Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 213 https://internationalpubls.com Skin Cancer Diagnosis with a Customized CNN Model using Deep Learning Approaches Kiran Likhar1, Dr. Sonali Ridhorkar2 1Research Scholar(Ph.D), Department of CSE, G H Raisoni University Amravati , India. Email: kiran.likhar24@gmail.com 2Associate Professor, Department of CSE, G H Raisoni Institute of Engineering and Technology, Nagpur, India. Email: sonaliridhorkar@gmail.com Article History: Received: 11-05-2024 Revised: 23-06-2024 Accepted: 05-07-2024 Abstract: Medical imaging has a significant challenge in accurately classifying skin lesions into benign and malignant classifications. To solve this issue, we have developed a technique that utilizes a custom convolutional neural network classifier with a support vector machine. Our customized CNN architecture is designed to address the core issue of skin cancer categorization. DenseNet121, DenseNet201, InceptionV3, InceptionResNetV2, MobileNet, ResNet50V2, ResNet101, VGG16, VGG19, and Xception are among the most prominent pre-trained models evaluated in our study. The customized CNN exceeds existing models on an average basis, displaying greater accuracy, recall, precision, and F1-Score for both benign and malignant cases. This technique has significant prospects for enhancing early skin cancer diagnosis, perhaps leading to better patient results and more efficient medical treatments. Keywords: Artificial intelligence, Skin cancer detection, Deep learning, Pretrained Models, CNN. 1. Introduction Skin cancer detection is a critical aspect of healthcare due to its increasing prevalence worldwide. The conventional methods of visual examination used by dermatologists for detecting skin cancer are subjective. The use of DL techniques has gained significant attention for improving the accuracy and efficiency of skin cancer detection. To offer an extensive analysis of Deep learning approaches in the identification of skin cancer through integrating various datasets and reviewing relevant literature [1]. By examining existing study, the objective is to identify the latest methodologies, address challenges, and propose potential solutions to increase the accuracy and generalizability of skin cancer models for detection. To accomplish this goal, Skin lesion datasets containing dermoscopy images, clinical photographs, and histopathological slides will be collected and curated. These datasets will be utilized in DL technique training and evaluation across different imaging techniques and patient demographics. The inclusion of a wide range of data sources will enhance the reliability and applicability of the developed models [2]. Moreover, an extensive literature review will be conducted to identify the advancements and limitations of existing Deep learning approaches in skin cancer detection. The analysis will focus on the methodologies employed, such as feature extraction techniques, classification algorithms, and model evaluation metrics. By understanding the advantages and disadvantages of certain strategies, opportunities for improvement and innovation can be identified [3]. mailto:kiran.likhar24@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 214 https://internationalpubls.com 1.1 Review of Deep Learning classifier for Skin Cancer Detection Deep learning has emerged as a revolutionary force within the DL sector, particularly within the last few decades. It is recognized as a sophisticated subfield that focuses on ANN methods, drawing inspiration from the structure and functionality within the human mind. DL techniques have been successfully applied in various domains, including speech recognition, pattern recognition, and bioinformatics, producing impressive results when compared to traditional DL approaches [4]. In recent years, DL approaches have gained significant traction in computer-based skin cancer detection. This research goal is to provide a thorough and structured survey of the literature on DL techniques employed in skin cancer detection. The focus is on classical DL approaches such as CNN. To ensure a valuable systematic review of neural network-based classification techniques for identifying skin cancer, a rigorous strategy was devised. 