XAI-MedNet: A Next-Generation Explainable AI Framework for Contrast-Enhanced Skin Lesion Classification via...

Area of Science:

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

Background:

• Melanoma detection is crucial for survival, yet current automated systems face challenges with accuracy and interpretability.
• Explainable Artificial Intelligence (XAI) is vital for clinical trust in AI-driven medical image analysis.
• Early-stage diagnosis of skin cancer significantly improves patient outcomes.

Purpose of the Study:

• To develop an enhanced XAI framework for melanoma segmentation and classification.
• To improve the accuracy and interpretability of AI models in skin lesion analysis.
• To address limitations in current automated diagnostic systems, including high error margins and lack of transparency.

Main Methods:

• A novel metaheuristic contrast-stretching method was used to enhance image quality and lesion boundary clarity.
• Features were extracted from pre-trained deep models (DenseNet-201, Inception-ResNet v2, NASNet-Mobile) and fused.
• An entropy-controlled whale optimization algorithm was employed for discriminative feature selection, followed by multi-classifier classification.

Main Results:

• The proposed framework demonstrated superior performance over existing methods in accuracy, sensitivity, specificity, and F1-score.
• The approach enhanced lesion boundary distinguishability and segmentation accuracy.
• Feature refinement using entropy-controlled whale optimization resulted in a compact and informative feature set.

Conclusions:

• The developed XAI framework offers a more explainable, transparent, and accurate diagnostic pipeline for melanoma detection.
• The study successfully addressed key challenges in AI-based skin lesion analysis, improving clinical decision support.
• The integration of image enhancement, feature fusion, and optimized feature selection provides a robust approach for early skin cancer diagnosis.