Development of deep clustering model to stratify occurrence risk of diabetic foot ulcers based on foot pressure patterns and clinical indices

Abstract

In recent years, the number of patients suffering from diabetes mellitus has continued to increase. When diabetes becomes severe, ulcers may form on the feet of the patient. In the past few years, several researchers have focused on the risk factors and avoidance of ulceration. One effective method to predict the occurrence of diabetic foot ulcers is developing a machine learning model. However, few studies combine both clinical indices and mechanical data as the attributes of the training datasets. In this study, we developed a composite model of a convolutional neural network and K-means clustering to extract features from diabetic patients with or without ulceration as well as healthy individuals. Using a deep clustering model, the center of pressure (CoP) trajectory images were divided into three clusters. Furthermore, we evaluated the performance of the clustering by extracting the features from the CoP trajectory images in each cluster and combining them with the clinical indices of the patients. The results showed that patients with ulcers when walking tend to contact the ground with a narrow area of the plantar and apply a small force. Furthermore, it was found that patients undergoing diabetic neuropathy or with a toe amputation have a high potential of suffering from ulcers.