Recent mental health studies have led to an expanded depth of multimodal brain imaging data, clinical assessments and physiological data. In addition, longitudinal studies have become increasingly important to capture the trajectory of disease progression, treatment response and relapse. This wealth of datasets provides an unprecedented opportunity for crosscutting investigations. However, much-needed statistical methods for exploring discoveries are lacking. In particular, there has been very limited development of advanced statistical methods for several important objectives: decompose observed brain connectivity measures to reveal underlying neural circuits which are key biomarkers for mental disorders, effectively extract low dimensional neural features from imaging to reliably predict clinical outcomes such as treatment response, and analyze longitudinal multidimensional data including neuroimaging, clinical and behavioral assessments to study the dynamic interplay between brain and behavior changes due to treatments. In this competing renewal proposal, we will build upon the theoretical and computational framework established in our previous award to develop rigorous and computationally efficient statistical methods to address the aforementioned objectives. Specifically, we plan to develop 1) a sparse and low rank ICA (SLR- ICA) framework for reliable and parsimonious decomposition of brain connectivity measures to reveal underlying neural circuits associated with specific clinical symptoms in mental disorders; 2) an ICA-Neural Network (ICA-NN) predictive model that effectively extracts relevant low dimensional linear and non-linear neural features to predict clinical outcomes; and (3) longitudinal multidimensional data analysis tools for investigating heterogeneous changes in neural circuits due to different treatments and disease subtypes, and disentangle the relationship between changes in neuroimaging phenotypes and clinical symptoms. The statistical methods will be applied to a major NIH funded longitudinal study of major depressive disorder (MDD) to help discover neural circuits underlying specific depressive symptoms (e.g. suicidal thoughts) and differential treatment response, and ultimately help lead to more effective treatment for individual MDD patients based on his/her own neural circuitry fingerprints and behavior. We plan to replicate the findings using an independent validation cohort from an R01 study of MDD. User-friendly software will be made available to general research communities. Our proposed method developments will directly benefit mental health research by providing innovative statistical tools to effectively extract reliable and highly relevant low dimensional features from neuroimaging to deepen mechanistic understanding and improve treatment of MDD and other mental disorders.
Recent mental health studies, both cross-sectional and longitudinal, have led to an expanded depth of neuroimaging, clinical assessments and physiological data, which provide an unprecedented opportunity for crosscutting investigations that may offer new insights to mechanisms and trajectory of mental disorders. However, much-needed statistical methods for exploring discoveries are lacking. In this project, we seek to develop rigorous and computationally efficient statistical methods for neural circuitry analysis, predictive modeling and longitudinal modeling of multi-dimensional data to facilitate diagnosis, deepen mechanistic understanding and improve treatment of mental disorders.
Showing the most recent 10 out of 36 publications
