Structuro-functional surrogates of response to subcallosal cingulate deep brain stimulation for depression

• Published in: Brain (London, England : 1878) England : 1878), 2022-03, Vol.145 (1), p.362-377
• Main Authors: Elias, Gavin J B, Germann, Jürgen, Boutet, Alexandre, Pancholi, Aditya, Beyn, Michelle E, Bhatia, Kartik, Neudorfer, Clemens, Loh, Aaron, Rizvi, Sakina J, Bhat, Venkat, Giacobbe, Peter, Woodside, D Blake, Kennedy, Sidney H, Lozano, Andres M
• Format: Article
• Language: English
• Subjects: Deep Brain Stimulation - methods; Depression; Depressive Disorder, Treatment-Resistant - diagnostic imaging; Depressive Disorder, Treatment-Resistant - therapy; Gyrus Cinguli; Humans; Retrospective Studies
• ISSN: 0006-8950; 1460-2156
• DOI: 10.1093/brain/awab284

Subcallosal cingulate deep brain stimulation produces long-term clinical improvement in approximately half of patients with severe treatment-resistant depression. We hypothesized that both structural and functional brain attributes may be important in determining responsiveness to this therapy. In a treatment-resistant depression subcallosal cingulate deep brain stimulation cohort, we retrospectively examined baseline and longitudinal differences in MRI-derived brain volume (n = 65) and 18F-fluorodeoxyglucose-PET glucose metabolism (n = 21) between responders and non-responders. Support vector machines were subsequently trained to classify patients' response status based on extracted baseline imaging features. A machine learning model incorporating preoperative frontopolar, precentral/frontal opercular and orbitofrontal local volume values classified binary response status (12 months) with 83% accuracy [leave-one-out cross-validation (LOOCV): 80% accuracy] and explained 32% of the variance in continuous clinical improvement. It was also predictive in an out-of-sample subcallosal cingulate deep brain stimulation cohort (n = 21) with differing primary indications (bipolar disorder/anorexia nervosa; 76% accuracy). Adding preoperative glucose metabolism information from rostral anterior cingulate cortex and temporal pole improved model performance, enabling it to predict response status in the treatment-resistant depression cohort with 86% accuracy (LOOCV: 81% accuracy) and explain 67% of clinical variance. Response-related patterns of metabolic and structural post-deep brain stimulation change were also observed, especially in anterior cingulate cortex and neighbouring white matter. Areas where responders differed from non-responders-both at baseline and longitudinally-largely overlapped with depression-implicated white matter tracts, namely uncinate fasciculus, cingulum bundle and forceps minor/rostrum of corpus callosum. The extent of patient-specific engagement of these same tracts (according to electrode location and stimulation parameters) also served as an independent predictor of treatment-resistant depression response status (72% accuracy; LOOCV: 70% accuracy) and augmented performance of the volume-based (88% accuracy; LOOCV: 82% accuracy) and combined volume/metabolism-based support vector machines (100% accuracy; LOOCV: 94% accuracy). Taken together, these results indicate that responders and non-responders to subcallosal cingulate deep brain stimu