Table 2 Performance metrics of multi-resolution AI model in internal cohort validation set, testing set, and external cohort testing set.

From: High throughput assessment of biomarkers in tissue microarrays using artificial intelligence: PTEN loss as a proof-of-principle in multi-center prostate cancer cohorts

Cohort

AUC

Average probability threshold p > 0.265

TP

TN

FP

FN

Sensitivity

Specificity

Accuracy

Internal cohort validation set (N = 2048)

0.989 (0.980–0.996)

153

1770

117

8

95.03% (90.4–98.4)

93.80% (92.0–95.6)

93.90% (92.2–95.5)

Internal cohort test set (N = 224)

0.993 (0.975–1.00)

19

194

11

0

100.00% (100–100)

94.63% (90.3–98.1)

95.09% (90.9–98.3)

External cohort (N = 428)

0.963 (0.909–0.994)

42

161

224

1

97.67% (91.1–100)

41.82% (35.9–48.0)

47.43% (41.9–53.3)

External cohort with fine-tuning (test set N = 357)

0.964 (0.902–0.998)

29

311

14

3

90.63% (75.0–100)

95.69% (93.6–97.6)

95.24% (93.0–97.3)

  1. External cohort performance is reported for all cores (N = 428) and with fine-tuning (N = 357). 95% confidence intervals calculated from bootstrap analysis on the patient level. AUC is reported for continuous probability of PTEN loss. TP = correctly classified as PTEN loss; TN = correctly classified as PTEN intact; FP = incorrectly classified as PTEN loss; FN = incorrectly classified as PTEN intact.
Back to article page