1.2 Analyzing Deep Learning Methods for the Identification of Skin Cancer DL techniques for skin cancer detection were explored. Skin cancer datasets were integrated, and a comprehensive literature review on DL methods was conducted [5]. Various DL models were applied, and their performances were compared. The potential of DL for enhancing skin cancer detection mechanisms was demonstrated. Accurate detection of skin cancer was achieved using these techniques. The significance of early detection in improving patient outcomes was emphasised. Key factors influencing accurate detection were identified and analysed. The role of DL in aiding medical professionals in making timely diagnoses and treatment decisions was highlighted [6]. 2. Literature review Skin cancer stands as a prevalent form of cancer on a global scale, posing a substantial concern for people annually. Early detection and accurate diagnosis play pivotal roles in improving patient outcomes and reducing the mortality and morbidity associated with this disease. Over the years, study and healthcare professionals have explored various innovative approaches to enhance skin cancer detection and treatment. In their study, Shi Wang et al. [7] suggested an innovative method for skin cancer detection, combining the Extreme Learning Machine (ELM) with an enhanced version of Thermal Exchange Optimization (TEO). By leveraging ELM and TEO, they achieved improved accuracy, sensitivity, and specificity in identifying malignant skin lesions, leading to reliable and timely diagnoses. The ELM-TEO system has shown great potential for lowering mortality and morbidity associated with skin cancer, proving its usefulness within the healthcare industry. In addressing the problem of cervical cancer detection, Umesh Kumar Lilhore et al. [8] developed a Model that integrated Causal Analysis and Deep Learning techniques. This novel approach allowed the identification of potential risk factors and their relationships through causal analysis, which, in turn, contributed to building a predictive model using Deep Learning algorithms. The developed Model showcased substantial improvements in accuracy and sensitivity for cervical cancer detection, providing valuable insights for better healthcare management and improved patient outcomes. However, not all innovations in skin cancer detection have been successful. Ahmad M. Khasawneh et al. [9] investigated the problem of immediate identification by DL-based medical picture Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 215 https://internationalpubls.com evaluation. Unfortunately, the study faced challenges with data quality, algorithmic limitations, and potential biases, leading to the retraction of the reported results. This highlights the importance of addressing such issues and ensuring the reliability and validity of study outcomes in the domain of medical diagnostics. On a different note, M. Shobana et al. [10] successfully addressed the challenge of mesothelioma cancer classification and detection using a Feature Selection-Enabled Deep Learning technique. By identifying the most informative and discriminative features from the dataset, their method achieved high-performance classification and early-stage detection of this aggressive form of cancer. This development showed greater sensitivity and accuracy in detecting mesothelioma malignancy, with positive implications for better patient treatment and healthier outcomes. CNN has displayed known potential in various fields, including skin cancer detection. Mohammed Rakeibul Hasan et al. [11] performed a comparative analysis using CNNs to separate equally benign and malignant skin cancer cases. Their approach demonstrated significantly improved accuracy, sensitivity, and specificity compared to conventional methods, proving the efficacy of CNN-based techniques in analyzing complex image patterns and identifying cancerous conditions. Beyond diagnosis, researchers have explored the impact of diet on skin cancer risk. Sreevidya R. C. et al. [12] suggested an innovative approach to identify potential correlations between antioxidant-rich diets and their impact on skin cancer risk. By utilizing AI and Deep Learning algorithms to analyze vast datasets of dietary information and skin cancer cases, this method revealed valuable insights into the potential preventive properties of antioxidants against skin cancer. Such AI-driven