642-P: Diagnosing Peripheral Neuropathy with a Remote Temperature Monitoring Mat

The aim of this study was to derive and validate a model to identify peripheral sensory neuropathy using remote temperature monitoring (RTM). We used data from a recent longitudinal study of a once-daily in-home RTM mat. Neurological status was assessed via 10g monofilament. We trained and cross-validated (repeated k-fold sampler) machine learning classifiers (support vector machine with radial basis function kernels). Accuracy was defined as the area under the receiver operator curve (AUC). We report on three final models that assess neuropathic status: (1) foot temperatures at the six plantar locations commonly used for RTM; (2) high-resolution foot temperature maps (thermograms); (3) high-resolution foot shape without associated temperatures. A total of 154 feet were included, of which 30 were insensate (19.5%). The AUCs of these three models were 0.78±0.13 (good), 0.92±0.07 (excellent), and 0.54±0.16 (poor), respectively. The results suggest the diagnostic usefulness of thermograms, which is supported by the fact that analysis feature importance of model 2 shows high diagnostic value for temperatures in the medial midfoot and distal toes. The latter is possibly due to vasodilation of the medial plantar artery, while the former can be explained by the progressive nature of neuropathy from distal to proximal. That model 3 had nearly no predictive value suggests that foot deformity secondary to motor neuropathy is not contributing diagnostic value. Interestingly, 22 feet showed neuropathic progression over the 34-week study. Results from model 1 reflected these changes when assessed at enrollment and at study completion for these feet. These results suggest the RTM mat may be useful for assessing peripheral neurological function and progression in patients with diabetes. These diagnostic models may also inform improved RTM standard and machine learning algorithms for predicting diabetic foot complications secondary to neuropathy.