IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 12, No. 4, December 2023, pp. 1864~1872

ISSN: 2252-8938, DOI: 10.11591/ijai.v12.i4.pp1864-1872  1864

Development of hybrid convolutional neural network and

autoregressive integrated moving average on computed
tomography image classification

Abdulrazak Yahya Saleh1, Chee Ka Chin2

1
Faculty of Cognitive Sciences and Human Development, Universiti Malaysia Sarawak (UNIMAS), Sarawak, Malaysia
2
Department of Electrical and Electronic, Faculty of Engineering, Universiti Malaysia Sarawak, Sarawak, Malaysia

Article Info ABSTRACT

Article history: One of the deadliest diseases in humans is lung cancer. Radiologists and
experienced doctors spend much more time investigating the pulmonary
Received Nov 14, 2022 nodules due to the high similarities between malignant and benign nodules.
Revised Jan 27, 2023 Recently, the computer-assisted diagnosis (CAD) tool for nodule detection
Accepted Mar 10, 2023 can provide a second opinion for the doctor to diagnose lung cancer. Although
machine learning technologies are extensively employed to identify lung
cancer, the process of these methods is complex. The numerous researches
Keywords: have sought to automate the diagnosis of pulmonary nodules using
convolutional neural networks (CNN) to aid radiologists in the lung screening
Artificial intelligence process. However, CNN still confronts some challenges, including a
Classification significant number of false positives and limited performance in detecting
Computed tomography image lung cancer from computed tomography (CT) images. In this work, we
Hybrid convolutional neural proposed a hybrid of CNN and auto-regressive integrated moving average
network and autoregressive (ARIMA) for lung nodule classification using CT images to address the
integrated moving average classification issue. The proposed hybrid CNN-ARIMA can classify CT
Lung cancer images successfully with test accuracy, average sensitivity, average precision,
average specificity, average F1-Score, and area under the curve (AUC) of
99.61%, 99.71%, 99.43%, 99.71%, 99.57%, and 1.000, respectively.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Chee Ka Chin
Department of Electrical and Electronic, Faculty of Engineering, Universiti Malaysia Sarawak
94300 Kota Samarahan, Sarawak, Malaysia
Email: [email protected]

1. INTRODUCTION
The most common cause of mortality worldwide is lung cancer. In comparison to other cancers, lung
cancer is responsible for roughly 1.8 million fatalities (18.0% of all cancer deaths) in 2020 [1]. If lung cancer
is detected earlier, it can be treated effectively and even cured. The pulmonary nodules are the abnormal
development of lung tissue and are generally the first sign of lung cancer. As a result, it is critical to detect and
diagnose lung nodules earlier to reduce lung cancer mortality [2]. Radiologists utilize a variety of diagnostic
techniques to detect lung cancer, including magnetic resonance imaging (MRI), computed tomography (CT),
chest X-ray, and other methods. However, a CT scan is the most reliable screening approach due to its low
noise and robustness in determining tumor size [3]. With the widespread use of CT scanning technology, the
need for medical image analysis has risen and causes an increased workload for radiologists [4]. Traditionally,
achieving an accurate radiologic diagnosis requires the expertise of the radiologist throughout the analysing
process. The emergence of computer-aided tools has relieved the burden on radiologists and offered a second
opinion for doctors to diagnose lung cancer [5]. In traditional methods, there exist various techniques for feature

