Project Summary Liver cancer is the fourth most common cause of cancer-related death worldwide. Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer and is on the rise in the western world. Minimally inva- sive, catheter-based locoregional therapies (LRT), such as transarterial chemoembolization (TACE), are now the mainstay treatments for intermediate to advanced stage HCC and are included in all management guidelines. TACE is a palliative therapy that prolongs survival by controlling intra-hepatic tumor progression via targeted is- chemic injury, paired with the delivery of highly concentrated chemotherapy into the tumor-feeding artery. More recently, systemic immunotherapies (IMT), specifically immune checkpoint inhibitors, have emerged as an im- portant treatment option for HCC to boost the body's own immune response against the tumor. While IMT is promising for many cancers, only 15-30% of HCC patients respond to this type of therapy. TACE is increasingly used in conjunction with IMT, both in neoadjuvant and adjuvant scenarios. Recent efforts show that TACE can dramatically alter the tumor microenvironment (TME) to become more immune-permissive, enabling more ef- fective immune cell recruitment against the tumor through IMT. Thus, the LRT+IMT combination is a likely path forward for HCC treatment strategies. In this context, there is an urgent and unmet clinical need for robust, non- invasive quantitative biomarkers to help guide therapeutic decision making and assess therapeutic outcome early during treatment. Previously, our team developed clinical and preclinical advanced imaging, image analysis, and imaging biomarkers to study, guide and assess HCC treatment with TACE alone using multiparameter magnetic resonance imaging (mpMRI) and magnetic resonance spectroscopic imaging (MRSI). We developed random forests and convolutional neural networks for liver segmentation, tissue classification and nonrigid registration to map these results into the clinical treatment environment. Using graph convolutional neural networks, we pre- dicted and assessed therapeutic outcomes. In a rabbit model of liver cancer (VX2), using Biosensor Imaging of Redundant Deviation in Shifts (BIRDS), we successfully characterized the metabolic state of the TME with respect to extracellular acidosis, before and after TACE. We now propose to develop robust quantitative biomarkers for combined LRT+IMT assessment and outcome prediction in humans. We will develop novel image analysis (Joint Domain Learning with Structure-Consistent Embedding by Disentanglement) and characterize the changing TME over the course of LRT+IMT by deriving information from longitudinal mpMRI (with liver-specific contrast) and/or multiphase computed tomography (mpCT), learning across modalities via domain adaptation. Since LRT+IMT is expected to reduce extracellular acidosis in treated liver tumors, we propose to develop high-resolution advanced BIRDS in the rabbit VX2 model w...