Schizophrenia (SZ) and bipolar disorder (BP) are two of the most challenging and costliest mental disorders in terms of human suffering and societal expenditure. Clinically, SZ and BP can present with similar symptomology during acute psychotic periods, raising issues of differential diagnosis, frontline medication regime, and treatment planning. Currently there are no definitive biological markers for either diseases, and their diagnosis relies upon longitudinal symptom assessment. Several studies have been published which compared SZ and BP within a single modality such as fMRI, sMRI, EEG, and DTI, and have identified brain alterations that discriminate the two conditions. However, this work has been hampered by small sample sizes, limited re-test reliability and general replicability. Each brain imaging technique provides a different view of brain function or structure, while multimodal fusion capitalizes on the strength of each and may uncover the hidden factors that can unify disparate findings. Here we seek to replicate and extend the search for biomarkers to reliably differentiate SZ from BP by using N-way multimodal fusion, e.g., fMRI, DTI, and sMRI data, which is expected to improve the group-differentiating ability beyond any single modality. We will develop a novel multivariate model and release a user-friendly toolbox, which enables people to combine multiple modalities freely, explore the joint information accurately and examine the relationship between brain patterns and clinical measures smartly, such as symptom scores etc. Another aim of this proposal is to study the trait versus state effect of SZ and BP, using longitudinal data and in a 3-way fMRI-DTI-sMRI fusion. We will access data from patients who were scanned immediately after discharge and again 5-7 weeks later. This time period is when clinicians have the most difficulties in distinguishing SZ from BP. Such a valuable dataset along with the use of a cutting-edge joint analysis model, will enable us to investigate multiple group-discriminating factors and the traits which may serve as potential biomarkers of SZ or BP. In addition, the modalities (and their combinations) will be ranked according to their ability to distinguish groups, resulting in a modal selection preference. We will further evaluate whether there are natural clusters in multimodal data that provide evidence compatible the clinical diagnoses and attempt to classify patients at the level of individual psychiatric patients based on the selected group-discriminative features and novel classification algorithms. We believe the group-differentiating information retrieved from 3 modalities will enhance the sensitivity and specificity of the classification and permit more reliable and valid biomarkers to be identified by fusing similar data types from other sites. The successful completion of this project will provide a powerful tool for N-way multimodal data fusion, help characterize the traits of SZ and BP which may serve as potential biomarkers and expedite their differential diagnosis in acute settings, leading to more appropriate treatment and improved outcomes for both patients.
Showing the most recent 10 out of 222 publications
