There is a pressing need for predicting antidepressant response early in treatment to reduce patient suffering
and economic burden. Conventional antidepressants typically require two months to determine efficacy, and
two-thirds of patients will not remit (be free of depression) while on their first-line treatment. No study to date has
identified clinically useful markers to predict antidepressant response early in treatment. Therefore, the long-
of this project is to develop a predictive algorithm for antidepressant treatment efficacy early in
treatment by using noninvasive brain imaging. The
of this proposal is that brain changes,
assessed by imaging, can be used as early predictors of antidepressant response.
Magnetic resonance imaging (MRI) can provide valuable information about brain structure and function
through various techniques early in treatment that may relate to the final response to antidepressant treatment.
Even though these imaging techniques have been used to predict antidepressant response, the findings have
been inconsistent, most likely due to variable study design and small sample size, and none of the imaging
markers have been clinically validated. To fill these gaps, I will use a recently acquired imaging data from a large
sample of patients at their initiation and first week of treatment, and their depression severity was quantified
regularly by expert clinicians, to build a prediction model for antidepressant efficacy through the following aims.
1) Aim 1: Compare brain images acquired before and after antidepressant treatment to identify
regions that need to change for the treatment to be effective. I will use imaging from a moderately large data
set where patients with major depressive disorder (MDD) were imaged before and after 8 weeks of
antidepressant treatment. I will measure brain structures and their activity in individuals who got better with
treatment and analyze if there is significant difference in any brain regions in their depressive state compared to
remitted state. I will then explore those regions in a large imaging data set to see if these necessary brain
changes can be detected early in the first week of treatment.
2) Aim 2: Examine brain changes from the first week of treatment based on brain imaging and
incorporate them into a predictive model for antidepressant efficacy. I will reduce the number of features
related to brain structure and activity without losing information about the data. The selected features will be
entered in a machine learning algorithm called XGBoost, which is time-efficient and cost-effective and has been
used for detecting depression with moderate success. The model will rank features based on their contribution
to prediction of antidepressant efficacy. If treatment response is found to be unrelated to imaging, this will inform
future alternative imaging (e.g., EEG) or non-imaging (e.g., sleep, motor activity or location) studies.
: If successful, the proposed work will have broad implications for early monitoring of
antidepressant efficacy and application of an effective clinical decision-making tool for treatment planning.
Page 1 of 1