A novel method for screening diabetic peripheral neuropathy using fused surface electromyogram signal and mechanomyography signal.

Abstract

Diabetes peripheral neuropathy (DPN), one of the most common complications in people with diabetes, can seriously undermine their quality of life. Early detection and treatment of DPN is of great significance to the diabetes population. Nerve conduction studies, the gold standard for diagnosing DPN, causes substantial discomfort for people and requires specialized personnel and expensive equipment, making it challenging to implement as a mass screening tool for DPN. Here, a novel, non-invasive and convenient screening method for DPN is proposed. In the proposed method, surface electromyography and mechanomyography data are acquired in a non-invasive and painless manner from people' dorsalis pedis muscles. The acquired data are then processed by means of downsampling, motion data extraction, and subsequently converted into images, which are utilized for diagnosing DPN or non-DPN by a convolutional neural network. The proposed method is developed based on actual data from 167 people with diabetes. After 4-fold cross validation of the method, the mean accuracy, sensitivity and specificity are evaluated to be 96.15 %, 91.39 % and 98.78 % with variances of 0.003 %, 0.017 % and 0.00014 %, respectively. Furthermore, the method is preliminary tested on 21 people with diabetes, resulting in accuracy, sensitivity and specificity of 95.48 %, 90.91 % and 98.88 %, respectively. Notably, the screening process for a single diabetic using this method can be completed in under 10 min. The results above demonstrate the efficacy of the method in diagnosing DPN. The proposed method has the considerable potential for noninvasive and convenient screening of DPN without requiring professionals or expensive equipment.