Copyright: ©Author(s) 2026.
World J Gastroenterol. Jan 21, 2027; 33(3): 125206
Published online Jan 21, 2027. doi: 10.3748/wjg.125206
Published online Jan 21, 2027. doi: 10.3748/wjg.125206
Table 1 Clinical characteristics of patients with different colorectal polyp subtypes, median (interquartile range)/n (%)
| Variables | Total | HP (n = 83) | SSL (n = 306) | TSA (n = 96) | VA (n = 115) | TA (n = 130) | Statistic | P value |
| Age, years | 58.00 (50.00, 65.00) | 57.00 (42.00, 64.00) | 54.50 (45.00, 62.00) | 58.00 (52.00, 63.25) | 62.00 (55.00, 68.50) | 62.00 (55.00, 68.00) | H = 56.18 | < 0.001 |
| Gender | χ2 = 35.72 | < 0.001 | ||||||
| Male | 373 (51.10) | 40 (48.19) | 120 (39.22) | 60 (62.50) | 69 (60.00) | 84 (64.62) | ||
| Female | 357 (48.90) | 43 (51.81) | 186 (60.78) | 36 (37.50) | 46 (40.00) | 46 (35.38) | ||
| Smoking | χ2 = 5.10 | 0.277 | ||||||
| No | 505 (72.87) | 53 (68.83) | 226 (74.83) | 60 (80.00) | 81 (71.68) | 85 (67.46) | ||
| Yes | 188 (27.13) | 24 (31.17) | 76 (25.17) | 15 (20.00) | 32 (28.32) | 41 (32.54) | ||
| Drinking | χ2 = 6.27 | 0.180 | ||||||
| No | 561 (80.72) | 61 (79.22) | 255 (84.44) | 62 (81.58) | 88 (77.88) | 95 (74.80) | ||
| Yes | 134 (19.28) | 16 (20.78) | 47 (15.56) | 14 (18.42) | 25 (22.12) | 32 (25.20) | ||
| BMI | 24.07 (21.72, 26.58) | 24.22 (21.63, 27.55) | 23.32 (21.25, 26.25) | 24.03 (22.04, 27.34) | 24.91 (22.51, 26.99) | 24.62 (22.24, 26.49) | H = 10.92 | 0.028 |
| Hypertension | χ2 = 12.98 | 0.011 | ||||||
| No | 505 (72.45) | 53 (68.83) | 239 (79.14) | 53 (68.83) | 72 (63.72) | 88 (68.75) | ||
| Yes | 192 (27.55) | 24 (31.17) | 63 (20.86) | 24 (31.17) | 41 (36.28) | 40 (31.25) | ||
| Diabetes | χ2 = 15.15 | 0.004 | ||||||
| No | 610 (87.52) | 64 (83.12) | 281 (93.05) | 64 (83.12) | 93 (82.30) | 108 (84.38) | ||
| Yes | 87 (12.48) | 13 (16.88) | 21 (6.95) | 13 (16.88) | 20 (17.70) | 20 (15.62) | ||
| Coronary heart disease | χ2 = 5.92 | 0.205 | ||||||
| No | 639 (91.68) | 68 (88.31) | 284 (94.04) | 72 (93.51) | 102 (90.27) | 113 (88.28) | ||
| Yes | 58 (8.32) | 9 (11.69) | 18 (5.96) | 5 (6.49) | 11 (9.73) | 15 (11.72) | ||
| Cerebral infarction | 0.038 | |||||||
| No | 684 (98.13) | 74 (96.10) | 300 (99.34) | 73 (94.81) | 111 (98.23) | 126 (98.44) | ||
| Yes | 13 (1.87) | 3 (3.90) | 2 (0.66) | 4 (5.19) | 2 (1.77) | 2 (1.56) | ||
| Stool pattern change | χ2 = 12.76 | 0.120 | ||||||
| Normal | 602 (85.88) | 61 (79.22) | 267 (88.41) | 67 (82.72) | 99 (87.61) | 108 (84.38) | ||
| Constipation | 57 (8.13) | 8 (10.39) | 15 (4.97) | 10 (12.35) | 11 (9.73) | 13 (10.16) | ||
| Diarrhea | 42 (5.99) | 8 (10.39) | 20 (6.62) | 4 (4.94) | 3 (2.65) | 7 (5.47) | ||
| Lesion size, cm | 1.20 (0.90, 1.50) | 0.90 (0.72, 1.20) | 1.20 (1.00, 1.50) | 1.00 (0.80, 1.50) | 1.50 (1.00, 2.45) | 1.00 (0.70, 1.20) | H = 114.00 | < 0.001 |
| Lesion location | χ2 = 144.91 | < 0.001 | ||||||
| Ileocecal and ascending | 340 (46.90) | 45 (54.88) | 205 (67.21) | 18 (18.95) | 32 (28.07) | 40 (31.01) | ||
| Transverse | 96 (13.24) | 12 (14.63) | 43 (14.10) | 12 (12.63) | 7 (6.14) | 22 (17.05) | ||
| Descending, sigmoid and rectum | 289 (39.86) | 25 (30.49) | 57 (18.69) | 65 (68.42) | 75 (65.79) | 67 (51.94) | ||
| Lesion morphology | < 0.001 | |||||||
| Protruded | 463 (63.69) | 62 (75.61) | 126 (41.18) | 86 (90.53) | 75 (65.22) | 114 (88.37) | ||
| Flat | 17 (2.34) | 1 (1.22) | 11 (3.59) | 2 (2.11) | 2 (1.74) | 1 (0.78) | ||
| Laterally spreading | 247 (33.98) | 19 (23.17) | 169 (55.23) | 7 (7.37) | 38 (33.04) | 14 (10.85) | ||
