12.0197 87.6603
Image 27 78.4169 88.049 21.5831 11.951 83.233
Image 28 81.1904 85.3374 18.8096 14.6626 83.2639
Image 29 83.385 83.557 16.615 16.443 83.471
Image 30 81.926 86.5422 18.074 13.4578 84.2341

Table 4. Numerical outcomes for PSO algorithm

Images True positive rate True negative rate False positive rate False negative rate Accuracy
Image 1 78.3473 83.7745 21.6527 17.4683 81.0609
Image 2 87.7165 82.5317 12.2835 11.8226 85.1241
Image 3 85.0788 88.1774 14.9212 12.504 86.6281
Image 4 80.7158 87.496 19.2842 18.8781 84.1059
Image 5 77.2816 81.1219 22.7184 15.186 79.20175
Image 6 85.578 84.814 14.422 15.1669 85.196
Image 7 77.5807 84.8331 22.4193 17.5781 81.2069
Image 8 81.6709 82.4219 18.3291 14.9456 82.0464
Image 9 79.424 85.0544 20.576 12.0061 82.2392
Image 10 81.1218 87.9939 18.8782 17.1153 84.55785
Image 11 83.3637 82.8847 16.6363 12.8556 83.1242
Image 12 83.5755 87.1444 16.4245 10.3926 85.35995
Image 13 88.0024 89.6074 11.9976 12.4266 88.8049
Image 14 88.1423 87.5734 11.8577 16.7596 87.85785
Image 15 77.2572 83.2404 22.7428 18.2952 80.2488
Image 16 88.4797 81.7048 11.5203 13.7834 85.09225
Image 17 83.8806 86.2166 16.1194 19.5936 85.0486
Image 18 80.8995 80.4064 19.1005 16.3412 80.65295
Image 19 81.5196 83.6588 18.4804 18.2952 82.5892
Image 20 78.789 86.0371 21.211 13.9629 82.41305
Image 21 86.727 80.0248 13.273 19.9752 83.3759
Image 22 84.945 81.7717 15.055 18.2283 83.35835
Image 23 87.1483 82.4668 12.8517 17.5332 84.80755
Image 24 79.8254 80.315 20.1746 19.685 80.0702
Image 25 81.0092 83.7164 18.9908 16.2836 82.3628
Image 26 77.3288 87.0294 22.6712 12.9706 82.1791
Image 27 79.8118 84.8031 20.1882 15.1969 82.30745
Image 28 81.0954 89.095 18.9046 10.905 85.0952
Image 29 84.5016 79.0836 15.4984 20.9164 81.7926
Image 30 85.4518 80.8443 14.5482 19.1557 83.14805

Lung cancer detection has been done in prior studies, 89.5% accuracy was achieved utilising PSO,
genetic optimisation, the SVM algorithm, and the Gabor filter [27]. A maximum accuracy of 90% was achieved
utilising the genetic algorithm and k-nearest neighbor (K-NN) classification to identify lung cancer [28].
Figure 3 displays the relative outcomes of the true positive rate value, Figure 4 displays the relative outcomes of
Optimizing pulmonary carcinoma detection through image segmentation using … (Moulieswaran Elavarasu)
2918  ISSN: 2252-8938

the true negative rate value, Figure 5 displays the relative outcomes of the false positive rate value, Figure 6
displays the relative outcomes of the false negative rate value. Figure 7 represents the outcomes of accuracy,
Figure 8 display the accuracy.

Table 5. Numerical outcomes for MGCPSO algorithm

Images True positive rate True negative rate False positive rate False negative rate Accuracy
Image 1 85.5901 97.894 14.4099 2.106 91.74205
Image 2 92.1561 97.4464 7.8439 2.5536 94.80125
Image 3 85.9277 95.4679 14.0723 4.5321 90.6978
Image 4 88.0733 97.7006 11.9267 2.2994 92.88695
Image 5 88.036 98.8191 11.964 1.1809 93.42755
Image 6 88.9129 97.8826 11.0871 2.1174 93.39775
Image 7 89.5582 97.1109 10.4418 2.8891 93.33455
Image 8 87.7662 95.9419 12.2338 4.0581 91.85405
Image 9 90.925 97.5847 9.075 2.4153 94.25485
Image 10 94.3613 97.6492 5.6387 2.3508 96.00525
Image 11 91.4426 98.4483 8.5574 1.5517 94.94545
Image 12 88.2684 97.1438 11.7316 2.8562 92.7061
Image 13 92.4882 98.5891 7.5118 1.4109 95.53865
Image 14 85.1011 98.428 14.8989 1.572 91.76455
Image 15 88.4775 98.4377 11.5225 1.5623 93.4576
Image 16 85.8022 95.6501 14.1978 4.3499 90.72615
Image 17 85.6168 98.8983 14.3832 1.1017 92.25755
Image 18 84.5809 95.9124 15.4191 4.0876 90.24665
Image 19 84.935 96.9602 15.065 3.0398 90.9476
Image 20 91.0548 98.9336 8.9452 1.0664 94.9942
Image 21 84.6432 95.7747 15.3568 4.2253 90.20895
Image 22 87.7662 96.7971 12.2338 3.2029 92.28165
Image 23 87.7271 98.4374 12.2729 1.5626 93.08225
Image 24 85.9182 96.7405 14.0818 3.2595 91.32935
Image 25 85.2057 95.9214 14.7943 4.0786 90.56355
Image 26 90.1859 97.8667 9.8141 2.1333 94.0263
Image 27 87.2679 97.837 12.7321 2.163 92.55245
Image 28 84.3414 96.8343 15.6586 3.1657 90.58785
Image 29 89.5071 96.0861 10.4929 3.9139 92.7966
Image 30 90.8882 95.9334 9.1118 4.0666 93.4108

