Classification of Prediabetes and Healthy Subjects in Plantar Infrared Thermal Imaging Using Various Machine Learning Algorithms

Abstract

In the course of recent years, the size of individuals with diabetes mellitus has been dramatically increased than before. There is a need for screening and interventions which could prevent the individuals from the serious diabetic complications. Prediabetes may be a forerunner of type two diabetes mellitus, as well as a risk factor for heart illness. The body temperature is an essential parameter used for indicating the abnormal activity of human tissues. The thermal imaging primarily uses the infrared radiation emitted from the body naturally. The aim of this study was to evaluate the potential of thermography in screening the prediabetes. Sixty subjects were recruited for this study. Group I: HbA1c is <5.7%, Group II: HbA1c is 5.7–6.4%, Group III: HbA1c is >6.5%. The plantar thermograms were captured, and the temperature was measured at toe, metatarsal 1, metatarsal 3, metatarsal 5, instep and heel, respectively. The HbA1c was measured using the standard biochemical method. Three groups were categorized based on the accuracy rate obtained by five different machine learning algorithms (support vector machine, random forest, Naïve Bayes, multilayer perceptron and k-nearest neighbour). In prediabetes group, HbA1c exhibited positive correlation with measured temperature at toe region (r = 0.917, p < 0.01) and the negative relationship with measured temperature at metatarsal 1 (r = −0.474, p < 0.05), metatarsal 3 and heel regions (r = −0.895, −0.901, p < 0.01). The support vector machine has outperformed the other classifiers with good accuracy rate as 81.6%. The findings from this preliminary study indicate that measured temperature from plantar thermograms may be useful in screening the population for prediabetes.