ENHANCING SKIN CANCER DETECTION THROUGH ADVANCED IMAGE SEGMENTATION AND FEATURE EXTRACTION

Abstract

Improving patient outcomes via early identification and therapy depends on accurate skin cancer identification and categorization. To distinguish between skin cancer, non-cancerous lesions, and healthy skin, this research investigates a method for picture segmentation and variable extraction. For precise lesion boundary delineation, the segmentation step makes use of MorphACWE and MorphGAC, as well as thresholding, identification of edges, clustering, and semantic segmentation. Variable extraction is carried out utilizing both conventional and cutting-edge techniques after segmentation. HOG as well as LBP are employed for color and texture variable extraction, respectively. As for texture analysis, GLCM and Gabor filters are employed. Lesions' complex patterns and geometries may be captured by shape descriptors such as Hu moments as well as Zernike moments. Employing these all-encompassing methods, skin disorders are identified and categorized according to their unique morphological and textural traits. By combining these techniques, we want to improve the accuracy of skin cancer identification algorithms using ABL prediction model, which will lead to more informed clinical decisions and improved treatment for skin conditions with with the highest accuracy (97.3%), sensitivity (96.8%), specificity (97.9%), F1-score (96.9%), and AUC (0.98), while maintaining a low–moderate computational cost. The significance of integrating several image processing methods to enhance the precision and dependability of skin lesion identification is highlighted by this multi-pronged strategy.