Table 6. Statistical comparative result of accuracy

Images K-means PSO MGCPSO
Image 1 85.7287 81.0609 91.74205
Image 2 83.7174 85.1241 94.80125
Image 3 83.793 86.6281 90.6978
Image 4 84.0653 84.1059 92.88695
Image 5 78.6684 79.20175 93.42755
Image 6 85.7138 85.196 93.39775
Image 7 82.6719 81.2069 93.33455
Image 8 87.5761 82.0464 91.85405
Image 9 83.6197 82.2392 94.25485
Image 10 78.4733 84.55785 96.00525
Image 11 79.8238 83.1242 94.94545
Image 12 82.4758 85.35995 92.7061
Image 13 80.1863 88.8049 95.53865
Image 14 79.9225 87.85785 91.76455
Image 15 87.6441 80.2488 93.4576
Image 16 83.2119 85.09225 90.72615
Image 17 79.8181 85.0486 92.25755
Image 18 87.57 80.65295 90.24665
Image 19 84.9876 82.5892 90.9476
Image 20 84.4733 82.41305 94.9942
Image 21 84.0637 83.3759 90.20895
Image 22 85.8618 83.35835 92.28165
Image 23 80.3696 84.80755 93.08225
Image 24 84.2435 80.0702 91.32935
Image 25 85.5169 82.3628 90.56355
Image 26 87.6603 82.1791 94.0263
Image 27 83.233 82.30745 92.55245
Image 28 83.2639 85.0952 90.58785
Image 29 83.471 81.7926 92.7966
Image 30 84.2341 83.14805 93.4108

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2912-2922

Int J Artif Intell ISSN: 2252-8938  2919

Table 7. Comparative evaluation of the predicted method’s accuracy

Numerous techniques Accuracy
PSO, Genetic Algorithm, and SVM algorithm [27] 89.5%
K-NN classification using genetic algorithm [28] 90%
MGCPSO 96.01%

Figure 3. Relative outcomes of the true positive rate value

Figure 4. Relative outcomes of the true negative rate value

Figure 5. Relative outcomes of the false positive rate value

Optimizing pulmonary carcinoma detection through image segmentation using … (Moulieswaran Elavarasu)
2920  ISSN: 2252-8938

Figure 6. Relative outcomes of the false negative rate value

Figure 7. Outcomes of accuracy

Figure 8. Accuracy

5. CONCLUSION
Optimizing pulmonary carcinoma detection through image segmentation using evolutionary
algorithms is a promising approach to improve the accuracy of diagnosis from medical images. By leveraging
the power of evolutionary algorithms, the segmentation parameters of the image segmentation algorithm can
be optimized to achieve better results. The process involves several key steps, including dataset preparation,
preprocessing, image segmentation, fitness function design, evolutionary algorithm design, chromosome

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2912-2922

Int J Artif Intell ISSN: 2252-8938  2921

encoding, initialization, fitness evaluation, selection, variation operators, fitness update, replacement,
termination criteria, and evaluation. These steps work together iteratively to refine the image segmentation
algorithm and enhance its ability to detect and differentiate pulmonary carcinoma regions from healthy lung
tissue. The use of evolutionary algorithms allows for the exploration of a vast parameter space, gradually
converging towards better solutions. By evaluating the fitness of each chromosome based on the accuracy of
the segmentation results, the algorithm can guide the optimization process towards more accurate and reliable
segmentation. It is important to note that the success of this approach relies on various factors, such as the
quality and representativeness of the training dataset, the design of the fitness function, and the selection of
appropriate evolutionary algorithm parameters. Additionally, the optimized image segmentation algorithm
should be evaluated on separate validation or testing images to ensure it s generalizability and effectiveness.
Overall, by leveraging evolutionary algorithms to optimize the image segmentation process, the accuracy and
efficiency of pulmonary carcinoma detection can be significantly improved, thereby aiding in early diagnosis
and improving patient outcomes. This approach holds great potential for advancing the field of medical
imaging and facilitating more precise and timely detection of pulmonary carcinoma.

