Machine Learning to Predict Clinical Response to TMS

Major Depressive Disorder (MDD) is a common and debilitating illness. It affects a person's
family and personal relationships, work, education, and life. It changes sleeping and eating
habits and significantly impairs patients' general health. The disorder affects Veterans more
than the general population, both as an isolated illness and in conjunction with
posttraumatic stress disorder (PTSD) and suicidality. Symptoms in a notable proportion of
patients (~30%) do not respond to behavioral and pharmacological interventions, and new
treatments are in great need. One such treatment, transcranial magnetic stimulation (TMS),
has been cleared by Food and Drug Administration for treatment in MDD. TMS is effective in
around 60% of patients with treatment-resistant MDD but is associated with significant
financial and time burden. Further insights into the neurobiological effects of TMS and
markers for functional recovery prediction and treatment progression are of great value.

The goal of this proposal is to use human electrophysiology (electroencephalography,
hereafter EEG, in particular) and machine learning to predict treatment response in
candidates for TMS treatment and also study TMS's mechanism of action. Doing so has several
benefits for patients, as prediction of treatment helps providers in screening out the
patients for whom TMS is ineffective and understanding the mechanism allows us to refine and
individualize the treatment.

The investigators will recruit 35 patients with treatment-resistant MDD and record resting
state EEG signal with a dense electrode array before and after a 6-week clinical course of
TMS treatment. The investigators will use machine learning (Sparse regressions) to predict
treatment outcome using functional connectivity (Coherence) maps derived from the EEG signal.
The investigators also will use classifiers to track changes in functional connectivity
through the course of treatment. Based on our preliminary data, the investigators hypothesize
that weaker functional connectivity between prefrontal cortex (where the stimulation is
delivered) and parietal/posterior midline sites predict better response to treatment and that
TMS treatment will enhance these connections.

The data collected here would be used as a seed and preliminary data for future federal (NIH
and the VA) career development awards which will focus on the use of EEG to better understand
brain function and neuromodulation treatments.