Skin Lesion Classification on Imbalanced Data Using Deep Learning with Soft Attention.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Industrial development increases air pollution, leading to a rise in skin diseases.
• Skin cancer poses a significant health threat, with high estimated incidence and mortality rates.
• Healthcare systems face challenges with overloaded doctors and varying experience levels, necessitating diagnostic support tools.

Purpose of the Study:

• To develop an AI-powered tool for rapid and accurate skin disease diagnosis.
• To enhance diagnostic capabilities by integrating deep learning with attention mechanisms.
• To address data imbalance issues in medical datasets with a novel loss function.

Main Methods:

• Employed deep learning models (e.g., InceptionResNetV2, MobileNetV3Large) combined with Soft-Attention.
• Utilized unsupervised extraction of heatmaps for key skin lesion features.
• Incorporated patient demographic data (age, gender) into the diagnostic model.
• Proposed and implemented a new loss function to handle imbalanced datasets.

Main Results:

• The InceptionResNetV2 model with Soft-Attention achieved 90% accuracy, with precision, F1-score, recall, and AUC of 0.81, 0.81, 0.82, and 0.99 respectively.
• MobileNetV3Large, despite having fewer parameters and layers, achieved 86% accuracy.
• The MobileNetV3Large model offered a diagnostic speed 30 times faster than InceptionResNetV2.
• The proposed loss function effectively addressed data imbalance issues.

Conclusions:

• AI, particularly deep learning with Soft-Attention, shows significant promise in improving skin disease diagnosis.
• The developed models offer a viable solution for supporting clinicians, especially in resource-limited settings.
• Optimized models like MobileNetV3Large provide a balance between performance and computational efficiency for faster diagnostics.