REFERENCES
[1] M. Kurkure and A. Thakare, “Introducing automated system for lung cancer detection using evolutionary approach,”
International Journal of Engineering and Computer Science, vol. 5, no. 5, pp. 16736-16739, 2016, doi: 10.18535/ijecs/v5i5.69.
[2] N. Kumar and M. Nachamai, “Noise removal and filtering techniques used in medical images,” Oriental Journal of Computer
Science and Technology, vol. 10, no. 1, pp. 103–113, 2017, doi: 10.13005/ojcst/10.01.14.
[3] A. C. G. and R. M. S, “Analysis of CT images and localization of lung tumor using k-means clustering,” Advances in
Computational Sciences and Technology, vol. 12, no. 1, pp. 45–51, 2019.
[4] P. Singh, P. Nanglia, and D. A. N. Mahajan, “Improved lung cancer segmentation using k-means and cuckoo search,”
International Journal of Innovative Technology and Exploring Engineering, vol. 9, no. 2, pp. 3746–3757, 2019, doi:
10.35940/ijitee.b6221.129219.
[5] A. K. Yadav, D. Tomar, and S. Agarwal, “Clustering of lung cancer data using foggy k-means,” 2013 International Conference
on Recent Trends in Information Technology, ICRTIT 2013, pp. 13–18, 2013, doi: 10.1109/ICRTIT.2013.6844173.
[6] M. Sabir, A. Ahmad, A. Ahmed, S. Siddique, and U. A. Hashmi, “A modified inertia weight control of particle swarm
optimization for optimal reactive power dispatch problem,” 2021 International Conference on Emerging Power Technologies,
ICEPT 2021, 2021, doi: 10.1109/ICEPT51706.2021.9435588.
[7] B. Tang, Z. Zhanxia, and J. Luo, “A convergence-guaranteed particle swarm optimization method for mobile robot global path
planning,” Assembly Automation, vol. 37, no. 1, pp. 114–129, 2017, doi: 10.1108/AA-03-2016-024.
[8] G. Jana, A. Mitra, S. Pan, S. Sural, and P. K. Chattaraj, “Modified particle swarm optimization algorithms for the generation of
stable structures of carbon clusters, Cn (n=3–6, 10),” Frontiers in Chemistry, vol. 7, 2019, doi: 10.3389/fchem.2019.00485.
[9] L. Sun, X. Song, and T. Chen, “An improved convergence particle swarm optimization algorithm with random sampling of
control parameters,” Journal of Control Science and Engineering, vol. 2019, 2019, doi: 10.1155/2019/7478498.
[10] A. G. N. Panpaliya, N. Tadas, S. Bobade, R. Aglawe, “A survey on early detection and prediction of lung cancer,” International
Journal of Computer Science and Mobile Computing, vol. 4, no. 1, pp. 175–184, 2015.
[11] K. J. Dreyer, J. H. Thrall, D. S. Hirschorn, and A. Mehta, “PACS: A guide to the digital revolution: second edition,” PACS: A
Guide to the Digital Revolution: Second Edition, pp. 1–579, 2006, doi: 10.1007/0-387-31070-3.
[12] P. P. Shetti and A. P. Patil, “Performance comparison of mean, median and wiener filter in MRI image de-noising,” International
Journal for Research Trends and Innovation, vol. 2, no. 6, 2017.
[13] J. Joseph and R. Periyasamy, “An image driven bilateral filter with adaptive range and spatial parameters for denoising magnetic
resonance images,” Computers and Electrical Engineering, vol. 69, pp. 782–795, 2018, doi: 10.1016/j.compeleceng.2018.02.033.
[14] D. Devasena, K. Srinivasan, B. Sharmila, and A. Booja, “Despeckling algorithms for removing noise in medical images,” Turkish
Journal of Computer and Mathematics Education (TURCOMAT), vol. 12, no. 6, pp. 87–94, 2021, doi:
10.17762/turcomat.v12i6.1271.
[15] G. Deng and L. W. Cahill, “Adaptive Gaussian filter for noise reduction and edge detection,” IEEE Nuclear Science Symposium
& Medical Imaging Conference, no. pt 3, pp. 1615–1619, 1994, doi: 10.1109/nssmic.1993.373563.
[16] A. Singh, “Comparative analysis of Gaussian filter with wavelet denoising for various noises present in images,” Indian Journal
of Science and Technology, vol. 9, no. 1, pp. 1–8, 2016, doi: 10.17485/ijst/2016/v9i47/106843.
[17] T. Barbu, “Gabor filter-based face recognition technique,” Proceedings of the Romanian Academy Series A - Mathematics Physics
Technical Sciences Information Science, vol. 11, no. 3, pp. 277–283, 2010.
[18] A. Kafuo, S. Diaf, A. Gonifeda, and A. Baba, “A literature survey of Gabor filter and its application,” Technical Report, 2017,
doi: 10.13140/RG.2.2.11079.50085.
[19] S. Paris, S. W. Hasinoff, and J. Kautz, “Local Laplacian filters: edge-aware image processing with a Laplacian pyramid,”
Communications of the ACM, vol. 58, no. 3, pp. 81–91, 2015, doi: 10.1145/2723694.
[20] M. Aubry, S. Paris, S. W. Hasinoff, J. Kautz, and F. Durand, “Fast local laplacian filters: theory and applications,” ACM
Transactions on Graphics, vol. 33, no. 5, 2014, doi: 10.1145/2629645.
[21] Y. Zhu and C. Huang, “An improved median filtering algorithm for image noise reduction,” Physics Procedia, vol. 25,
pp. 609–616, 2012, doi: 10.1016/j.phpro.2012.03.133.
[22] T. M. Sheeba and S. A. A. Raj, “Analysis of noise removal techniques on retinal optical coherence tomography images,”
International Journal of Advanced Computer Science and Applications, vol. 13, no. 9, pp. 422–427, 2022, doi:
10.14569/IJACSA.2022.0130948.
[23] K. Venkatalakshmi and S. Mercy Shalinie, “Classification of multispectral images using support vector machines based on PSO
and K-means clustering,” Proceedings-2005 International Conference on Intelligent Sensing and Information Processing,
ICISIP’05, vol. 2005, pp. 127–133, 2005, doi: 10.1109/ICISIP.2005.1529435.
[24] K. Venkatalakshmi and S. M. Shalinie, “Multispectral image classification using modified k-means algorithm,” Neural Network
World, vol. 17, no. 2, pp. 113–120, 2007.

