Skin Lesion Segmentation and Multiclass Classification Using Deep Learning Features and Improved Moth Flame...

Area of Science:

• Dermatology
• Computer Vision
• Artificial Intelligence

Background:

• Manual skin cancer diagnosis is slow and costly.
• Automated methods are needed for accurate multiclass skin lesion classification.

Purpose of the Study:

• To develop a fully automated approach for skin lesion segmentation and classification.
• To improve diagnostic accuracy using deep features and optimized selection.

Main Methods:

• Image enhancement using local color-controlled histogram intensity values (LCcHIV).
• Deep Saliency Segmentation with a custom 10-layer CNN for saliency estimation.
• Feature extraction via a deep pre-trained CNN, followed by improved moth flame optimization (IMFO) for feature selection.
• Feature fusion using multiset maximum correlation analysis (MMCA) and classification with Kernel Extreme Learning Machine (KELM).

Main Results:

• High segmentation accuracy achieved across multiple datasets: ISBI 2016 (95.38%), ISBI 2017 (95.79%), ISIC 2018 (92.69%), and PH2 (98.70%).
• Classification accuracy of 90.67% on the HAM10000 dataset.
• Demonstrated effectiveness through comparison with state-of-the-art techniques.

Conclusions:

• The proposed automated method offers a robust and accurate solution for multiclass skin lesion segmentation and classification.
• The integration of deep features, optimized selection, and advanced classification techniques significantly enhances diagnostic capabilities.