Cardiac Magnetic Resonance Image Analysis and Biomarker Measurements for Cardiovascular Disease Care

Speaker: Dr. Fumin Guo (UofT)

Abstract: Cardiovascular disease represents a huge burden on patients, economy, and healthcare systems worldwide. In 2016, it caused 18 million or just accounted for 31% of global deaths with ventricular tachycardia being the most prevalent and life-threatening cardiovascular disorder. Cardiac magnetic resonance imaging (MRI) is the method of choice for cardiovascular disease diagnosis, characterization and therapeutic guidance. Today I will focus on cine and multi-contrast late Gd enhancement (MCLE) MRI and introduce a number of CMR image analysis methods that we developed for broad research and clinical applications of CMRI.
Cine images were segmented using a U-net neural network and the derived labeling probability maps were refined using a continuous max-flow segmentation model, where we proposed new image features combining image signal intensity and spatial location information to compensate signal intensity inhomogeneity. The high-dimensional image features were classified in the kernel space using a kernel k-means with a smoothness prior. The resulting high-order energy function was minimized iteratively in a continuous max-flow frame under an upper-bound relaxation perspective. The segmentation masks were used to quantify clinically-relevant imaging biomarkers of the left ventricle (LV), including LV myocardium mass, LV cavity dimension and regional wall thickness. We also developed an approach for cine-MCLE image registration and information fusion for novel CMR applications such as image-guided cardiac interventions and computational modeling of cardiac function. Cine-MCLE registration was initialized using a Block-matching scheme followed by a dual-optimization based deformable step. The resulting deformation field was used to warp the cine segmentation that were fused with MCLE-derived tissue heterogeneity information to generate high-fidelity cardiac models. Our approaches demonstrated high performance, suggesting the promising potential for research and clinical investigations of cardiovascular disease.