Optimizing pulmonary carcinoma detection through image segmentation using … (Moulieswaran Elavarasu)
2922  ISSN: 2252-8938

[25] H. Gao, C. M. Pun, and S. Kwong, “An efficient image segmentation method based on a hybrid particle swarm algorithm with
learning strategy,” Information Sciences, vol. 369, pp. 500–521, 2016, doi: 10.1016/j.ins.2016.07.017.
[26] P. K. Patel, V. Sharma, and K. Gupta, “Guaranteed convergence particle swarm optimization using personal best,” International
Journal of Computer Applications, vol. 73, no. 7, pp. 6–10, 2013, doi: 10.5120/12751-9694.
[27] A. Asuntha, N. Singh, A. Srinivasan, and A. Professor, “PSO, genetic optimization and SVM algorithm used for lung cancer
detection,” Available online www.jocpr.com Journal of Chemical and Pharmaceutical Research, vol. 8, no. 6, pp. 351–359, 2016.
[28] P. Bhuvaneswari and A. B. Therese, “Detection of cancer in lung with K-NN classification using genetic algorithm,” Procedia
Materials Science, vol. 10, pp. 433–440, 2015, doi: 10.1016/j.mspro.2015.06.077.

BIOGRAPHIES OF AUTHORS

Moulieswaran Elavarasu is pursuing Ph.D. in Computer Science from Faculty of

Science and Humanities, SRM Institute of Science and Technology, Chennai. He completed
his B.Sc. Computer Science from Periyar University, Salem, Completed his M.C.A., from
Anna University, Coimbatore and completed his M.Phil. Computer Science in Periyar
University, Salem. Previously he had 8.5 years of teaching experience at M.G.R. College,
Hosur. His research work mainly focuses on machine learning techniques in detection of
pulmonary carcinoma. He can be contacted at email: [email protected].

Kalpana Govindaraju has completed her Ph.D. from SRM Institute of Science
and Technology. Her Current research focuses on parallel and distributed computing, block
chain technology, machine learning, and cloud security. Currently, she is working as a
Professor and Head in the Department of Computer Science, Faculty of Science and
Humanities, SRM Institute of Science and Technology, Chennai. She has published more than
15 papers in journals and conferences. She has also filed an Indian patent in the year 2021. She
can be contacted at email: [email protected].

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2912-2922