Understanding and Improving Risk Assessment After Myocardial Infarction Using Automated Left Ventricular Shape Analysis

Figure 1. Shape Analysis Pipeline Shape analysis pipeline (left) along sample case (right), consisting of the following steps: 1) short-axis stacks (SAx) preprocessing (intensity normalization, heart detection, cropping of region of interest [ROI]); 2) fine segmentation of left ventricular myocardium; 3) volumetric segmentation at end-systole (ES) and end-diastole (ED) phases; and 4) personalized mesh generation. 3-dimensional (3D) contraction is then calculated from the ES and ED meshes, and principal component analysis (PCA) dimensionality reduction is applied on the meshes to facilitate the supervise learning of major adverse cardiac events (MACE) occurrence.

Figure 2. 3D LV Shape and Contraction Risk-Related Patterns The 3 main ES shape features (ES1, ES5, ES6) and contraction patterns (C3, C5, C16) related to MACE occurrence are illustrated. Meshes shown in anterior and septal views, and as differential thickness maps (ED-ES thickness) on polar plots of the American Heart Association model. To facilitate comparisons, the contractions are applied on the mean ED shape (reference transparent surface) and visualized as resulting ES shapes. No MACE (blue, class 0) and MACE (red, class 1) representations correspond to the 5th and 95th percentiles in the LDA direction. P values and AUCs shown along MACE and No MACE distributions, further stratified into infarct etiology (ST-segment elevation myocardial infarction [STEMI] and non–ST-segment elevation myocardial infarction [NSTEMI]). AUC = area under the receiver-operating characteristic curve; LDA = linear discriminant analysis; other abbreviations as in Figure 1.

Central Illustration. Prognostic Contribution and Representative Risk-Related Shape and Contraction Modes (Left) MACE prediction comparison between ESV and LVEF and the proposed 3D disentanglement into ES shape and contraction. Assessment based on AUC resubstitution (AUCrs, blue) and 10-cross-fold validation (AUCk, orange), repeated for 100 random data splits (black distributions). All prediction differences were significant (P < 0.001); y-axis origin set to 0.5 (random classifier). (Right) Anterior views of representative No MACE (blue) vs MACE (red) ES shape and 3D contraction, and differential thickness maps (ED-ES thickness) in American Heart Association model representation. Contractions are applied on the mean ED shape (transparent surface) and visualized as resulting ES shapes to ease comparisons. 3D = 3-dimensional; AUC = area under the receiver-operating characteristic curve; ED = end-diastole; ES = end-systole; ESV = end-systolic volume; LVEF = left ventricular ejection fraction; MACE = major adverse cardiac events.