approaches hold promise in supporting healthcare professionals and individuals in making informed dietary choices to mitigate skin cancer risks and enhance overall health outcomes. Hamza Abu Owida et al. [13] focused on skin cancer therapy and detection using Biomimetic Nanoscale Materials. These nanomaterials were designed to mimic biological processes, effectively targeting and treating skin cancer cells. Additionally, they were utilized in the initial identification of skin cancer by selectively binding to cancer-specific biomarkers or signaling molecules. The data proven important advancements in the area of skin cancer diagnosis and treatment, showing the potential of biomimetic nanoscale materials as a viable and novel strategy to improve therapeutic results and boost skin cancer early detection rates. Taher M. Ghazal et al. [14] employed Transfer Learning to address the challenge of detecting benign and malignant tumors in the skin. Their approach involved fine-tuning pre-trained deep learning models on large datasets from unrelated tasks, leading to significant improvements in accuracy and efficiency for skin tumor classification. The use of Transfer Learning empowered the method with valuable insights from unrelated datasets, offering a valuable tool for healthcare professionals in early diagnosis and effective treatment planning for patients with skin cancer. Machine Vision with Texture Features was employed by Syeda Shamaila Zareen et al. [15] to address skin cancer classification. The extraction and combination of various texture features from dermatoscopic images of skin lesions significantly enhanced the accuracy of skin cancer classification. This Machine Vision-based method demonstrated notable improvements in differentiating benign and malignant skin tumors, indicating its potential in assisting healthcare professionals in making more informed and timely decisions for skin cancer diagnosis and treatment, ultimately leading to improved patient outcomes. Muhammad Arif et al. [16] will propose another notable advancement in skin cancer Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 216 https://internationalpubls.com evaluation. They developed an automated system using DCNN to detect nonmelanoma skin cancer. A convolutional neural network has learned from an extensive dataset of dermatoscopic pictures, enabling the automated identification and differentiation of nonmelanoma skin cancer cases based on benign lesions. Substantial advancements in enhancing the precision and reliability of nonmelanoma skin cancer detection were demonstrated, showcasing the effectiveness of Deep CNN in analyzing complex image patterns and identifying cancerous conditions. Automated detection systems can assist healthcare professionals in the initial stages of detection and timely intervention, thus contributing to improved patient care and better outcomes for individuals with nonmelanoma skin cancer. 3. Methodology Using deep learning technology, our study improves skin cancer diagnosis by combining pre-trained CNN models, data augmentation, and image normalization. Relevant characteristics are extracted using CNN models such as DenseNet121, DenseNet201, InceptionV3, InceptionResNetV2, MobileNet, ResNet50V2, ResNet101, VGG16, VGG19, Xception, and bespoke CNNs. These attributes are then used to classify skin cancers as benign or malignant using deep learning techniques such as SVMs or random forests. This complete method creates an effective and accurate structure for detecting skin cancer. 3.1 Summary of dataset Table 1 presents a summary of the image distribution within a medical imaging dataset. This dataset is divided into two main folders: 'benign' and 'malignant,' and further categorized into three subfolders: 'training,' 'validation,' and 'testing.' The 'training' subfolder is dedicated to images used for classifying them as benign or malignant. Model efficiency is analyzed within the "validation" subfolder during the training process, and the model's effectiveness is assessed using previously unseen data in the 'testing' subfolder. The dataset comprises a total of 2360 images, with 1180 classified as benign and 1200 as malignant. This medical imaging data set is organized according to several subfolders: 'training,' 'validation,' and 'testing,' with each subfolder