(A) Experimental conditions. (B) Intensities and movements of the correctly tracked cells with large size (26 cells) or small size (40 cells). (C) 2D and 3D images and its segmentation result in volume #1. The large bright areas in Z110 and Z120 indicated by arrows are irrelevant for the tracking task, and our 3D U-net correctly classified them as background. Orange and blue outlines in bottom panels indicate the regions of the ventricle and atrium, respectively. (D) Tracking results. Tracked cells in volume #531 and #1000 are shown; we showed #531 because #500 looks similar to #1. Arrows indicate two correctly tracked representative cells in the ventricle and atrium. The sizes of the ventricle and atrium changed periodically (orange and blue outlines). (E) Tracking accuracy in 30 large cells or in all 98 cells through time. (F) The positions of 98 cells with/without tracking mistakes. (G) Movements of the two representative cells N9 and N95, indicating regular oscillations of cells in whole 3D space. (H) Dynamics of ventricle size, atrium size, and calcium signals in ventricle cells and atrium cells. The sizes of the ventricle and atrium cannot be directly measured, so we instead estimated them as , where sd is standard deviation (more robust than range of x, y, z), and x, y, z are coordinates of correctly tracked cells in the ventricle or in the atrium. To improve visibility, these sizes were normalized by . Calcium signals (GCaMP) were also normalized by . (I) Phase differences between intracellular calcium dynamics and the reciprocal of segment sizes in ventricle and atrium were estimated. Here we used reciprocal of segment sizes because we observed the anti-synchronization relationships in (H). The phase differences were estimated using cross-correlation as a lag with the largest correlation. Most cells showed similar phase differences (mean = −0.110 π; standard deviation = 0.106 π). All scale bars, 40 µm.