Committee:

Dr. Vince Calhoun, ECE, Chair , Advisor

Dr. Mark Davenport, ECE

Dr. Eva Dyer, BME

Dr. Rogers Silva, TReNDS

Dr. Jungyu Liu, GSU CS

Abstract: This dissertation aims to develop novel algorithms that leverage sparsity and mutual information across data modalities built upon the independent component analysis (ICA) framework to improve the performance of current multimodal fusion approaches. To alleviate the signal-background separation difficulties in sources of genetic data, we propose a sparse ICA by enhancing a robust sparsity measure, the Hoyer index. Hoyer index is scale-invariant and well suited for ICA frameworks since the scale of decomposed sources is arbitrary. The proposed sparse ICA is further extended into two data modalities as a sparse parallel ICA for applications to imaging genomics in order to investigate the association between brain imaging and genomics. Moreover, to increase the flexibility and robustness in mining multimodal data, we propose aNy-way ICA, which optimizes the entire correlation structure of linked components across any number of modalities via Gaussian independent vector analysis and simultaneously optimizes independence via separate (parallel) independent component analyses. The proposed aNy-way ICA is applied to multimodal brain imaging data fusion for the Philadelphia Neurodevelopmental Cohort (PNC) to extract multi-aspect coherent brain functional and anatomical patterns. Finally, we extend aNy-way ICA with a reference constraint to enable prior-guided multimodal fusion. We then apply aNy-way ICA with reference to multimodal neuroimaging data in the PNC to investigate covarying structural and functional brain patterns underlying intelligence quotient score.