Table 4. Model Discrimination (AUROC) for AS Screening Tasks.

Model Discrimination for three binary screening tasks: 1) AS absent vs AS present, 2) Early AS vs. Significant AS, 3) no significant AS vs Significant AS. We report the Area under the Receiver Operator Curves (AUROC) for each task, averaged over 3 random training-validation-test splits of the data. The 95% bootstrap CI of this average is in parentheses. Methods two methods of aggregating image-level predictions to a study-level diagnosis: simple averaging or a weighted averaged that prioritizes specific views (PLAX or PSAX) that depict the aortic valve and are thus relevant for AS diagnosis.

Model	AS absent vs AS present	Early AS vs Sig AS	Sig AS vs No Sig AS
Simple Average	0.86 (0.81 – 0.91)	0.73 (0.67 – 0.79)	0.84 (0.79 – 0.88)
Prioritized View	0.96 (0.93 – 0.97)	0.75 (0.68 – 0.81)	0.86 (0.82 – 0.90)