A None-Invasive Diabetes Detection Machine Learning-Based Tongue Image Analysis Tool

Diabetes is a pervasive global health issue, affecting millions of individuals worldwide. Early detection is critical for effective management and preventing severe complications. However, the invasiveness of conventional blood-based testing methods often discourages regular screening, leading to delayed diagnosis and intervention. Traditional diagnostic methods for diabetes typically involve invasive procedures, such as blood sampling, which can be uncomfortable and inconvenient for patients. This impediment to regular screening can result in delayed diagnosis and less effective disease management. This article proposes a non-invasive, machine-learning-driven diagnostic tool that can detect diabetes by analyzing tongue images. This solution is based on the hypothesis that the tongue exhibits physiological changes associated with diabetes, such as alterations in color and texture. The article trains the models on a large set of labeled tongue images using two main machine-learning algorithms: support vector machines (SVM) and k-nearest neighbors (KNN). This allows the models to correctly identify diabetic patients. The classification process showed that both SVM and KNN demonstrate the efficacy of the proposed approach. The SVM model's performance metrics and accuracy were precision 98%, recall 87%, F1_score 92%, and accuracy 91.9%, whereas the KNN classification report provides similar insights for the KNN model with precision 96%, recall 94%, F1_score 95%, and accuracy 94.9%. Our findings underscore the potential of this non-invasive diagnostic tool to transform diabetes detection, making it more accessible and less burdensome for patients. This innovative work seeks to transform diabetes diagnosis by providing a practical and comfortable alternative that improves early detection and, in turn, improves patient safety and satisfaction.