Revolutionizing Skin Cancer Diagnosis with AI: Integrating YOLO-Based Lesion Detection into a Telegram Chatbot

Abstract

Skin cancer is one of the leading types of cancer globally and the early detection and treatment of skin lesions can drastically decrease mortality. Traditional diagnostic methods require specialized dermatological knowledge and facilities, which are inaccessible in remote and underserved areas. In this paper we develop an Artificial Intelligence-based skin cancer screening framework which is comprised of a YOLO-based object detection model integrated with a Telegram bot for real-time lesion analysis. Using the ISIC (International Skin Imaging Collaboration) dataset, we performed data labelling using Roboflow and developed a YOLO-based object detection model trained to detect nine classes of skin lesions namely: melanoma, basal cell carcinoma, squamous cell carcinoma, actinic keratosis, among others clinically relevant types of lesions. We then integrated our trained model into a Telegram bot which allows for the real-time assessment of uploaded lesion images. Through rigorous testing on the validation set, we report an mAP@50 of 66.4% and an accuracy of recall of 81.0%, precision of 52.5% and an F1 score of 63.7%. This work suggests a feasible approach in using Computer Vision and conversational AI for remote preliminary dermatological screening.