Statistical Learning Of Multimodal Medical Imaging Data With Block-Wise Missing Structure

This dissertation addresses the pervasive challenge of block-wise missingness in multimodal medical imaging data. To address this, we propose two novel frameworks: BaMM and IMMR. Through rigorous theoretical development, extensive simulations, and real-world applications on the ADNI dataset, these frameworks demonstrate competitive performance in prediction accuracy and uncertainty quantification. These contributions significantly advance the field of multi-modal medical data analysis by providing effective solutions for handling missing data and improving the reliability of diagnostic predictions.