Fig. 9
From: Automatic segmentation of the great arteries for computational hemodynamic assessment
Flow errors against similarity metrics. The figure shows a scatter-plot matrix where each point corresponds to a subject. In the abscissas, two confusion-based metrics, Dice and IoU, and two distance-based metrics, the Hausdorff distance and the average surface distance, measured in pixels. In the ordinates, the pressure and velocity mean average percentage errors (MAPE). All values are for the ML vs GT comparison. Red and blue colours identify aorta and pulmonary artery data, respectively. Trend lines are least-squares polynomial fits of degree 1. For Dice and IoU, higher is better (more similar). For Hausdorff distance, average surface distance and pressure and velocity MAPEs, lower is better