Skip to main content

Table 2 Detection accuracy analysis in both study scenarios

From: Breast cancer detection accuracy of AI in an entire screening population: a retrospective, multicentre study

 

Sensitivity

(95% CI); p value*

Specificity

(95% CI); p value*

PPV

(95% CI); p value

NPV

(95% CI); p value

Recall rate

(95% CI); p value

Arbitration rate

(95% CI); p value*

Standalone AI scenario

First reader

63.7 (61.6–65.8); ref.

97.8 (97.7–97.8); ref.

18.7 (17.8–19.6); ref.

99.7 (99.7–99.7); ref.

2.7 (2.6–2.8); ref.

NA

Standalone AIsens

63.7 (61.6–65.8); >0.99

96.5 (96.4–96.5); <0.0001

12.6 (11.9–13.2); <0.0001

99.7 (99.7–99.7); 0.71

4.0 (3.9–4.1); <0.0001

NA

Standalone AIspec

58.6 (56.5–60.8); <0.0001

97.8 (97.7–97.8); 0.95

17.4 (16.5–18.3); 0.01

99.7 (99.6–99.7); 0.0002

2.7 (2.6–2.7); 0.24

NA

AI-integrated screening scenario

Combined reading

73.9 (72.0-75.8); ref.

97.9 (97.9–98.0); ref.

22.0 (21.0–23.0); ref.

99.8 (99.8–99.8); ref.

2.7 (2.6–2.7); ref.

2.9 (2.8-3.0); ref.

Integrated AIsens

76.2 (74.3–78.0); 0.0004

97.3 (97.2–97.3); <0.0001

18.1 (17.3–19.0); <0.0001

99.8 (99.8–99.8); 0.07

3.3 (3.3–3.4); <0.0001

5.1 (5.1–5.2); <0.0001

Integrated AIspec

74.6 (72.6–76.4); 0.32

97.9 (97.8–97.9); 0.54

22.0 (21.0–23.0); 0.99

99.8 (99.8–99.8); 0.60

2.7 (2.6–2.7); 0.49

4.0 (3.9–4.1); <0.0001

  1. Data are % (95% CI); p value. PPV = positive predictive value. NPV = negative predictive value. AIsens=artificial intelligence score cut-off point matched at mean first reader sensitivity. AIspec=artificial intelligence score cut-off point matched at mean first reader specificity. *p values were calculated using McNemar’s test. †p values were calculated using exact binomial test