Table 4 Diagnostic performance of AI system for detecting and displaying early stage colorectal cancers and precursor lesions in 77 videos frames.

From: Development of a real-time endoscopic image diagnosis support system using deep learning technology in colonoscopy

 

With lesions§

Without lesion

AI flag

Sensitivity

False positive

False positive

Present

74.6% (47.0–85.1)

1.1% (0.1–8.3)

5.5% (1.9–10.6)

Def. 1*

100% (56/56 lesions)

0% (0.0–1.6)

2.0% (0.5–5.4)

Def. 2Ɨ

69.6% (39/56 lesions)

1.8% (1/56 lesions)

0% (0/77 colonoscopies)

Def. 3ǂ

Absent

False negative

—

Specificity

Def. 1*

25.4% (14.9–53.0)

94.5% (89.4–98.1)

Def. 2Ɨ

0% (0/56 lesions)

98.0% (94.6–99.5)

Def. 3ǂ

30.4% (17/56 lesions)

100% (77/77 colonoscopies)

  1. Data shows median (interquartile range).
  2. *Definition 1 = correct frame number/all frame number.
  3. ƗDefinition 2 = consecutive 5 or more correct frame number/all frame number.
  4. ǂDefinition 3 = correct when 50% or more of the entire frame is a correct frame, and calculate number of correct videos/number of lesions.
  5. §56 colonic lesions were included in the 77 videos.
  6. Def., definition.
Back to article page