Machine learning models for diabetic neuropathy diagnosis using microcirculatory parameters in type 2 diabetes patients.

Background: Diabetic peripheral neuropathy (DPN) is a primary cause of diabetic foot, early detection of DPN is essential. This study aimed to construct a machine learning model for DPN diagnosis based on microcirculatory parameters, and identify the most predictive parameters for DPN.

Methods: Our study involved 261 subjects, including 102 diabetics with neuropathy (DMN), 73 diabetics without neuropathy (DM), and 86 healthy controls (HC). DPN was confirmed by nerve conduction velocity and clinical sensory tests. Microvascular function was measured by postocclusion reactive hyperemia (PORH), local thermal hyperemia (LTH), and transcutaneous oxygen pressure (TcPO2). Other physiological information was also investigated. Logistic regression (LR) and other machine learning (ML) algorithms were used to develop the model for DPN diagnosis. Kruskal-Wallis Test (non-parametric) were performed for multiple comparisons. Several performance measures, such as accuracy, sensitivity and specificity, were used to access the efficacy of the developed model. All the features were ranked based on the importance score to find features with higher DPN predictions.

Results: There was an overall decrease in microcirculatory parameters in response to PORH and LTH, as well as TcPO2, in DMN group compared to DM group and HC group. Random forest (RF) was found to be the best model, and achieved 84.6% accuracy along with 90.2% sensitivity and 76.7% specificity. RF_PF% of PORH was the main predictor of DPN. In addition, diabetic duration was also an important risk factor.

Conclusions: PORH Test is a reliable screening tool for DPN, which can accurately distinguish DPN from diabetics using RF.