Optimal Skin Cancer Detection Model Using Transfer Learning and Dynamic-Opposite Hunger Games Search.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Deep learning (DL) models are increasingly used for medical image analysis, including skin cancer detection.
• Existing DL models often struggle to achieve high accuracy in skin cancer classification.
• There is a need for more robust and accurate DL frameworks for early and reliable skin cancer diagnosis.

Purpose of the Study:

• To propose a novel deep learning framework to enhance the accuracy of skin cancer detection.
• To improve image representation learning and feature selection for better classification performance.
• To develop a more efficient and accurate system for identifying skin lesions.

Main Methods:

• Utilized a MobileNetV3 architecture for extracting relevant image representations from skin lesion images.
• Developed a modified Hunger Games Search (HGS) algorithm incorporating Particle Swarm Optimization (PSO) and Dynamic-Opposite Learning (DOLHGS) for feature selection.
• Evaluated the framework on the ISIC-2016 and PH2 datasets, comprising two and three categories of skin lesions, respectively.

Main Results:

• The proposed framework achieved an accuracy of 88.19% on the ISIC-2016 dataset.
• The model demonstrated a high accuracy of 96.43% on the PH2 dataset.
• Experimental results indicate superior performance compared to existing algorithms in terms of classification accuracy and feature optimization.

Conclusions:

• The developed framework, integrating MobileNetV3 and DOLHGS, offers a robust and accurate solution for skin cancer detection.
• The novel feature selection method effectively enhances the model's diagnostic capabilities.
• This approach represents a significant advancement in applying deep learning for improved skin cancer classification accuracy.