An attention based optimized network for the classification of skin lesions

Area of Science:

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

Background:

• Skin lesions require early identification to prevent progression to skin cancer.
• Limited dermatological resources and healthcare access hinder timely diagnosis, especially in rural areas.
• Accurate skin lesion classification is vital for effective early intervention and treatment.

Purpose of the Study:

• To propose a novel and optimal method for classifying diverse skin lesions using deep learning and optimization.
• To enhance the accuracy and efficiency of automated dermatological diagnosis systems.
• To improve early detection and treatment of skin cancer through advanced image analysis.

Main Methods:

• Utilized the RegNetY032 model with a modified classification head for feature extraction and classification of dermoscopy images.
• Integrated a Soft Attention Block to focus on salient lesion features and ignore artifacts like hair and veins.
• Employed Harris-Hawks Optimization (HHO) for hyperparameter tuning to boost classification performance.

Main Results:

• The proposed HHO-Reg-SA-Net achieved a superior classification accuracy of 99.27% on the HAM10000 benchmark dataset.
• Demonstrated the model's robustness in discerning and prioritizing critical features of skin lesions.
• Successfully minimized the impact of common artifacts in dermoscopy images.

Conclusions:

• The HHO-Reg-SA-Net model shows significant promise for advancing automated dermatological diagnosis.
• This approach can contribute to improved patient care and earlier intervention in skin cancer detection.
• The integration of deep learning and optimization techniques offers a powerful tool for medical image analysis in dermatology.