Leveraging artificial intelligence to predict ERG gene fusion status in prostate cancer

BMC Cancer

Table 3 Algorithm performance metrics based on tumor grade

Magnification	Grade Group	TP	FP	TN	FN	Sensitivity	Specificity	PPV	NPV	Accuracy	F1 score	Cut-off
10 ×	1–2	33	8	25	1	97.1%	75.8%	80.5%	96.2%	86.6%	0.88	0.4
10 ×	3–5	12	5	33	14	46.2%	86.8%	70.6%	70.2%	70.3%	0.56	0.4
20 ×	1–2	28	3	30	6	82.4%	90.9%	90.3%	83.3%	86.6%	0.86	0.5
20 ×	3–5	17	9	29	9	65.4%	76.3%	65.4%	76.3%	71.9%	0.65	0.5
40 ×	1–2	28	5	28	6	82.4%	84.8%	84.8%	82.4%	83.6%	0.84	0.35
40 ×	3–5	17	8	30	9	65.4%	78.9%	68.0%	76.9%	73.4%	0.67	0.35

TP True positive (correctly classified as ERG-positive), FP False positive (incorrectly classified as ERG-positive), TN True negative (correctly classified as ERG-negative), FN False negative (incorrectly classified as ERG-negative), PPV Positive predictive value, NPV: Negative predictive value, Cut-off indicates cut-off value of proportion of positive patches that gives best accuracy

ISSN: 1471-2407