455-P: Artificial Intelligence Approach to Treatment Classification in Painful Diabetic Neuropathy

Aim: Our study predicted treatment responses in patients with painful DPN (diabetic peripheral neuropathy) by developing a deep learning model using resting state functional magnetic resonance imaging (fMRI) neuroimaging datasets. Methods: Forty-three consecutive patients who received intravenous lidocaine treatment for painful DPN were assessed. All subjects (responders n=29 and non-responders n=14) underwent detailed clinical and neurophysiological assessments to phenotype their pain sensory profile. Subjects also underwent brain resting-state fMRI. After pre-processing we performed a group concatenated independent component analysis (ICA) set to 30 components and automatically chose 7 highly correlated (p Results: The deep learning treatment response classification in a ten-fold cross validation experiment using 7 ICA spatial maps has a mean AUC of 0.85 and an F1-Score of 0.90. However, with the extra information of all 30 ICA maps the mean AUROC increased to 0.97 with an F1-Score of 0.95. Using only pre-processed resting-state fMRI data achieved suboptimal F1-Score of 69% and AUC score of 44%. Conclusion: Through the use of our deep learning model, we have demonstrated high classification performance. Our method improves painful DPN treatment efficiency by stratifying patients to receive the correct treatment from the outset. We believe to our knowledge this is the first study utilising deep learning methods to classify treatment response in painful DPN. Disclosure K.Teh: None. I.D.Wilkinson: None. G.P.Sloan: None. S.Tesfaye: Advisory Panel; Astellas Pharma Inc., Bayer AG, Grünenthal Group, Nevro Corp., Wörwag Pharma GmbH & Co. KG, Speaker's Bureau; Eva Pharma, Pfizer Inc., Viatris Inc. D.Selvarajah: None. Funding National Instituteof Health Research Efficacy and Mechanism Evaluation Programme (NIHR 129921) European Foundation for the Study of Diabetes (Microvascular Complications Project Grant) University of Sheffield, Health Education England, Knowledge Exchange