Automated Detection of Diabetic Foot Using Thermal Images by Neural Network Classifiers

Diabetes is one of the major problems worldwide. It is a metabolic disease where the improper management of
blood glucose level lead to the risk of many diseases like foot complications, amputation, kidney diseases etc.
Early detection of diabetes and their precursors is essential in preventing their devastating consequences such
as foot infection and amputation. There are various methods to diagnose diabetes which are invasive
techniques. Infrared thermography is a promising modality for such a system from which diabetes is noninvasively detected using the thermal foot images. The temperature differences between corresponding areas
on feet are the clinically significant parameters. The thermal images are preprocessed, segmented using kmeans clustering and then the textural features (GLCM) are extracted and then classified using classifiers. In
this paper, to diagnose diabetic foot, three models like Probabilistic Neural Network(PNN), K-nearest
Neighbor Network(KNN), and Support vector machine(SVM) are described and their performances are
compared. Experimental results show that KNN has an accuracy of 95.66 %. This infers that KNN model
outperforms the other two models.