containing 300 images. Table 1. Medical Imaging Dataset Label Benign Malignant Total Images Training Folder 1180 1180 2360 Validation Folder 1180 1180 2360 Test Folder 300 300 600 3.2 Balanced Skin Lesion Dataset for Binary Classification The data appears to be organized into three main categories: Training, Validation, and Test sets, as seen in Figure 1. Each folder contains images of skin lesions, which have been classified into two categories: Benign and Malignant. In each of the Training and Validation folders, there are 1,180 pictures depicting benign cancers and 1,180 pictures depicting malignant cancer. This balanced distribution is important for training and evaluating deep learning models, as it helps prevent bias. In the Test folder, 300 pictures of benign lesions and 300 pictures of malignant cancers are maintained in a balanced distribution. The maximum total of pictures in the Training and Validation folders is Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 217 https://internationalpubls.com 2,360 each, and there are 600 images in the Test folder. This dataset is going to be used for Skin Cancer Binary Classification. Figure 1. Skin Cancer Data Distribution 3.3 Comparative Analysis of Pre-trained Deep Learning Models by Model Complexity and Parameters Table 2 provides a comprehensive analysis of various pre-trained deep learning models, highlighting their model complexity in terms of modifiable and fixed parameters. Trainable parameters represent those components that are fine-tuned during training, while non-trainable parameters are fixed weights that are usually pre-trained on large datasets. The overall parameter count in a model equals the sum of these two components. DenseNet121 and DenseNet201 demonstrate relatively compact architectures with lower trainable and non-trainable parameters. DenseNet121 features 32,770 trainable parameters, making it computationally efficient. InceptionV3 is characterized by a moderate number of trainable parameters (16,386) and a substantial number of non-trainable parameters (21,802,784). InceptionResNetV2 features a comparatively low number of trainable parameters (12,290) but possesses a significant number of non-trainable parameters (54,336,736). This points to its complex architecture. MobileNet is notable for having no trainable parameters, offering efficient inference with a smaller parameter footprint. ResNet50V2 and ResNet101 are more parameter-heavy models, boasting 65,538 trainable parameters each, combined with a substantial number of non-trainable parameters. VGG16 and VGG19 exhibit balance with moderate trainable and non-trainable parameters, offering a good trade-off between model complexity and computational requirements. Xception is characterized by a considerable number of trainable parameters (65,538) and non-trainable parameters (20,861,480), reflecting its complex architecture. Table 2. Comparison of Pre-Trained Deep Learning Models Based on Model Complexity Pre-Trained Deep Learning Models Trainable Parameters Non-Trainable Parameters Total Parameters DenseNet121 32,770 7,037,504 7,070,274 DenseNet201 61,442 18,321,984 18,383,426 InceptionV3 16,386 21,802,784 21,819,170 InceptionResNetV2 12,290 54,336,736 54,349,026 MobileNet 0 3,228,864 3,228,864 ResNet50V2 65,538 23,564,800 23,630,338 ResNet101 65,538 42,658,176 42,723,714 VGG16 16,386 14,714,688 14,731,074 VGG19 16,386 20,024,384 20,040,770 Xception 65,538 20,861,480 20,927,018 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 218 https://internationalpubls.com Deep Learning Approach for Skin Cancer Detection Technology employs DL to detect skin cancer using an advanced approach. It leverages pre-trained weights from a CNN model recently adapted for image categorization tasks. Methods for enhancing data, including techniques like resizing and orientation adjustment, are used to augment the training dataset with skin lesion images. All images are standardized to a uniform size of 224x224x4 pixels. Image normalization is used to ensure consistent pixel values. Various convolutional neural network frameworks, including DenseNet121, DenseNet201, InceptionV3, InceptionResNetV2, MobileNet, ResNet50V2, ResNet101, VGG16, VGG19, Xception, and custom CNN models, are employed to extract features relevant to skin cancer identification. These features undergo a decision-making process, typically implemented using deep learning techniques like SVMs or random forests, to determine the category of skin lesions as benign or malignant. The use of pre-trained weights accelerates the process, data augmentation mitigates overfitting, image normalization ensures consistency, and a robust framework for accurate skin cancer detection is established through feature extraction and decision-making, as illustrated in the figure 2. 