Detection of Skin Cancer Using SVM, Random Forest and kNN Classifiers.

Area of Science:

• Dermatology
• Medical Imaging
• Computational Biology

Background:

• Melanoma is the deadliest form of skin cancer, necessitating early and accurate diagnosis.
• Identifying melanocytes in the epidermis is crucial for melanoma detection and treatment.
• Early-stage melanoma is curable, highlighting the importance of effective diagnostic tools.

Purpose of the Study:

• To implement and evaluate image segmentation and feature extraction techniques for melanoma classification.
• To compare the performance of k-Nearest Neighbor (kNN), Random Forest, and Support Vector Machine (SVM) classifiers for skin lesion diagnosis.

Main Methods:

• Watershed segmentation was applied to segment skin lesion images.
• Features including shape, the ABCD rule, and Gray-Level Co-occurrence Matrix (GLCM) were extracted.
• kNN, Random Forest, and SVM classifiers were employed for classification.

Main Results:

• Feature extraction successfully identified key characteristics of skin lesions.
• The Support Vector Machine (SVM) classifier demonstrated superior performance in classifying skin lesions compared to kNN and Random Forest.
• The implemented methods show promise for automated melanoma detection.

Conclusions:

• The combination of watershed segmentation, feature extraction (shape, ABCD, GLCM), and SVM classification offers a robust approach for melanoma diagnosis.
• SVM provides a reliable method for classifying skin lesions, aiding in the early detection of melanoma.
• This automated approach can support dermatologists in improving diagnostic accuracy and patient outcomes for skin cancer.