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

Int J Artif Intell ISSN: 2252-8938  1865

extraction and classification in the diagnosis of lung cancer. Among feature extraction techniques are the
threshold method, the region growth method, the morphology approach, and so on [6]. Machine learning
technologies, such as support vector machines (SVM) and k-nearest neighbor, are extensively employed to
distinguish the false-positive nodules [6]. However, the process of these traditional methods is complex, which
leads to poor generalization. This issue is due to the pulmonary nodule features needing to extract manually,
which causes the automation degree to be limited [7]. Motivated by the lung nodule analysis 16 (LUNA16)
challenge [8], numerous research has sought to automate the diagnosis of pulmonary nodules using deep
learning, particularly convolutional neural networks (CNN), to aid radiologists in the lung screening process
[9]. These techniques have substantially enhanced the treatment’s quality and dependability, particularly the
early detection of lung cancer. While lung cancer is slow and unreliable in survival rates compared to other
cancers, CNN techniques show promise performance. The computer-aided diagnosis (CAD) systems that
focused on CNN strategies can reduce lung cancer death rates by a factor of 25% in the last five years [10].
Despite the advancements in using CNN for lung cancer detection, there are still several issues that need to be
addressed. These include a high false-positive rate that limits its clinical application and the challenge of
surpassing the performance of conventional CAD systems in detecting lung cancer from CT images [11].
Moreover, the approach of adding more training data or fine-tuning hyperparameters to improve CNN’s
performance faces three main challenges. Firstly, there is observer variability among radiologists, which can
significantly affect the interpretation of the images. Secondly, detection networks may miss nodules that are
under-represented in the training set. Finally, neural networks are susceptible to unanticipated image distortions
[9].
To address the challenges of lung nodule classification, we propose a hybrid approach that combines
CNN and auto-regressive integrated moving average (ARIMA) for lung nodule classification using CT scan
images. This approach was motivated by previous research that used the hybrid CNN-ARIMA model for skin
cancer classification, which demonstrated promising results [12], [13]. In this approach, the CNN algorithm
serves as a feature extractor, while the ARIMA model functions as a classifier. The aim of this study is to
assess the performance of the hybrid CNN-ARIMA model for lung cancer classification.

2. RELATED WORKS
Several deep learning approaches had effectively applied in medical domains to address classification
problems [12]–[15]. A recent study showed that deep learning methods outperform conventional CAD systems
in detecting lung cancer with CT scans [11]. As a result, deep learning algorithms become beneficial for nodule
detection and classification, with very intriguing results.
Haiying et al. [16] proposed a new deformable convolutional neural network (DCNN) architecture to
address the problem of the similarity between true and false-positive nodules in early morphology in China.
The efficacy of the DCNN model was evaluated on the Lung Nodule Analysis 2016 dataset, achieving an
average competitive performance metric score of 0.835 and a high sensitivity of 0.941 and 0.958 to 4 and 8
false positives per scan, respectively. Sheng et al. proposed a method for detecting and diagnosing lung nodules
using two networks: Prediction network (PredNet) and diagnosis network (DiagNet). The PredNet uses
convolutional networks to predict the growth of nodules between consecutive CT scans through spatial
transformation. The DiagNet categorizes nodules based on their growth and previous diagnosis. The model
was tested on a dataset of 615 low-dose computed tomography (LDCT) images of 153 early-stage lung nodules
from 125 individuals. The proposed method demonstrated promising performance with a classification
accuracy of 90%. Many researchers have also shown interest in hybridization techniques due to their
outstanding performance in addressing cancer problems by enhancing algorithm performance in recent times
[10], [12], [13], [17]–[19]. Agarwal et al. [10] proposed a method to classify lung cancers as either malignant
or benign by combining a CNN with the AlexNet network model whereas Ardimento et al. [17] suggested a
novel ensemble-based method for classifying lung cancer more accurately using CT scan images.
Saleh et al. [20] developed an approach for classifying CT images of the lung using a hybrid of CNN and SVM
algorithms. The goal was to effectively detect the presence of cancer cells. The results showed that the proposed
CNN-SVM approach achieved an accuracy of 97.91% and had the ability to accurately classify lung cancer in
CT scans.

3. METHODOLOGY
This section discusses the study's methodology. The proposed approach in this study does not require
any prior techniques because the CNN can classify lesions based on high-level features without the need
for low-level nodule visual information and segmentation steps as required in traditional methods [10], [19].
Figure 1 depicts the image classification algorithm used in this study to detect lung cancer.

Development of hybrid convolutional neural network and autoregressive …(Abdulrazak Yahya Saleh)
1866  ISSN: 2252-8938

3.1. Dataset
In this study, the doctor from Sarawak General Hospital provided the dataset for the experiment. These
datasets contain 1,286 digital imaging and communications in medicine (DICOM) images divided between
normal lung and lung cancer images. There were 890 images of lung cancer and 396 images of normal lung.
These DICOM-format datasets need to convert to JPEG format for deep learning applications. Furthermore,
all images have a resolution of 512×512 pixels. Figure 2(a) shows the examples of normal lung. Meanwhile,
Figure 2(b) illustrates the lung cancer images and the red circle indicates the nodules.