3.4 Convolutional Neural Network Based Skin Cancer Detection CNNs are a vital type of deep neural network that finds effective applications in visual recognition. It is employed for image categorization, creating a compilation of the supplied images, and performing picture identification. A convolutional neural network is an excellent method of collecting and processing local and global data because it combines basic features such as curves and edges to build more intricate elements such as shapes and edges. Convolutional Neural Network intermediate layers are made up of convolutional, fully connected, and nonlinear pooling layers. CNN may contain several convolutional layers, preceded by several fully linked layers. The three primary types of layers utilized in CNN are convolution, pooling, and full-connected layers. Fig 3 presents CNN's fundamental architecture [19]. Figure 2. Deep Learning-based Skin Cancer Detection Pipeline Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 219 https://internationalpubls.com Figure 3. Skin Cancer Detection CNN Architecture 4. Experimental Results and Data Analysis 4.1 Model Performance in Skin Cancer Classification The utilization of pre-trained deep learning models has become a prominent approach in various domains, particularly in the field of computer-aided diagnosis for skin cancer. This study offers a comprehensive performance evaluation of several pre-trained models when employed in skin cancer classification. The focus lies on five different deep learning classifiers: SVM, KNN, Decision Tree, Gradient Boosting, and Random Forest. Analyzing the models mentioned as follows: DenseNet121, DenseNet201, VGG16, VGG19, Xception, MobileNet, ResNet50V2, InceptionV3, and InceptionResNetV2. Table 3 further shows that pre-trained DL algorithms demonstrate quite distinct when applied to the categorization of skin cancer. The presented table illustrates the classification performance of each model using the five deep learning classifiers, evaluated in terms of accuracy. DenseNet121 exhibits relatively low performance across all classifiers. It struggles to achieve high accuracy in most cases. DenseNet201 demonstrates improved performance compared to DenseNet121, especially with the KNN and Decision Tree classifiers. InceptionV3 achieves consistent performance across all classifiers, indicating its robustness and suitability for various tasks. InceptionResNetV2 performs well with SVM and Decision Tree, showcasing its versatility. MobileNet achieves modest accuracy and shows potential for lightweight applications due to its efficient architecture. ResNet50V2 and ResNet101 deliver strong performance, especially with the Gradient Boosting classifier. Xception exhibits a moderate level of performance, with availability for higher-level challenges. VGG19 and VGG16 show relatively high accuracy with most classifiers, particularly with Gradient Boosting. This study highlights the impact of model complexity on classification accuracy, illustrating the drawbacks between model efficiency and performance in the context of skin cancer diagnosis. Table 3. Performance Evaluation of Pre-Trained Deep Learning Models with DL Classifiers in Skin Cancer Classification Pre-Trained Model \ Names of Classifiers SVM KNN Decision Tree Gradient Boosting Random Forest DenseNet121 0.0144 0.4766 0.40084 0.3203 0.4245 DenseNet201 0.0427 0.5152 0.6644 0.6072 0.6453 InceptionV3 0.4949 0.5084 0.4911 0.5127 0.4911 InceptionResNetV2 0.5419 0.4991 0.4949 0.4974 0.4949 MobileNet 0.0347 0.4355 0.2457 0.2042 0.2487 ResNet50V2 0.5322 0.4889 0.4936 0.5173 0.4944 ResNet101 0.7402 0.4699 0.4966 0.5245 0.4966 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 220 https://internationalpubls.com VGG19 0.1038 0.4762 0.4838 0.4101 0.4838 VGG16 0.0919 0.4635 0.4877 0.3292 0.4868 Xception 0.0347 0.4648 0.2313 