Table 3.

CAD software performance when matching the sensitivity of the Expert Reader.

	Cut-off Score	TP	FP	FN	TN	Sensitivity (95% CI)	Specificity (95% CI)	Accuracy (95% CI)
Expert Reader	N/A	127	520	6	379	95.5% (90.4–98.3)	42.2% (38.9–45.5)	49.0% (45.9–52.1)
Abnormality scores obtained by FIT
Qure.ai	44.1	127	461	6	438	95.5% (90.4–98.3)	48.7% (45.4–52.0)	54.7% (51.7–57.8)
DeepTek	31.1	127	483	6	416	95.5% (90.4–98.3)	46.3% (43.0–49.6)	52.6% (49.5–55.7)
Delft imaging	46.7	127	492	6	407	95.5% (90.4–98.3)	45.3% (42.0–48.6)	51.7% (48.7–54.8)
Abnormality scores provided by CAD company
JF Healthcare	83.4	127	530	6	369	95.5% (90.4–98.3)	41.0% (37.8–44.3)	48.1% (45.0–51.2)
OXIPIT	15.4	127	532	6	367	95.5% (90.4–98.3)	40.8% (37.6–44.1)	47.9% (44.8–51.0)
Lunit	3.0	127	551	6	348	95.5% (90.4–98.3)	38.7% (35.5–42.0)	46.0% (43.0–49.1)
InferVision	53.8	127	661	6	238	95.5% (90.4–98.3)	26.5% (23.6–29.5)	35.4% (32.5–38.4)
Artelus	1.2	127	691	6	208	95.5% (90.4–98.3)	23.1% (20.4–26.0)	32.5% (29.6–35.4)
Dr CADx	27.8	127	790	6	109	95.5% (90.4–98.3)	12.1% (10.1–14.4)	22.9% (20.3–25.6)
SemanticMD	0.4	127	808	6	91	95.5% (90.4–98.3)	10.1% (7.2–10.8)	21.1% (16.6–21.2)
EPCON	0.6	127	815	6	84	95.5% (90.4–98.3)	9.3% (6.6–10.0)	20.4% (16.0–20.6)
COTO	1.5	127	842	6	57	95.5% (90.4–98.3)	6.3% (4.8–8.1)	17.8% (15.5–20.3)

TP true positive, FP false positive, FN false negative, TN true negative.