3.2. Image pre-processing

Most of the collected images are not clear. Therefore, the pre-processing process needs to perform.
To enhance image clarity, the contrast limited adaptive histogram equalization (CLAHE) method was used via
OpenCV as the pre-processing technique [20]. CLAHE is also often used to improve the intensity of the lung
image by adding smoothing filters [21]. The final pre-processed images will generate after the CLAHE method.
Figure 3 summarizes the process of the CLAHE technique on lung cancer images.

Figure 1. The proposed lung cancer diagnostic CAD system

(a) (b)

Figure 2. Type of lungs: (a) Normal lung and (b) Lung cancer – red circle indicates the nodules

Int J Artif Intell, Vol. 12, No. 4, December 2023: 1864-1872

Int J Artif Intell ISSN: 2252-8938  1867

Figure 3. CLAHE using OpenCV framework

3.3. Image Augmentation

An augmentation technique used in this study is an affine transformation [22]. This technique aims to
avoid overfitting issues and improve neural network accuracy in this study. This process can map an object
from one affine space to another while preserving its structures [23]. In this study, various affine transformation
techniques were used, such as width shift, height shift, zoom, stretch, and rotation. Table 1 provides a detailed
overview of the chosen parameters for data augmentation through affine transformation on the skin cancer
dataset.

Table 1. Parameters and values of selected affine transformation techniques

Parameters Value of Parameter Action
width_shift_range 0.1 Shifts the image size horizontally by 10% at random.
height_shift_range 0.1 Shifts the image size vertically by 10% at random.
zoom_range 0.2 Zoom out by 0.2 from the middle.
shear_range 10 Extend the image by a factor of 10.
rotation_range 10 Rotate from -10 to 10 degrees.

3.4. CNN-ARIMA
The aim of this study was to combine the CNN and ARIMA algorithms to extract image features and
classify them. The CNN algorithm was used as a feature extractor, while the ARIMA algorithm acted as a
classifier by being integrated into a fully connected layer, which transformed the extracted features non-
linearly. To output two classes (normal lung and lung cancer), the traditional CNN algorithm was modified by
placing the ARIMA model between the flatten and dense layers of the fully connected layer. This hybridization
also evaluated the similarity between images of different categories by calculating the performance evaluation
for all classes when the proposed CNN-ARIMA model predicted a new image. Based on the approach
presented in [13], the decision-making process of the CNN-ARIMA model proposed in this study is achieved
through the implementation of a modified ARIMA algorithm in the fully connected layer.

𝐴𝑅𝐼𝑀𝐴 (9, (𝑝, 𝑑, 𝑞)) (1)

The value 9 was used as the average out-shape value in the fully connected layer, which showed the
best performance in [13]. In addition, 𝑝, 𝑑, and 𝑞 are the auto-regressive, differencing, and moving average
values, respectively. However, the performance in this experiment was not affected by the specific values
chosen for 𝑝, 𝑑, and 𝑞. Therefore, these 𝑝, 𝑑, and 𝑞 set to a value of 1. Figure 4 illustrates the architecture of
CNN-ARIMA.
The datasets are resized to a 32×32 shape and converted into feature maps, which are arrays of every
pixel in the images, before being pre-processed and augmented for use in the model. The CT image, represented
as a 2D array, is then input into three components, each consisting of two convolutional layers with 64, 32, and
16 filters, respectively. After each convolutional layer in the feature extractor, a rectified linear unit (ReLU) is
applied as an activation function, and dropout with a ratio of 0.5 is utilised.
In the classification stage, the output obtained from the last dropout layer in the feature extraction stage
is flattened into a single array. Then, the ARIMA layer is used to improve the classification accuracy. A dense
layer with 512 units and a dropout ratio of 0.5 is added, followed by a final layer with two nodes for classifying
the CT images into two categories: lung cancer and normal lung. The SoftMax activation function is used in the
last three layers for classification.
Development of hybrid convolutional neural network and autoregressive …(Abdulrazak Yahya Saleh)
1868  ISSN: 2252-8938

Figure 4. CNN-ARIMA architecture

3.5. Parameters and considerations

The model’s accuracy was evaluated on a test set comprising 20% of the overall dataset, while the
remaining data was split randomly into training and validation sets in a ratio of 80:20 to prevent overfitting. A
learning rate of 0.0001 was used to assess overall accuracy performance, and other key parameters such as
batch size and epoch values were carefully chosen for model training. A batch size employed in this research
is 64 to perform model training because this batch size could result in faster training progress [24]. Furthermore,
epochs values used in this work are 30 for model training. If the value of the epoch is too large, the training
process takes too long [25]. The model training was also carried out with 2,000 steps per epoch since this
parameter value could provide an excellent classification performance [26]. The optimizer used in this study
is Adaptive Moment Estimation (Adam) for neural work optimization. Adam is considered a famous optimizer
in the deep learning field due to its ability to perform faster and more reliably to reach a global minimum of
the objective function during the optimization [27]. Adam was chosen as the optimizer in this study because it
is highly suited to the classification problems, which involve vast amounts of data. Furthermore, this optimizer
also has high computational efficiency with a low memory demand [25]. Therefore, this optimizer can reduce
the loss in the model training.