0.2233 0.2368 4.2 Evaluation of Binary Skin Cancer Classification Models The custom CNN algorithm outperformed the trained methods in measures of accuracy, recall, precision, and F1-score for both benign and malignant classes. Among the pre-trained models, DenseNet121, DenseNet201, MobileNet, and ResNet50V2 showed relatively good performance. In terms of Accuracy, the Custom CNN, DenseNet121, VGG16, and VGG19 models performed the best. Regarding Recall, the MobileNet model performed exceptionally well for malignant cases, but DenseNet121, VGG19, and the Custom CNN also had high Recall values. In terms of Precision, the MobileNet model showed high precision for benign cases, while DenseNet121, VGG19, and the Custom CNN had good precision values for malignant cases. The F1-Score, which balances precision and recall, demonstrated that the Custom CNN and DenseNet121 performed well for both benign and malignant cases. Table 4. Testing for Evaluation Metric based on Deep Learning Models in Skin Cancer Binary Classification Sr. No Model Name Metric Class Benign Class Malignant 1 Densenet121 Accuracy 0.8400 0.8400 Recall 0.8033 0.8767 Precision 0.8669 0.8168 F1-Score 0.8339 0.8457 2 DenseNet201 Accuracy 0.8217 0.8217 Recall 0.9167 0.7267 Precision 0.7703 0.8971 F1-Score 0.8371 0.8029 3 InceptionResNetV2 Accuracy 0.7783 0.7783 Recall 0.7033 0.8533 Precision 0.8275 0.7420 F1-Score 0.7604 0.7938 4 InceptionV3 Accuracy 0.7933 0.7933 Recall 0.7867 0.8000 Precision 0.7973 0.7895 F1-Score 0.7919 0.7947 5 MobileNet Accuracy 0.8217 0.8217 Recall 0.6667 0.9767 Precision 0.9662 0.7455 F1-Score 0.7890 0.8456 6 ResNet50V2 Accuracy 0.8217 0.8217 Recall 0.8733 0.7700 Precision 0.7915 0.8587 F1-Score 0.8304 0.8120 7 ResNet101 Accuracy 0.7450 0.7450 Recall 0.8300 0.6600 Precision 0.7094 0.7952 F1-Score 0.7650 0.7213 8 VGG16 Accuracy 0.8250 0.8250 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 221 https://internationalpubls.com Recall 0.8867 0.7633 Precision 0.7893 0.8707 F1-Score 0.8352 0.8135 9 VGG19 Accuracy 0.8367 0.8367 Recall 0.7467 0.9267 Precision 0.9106 0.7853 F1-Score 0.8205 0.8502 10 Xception Accuracy 0.7833 0.7833 Recall 0.8233 0.7433 Precision 0.7623 0.8080 F1-Score 0.7917 0.7743 11 Custom CNN Accuracy 0.8617 0.8617 Recall 0.8533 0.8700 Precision 0.8678 0.8557 F1-Score 0.8605 0.8628 4.3 Comparative analysis The confusion matrices for several DL algorithms with the task of categorizing skin lesions as benign or malignant give fascinating perspective into their performance. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 222 https://internationalpubls.com Figure 4. Performance Comparison of Deep Learning Models for Skin Cancer Categorization Xception produced a moderately satisfactory categorization, properly detecting 225 benign and 223 malignant lesions, with a few misclassifications in each group. DenseNet201, exhibited a greater percentage of misclassification, notably for benign lesions, where 200 were correctly recognized. The frequency of misclassifications was much greater for VGG19 and VGG16, suggesting a less accurate performance in discriminating between benign and malignant cases. DenseNet121 and MobileNet produced a significantly improved misclassifications, particularly DenseNet121, which showed a high accuracy in properly diagnosing benign lesions. InceptionV3 and InceptionResNetV2 performed well, with a very even distribution of properly and wrongly categorized lesions in both categories. ResNet-101 had more difficulty with benign lesions, misclassifying a significant Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 223 https://internationalpubls.com percentage of them as malignant. ResNet50V2 also had difficulty identifying benign instances. Finally, the Custom CNN model produced mixed results, with a large number of misclassifications, particularly in the case of malignant lesions. These findings imply that the deep learning algorithm used has a considerable impact on the precision of skin cancer lesion classification, with certain models outperforming others in this task. 5. Conclusion This research findings present a unique and accurate mechanism for accurately classifying skin cancer lesions into benign and malignant categories. Our technique consistently outperforms renowned pre-trained models by merging a custom CNN with a SVM model. For both