Integration of morphological preprocessing and fractal based feature extraction with recursive feature elimination...

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Skin cancer is a prevalent global health concern.
• Early detection of melanoma and other skin malignancies is crucial.
• Differentiating visually similar skin lesions like melanoma, dysplastic nevi, and basal cell carcinoma (BCC) is challenging.

Purpose of the Study:

• To develop an integrated computer-aided method for identifying melanoma, dysplastic nevi, and BCC from dermoscopic images.
• To overcome diagnostic challenges posed by visually similar skin lesions.
• To enhance early detection and prevention of skin cancer.

Main Methods:

• Employed a recursive feature elimination (RFE) based layered structured multiclass image classification technique.
• Extracted quantitative features (shape, border, texture, color) using image processing tools.
• Utilized Gray Level Co-occurrence Matrix (GLCM) and fractal-based regional texture analysis (FRTA) for texture quantification, with Support Vector Machine (SVM) and Radial Basis Function (RBF) classifiers.

Main Results:

• Achieved high performance in skin lesion segmentation with sensitivity (0.9172), specificity (0.9788), and accuracy (0.9521).
• Image similarity indices included Jaccard (0.8562) and Dice (0.9142).
• The layered classification model achieved high accuracies: melanoma (98.99%), dysplastic nevi (97.54%), and BCC (99.65%).

Conclusions:

• The integrated system accurately quantifies features and identifies skin diseases, overcoming visual diagnostic limitations.
• This combined quantitative and qualitative analysis enhances diagnostic accuracy.
• The system provides valuable information beyond traditional qualitative assessment.