DESIGNING A WEB APPLICATION FOR EXPRESS DIAGNOSTICS OF MELANOMA BASED ON THE STOLZ ALGORITHM

Korobkova Tamara

The article discusses the process of designing a web application for express diagnosis of melanoma using computer vision algorithms and a neural classifier. Based on the dermatoscopic method for determining V. Stolz's melanoma, an image with a skin neoplasm is pre-processed: a neoplasm is highlighted in the image; parameters for Stolz's formula and a total dermatoscopic score are calculated, according to which a preliminary diagnosis is calculated. Next, the optimization problem is solved to clarify the boundaries of the areas of differentiation of neoplasms and adjust the coefficients of the Stolz formula, a balanced set for training a neural network is generated. Based on the obtained parameter values, a trained neural network (multilayer perceptron) classifies the neoplasm as benign or melanoma. The approbation was carried out on two data sets, one of which is balanced. During testing, the average accuracy of the algorithm on two independent datasets is 86%.

Keywords:melanoma, Stolz algorithm, computer vision in medicine, multilayer perceptron, diagnosis of skin tumors, ABCD rule, melanoma diagnostic app.