3.6. Implementation requirements

In this work, Keras (2.3.1) [28] and TensorFlow (2.2.0) [29] are used to train the model. Keras is a
simple, flexible, and robust framework for deep learning applications in Python language. This framework is
also on top of TensorFlow, which provides an abstraction layer to give the configuration for the neural network
for building the deep learning model. The Python environment allows the user to create a deep learning model
using external Python libraries. Additionally, the model is trained using an Intel Core i5 10th Gen processor
with 16 GB of memory and NVIDIA GeForce RTX 2060 GPU support with 6 GB of memory. This graphics
processing unit (GPU) can train the model faster and more effectively.

3.7. Result interpretation and evaluation

The analysis performance of the CNN-ARIMA model on CT image will provide the test accuracy and
then compute the confusion matrix, performance metrics, receiver operating characteristic (ROC) curve, and
prediction performance to evaluate the classification performance in this work. The confusion matrix is a tool
used to evaluate the performance metrics such as sensitivity, precision, specificity, and F1-score. It is a table
that reports the number of true positive (TP), false positive (FP), true negative (TN), and false negative (FN)
predictions. These four categories can be used to calculate four performance metrics, namely sensitivity,
precision, specificity, and F1-score, using the following formulas [30],
𝑇𝑃
𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 = (2)
(𝑇𝑃+𝐹𝑁)

𝑇𝑃
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (3)
(𝑇𝑃+𝐹𝑃)

𝑇𝑁
𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 = (4)
(𝑇𝑁+𝐹𝑃)

Int J Artif Intell, Vol. 12, No. 4, December 2023: 1864-1872

Int J Artif Intell ISSN: 2252-8938  1869
2𝑇𝑃
𝐹1 − 𝑠𝑐𝑜𝑟𝑒 = (5)
(2𝑇𝑃+𝐹𝑃+𝐹𝑁)

after that, the ROC curve will show the performance of the binary classification model at all classification
thresholds due to their decision-making ability in the classification issues. The prediction performance will
show the output of an algorithm after being trained on the dataset.

4. RESULTS AND DISCUSSIONS

The method used for image pre-processing in this work is the CLAHE technique through OpenCV.
The CT images collected were subjected to this technique to produce pre-processed images. This CLAHE
technique can apply to lung skin cancer images to enhance the contrast of the images. Figure 4 illustrates the
process of the CLAHE technique from CT image to pre-processed image that applies on the sample of lung
cancer image, which is unclear. Based on Figure 4, the contrast of raw lung cancer images can enhance
effectively using the CLAHE method. Compared to the lung cancer image in Figure 5(a), the pre-processed
image in Figure 5(b) after implementation of CLAHE is much clearer.

(a) (b)

Figure 5. CLAHE implementation: (a) Sample of lung cancer image and (b) Clean image after CLAHE

This study achieved a test accuracy performance of 99.61% for CT image classification using the
CNN-ARIMA model. The loss performance decreased exponentially while the accuracy performance
increased exponentially for training and validation, as shown in Figure 6(a) and Figure 6(b), respectively,
leading to high accuracy. The results demonstrated that the training and validation performance of the CNN-
ARIMA on CT image classification were almost identical at every epoch, as illustrated in Figure 6. This
condition can deliver high classification accuracy in the proposed method.

(a) (b)

Figure 6. Performance during training and validation: (a) Loss and (b) Accuracy

Development of hybrid convolutional neural network and autoregressive …(Abdulrazak Yahya Saleh)
1870  ISSN: 2252-8938

According to Figure 7(a), the test datasets are labelled as “True label” in the confusion matrix. A total
of 258 test images (20% of the overall dataset) are distributed randomly, of which 172 images were distributed
in lung cancer and 86 images distributed in normal lung. Moreover, Figure 7(a) also shows the outstanding
confusion matrix performance in CT images classification, in which 171 images categorize as lung cancer and
86 images categorize as normal lung. According to the results presented in Table 2, the performance metrics
for CT image classification using the CNN-ARIMA model achieved an average sensitivity of 99.71%, an
average precision of 99.43%, an average specificity of 99.71%, and an average F1-score of 99.57%.
Furthermore, the ROC curve, as shown in Figure 7(b), demonstrated excellent performance with an area under
the curve (AUC) of 1.000.

