Classification of Skin Cancer Segmentation using Hybrid Partial Differential Equation with Fuzzy Clustering based on Machine Learning Techniques

Abstract

Skin cancer, particularly melanoma skin cancer is a serious public health concern today. The majority of skin cancers may be treated if caught early enough. The fast proliferation of skin cancer necessitates the development of an automatic computerized detection system for skin cancer in its primary phases. The visual qualities of many skin cancer images are similar. The process of extracting characteristics from skin cancer image is a difficult one. The automated computerized diagnosis process aids dermatologists in improving the accuracy of skin disease analysis, allowing them to save diagnostic time and provide better therapy for their patients. This research study presents an intelligent method for detecting malignant and non-cancerous cells using image processing techniques. The Median Filter is first used to reduce artifacts, skin color, hair, and other characteristics of the produced images by removing noise from the skin lesion. The lesion section is then segmented individually using the suggested Hybrid Partial Differential Equation with Fuzzy Clustering (HPDE-FC) technique, which is also effective for feature extraction. Asymmetry, Border, Color and Diameter characteristics are extracted using ABCD scoring approach. Using several machine learning approaches like K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF) and Nave Bayes (NB) classifiers, the collected features are immediately fed to classifiers to categorize skin lesions amongst cancerous (malignant) and non-cancerous(melanoma). This research work, 325 images of normal skin lesions as well as 572 images of malignant skin lesions are downloaded from the International Skin Imaging Collaboration (ISIC) for this study. Using SVM classifiers, a classification result of 97.7% accuracy is obtained. Our goal is to evaluate the suggested segmentation technique’s performance, excerpt the most applicable features and associate the categorization results against those of other algorithms in literature.