Neural Classifier of Deep Brain Stimulation Evoked Emotion

Vineet R Tiruvadi

Keywords: Deep Brain Stimulation, Vector Machines, Classifier, Neural Recordings, Dense EEG, Dynamics

TL;DR: We build a machine learning (Regularized SVM) classifier on a deep-phenotyped set of dense scalp (EEG) neural recordings from depression patients treated successfully for psychiatric depression.

Abstract:

Precise deep brain stimulation (DBS) of Subcallosal Cingulate White Matter (SCCwm) alleviates symptoms of treatment resistant depression (TRD). Objective signatures from neurl recordings are needed to optimize implantation and programming of antidepressant brain stimulation, and recent advances in machine learning help identify these in noisy patient recordings.
In this study, we present a machine learning classifier build from previously reported dense-array scalp EEG taken during active DBS at therapeutic (OnTarget) and non-therapeutic (OffTarget) targets. Using combined emotion self-report and EEG measurements alongside OnTarget stimulation of SCCwm and OffTarget stimulation \SI{1.5}{\milli\meter} away, we trained a \textit{support vector machine} (SVM) capable of confirming precise stimulation. We demonstrate that the learned model coefficients align with \textit{engaged tractography} predicted through volume of tissue activated (VTA) modeling, and that these VTA models select for informative streamlines in large-scale Human Connectome Project (HCP) structural connectomes. Our compound model will enable implementation, study, and improvement of adaptive SCCwm-DBS, particularly in TRD, more systematic. The classifier is released open source to the community for further validation, refinement, and extension; the dataset is released as a part of a multimodal foundation for antidepressant DBS.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 22778