Diabetes mellitus early detection using deep learning on fingernail images

Abstract

Diabetes Mellitus (DM) is a global health concern, necessitating early diagnosis to mitigate its long-term complications. This paper presents a novel approach to DM diagnosis through fingernail image analysis, leveraging Deep Learning (DL) techniques. By focusing on color, texture, and geometric features of nails, this study proposes a non-invasive, accessible, and scalable diagnostic tool. A comparison of several standard Convolutional Neural Network (CNN) models were conducted as the classification process, distinguishing between non-diabetic and diabetic categories. Then ultimately a transfer learning model of ResNet50 architecture was finetuned to obtain an accuracy of 80.6%. This approach has shown the potential to address the limitations of traditional methods, emphasizing early detection, affordability and patient convenience.