Support vector regression correlates single-sweep evoked brain potentials to gastrointestinal symptoms in diabetes mellitus patients

Diabetes mellitus (DM) is a multi-factorial and complex disease causing autonomic neuropathy and gastrointestinal symptoms in some patients. The neural mechanisms behind these symptoms are poorly understood, but it is believed that both peripheral and central mechanisms are involved. To gain further knowledge of the central mechanisms, the aim of this study was to identify biomarkers for the altered brain activity in type-1 DM patients compared to healthy volunteers (HV), and to correlate the obtained biomarkers to clinical patient scores. The study included 14 DM patients and 15 HV, with brain activity recorded as multi-channel electroencephalography evoked brain potentials (EPs) elicited by painful electrical stimulations in the esophagus. The single-sweep EPs were decomposed by an optimized discrete wavelet transform (DWT), and averaged for each channel. The DWT features from the DM patients were discriminated from the HV by a support vector machine (SVM) applied in regression mode. For the optimal DWT, the discriminative features were extracted and the SVM regression value representing the overall alteration of the EP was correlated to the clinical scores. A classification performance of 86.2% (P=0.01) was obtained by applying a majority voting scheme to the 5 best performing channels. The biomarker was identified as decreased theta band activity. The regression value was correlated to symptoms reported by the patients (P=0.04). The methodology is an improvement of the present approach to study central mechanisms in diabetes mellitus, and may provide a future application for a clinical tool to optimize treatment in individual patients.