326-OR: A Novel Machine Learning Analysis of Brain Multimodal Magnetic Resonance Imaging Classifies Painful Diabetic Neuropathic Pain Severity with High Accuracy

• Published in: Diabetes (New York, N.Y.) Vol. 68; no. Supplement_1
• Main Authors: SELVARAJAH, DINESH, TEH, KEVIN, SLOAN, GORDON P., SHILLO, PALLAI RAPPAI, WILKINSON, IAIN D., TESFAYE, SOLOMON
• Format: Journal Article
• Language: English
• Published: New York American Diabetes Association 01.06.2019
• Subjects: Accuracy; Analgesics; Artificial intelligence; Biomarkers; Clinical trials; Diabetes; Diabetes mellitus; Diabetic neuropathy; Functional magnetic resonance imaging; Learning algorithms; Machine learning; Neural networks; Neuroimaging; NMR; Nuclear magnetic resonance; Pain; Structure-function relationships
• ISSN: 0012-1797; 1939-327X
• DOI: 10.2337/db19-326-OR

Using advanced MR neuroimaging we have demonstrated altered brain structure and functional connectivity that could serve as a potential Central Pain Signature (CPS) for painful diabetic neuropathy (DN). However, the key challenge is how to apply this potential biomarker for routine diagnostic purposes. The aim of this study was to examine if the CPS can accurately classify painful DN patients based on a patient’s pain severity. Methods: 53 painful DN patients underwent detailed clinical and neurophysiological assessments. The NTSS-6, a validated questionnaire that measures both the frequency and the intensity of neuropathic pain was used to assess pain severity. Patients were divided into high pain (NTSS-6 score >7) and low pain (<7). All subjects underwent brain volume and resting state functional MRI at 3T. We used novel and validated machine learning algorithms [support vector machines] to differentiate pain severity based on functional connectivity and structural/volume data. A nested cross validation method was used to determine the accuracy (sensitivity and specificity). Results: There was no age or gender difference (p > 0.05) between the high and low pain groups. The CPS classified painful DN patients based on their pain intensity with 94% accuracy (AUC 0.98). The positive and negative predictive values were 0.80 and 1.00 respectively. The F1 scores for predicting high pain and low pain were 0.89 and 0.96 respectively. Brain regions identified as the best classifier were the left and right postcentral gyri, thalami, and anterior and posterior cingulate cortices. Conclusions: This novel study demonstrates that a simple, 15-minute MR brain scan can accurately classify painful DN patients according to pain intensities with high accuracy. This assessment tool has a great potential as a biomarker of diabetic neuropathic pain and may serve as a target for future trials of analgesic compounds for painful DN.