Fig. 6: Differences in confidence scores for predictions and false positives.

Plotted confidence scores for the vessel pathology detected by the ANN in each of the observed categories within the FAST and UKB dataset, demonstrating the utility of the confidence scores for prioritization of findings within a simulated clinical setting. Groups are compared using the two-sided Wilcoxon rank-sum test. Statistical significance is highlighted between the groups. In both datasets the ANN demonstrated significantly higher confidence when detecting vessel occlusions (VO, depicted in blue color on the left, median FAST (n = 47) 0.90 [0.81–0.95] and UKB (n = 71) 0.92 [0.85–0.96]) as compared to high-grade stenosis (HGS, depicted in green color in the middle, median FAST (n = 32) 0.81 [0.74–0.85], p = 0.002 and UKB (n = 22) 0.84 [0.74–0.88], p < 0.001), as well as compared to false positive results (FP, depicted in salmon color on the right, median FAST (n = 48) 0.72 [0.67–0.78], p < 0.001 and UKB (n = 44) 0.73 [0.68–0.77], p < 0.001). The ANN demonstrated also higher confidence when detecting HGS as compared to FP (FAST, p < 0.001 and UKB, p = 0.001). Data boxes depict median values (center) ± interquartile range (bounds of box), bars depict data range. P values are reported without correction for multiple comparisons.