(a) (b)

Figure 7. Performance: (a) confusion matrix and (b) ROC

Table 2. Performance measures

Type of CT images Sensitivity Precision Specificity F1-Score
Lung cancer 99.42% 100.00% 100.00% 99.71%
Normal 100.00% 98.85% 99.42% 99.42%
Average 99.71% 99.43% 99.71% 99.57%

The prediction performance of the proposed work on the testing dataset shows remarkable
performance with accurate prediction. Figure 8(a) and Figure 8(b) show the examples of predicted lung cancer
and normal lung images from the testing dataset respectively. The comparative analysis in Table 3 demonstrates
that the performance of the proposed CNN-ARIMA model surpasses that of the existing techniques. Only work
from Saleh et al. [20] obtained a similar result with the proposed work in terms of AUC.

(a) (b)

Figure 8. Detection Performance: (a) Predict: Lung cancer, Actual: Lung cancer and
(b) Predict: Normal lung, Actual: Normal lung

Int J Artif Intell, Vol. 12, No. 4, December 2023: 1864-1872

Int J Artif Intell ISSN: 2252-8938  1871

Table 3. Comparative studies

Works Method Accuracy Sensitivity Precision Specificity F1-Score AUC
Haiying et al. [16] DCNN - 95.80% at 8 false - - - -
positive per image
Sheng et al. [18] DiagNet 90.00% - - - - -
Agarwal et al. [10] Alexnet CNN 96.00% - - - - -
Ardimento et al. [17] Ensemble learning 96.49% 98.73% 80.62% - - -
(VGG, Xception,
and ResNet)
Saleh et al. [20] Hybrid CNN and 97.91% 97.90% 97.96% 99.32% - 1.000
SVM
Proposed Work CNN-ARIMA 99.61% 99.71% 99.43% 99.71% 99.57% 1.000

5. CONCLUSION
This study presented the results of using a hybrid approach of CNN and ARIMA for CT image
classification. The hybrid of CNN and ARIMA shows better performance than the state-of-art deep learning
method. This study can further establish a CAD system, which assists radiologists and doctors in automatically
detecting and predicting lung conditions effectively. Nevertheless, this proposed work did not localize and
classify the pulmonary nodules in lung cancer image in depth. To further enhance the performance, the
segmentation step will consider as a future work because the efficiency of segmentation performance will
increase the diagnosis performance.

ACKNOWLEDGEMENTS
The authors express their gratitude towards the Ministry of Higher Education Malaysia for providing
financial assistance through the Fundamental Research Grant Scheme (FRGS)
FRGS/1/2021/SS0/UNIMAS/02/4. Additionally, the authors would like to extend their appreciation to the
Faculty of Cognitive Sciences and Human Development at Universiti Malaysia Sarawak (UNIMAS) for their
support and funding of the publication.

