A Pilot Study for Profiling Diabetic Foot Ulceration Using Machine Learning Techniques

Abstract

Early diagnosis of risk of diabetic foot ulceration (DFU) may allow for earlier care to avoid foot ulcers, amputation, and death. Thermography is a non-invasive imaging technique that is used to detect thermal changes in the diabetic foot. This study illustrates the comparative analysis of ML techniques (KNN, Naïve Bayse, Decision Tree, Random Forest, Logistic Regression, SVM, Ada Boost) when performed on the publicly available thermogram database with 122 diabetic and 45 non-diabetic subjects. SVM and Random forest surpass other machine learning techniques in terms of accuracy. Age and IMC (Body Mass Index, in French its Indice de Masse Corporelle–IMC) are the major factors contributing to the accurate profiling of the patients in DFU. In this pilot study, five cases have been considered on the dataset used with different training and testing splits. The comparative analysis done in this paper shows that a 70: 30 ratio of train and test data is giving optimum generalized results. The paper provides a detailed description of the importance of features corresponding to various machine learning techniques. The paper concludes that Random Forest and SVM are performing better in comparison to all the machine learning techniques. This analysis helps to understand the importance of various features and hence can help in localizing the foot regions which are more sensitive towards ulcer formation.