Game-based assessment of peripheral neuropathy in patients with diabetes by combining sensor-equipped insoles with video games and machine learning algorithms

Abstract

Diagnosis of diabetic peripheral neuropathy (DPN) is essential to prevent complications, such as the diabetic foot syndrome. Diagnosis mostly relys on a time-consuming clinical examination by standardized procedures (pinprick test, vibration perception, Tip Therm, reflexes, muscle function). Furthermore, investigator-related bias confounds findings. To explore the potentials of a video game-based approach to diagnose polyneuropathy, a gaming platform (“Gamidiagnostics”) was set up. Participants utilized pressure sensor- equipped insoles as control units and played four games that were specifically designed to test for reaction time, sensation, skillfulness, endurance, balance, and muscle strength. A pilot study with 71 healthy volunteers and 112 patients diagnosed with DPN by clinical examination (neuropathy deficit score, NDS) evaluated the feasibility of this approach. Unbiased training of prediction algorithms with data sets identified 15 independent variables with discriminatory functions that indicated DPN. In age-matched cohorts, the support vector machines achieved a training accuracy of 87.8% (AUC-ROC 0.91) and an adjusted accuracy of 85.2% on a held- out testing data set (sensitivity 92.6%, specificity 77.8%). Distinct variables were identified for each nerve fiber deficit and allowed correct classification with adjusted accuracies of 88.1%, 91.9%, and 95.3% for Achilles tendon reflex, Aδ-/C-fiber, and Aβ-fiber impairment, respectively. Thus, a video game-based approach with smart footwear sensors was able to diagnose advanced peripheral nerve malfunction with high accuracy. This was set up in an examiner- independent manner and may be established as telemedical device.