| Advanced lesion | χ2 = 107.49 | < 0.001 | ||||||
| No | 651 (89.18) | 83 (100.00) | 300 (98.04) | 84 (87.50) | 75 (65.22) | 109 (83.85) | ||
| Yes | 79 (10.82) | 0 (0.00) | 6 (1.96) | 12 (12.50) | 40 (34.78) | 21 (16.15) | ||
| H. pylori infection | 0.016 | |||||||
| No | 27 (24.11) | 4 (22.22) | 21 (36.84) | 1 (8.33) | 0 (0.00) | 1 (5.88) | ||
| Yes | 85 (75.89) | 14 (77.78) | 36 (63.16) | 11 (91.67) | 8 (100.00) | 16 (94.12) | ||
| CEA, ng/mL | 2.02 (1.54, 2.80) | 1.81 (1.50, 2.35) | 2.02 (1.48, 2.64) | 1.73 (1.53, 2.41) | 2.06 (1.54, 2.94) | 2.05 (1.56, 3.07) | H = 3.58 | 0.466 |
| SII | 425.89 (318.28, 655.39) | 522.13 (326.49, 799.84) | 411.43 (318.18, 623.64) | 435.12 (331.92, 687.01) | 425.89 (327.88, 621.25) | 409.20 (316.06, 544.83) | H = 6.58 | 0.160 |
| NLR | 1.98 (1.49, 2.71) | 2.38 (1.58, 3.14) | 1.89 (1.44, 2.65) | 2.32 (1.59, 3.24) | 1.92 (1.48, 2.58) | 2.04 (1.51, 2.54) | H = 9.12 | 0.058 |
| LMR | 4.80 (3.60, 6.25) | 5.00 (3.33, 6.27) | 5.00 (3.80, 6.22) | 4.00 (3.33, 6.00) | 4.73 (3.75, 6.00) | 4.63 (3.50, 6.06) | H = 6.16 | 0.187 |
| PLR | 130.68 (102.84, 161.07) | 136.00 (104.83, 174.74) | 134.44 (104.50, 157.69) | 125.70 (103.09, 164.93) | 130.59 (102.35, 171.33) | 118.82 (99.09, 144.05) | H = 9.06 | 0.060 |
Table 2 Multivariable logistic regression analysis of risk factors for advanced colorectal lesions
| Variables | β | SE | Z | P value | OR (95%CI) |
| Age (per 1-year increase) | 0.03 | 0.01 | 2.55 | 0.011 | 1.03 (1.01-1.06) |
| Gender | |||||
| Male | 1.00 (reference) | ||||
| Female | -0.47 | 0.27 | -1.74 | 0.082 | 0.63 (0.37-1.06) |
| Lesion size (per 1 cm increase) | 1.16 | 0.17 | 6.67 | < 0.001 | 3.19 (2.27-4.48) |
| Lesion location | |||||
| Ileocecal & ascending | 1.00 (reference) | ||||
| Transverse | 0.48 | 0.47 | 1.03 | 0.305 | 1.62 (0.65-4.05) |
| Descending/sigmoid/rectum | 1.03 | 0.30 | 3.48 | < 0.001 | 2.79 (1.56-4.98) |
Table 3 Threshold effect analysis of lesion size and age on high-grade dysplasia
| Outcome | Model | OR (95%CI) | P value |
| Lesion size (per 1 cm increase) | 3.17 (2.25-4.45) | < 0.001 | |
| Inflection point | 2.48 | ||
| < 2.48 | 5.51 (2.84-10.69) | < 0.001 | |
| ≥ 2.48 | 1.17 (0.66-2.10) | 0.588 | |
| P value for likelihood test | 0.004 | ||
| Age (per 1-year increase) | Linear model | 1.04 (1.01-1.06) | 0.001 |
| Two-piecewise model | |||
| Inflection point | 48.53 | ||
| < 48.53 | 1.44 (0.93-2.23) | 0.101 | |
| ≥ 48.53 | 1.00 (0.97-1.03) | 0.967 | |
| P value for likelihood test | < 0.001 |
Table 4 Performance of the white-light imaging-based deep learning model and comparator approaches for advanced lesion prediction
| Model | Cohort | n | 5-class accuracy | Advanced AUC (95%CI) | Sensitivity | Specificity |
| ResNet18 (baseline) | Internal test | 146 | 0.4795 | 0.7896 (0.7200-0.8600) | 0.8824 | 0.6279 |
| MobileNetV2 | Internal test | 146 | 0.5205 | 0.8789 (0.7948-0.9523) | 0.8235 | 0.8217 |
| MobileNetV2 | External validation A | 159 | 0.8138 (0.7192-0.8964) | 0.7500 | 0.7914 | |
| MobileNetV2 | External validation B | 182 | 0.8033 (0.7087-0.8863) | 0.8077 | 0.7949 | |
| Clinical model | Internal test | 146 | 0.8210 (0.7701-0.8728) | 0.8235 | 0.7674 |
- Citation: Zhang Y, Yan S, Li YJ, Li QX, Wei LJ. Deep learning for predicting high-grade dysplasia in colorectal polyps using white-light endoscopy. World J Gastroenterol 2027; 33(3): 125206
- URL: https://www.wjgnet.com/1007-9327/full/v33/i3/125206.htm
- DOI: https://dx.doi.org/10.3748/wjg.125206