Scalable HMO-CNN-SVM Framework for Skin Lesion Classification: A Metaheuristic-Driven Approach With Parallelizable...

Area of Science:

• Medical Image Analysis
• Computational Dermatology
• Artificial Intelligence in Healthcare

Background:

• Accurate skin lesion categorization is challenging in medical image analysis, especially with limited data and computational constraints.
• Automated skin cancer detection requires robust and efficient classification models.

Purpose of the Study:

• To develop a scalable hybrid model for improved skin lesion categorization and automated skin cancer detection.
• To optimize Convolutional Neural Network (CNN) hyperparameters using the Harmonic Mean Optimizer (HMO) for enhanced classification performance.
• To leverage Support Vector Machine (SVM) on CNN feature embeddings for sharper decision boundaries.

Main Methods:

• A hybrid model termed HMO-CNN-SVM was developed, combining CNN, HMO, and SVM.
• HMO was employed to optimize key CNN hyperparameters (learning rate, batch size, kernel configuration).
• SVM was applied to CNN feature embeddings, further refined by HMO, to enhance classification accuracy.

Main Results:

• The HMO-CNN-SVM model achieved a 95.02% accuracy on the ACS skin lesion dataset and ISIC 2018 benchmarks.
• Experiments demonstrated robust performance and consistent generalization across 5-fold cross-validation.
• The model exhibits significant parallelism potential, suitable for GPU clusters and cloud-based training.

Conclusions:

• The proposed HMO-CNN-SVM model offers a scalable and effective solution for skin lesion classification.
• The hybrid approach provides both diagnostic dependability and computational tractability for automated skin cancer detection.
• The framework is well-suited for high-performance and distributed computing environments.