They used the AUC (area under the curve) to measure accuracy, with a value of 1 being 100% and 0.5 being random. The AI scored an AUC of 0.87, higher than the Brock score, which clinics currently use, at 0.67.