Table 2 The basic characteristics of the lesions in training and validation cohorts

From: Development of deep learning-based narrow-band imaging endocytoscopic classification for predicting colorectal lesions from a retrospective study

Characteristic

Total (n = 1056)

Cohort

P-value

Training Cohort (n = 406)

Internal Validation Cohort (n = 498)

External Validation Cohort (n = 152)

 

Lesion Morphology, n (%)

     

Protruding

433 (41.00)

165 (40.64)

200 (40.16)

68 (44.73)

0.180

Flat elevated

420 (39.77)

166 (40.89)

191 (38.35)

63 (41.45)

 

Depressed

13 (1.23)

6 (1.48)

6 (1.20)

1 (0.66)

 

LST-NG

47 (4.45)

9 (2.22)

32 (6.43)

6 (3.95)

 

LST-G

16 (1.52)

8 (1.97)

7 (1.41)

1 (0.66)

 

Advanced-type

127 (12.03)

52 (12.80)

62 (12.45)

13 (8.55)

 

Lesion Location, n (%)

    

0.112

Cecum

47 (4.45)

20 (4.93)

22 (4.42)

5 (3.29)

 

Ascending Colon

153 (14.49)

65 (16.01)

62 (12.45)

26 (17.11)

 

Transverse Colon

277 (26.23)

106 (26.11)

124 (24.90)

47 (30.92)

 

Descending Colon

90 (8.52)

25 (6.16)

52 (10.44)

13 (8.55)

 

Sigmoid Colon

255 (24.15)

110 (27.09)

115 (23.09)

30 (19.74)

 

Rectum

234 (22.16)

80 (19.70)

123 (24.70)

31 (20.39)

 

Lesion Size, M (Q1, Q3)

0.70 (0.50, 1.20)

0.70 (0.50, 1.20)

0.70 (0.50, 1.50)

0.60 (0.50, 1.00)

0.008

  1. Data are represented by n (%). The p-values are calculated using Pearson χ2 test or Kruskal Wallis test. All statistical tests performed were two-sided.