From Pixels to Diagnoses: Deep Learning in Diabetes Detection

Abstract

Diabetes mellitus is a chronic metabolic disease with a rising prevalence worldwide, affecting millions of individuals. Early detection and accurate classification of diabetes are crucial to reduce mortality rates and enhance the quality of life for affected individuals. Traditional diagnostic techniques for diabetes, such as blood testing and glucose tolerance tests, are costly, time-consuming, and often require substantial resources.This study proposes a deep learning model that utilizes the Pima Indian Diabetes dataset, consisting of health information from 768 individuals with attributes including blood pressure, glucose levels, BMI, etc. The aim is to overcome the limitations of traditional detection methods and develop a model capable of early detection and precise classification of diabetes.The classification of the data into diabetes and non-diabetic groups is done using a convolutional neural network (CNN) model. The experimental outcomes show the effectiveness of the proposed deep learning model, obtaining an accuracy of 94.2%, precision of 90.18%, recall of 98.9%, F1-score of 94.3%, Cohen's kappa of 88.5%, and ROC AUC of 94.4%. These findings indicate that a deep learning approach can be utilized to develop a model capable of accurately identifying diabetes in its early stages.Early identification of diabetes through the suggested deep learning framework holds promise for reducing the risk of complications associated with the disease. By leveraging the power of deep learning techniques, healthcare professionals can enhance their ability to detect and manage diabetes more efficiently, leading to improved patient outcomes and an overall reduction in the burden of this chronic condition.