Detection of Foot-Ulcer from Digital Photographs using MobileNet Variants with Features Fusion

Diabetes is a chronic illness usually caused by elevated blood glucose levels. If left untreated, diabetes can have serious repercussions. Diabetic foot ulcers (DFUs) is a chief complications of diabetes, which can result in foot wounds and, if left untreated, can amputation of a limb or leg. A visual examination by the physician is typically used to identify the DFU at the clinical level and assess its severity. This research aims to create a computer program that can identify DFU from digital photos taken using a camera. The stages of the developed technique are as follows: gathering and processing DFU images, extracting features using a selected deep-learning (DL) scheme, reducing and fusing features, and classifying the results using five-fold cross-validation. In this research, the benchmark DFU images were taken into consideration, and MobileNet (MN) and its variations were used to carry out the classification task. Using the individual deep-features, the proposed DFU detection task is first carried out. Afterwards, the selected features are serially fused to obtain a new feature vector, and the proposed work is repeated. The study's experimental results validate the developed technique's merit, as the K-Nearest Neighbor classifier achieves 97% detection accuracy.