Figure 4.

Representative examples of the ventricular septum segmentation in test data for each module combination. One horizontal row presents the segmentation results obtained using each method for the same case. The white pixels are estimated as the ventricular septum, and the degree of whiteness indicates the confidence level. Among the various module combinations, the segmentation results of U-net + YOLO (You Look Only Once) + ED + VGG (CSC) were most in accordance with the ground truth. The YOLO significantly contributed to the improvement of the segmentation accuracy, and ED improved the segmentation. Moreover, the addition of VGG slightly narrowed the prediction section.