REFERENCES
[1] H. Sung et al., “Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185
countries,” CA: A Cancer Journal for Clinicians, vol. 71, no. 3, pp. 209–249, 2021, doi: 10.3322/caac.21660.
[2] Y. Chen, C. Liu, W. Peng, and S. Xia, “Thyroid nodule detection using attenuation value based on non-enhancement CT images,”
IET Conference Publications, vol. 2017, no. CP718, 2017, doi: 10.1049/cp.2017.0079.
[3] A. P. A. Thabsheera, T. M. Thasleema, and R. Rajesh, “Lung cancer detection using CT scan images: A review on various image
processing techniques,” Lecture Notes in Networks and Systems, vol. 43, pp. 413–419, 2019, doi: 10.1007/978-981-13-2514-4_34.
[4] R. J. McDonald et al., “The effects of changes in utilization and technological advancements of cross-sectional imaging on
radiologist workload,” Academic Radiology, vol. 22, no. 9, pp. 1191–1198, 2015, doi: 10.1016/j.acra.2015.05.007.
[5] A. Hosny, C. Parmar, J. Quackenbush, L. H. Schwartz, and H. J. W. L. Aerts, “Artificial intelligence in radiology,” Nature Reviews
Cancer, vol. 18, no. 8, pp. 500–510, 2018, doi: 10.1038/s41568-018-0016-5.
[6] L. Haibo, T. Shanli, S. Shuang, and L. Haoran, “An improved yolov3 algorithm for pulmonary nodule detection,” IMCEC 2021 -
IEEE 4th Advanced Information Management, Communicates, Electronic and Automation Control Conference, pp. 1068–1072,
2021, doi: 10.1109/IMCEC51613.2021.9482291.
[7] J. Liu, J. Li, F. Xue, and C. Wu, “Dense attention module for accurate pulmonary nodule detection,” ICASSP, IEEE International
Conference on Acoustics, Speech and Signal Processing - Proceedings, vol. 2021-June, pp. 1220–1224, 2021, doi:
10.1109/ICASSP39728.2021.9413936.
[8] A. A. A. Setio et al., “Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in
computed tomography images: The LUNA16 challenge,” Medical Image Analysis, vol. 42, pp. 1–13, 2017, doi:
10.1016/j.media.2017.06.015.
[9] S. Liu et al., “No surprises: Training robust lung nodule detection for low-dose CT scans by augmenting with adversarial attacks,”
IEEE Transactions on Medical Imaging, vol. 40, no. 1, pp. 335–345, 2021, doi: 10.1109/TMI.2020.3026261.
[10] A. Agarwal, K. Patni, and D. Rajeswari, “Lung cancer detection and classification based on AlexNet CNN,” Proceedings of the 6th
International Conference on Communication and Electronics Systems, ICCES 2021, pp. 1390–1397, 2021, doi:
10.1109/ICCES51350.2021.9489033.
[11] S. Das and S. Majumder, “Lung cancer detection using deep learning network: A comparative analysis,” Proceedings - 2020 5th
International Conference on Research in Computational Intelligence and Communication Networks, ICRCICN 2020, pp. 30–35,
2020, doi: 10.1109/ICRCICN50933.2020.9296197.
[12] C. K. Chin, D. A. B. A. Mat, and A. Y. Saleh, “Hybrid of convolutional neural network algorithm and autoregressive integrated
moving average model for skin cancer classification among Malaysian,” IAES International Journal of Artificial Intelligence, vol.
10, no. 3, pp. 707–716, 2021, doi: 10.11591/ijai.v10.i3.pp707-716.
[13] C. K. Chin, D. A. Binti Awang Mat, and A. Y. Saleh, “Skin cancer classification using convolutional neural network with
autoregressive integrated moving average,” ACM International Conference Proceeding Series, pp. 18–22, 2021, doi:
10.1145/3467691.3467693.
[14] A. Tripathi, A. Arora, and A. Bhan, “Classification of cervical cancer using deep learning algorithm,” Proceedings - 5th
International Conference on Intelligent Computing and Control Systems, ICICCS 2021, pp. 1210–1218, 2021, doi:
10.1109/ICICCS51141.2021.9432382.

Development of hybrid convolutional neural network and autoregressive …(Abdulrazak Yahya Saleh)
1872  ISSN: 2252-8938

