A skin disease classification model based on multi scale combined efficient channel attention module.

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computer Vision

Background:

• Skin disease diagnosis is challenging, often leading to high misdiagnosis rates.
• Accurate classification of skin conditions is crucial for effective clinical treatment.
• Deep learning offers a promising avenue for improving diagnostic accuracy in dermatology.

Purpose of the Study:

• To develop and validate a novel deep learning model for skin disease classification.
• To enhance multi-scale feature extraction for improved image analysis.
• To assess the model's performance on established dermatological datasets.

Main Methods:

• A deep learning model incorporating a multi-scale channel attention mechanism was designed.
• The architecture features an improved pyramid segmentation attention module for comprehensive feature extraction.
• Reverse residual structures and integrated attention modules were employed in the backbone network.

Main Results:

• The model achieved 77.6% accuracy on the ISIC2019 skin disease dataset.
• The model demonstrated 88.2% accuracy on the HAM10000 skin disease dataset.
• External validation confirmed the model's effectiveness and robustness.

Conclusions:

• The proposed multi-scale channel attention deep learning model shows significant potential for accurate skin disease classification.
• The model's performance on ISIC2019 and HAM10000 datasets validates its clinical utility.
• This approach can aid clinicians in reducing misdiagnosis rates and improving patient care.