Attention Cost-Sensitive Deep Learning-Based Approach for Skin Cancer Detection and Classification

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Deep learning models struggle with imbalanced datasets common in rare skin disease detection.
• Existing models often fail to identify subtle, small affected skin regions in medical images.
• Accurate and early detection of skin cancer is crucial for effective treatment outcomes.

Purpose of the Study:

• To develop an advanced deep learning model for precise skin cancer detection and classification, addressing limitations with rare diseases.
• To enhance the model's ability to identify small, affected skin areas and manage data imbalance.
• To create a robust computer-aided diagnosis (CAD) tool for early skin cancer identification.

Main Methods:

• An attention-cost-sensitive deep learning ensemble meta-classifier was developed.
• Cost weights were integrated to mitigate data imbalance during training.
• Attention mechanisms were incorporated for optimal feature learning from small skin lesions.
• Kernel-based Principal Component Analysis (KPCA) reduced feature dimensionality before fusion and classification.

Main Results:

• The proposed approach achieved 99% accuracy for both skin disease detection and classification.
• It demonstrated a significant performance improvement over existing methods, with 4% and 9% higher accuracy in detection and classification, respectively.
• The model effectively handled imbalanced datasets and identified subtle disease indicators.

Conclusions:

• The attention-cost-sensitive deep learning ensemble meta-classifier offers a highly accurate solution for skin cancer detection and classification.
• This approach significantly improves upon existing methods, particularly for rare skin diseases.
• The developed model shows promise as a valuable computer-aided diagnosis (CAD) tool in clinical settings for early skin cancer diagnosis.