Skin Melanoma diagnosis using machine learning and ABCD rule

Skin cancer, being the most predominant type of malignant growth, represents more than half of all cancer instances. While melanoma is less common than other skin cancers, it poses the highest risk due to its tendency to metastasize and potentially result in mortality. Nonetheless, early detection significantly increases the chances of curing melanoma. The detection of melanoma poses a considerable challenge, primarily attributed to the minimal interclass variance observed between images of melanoma and non-melanoma, alongside the heightened interclass variance within melanoma images. Hence, it is imperative to formulate a Computer-aided Diagnosis (CAD) to identify such lesions and facilitate early disease diagnosis to enhance the probability of patient survival. The proposed methodology consists of a sequence of four distinct phases. The initial stage involves preprocessing, which incorporates the application of filtering and contrast enhancement algorithms such as the Median filter, Dull razor, and Histogram Equalization. The segmentation process proceeds by utilizing Otsu methods in order to detect any anomalies. Subsequently, the following phase involves the extraction of features through GLCM for textural attributes and ABCD base parameters, encompassing asymmetry, border irregularity, color, and diameter. Following this, a model based on artificial neural networks (ANN) is developed by leveraging these extracted traits to discern instances of skin cancer within the provided dataset. This methodology achieves a commendable overall accuracy rate of 99%. Dermatologists may derive significant assistance in their clinical decision-making procedures from this promising breakthrough.