Deep Learning Based Skin Lesion Segmentation and Classification of Melanoma Using Support Vector Machine (SVM)

Area of Science:

• Medical image analysis
• Artificial intelligence in dermatology
• Computational pathology

Background:

• Melanoma diagnosis relies on accurate skin lesion analysis.
• Deep learning offers potential for automated diagnostic tools.
• Segmentation is crucial for feature extraction in dermoscopy images.

Purpose of the Study:

• To improve melanoma classification accuracy through deep learning-based segmentation.
• To assist medical experts in the early diagnosis of melanoma using dermoscopy images.
• To evaluate the effectiveness of the U-Net algorithm for skin lesion segmentation.

Main Methods:

• Utilized a Convolutional Neural Network (CNN) U-Net algorithm for automatic skin lesion segmentation.
• Extracted color, texture, and shape features using Local Binary Pattern (LBP), Edge Histogram (EH), Histogram of Oriented Gradients (HOG), and Gabor methods.
• Classified segmented lesions as melanoma or benign using Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbor (KNN), and Naïve Bayes (NB) classifiers.

Main Results:

• Achieved a Dice coefficient of 77.5% for image segmentation.
• The Support Vector Machine (SVM) classifier yielded an accuracy of 85.19% for melanoma diagnosis.
• Demonstrated the effectiveness of the proposed deep learning approach in improving classification performance.

Conclusions:

• The U-Net segmentation algorithm is highly effective in deep learning environments for skin lesion segmentation.
• Improved segmentation significantly contributes to enhanced melanoma classification accuracy.
• The integrated approach aids in the early and accurate diagnosis of melanoma.