benign and malignant instances, the custom CNN outperforms in terms of accuracy, recall, precision, and F1- Score. Models such as DenseNet121, DenseNet201, MobileNet, and ResNet50V2 also performed well, with the custom CNN, DenseNet121, VGG16, and VGG19 dominating in accuracy. For malignant cases, the MobileNet model offers the best recall, but DenseNet121, VGG19, and the custom CNN also offer excellent recall scores. In terms of accuracy, MobileNet performs in benign situations, whereas DenseNet121, VGG19, and the custom CNN excel in malignant ones. The custom CNN's and DenseNet121's outstanding F1-Score performance demonstrates the dependability of our technique. This novel technology has the potential to enhance patient outcomes and medical treatment efficiency, consequently contributing to the evolution of healthcare imagery and improving skin cancer detection and treatment quality. References [1] Dhatri Raval, Jaimin N. Undavia, A Comprehensive assessment of Convolutional Neural Networks for skin and oral cancer detection using medical images, Healthcare Analytics, Volume 3, 2023, 100199, ISSN 2772-4425, https://doi.org/10.1016/j.health.2023.100199. [2] Vipin Venugopal, Navin Infant Raj, Malaya Kumar Nath, Norton Stephen, A deep neural network using modified EfficientNet for skin cancer detection in dermoscopic images, Decision Analytics Journal, Volume 8, 2023, 100278, ISSN 2772-6622, https://doi.org/10.1016/j.dajour.2023.100278. [3] Devakishan Adla, G. Venkata Rami Reddy, Padmalaya Nayak, G. Karuna, A full-resolution convolutional network with a dynamic graph cut algorithm for skin cancer classification and detection, Healthcare Analytics, Volume 3, 2023, 100154, ISSN 2772-4425, https://doi.org/10.1016/j.health.2023.100154. [4] Harsh Bhatt, Vrunda Shah, Krish Shah, Ruju Shah, Manan Shah, State-of-the-art machine learning techniques for melanoma skin cancer detection and classification: a comprehensive review, Intelligent Medicine, 2022, ISSN 2667-1026, https://doi.org/10.1016/j.imed.2022.08.004. [5] Ashutosh Lembhe, Pranav Motarwar, Rudra Patil, Susan Elias, Enhancement in Skin Cancer Detection using Image Super Resolution and Convolutional Neural Network, Procedia Computer Science, Volume 218, 2023, Pages 164- 173, ISSN 1877-0509, https://doi.org/10.1016/j.procs.2022.12.412. [6] K. Mridha, M. M. Uddin, J. Shin, S. Khadka and M. F. Mridha, "An Interpretable Skin Cancer Classification Using Optimized Convolutional Neural Network for a Smart Healthcare System," in IEEE Access, vol. 11, pp. 41003- 41018, 2023, doi: 10.1109/ACCESS.2023.3269694. [7] Shi Wang, Melika Hamian, "Skin Cancer Detection Based on Extreme Learning Machine and a Developed Version of Thermal Exchange Optimization", Computational Intelligence and Neuroscience, vol. 2021, Article ID 9528664, 13 pages, 2021. https://doi.org/10.1155/2021/9528664 [8] Umesh Kumar Lilhore, M. Poongodi, Amandeep Kaur, Sarita Simaiya, Abeer D. Algarni, Hela Elmannai, V. Vijayakumar, Godwin Brown Tunze, Mounir Hamdi, "Hybrid Model for Detection of Cervical Cancer Using Causal Analysis and Machine Learning Techniques", Computational and Mathematical Methods in Medicine, vol. 2022, Article ID 4688327, 17 pages, 2022. https://doi.org/10.1155/2022/4688327 [9] Ahmad M. Khasawneh, Amal Bukhari, Mahmoud Ahmad Al-Khasawneh, "Early Detection of Medical Image Analysis by Using Machine Learning Method", Computational and Mathematical Methods in Medicine, vol. 2022, Article ID 3041811, 11 pages, 2022. https://doi.org/10.1155/2022/3041811 https://doi.org/10.1016/j.health.2023.100199 https://doi.org/10.1016/j.dajour.2023.100278 https://doi.org/10.1016/j.health.2023.100154 https://doi.org/10.1016/j.imed.2022.08.004 https://doi.org/10.1016/j.procs.2022.12.412 https://doi.org/10.1155/2021/9528664 https://doi.org/10.1155/2022/4688327 https://doi.org/10.1155/2022/3041811 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 5s (2024) 224 https://internationalpubls.com [10] M. Shobana, V. R. Balasraswathi, R. Radhika, Ahmed Kareem Oleiwi, Sushovan Chaudhury, Ajay S. Ladkat, Mohd Naved, Abdul Wahab Rahmani, "Classification and Detection of Mesothelioma Cancer Using Feature Selection-Enabled Machine Learning Technique", BioMed Research International, vol. 2022, Article ID 9900668, 6 pages, 2022. https://doi.org/10.1155/2022/9900668 [11] Mohammed