AI Recognition in Skin Pathologies Detection

Abstract

Skin cancer is the most common type of cancer [1]. Between different malignant skin pathology melanoma is the most fleeting and mortality. Despite the superficial location of pathologies, only half of patients seek medical assistance on the early stages[2]. Treatment on the early (epidermal) stage provides a significantly higher chance of recovery. To assist a wide range of people in the early skin cancer detection, a software package was developed. The software based on deep convolutional neural networks technology. This complex allows to classify normal and malignant pathology on the uploaded photos. In clinical practice doctors use the ABCDE symptom's complex. This complex characterizes the observation of pigment spot asymmetry, border irregularities, color unevenness, diameter, and evolution [3]. The machine learning approach involves the computer evaluating similar factors when processing multiple images of different skin formations. The paper presents an algorithm for classification of skin lesions into pathology and norm using convolutional neural network architecture Xception with prior images segmentation. The upper classifying layers were frozen and new ones were added to classify skin diseases in the pre-trained neural network Xception. As a result, the classification of benign and malignant skin tumors provided at least 89% accuracy. At the moment, the result of research work is designed in form of application software that allows to download the image of pigmented skin spots from the camera. It is available on https://skincheckup.online