[15] A. Naeem, M. S. Farooq, A. Khelifi, and A. Abid, “Malignant melanoma classification using deep learning: Datasets, performance
measurements, challenges and opportunities,” IEEE Access, vol. 8, pp. 110575–110597, 2020, doi:
10.1109/ACCESS.2020.3001507.
[16] Y. Haiying, F. Zhongwei, D. Ding, and S. Zengyang, “False-positive reduction of pulmonary nodule detection based on deformable
convolutional neural networks,” 2021 IEEE 9th International Conference on Bioinformatics and Computational Biology, ICBCB
2021, pp. 130–134, 2021, doi: 10.1109/ICBCB52223.2021.9459209.
[17] P. Ardimento, L. Aversano, M. L. Bernardi, and M. Cimitile, “Deep neural networks ensemble for lung nodule detection on chest
CT scans,” Proceedings of the International Joint Conference on Neural Networks, vol. 2021-July, 2021, doi:
10.1109/IJCNN52387.2021.9534176.
[18] J. Sheng, Y. Li, G. Cao, and K. Hou, “Modeling nodule growth via spatial transformation for follow-up prediction and diagnosis,”
Proceedings of the International Joint Conference on Neural Networks, vol. 2021-July, 2021, doi:
10.1109/IJCNN52387.2021.9534163.
[19] P. Musa, F. Al Rafi, and M. Lamsani, “A review: Contrast-limited adaptive histogram equalization (CLAHE) methods to help the
application of face recognition,” Proceedings of the 3rd International Conference on Informatics and Computing, ICIC 2018, 2018,
doi: 10.1109/IAC.2018.8780492.
[20] A. Y. Saleh, C. K. Chin, V. Penshie, and H. R. H. Al-Absi, “Lung cancer medical images classification using hybrid CNN-SVM,”
International Journal of Advances in Intelligent Informatics, vol. 7, no. 2, pp. 151–162, 2021, doi: 10.26555/ijain.v7i2.317.
[21] G. Yadav, S. Maheshwari, and A. Agarwal, “Contrast limited adaptive histogram equalization based enhancement for real time
video system,” Proceedings of the 2014 International Conference on Advances in Computing, Communications and Informatics,
ICACCI 2014, pp. 2392–2397, 2014, doi: 10.1109/ICACCI.2014.6968381.
[22] A. Mikołajczyk and M. Grochowski, “Data augmentation for improving deep learning in image classification problem,” 2018
International Interdisciplinary PhD Workshop, IIPhDW 2018, pp. 117–122, 2018, doi: 10.1109/IIPHDW.2018.8388338.
[23] T. Tsao and D. Y. Tsao, “A topological solution to object segmentation and tracking,” Proceedings of the National Academy of
Sciences of the United States of America, vol. 119, no. 41, 2022, doi: 10.1073/pnas.2204248119.
[24] P. M. Radiuk, “Impact of training set batch size on the performance of convolutional neural networks for diverse datasets,”
Information Technology and Management Science, vol. 20, no. 1, 2018, doi: 10.1515/itms-2017-0003.
[25] A. M. Saleh and T. Hamoud, “Analysis and best parameters selection for person recognition based on gait model using CNN
algorithm and image augmentation,” Journal of Big Data, vol. 8, no. 1, 2021, doi: 10.1186/s40537-020-00387-6.
[26] O. G. Ajayi and J. Ashi, “Effect of varying training epochs of a faster region-based convolutional neural network on the accuracy
of an automatic weed classification scheme,” Smart Agricultural Technology, vol. 3, 2023, doi: 10.1016/j.atech.2022.100128.
[27] D. P. Kingma and J. L. Ba, “Adam: A method for stochastic optimization,” 3rd International Conference on Learning
Representations, ICLR 2015 - Conference Track Proceedings, 2015.
[28] H. Benbrahim, H. Hachimi, and A. Amine, “Deep convolutional neural network with TensorFlow and Keras to classify skin cancer
images,” Scalable Computing: Practice and Experience, vol. 21, no. 3, pp. 379–390, 2020, doi: 10.12694/scpe.v21i3.1725.
[29] M. Abadi et al., “TensorFlow: A system for large-scale machine learning,” Proceedings of the 12th USENIX Symposium on
Operating Systems Design and Implementation, OSDI 2016, pp. 265–283, 2016.
[30] M. Stojanović et al., “Understanding sensitivity, specificity and predictive values,” Vojnosanitetski Pregled, vol. 71, no. 11, pp.
1062–1065, 2014, doi: 10.2298/VSP1411062S.

BIOGRAPHIES OF AUTHORS

Dr Abdulrazak Yahya Saleh is a Senior Lecturer in Faculty of Cognitive Sciences

and Human Development, Universiti Malaysia Sarawak (UNIMAS). He received his PhD from
Universiti Teknologi Malaysia, MSc and BSc from University of Science and Technology,
Yemen. His research is mainly on the cognitive science, artificial intelligence, neural networks,
spiking neural networks, brain modeling, optimization methods, multi-objective optimization
algorithms and machine learning techniques and their applications. The current focus of his
research is on deep learning applications and challenges in big data analytics. He can be contacted
at email: [email protected].

Mr Chee Ka Chin received his MSc and BSc from Faculty of Engineering, University
Malaysia Sarawak (UNIMAS). His research is mainly on the artificial intelligence and neural
network. The current focus of his research is on deep learning applications, data science and
computer vision. He can be contacted at email: [email protected].

Int J Artif Intell, Vol. 12, No. 4, December 2023: 1864-1872