Open Collections

Artificial neural network based prediction of treatment response to repetitive transcranial magnetic stimulation for major depressive disorder patients

University of British Columbia

Major Depressive Disorder (MDD) is a severe medical condition that affects thousands of people every year. Therapy in MDD includes medication and psychotherapy, and is prescribed on the basis of the type and severity of depressive episodes. Treatment-resistance is common among MDD patients. Repetitive Transcranial Magnetic Stimulation (rTMS) is a form of deep brain stimulation used for relieving depressive symptoms. Due to its high cost and lengthy procedure, it’s reserved for patients showing treatment-resistance to at least 2 trials of antidepressants. Of all MDD patients, only 50% show response to rTMS, which leads to unnecessary patient frustration and additional costs. Prediction of resistance to rTMS treatment can thus help physicians decide on the best treatment course for each patient. This thesis presents a machine-learning based clinical assistive tool that predicts the probability of a patient to respond to rTMS treatment and if so, predict the probability whether they are likely to achieve remission. The most relevant clinical and sociodemographic variables associated with predicting treatment outcomes were selected on the basis of importance scores ranked using a Random Forest (RF) algorithm, and an elaborative analysis of their significance was presented. The most important variables were fed into a Deep Artificial Neural Network (DANN) for classification of patients who will respond to rTMS treatment. Two DANN variants were designed, trained, optimized and tested to predict each of rTMS treatment response and remission outcomes. Our model is based on the pre-treatment clinical and sociodemographic data which had been collected from 414 patients diagnosed with treatment-resistant MDD. Results show that our DANN model outperforms existing clinical procedures and yields an accuracy of 84.4% in predicting remission and 73.8% in distinguishing responders form non-responders. Additionally, a thorough evaluation and comparison with other methods that have used machine learning algorithms to predict rTMS treatment outcome was carried and discussed in detail. Findings in this thesis signify the potential of individual-based assessments that can improve rTMS treatment procedure.

Text

2019-01-04

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/68184

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/