cardiac_shape_analysis
We proposed a new framework for cardiac shape analysis and dysfunction classification in this paper. While most of the previous work are aiming to help comparing groups with global LV morphology, our proposed method focused on local information quantification and can improve the efficiency of cardiac MR images diagnosis by directly pointing out abnormal places on the cardiac surface. We linked multilinear principle analysis (MPCA) and linear discriminant analysis (LDA) together with large deformation diffeomorphic metric mapping (LDDMM) for both the purpose of dimension deduction on 3D surface and denoting the parts that contribute the most to group-wise variance, which, as we found also represented the most meaningful part to disease diagnosis. We achieved higher accuracy (leave-one-out accuracy: 94% 100%) than previous proposed methods.