Reconstructing ventricular activation sequences from epicardial data: Insights from Geodesic Back-Propagation optimization in porcine models

Figures

• 

  Fig. 1.. Electrode locations for Geometries A through E (columns). The top panel shows the epicardial sock electrodes, showing two views per experiment: the ventricular base (top) and the anterior surface (bottom). The bottom panel shows intramural needle electrode locations with matching views: base (top) and anterior (bottom).

• 

  Fig. 2.. Box plots of mean absolute differences. Mean absolute differences between optimized and experimental activation times are shown for measurements from the epicardial socks (top) and intramural needle electrodes (bottom) across five different experiments. The x-axis indicates the number of optimized EASs: 5, 50, 100, and 200.

• 

  Fig. 3.. Epicardial surface activation maps and corresponding difference maps for the best-performing (Geometry D, Run 6 with 200 EASs) and worst-performing (Geometry E, Run 0 with 5 EASs) optimizations. The best and worst performances correspond to the minimum and maximum mean absolute differences on the epicardial sock. The columns show, from left to right: the experimental ground truth, the optimized simulation, and the absolute differences in activation times between experimental and optimized cases. The first and second rows display posterior and anterior views of the epicardial surface, respectively. Spheres indicate the locations of the sock epicardial electrodes and are colored according to either activation time or absolute difference in activation time, depending on the column. Activation time color bars range from 0–54 ms for the best case and 0–61 ms for the worst case. The color bars for absolute differences ranged from 0–4 ms for the best case and 0–15 ms for the worst case.

• 

  Fig. 4.. Optimized initiation and acceleration sites for experimental activation times. The top row shows the runs with the best-performing (minimum mean absolute difference on the epicardial sock) optimized EAS locations on Geometry D, while the bottom row shows the runs with the worst-performing (maximum mean absolute difference on the epicardial sock) optimized EAS locations on Geometry E, each selected from initializations with 5, 50, 100, and 200 EASs. In each panel, color-coded spheres indicate the optimized stimulus (EAS) sites, with color representing activation start time. The top and bottom rows display basal and anterior views of the myocardium, respectively, both shown as transparent volumes. Activation times ranged from 0–54 ms in the best-performing run and 0–63 ms in the worst-performing run, as indicated by the color bar.

• 

  Fig. 5.. Epicardial and intramural maps of mean activation times and corresponding standard deviations for Geometry E. The top panel shows results for 5 EASs, and the bottom panel shows results for 200 EASs. In each panel, the first row presents the posterior view, and the second row presents the anterior view. Columns from left to right display: (1) mean activation times on the epicardial surface, (2) mean activation times on an intramural cut plane, (3) standard deviations on the epicardial surface, and (4) standard deviations on the cut plane. The color of the spheres corresponds to the mean activation time in the first and second columns, and to the standard deviation in the third and fourth columns, matching the corresponding epicardial sock recordings. For 5 EASs, mean activation times range from 0–54 ms and standard deviations range from 0–11 ms. For 200 EASs, mean activation times ranged from 0–49 ms and standard deviations from 0–8 ms.

• 

  Fig. 6.. Box plots of mean absolute differences. Mean absolute differences between optimized and synthetic activation times are shown for measurements from the epicardial socks (top) and for the ventricles (bottom) across five different experiments. The x-axis indicates the number of optimized EASs (5, 50, 100, and 200); note that the number of ground truth EAS remains constant.

• 

  Fig. 7.. Ventricular activation maps and corresponding difference maps for the best-performing (Geometry D, Run 0 with 200 EASs) and worst-performing (Geometry D, Run 9 with 5 EASs) optimizations. The best and worst performances correspond to the minimum and maximum mean absolute differences on the epicardial sock. The columns show, from left to right: the synthetic ground truth on the epicardium (column 1) and intramurally (column 2), the optimized simulation on the epicardium (column 3) and intramurally (column 4), and the absolute differences between synthetic and optimized activation times on the epicardium (column 5) and intramurally (column 6). The first and second rows display posterior and anterior views of the epicardial surface, respectively. Spheres indicate the locations of the sock epicardial electrodes and are colored according to either activation time or absolute difference in activation time, depending on the column. Activation time color bars range from 0–55 ms for the best case and 0–58 ms for the worst case. Color bars for absolute differences ranged from 0–51 ms for the best case and 0–40 ms for the worst case.

• 

  Fig. 8.. Optimized initiation and acceleration sites for synthetic activation times. The top row shows the runs with the best-performing (minimum mean absolute difference on the epicardial sock) optimized EAS locations and the bottom row shows the runs with the worst-performing (maximum mean absolute difference on the epicardial sock) optimized EAS locations on Geometry D, each selected from initializations with 5, 50, 100, and 200 EASs. In each panel, color-coded spheres indicate the optimized stimulus (EAS) sites, with color representing activation time. Additionally, color-coded cubes indicate the Durrer ground truth stimulus sites, with color representing activation time (all sites are pink, 0 ms). The top and bottom rows display basal and anterior views of the myocardium, respectively, both shown as transparent volumes. Activation times ranged from 0–60 ms, as indicated by the color bar.

• 

  Fig. 9.. Epicardial surface and intramural maps of mean activation times and corresponding standard deviations for Geometry D. The top panel shows results for 5 EAS, and the bottom panel shows results for 200 EASs. In each panel, the first row presents the posterior view, and the second row presents the anterior view. Columns from left to right display: (1) mean activation times on the epicardial surface, (2) mean activation times on an intramural cut plane, (3) standard deviations on the epicardial surface, and (4) standard deviations on the cut plane. The color of the spheres corresponds to the mean activation time in the first and second columns, and to the standard deviation in the third and fourth columns, matching the corresponding epicardial sock recordings. For the 5 EASs case, mean activation times range from 0–53 ms and standard deviations range from 0–12 ms. For the 200 EASs case, mean activation times ranged from 0–53 ms and standard deviations from 0–7 ms.

• 

  Fig. 10.. Histograms of stimulus start times for Geometries A–E. Each panel displays the distribution of start times across 10 optimization runs for a given geometry. Red bars correspond to the 5 EASs optimization setup, and blue bars correspond to the 200 EASs optimization setup.