Schizophrenia is a complex, chronic and disabling mental disorder that affects about one percent of the adult population. The etiology of schizophrenia remains elusive and to date there are no image based tools to diagnose it. Advancements in magnetic resonance imaging (MRI) have enabled researchers to develop less invasive and in vivo techniques, such as structural MRI (sMRI), functional MRI (fMRI) and diffusion tensor imaging (DTI), to construct theories about the neural underpinnings of schizophrenia. With sMRI, fMRI and DTI the distribution of tissues, the functional activity and the brain network are imaged respectively. Subjects with schizophrenia (SZ) and healthy controls (HC) are scanned with different modalities to identify differences, but the analysis of each modality has traditionally been carried out separately. Data fusion of multimodal data and an analysis of the joint information may hold the key to reveal hidden traces of this subtle disorder. In this work we develop techniques to correlate sMRI with fMRI, fMRI with other fMRI and DTI with symptom scores. The brain is a highly interconnected organ and local morphology can influence functional activity at distant regions. Through our methods it is possible to perform a cross correlation analysis between modalities incorporating all brain voxels. By reducing the large cross correlation matrix to useful statistics new aspects of schizophrenia are revealed. The methods introduced are simple, easy to implement and efficient. In another effort we modify canonical correlation analysis (CCA) to fuse two sets of brain data to locate brain regions with significant correlations. The new differential features identified through our fusion methods are used to classify subjects. The sMRI–fMRI fusion indicates that the linkage between gray matter and functional activity probed by a sensorimotor task is weaker in SZ than in HC. Linkages between functional activity and structural regions in the cerebellum and the prefrontal cortex are found to be aberrant in SZ. The pair wise fusion of four different fMRI tasks shows that SZ activate to different tasks less uniquely than do HC. The above results support the ‘disconnection hypothesis’ of schizophrenia and the ‘theory of cognitive dysmetria’. DTI–symptom score fusion indicates that regions in the superior longitudinal fasciculus have high DTI–symptom correlations. Our preliminary classification efforts show high success rates in the leave–one–out scheme. The results presented in this work reveal several novel and interesting findings to better understand schizophrenia. The methods introduced are general, and can be easily applied to healthy and other pathological brain data to explore brain behavior.
Library of Congress Subject Headings
Schizophrenia--Diagnosis; Multisensor data fusion; Magnetic resonance imaging; Diffusion tensor imaging
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Michael, Andrew M., "Imaging schizophrenia: data fusion approaches to characterize and classify" (2009). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus