Microscopic skin laceration segmentation and classification: A framework of statistical normal distribution and...

Area of Science:

• Medical Image Processing
• Computer Vision
• Oncology

Background:

• Skin cancer diagnosis relies heavily on dermoscopic images.
• Early detection of skin cancer significantly reduces mortality rates.
• Automated systems offer potential for improved diagnostic accuracy and efficiency.

Purpose of the Study:

• To propose an automated system for skin lesion detection and classification.
• To enhance early diagnosis of skin cancer using dermoscopic images.
• To improve the accuracy of skin cancer detection through advanced image processing techniques.

Main Methods:

• Utilized computer vision (CV) techniques for analyzing dermoscopic images.
• Implemented a statistical normal distribution approach for image segmentation.
• Employed optimized histogram, color, and Gray Level Co-occurrence Matrix (GLCM) features with covariance-based fusion.
• Applied a binary grasshopper optimization algorithm for optimal feature selection.
• Evaluated classification performance using Support Vector Machine (SVM) with various kernel functions.

Main Results:

• Achieved average segmentation accuracies of 93.79% (PH2) and 96.04% (ISBI 2016).
• Obtained classification accuracies of 93.80% (ISBI 2016) and 93.70% (ISBI 2017).
• The cubic SVM kernel function yielded the best classification accuracy.
• The proposed system demonstrated superior performance compared to existing methods on benchmark datasets.

Conclusions:

• The developed automated system shows high accuracy in skin lesion detection and classification.
• The integration of statistical methods and optimal feature selection enhances diagnostic capabilities.
• This approach holds promise for improving early skin cancer diagnosis and patient outcomes.