TABLE 5.

Neural network model vs. physician stratification of 50 women with cancer (ground truth positives) and 50 women without cancer (ground truth negatives) into below-, at-, or above-average risk groups.

	Below average risk		Average risk		Above average risk
	Model	Physicians	Model	Physicians	Model (%)	Physicians
Ground truth positives	0.0	22.0% (17%)	30.0	40.0% (16%)	70.0	38.0% (24%)
Ground truth negatives	24.0	39.5% (22%)	68.0	32.6% (16%)	8.0	27.9% (20%)

Inter-observer variability for the physicians is captured by a standard deviation in parenthesis.