Table 3 Diagnostic performance of the CAD in each histological group in comparison with the diagnostic performance of endoscopists.

From: Endoscopic diagnosis and treatment planning for colorectal polyps using a deep-learning model

All polyps

Test set I

Test set II

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Overall accuracy

82.4 (450/546)

81.3 (148/182)

71.8 (392/546)

84.2 (460/546)

0.724

0.005

<0.001

87.3 (951/1089)

82.4 (299/363)

63.8 (695/1089)

82.7 (901/1089)

0.005

<0.001

<0.001

Serrated polyp

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Sensitivity, % (fraction)

88.9 (104/117)

82.1 (32/39)

55.6 (65/117)

82.1 (96/117)

0.179

<0.001

<0.001

81.7 (245/300)

74.0 (74/100)

92.0 (276/300)

81.3 (244/300)

0.059

<0.001

0.0003

Specificity, % (fraction)

92.1 (395/429)

93.7 (134/143)

90.4 (388/429)

94.9 (407/429)

0.498

0.210

0.057

94.6 (746/789)

93.5 (246/263)

61.0 (481/789)

89.2 (704/789)

0.452

<0.001

<0.001

PPV, % (fraction)

75.4 (104/138)

78.0 (32/41)

61.3 (65/106)

81.4 (96/118)

0.666

0.018

0.003

85.1 (245/288)

81.3 (74/91)

47.3 (276/584)

74.2 (244/329)

0.250

<0.001

<0.001

NPV, % (fraction)

96.8 (395/408)

95.0 (134/141)

88.2 (388/440)

95.1 (407/428)

0.193

0.001

0.001

93.1 (746/801)

90.4 (246/272)

95.2 (481/505)

92.6 (704/760)

0.046

0.003

0.050

BA/MSMC

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Sensitivity, % (fraction)

83.1 (314/378)

84.1 (106/126)

81.7 (309/378)

88.1 (333/378)

0.775

0.508

0.040

92.2 (647/702)

88.5 (207/234)

53.0 (372/702)

85.8 (602/702)

0.041

<0.001

<0.001

Specificity, % (fraction)

81.0 (136/168)

75.0 (42/56)

51.8 (87/168)

75.6 (127/168)

0.304

0.001

<0.001

78.6 (304/387)

72.1 (93/129)

84.5 (327/387)

77.8 (301/387)

0.083

<0.001

0.0167

PPV, % (fraction)

90.8 (314/346)

88.3 (106/120)

79.2 (309/390)

89.0 (333/374)

0.330

0.002

<0.001

88.6 (647/730)

85.2 (207/243)

86.1 (372/432)

87.5 (602/688)

0.043

0.655

0.429

NPV, % (fraction)

68.0 (136/200)

67.7 (42/62)

55.8 (87/156)

73.8 (123/172)

0.961

0.028

<0.001

84.7 (304/359)

77.5 (93/120)

49.8 (327/657)

75.1 (301/401)

0.016

<0.001

<0.001

DSMC

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Experts

CAD

Trainees

CAD+trainees

P*

P†

P‡

Sensitivity, % (fraction)

62.7 (32/51)

58.8 (10/17)

35.3 (18/51)

58.8 (30/51)

0.839

0.043

0.013

67.8 (59/87)

62.1 (18/29)

54.0 (47/87)

63.2 (55/87)

0.532

0.268

0.161

Specificity, % (fraction)

93.9 (465/495)

93.3 (154/165)

93.5 (463/495)

92.9 (472/495)

0.771

>0.999

0.678

98.8 (990/1002)

96.7 (323/334)

97.4 (976/1002)

98.3 (985/1002)

0.004

0.436

0.201

PPV, % (fraction)

51.6 (32/62)

47.6 (10/21)

36.0 (18/50)

46.2 (31/54)

0.655

0.215

0.274

83.1 (59/71)

62.1 (18/29)

64.4 (47/74)

76.4 (55/72)

0.004

0.760

0.094

NPV, % (fraction)

96.1 (465/484)

95.7 (154/161)

93.3 (463/496)

95.6 (472/492)

0.742

0.114

0.117

97.2 (990/1018)

96.7 (323/334)

96.1 (976/1016)

96.9 (985/1017)

0.471

0.324

0.186

  1. CAD, computer-aided diagnostic system; PPV, positive predictive value; NPV, negative predictive value; BA, benign conventional adenoma; MSMC, mucosal or superficial submucosal tumor; DSMC, deep submucosal cancer.
  2. The diagnostic performance of the three experts and trainees was evaluated by combining the results of all the endoscopists. Therefore, the total number of examined polyps was 546 (3 times 182) in test set I and 1089 (3 times 363) in test set II.
  3. *P value in the comparison between CAD vs. experts.
  4. †P value in the comparison between CAD vs. trainees.
  5. ‡P value in the comparison between trainees vs. CAD+trainees.
Back to article page