Rakeibul Hasan, Mohammed Ishraaf Fatemi, Mohammad Monirujjaman Khan, Manjit Kaur, Atef Zaguia, "Comparative Analysis of Skin Cancer (Benign vs. Malignant) Detection Using Convolutional Neural Networks", Journal of Healthcare Engineering, vol. 2021, Article ID 5895156, 17 pages, 2021. https://doi.org/10.1155/2021/5895156 [12] Sreevidya R. C., Jalaja G, Sajitha N, D. Lakshmi Padmaja, S. Nagaprasad, Kumud Pant, Yekula Prasanna Kumar, "Role of Artificial Intelligence and Deep Learning in Easier Skin Cancer Detection through Antioxidants Present in Food", Journal of Food Quality, vol. 2022, Article ID 5890666, 12 pages, 2022. https://doi.org/10.1155/2022/5890666 [13] Hamza Abu Owida, "Biomimetic Nanoscale Materials for Skin Cancer Therapy and Detection", Journal of Skin Cancer, vol. 2022, Article ID 2961996, 12 pages, 2022. https://doi.org/10.1155/2022/2961996 [14] Taher M. Ghazal, Sajid Hussain, Muhammad Farhan Khan, Muhammad Adnan Khan, Raed A. T. Said, Munir Ahmad, "Detection of Benign and Malignant Tumors in Skin Empowered with Transfer Learning", Computational Intelligence and Neuroscience, vol. 2022, Article ID 4826892, 9 pages, 2022. https://doi.org/10.1155/2022/4826892 [15] Syeda Shamaila Zareen, Sun Guangmin, Yu Li, Mahwish Kundi, Salman Qadri, Syed Furqan Qadri, Mubashir Ahmad, Ali Haider Khan, "A Machine Vision Approach for Classification of Skin Cancer Using Hybrid Texture Features", Computational Intelligence and Neuroscience, vol. 2022, Article ID 4942637, 11 pages, 2022. https://doi.org/10.1155/2022/4942637 [16] Muhammad Arif, Felix M. Philip, F. Ajesh, Diana Izdrui, Maria Daniela Craciun, Oana Geman, "Automated Detection of Nonmelanoma Skin Cancer Based on Deep Convolutional Neural Network", Journal of Healthcare Engineering, vol. 2022, Article ID 6952304, 15 pages, 2022. https://doi.org/10.1155/2022/6952304. [17] R. Schiavoni, G. Maietta, E. Filieri, A. Masciullo and A. Cataldo, "Microwave Reflectometry Sensing System for Low-Cost in-vivo Skin Cancer Diagnostics," in IEEE Access, vol. 11, pp. 13918-13928, 2023, doi: 10.1109/ACCESS.2023.3243843. [18] N. Andreasen et al., "Skin Electrical Resistance as a Diagnostic and Therapeutic Biomarker of Breast Cancer Measuring Lymphatic Regions," in IEEE Access, vol. 9, pp. 152322-152332, 2021, doi: 10.1109/ACCESS.2021.3123569. [19] H. L. Gururaj, N. Manju, A. Nagarjun, V. N. M. Aradhya and F. Flammini, "DeepSkin: A Deep Learning Approach for Skin Cancer Classification," in IEEE Access, vol. 11, pp. 50205-50214, 2023, doi: 10.1109/ACCESS.2023.3274848. [20] Imran, A. Nasir, M. Bilal, G. Sun, A. Alzahrani and A. Almuhaimeed, "Skin Cancer Detection Using Combined Decision of Deep Learners," in IEEE Access, vol. 10, pp. 118198-118212, 2022, doi: 10.1109/ACCESS.2022.3220329. [21] Z. Lan, S. Cai, X. He and X. Wen, "FixCaps: An Improved Capsules Network for Diagnosis of Skin Cancer," in IEEE Access, vol. 10, pp. 76261-76267, 2022, doi: 10.1109/ACCESS.2022.3181225. [22] Y. Tian, Z. Wu, J. Zhao, H. Lui and H. Zeng, "Cutaneous Porphyrin Exhibits Anti-Stokes Fluorescence Emission Under Continuous Wave Laser Excitation," in IEEE Journal of Selected Topics in Quantum Electronics, vol. 29, no. 4: Biophotonics, pp. 1-6, July-Aug. 2023, Art no. 7000206, doi: 10.1109/JSTQE.2022.3227557. [23] N. Shafi et al., "A Portable Non-Invasive Electromagnetic Lesion-Optimized Sensing Device for the Diagnosis of Skin Cancer (SkanMD)," in IEEE Transactions on Biomedical Circuits and Systems, vol. 17, no. 3, pp. 558-573, June 2023, doi: 10.1109/TBCAS.2023.3260581. [24] K. Sharma et al., "Dermatologist-Level Classification of Skin Cancer Using Cascaded Ensembling of Convolutional Neural Network and Handcrafted Features Based Deep Neural Network," in IEEE Access, vol. 10, pp. 17920- 17932, 2022, doi: 10.1109/ACCESS.2022.3149824. [25] Y. Tang, L. -Y. Chen, A. Zhang, C. -P. Liao, M. E. Gross and E. S. Kim, "In Vivo Non-Thermal, Selective Cancer Treatment with High-Frequency Medium-Intensity Focused Ultrasound," in IEEE Access, vol. 9, pp. 122051- 122066, 2021, doi: 10.1109/ACCESS.2021.3108548. https://doi.org/10.1155/2022/9900668 https://doi.org/10.1155/2021/5895156 https://doi.org/10.1155/2022/5890666 https://doi.org/10.1155/2022/2961996 https://doi.org/10.1155/2022/4826892 https://doi.org/10.1155/2022/4942637 https://doi.org/10.1155/2022/6952304