Table 2: AS Diagnosis Label Cohorts across Train/Test Splits.

AS Diagnosis Label Cohorts across Train/Test Splits in TMED-2. We show the number of echocardiogram studies assigned to train/valid/test sets across all 3 possible aortic stenosis (AS) severity levels for the fully-labeled dataset of 599 studies. Image counts represent the mean over 3 splits, as the exact number of images per study differs across splits. No split deviates more than 12% from the reported mean here. Each patient’s data were assigned to exactly one set to properly assess generalizability to new patients, while preserving similar proportions of each AS severity level across train and test.

	Number of Studies				Number of Labeled Images
	Total	None	Early AS	Sig AS	Total	None	Early AS	Sig AS
Train*	360	76	103	181	10253	999	1316	7938
Valid	119	25	34	60	3505	344	402	2759
Test	120	26	34	60	3511	339	408	2764

^*indicates that the training set also included an additional set with only view labels but no AS diagnosis label. This view-only set contained 705 studies representing 7694 labeled images, see Supplement Table 3.

Abbreviations: None represents no AS, Early AS represents mild and mild/mod AS, Sig AS means moderate, moderate/severe, and severe AS.