AI-Powered Skin Lesion Diagnosis using Whale Optimization Algorithm Enhanced ResNet 50 for Cancer Prediction.

Area of Science:

• Dermatology
• Computer Science
• Artificial Intelligence

Background:

• Binary classification of skin lesions is crucial for early cancer detection.
• Deep learning models like ResNet-50 show promise but require optimization for clinical accuracy and efficiency.

Purpose of the Study:

• To enhance the performance of the ResNet-50 model for binary skin lesion classification.
• To optimize ResNet-50 hyperparameters using the Whale Optimization Algorithm (WOA).

Main Methods:

• Compared five Convolutional Neural Network (CNN) architectures: AlexNet, GoogleNet, VGG16, ResNet-50, and WOA-optimized ResNet-50.
• Trained models on a balanced dataset of 3,600 dermoscopic images (1,800 benign, 1,800 malignant).
• Optimized ResNet-50 hyperparameters (learning rate, weights, bias) using WOA and evaluated performance using accuracy, precision, recall, F1-score, MCC, log loss, AUC-ROC, and inference time.

Main Results:

• WOA-optimized ResNet-50 achieved 98.29% accuracy, outperforming standard ResNet-50 (90.13%) and other CNNs.
• The optimized model demonstrated superior recall (99.31%), specificity (97.07%), and AUC-ROC (99.84%).
• Achieved faster inference time (0.1488s) compared to standard ResNet-50 (1.029s) with minimal misclassifications.

Conclusions:

• WOA-optimized ResNet-50 significantly improves accuracy, recall, specificity, and computational efficiency in skin lesion classification.
• This approach offers superior predictive performance and fast inference, aiding dermatological diagnostics and clinical decision-making.
• Future work could extend this method to multi-class classification and real-time medical imaging systems.