Research on Assessment of Diabetic Foot Neuropathy Based on Multi-subdomain Classification Algorithm

Diabetic foot peripheral neuropathy is one of the most common complications of diabetes, which is extremely difficult to treat and carries a high risk of amputation and mortality rate. There is a need to develop a model capable of long-term monitoring, automatic, and accurate assessment, which can save medical resources and effectively prevent the occurrence of diabetic foot. Research has shown that the plantar pressure of diabetic patients will change with the condition of foot neuropathy, so foot ulcers caused by peripheral foot neuropathy can be prevented by monitoring changes in plantar pressure of diabetic patients. However, current studies in this area face the following issues: (1) The feature extraction methods for dynamic plantar pressure in current diabetic neuropathy research are relatively conventional, leading to inaccurate construction of the feature space. (2) Individual variations exist in plantar pressure data, and existing research models lack cross-subject studies, resulting in poor generalization performance. In order to solve the above problems, this paper extracts new features such as foot pressure stage and foot pressure center trajectory, which are used to construct a more complete feature space. By dividing the single domain space of the training set into five sub-domain spaces that are easier to perform classification tasks, followed by classifying the test set data into the corresponding sub-domain spaces, a multi-subdomain classification algorithm model is constructed. This algorithm achieved a cross-subject accuracy of 88.43% on a data set of 147 subjects, which was approximately 12% higher than the traditional machine learning method. This study only requires the patient’s plantar pressure data to accurately and effectively identify diabetic foot neuropathy, provide better decision support for clinicians, and improve the rehabilitation and treatment effects of diabetic patients.