Prediction of delayed post-polypectomy bleeding using a multimodal model

doi:10.3969/j.issn.1674-8115.2026.04.011

Abstract

Abstract:

Objective ·To develop a multimodal prediction model that integrates endoscopic wound images with clinical features, leveraging deep learning and machine learning techniques to predict the risk of delayed post-polypectomy bleeding (DPPB). Methods ·The clinical data and postoperative endoscopic wound images of patients who underwent endoscopic colorectal polypectomy at two hospitals were retrospectively collected. The study was designed in three stages. In the first stage, a YOLOv11 model trained via transfer learning was used to automatically detect and segment wound areas, extracting regions of interest (ROI). In the second stage, ROI images were input into a deep neural network based on the ResNet50 architecture. Supervised learning was performed using DPPB occurrence as the label, and a standardized endoscopic imaging score (E-Score) was generated. In the third stage, the E-Score was combined with clinical features. Feature selection was conducted using LASSO regression, and the selected variables were input into multiple machine learning algorithms to construct a multimodal DPPB prediction model. Finally, a web-based application was developed using the Streamlit framework to facilitate clinical use. Results ·A total of 2 782 patients who underwent colorectal polypectomy were included, among whom 228 (8.20%) developed DPPB. In the test set, the multimodal prediction model built using the XGBoost algorithm (PrismDPPB) achieved the best performance, with an area under the curve(AUC) of 0.831 (95%CI 0.81‒0.85), outperforming other models. It also showed superior performance in accuracy (79.78%), sensitivity (77.28%), positive predictive value (81.32%), and F1 score (79.25%). Feature importance analysis revealed that the six most contributive variables were E-Score, maximum polyp base diameter, sex, presence of a stalk, body mass index, and polyp location. For model interpretability, SHAP plots were used to visualize and explain the prediction results. Conclusion ·The multimodal prediction model developed by integrating image-based features with clinical characteristics, along with the accompanying web application, demonstrates strong practicality and clinical potential for DPPB risk assessment.

Key words: colorectal polypectomy, delayed post-polypectomy bleeding, web application, deep learning, machine learning, multimodal model

CLC Number:

R574.62

Zhu Menglin, Liu Xiao, Xu Xiaodan, Wang Ganhong, Xia Kaijian, Chen Jian. Prediction of delayed post-polypectomy bleeding using a multimodal model[J]. Journal of Shanghai Jiao Tong University (Medical Science), 2026, 46(4): 509-520.

Figures/Tables 12

TrendMD

References 21

[1]	Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2024, 74(3): 229-263.
[2]	Lopes S R, Martins C, Santos I C, et al. Colorectal cancer screening: a review of current knowledge and progress in research[J]. World J Gastrointest Oncol, 2024, 16(4): 1119-1133.
[3]	Brenner H, Chang-Claude J, Seiler C M, et al. Protection from colorectal cancer after colonoscopy: a population-based, case-control study[J]. Ann Intern Med, 2011, 154(1): 22-30.
[4]	Shaukat A, Levin T R. Current and future colorectal cancer screening strategies[J]. Nat Rev Gastroenterol Hepatol, 2022, 19(8): 521-531.
[5]	Gupta S, Lieberman D, Anderson J C, et al. Recommendations for follow-up after colonoscopy and polypectomy: a consensus update by the US multi-society task force on colorectal cancer[J]. Gastrointest Endosc, 2020, 91(3): 463-485.e5.
[6]	Jaruvongvanich V, Prasitlumkum N, Assavapongpaiboon B, et al. Risk factors for delayed colonic post-polypectomy bleeding: a systematic review and meta-analysis[J]. Int J Colorectal Dis, 2017, 32(10): 1399-1406.
[7]	王禹毅, 卜志军, 李元晞, 等. 临床预测模型变量筛选方法及比较[J]. 现代中医临床, 2024, 31(2): 6-12.
	Wang Y Y, Bu Z J, Li Y X, et al. Variable selection methods and comparison in clinical prediction models[J]. Modern Clinical Chinese Medicine, 2024, 31(2): 6-12.
[8]	Cao J, Long S K, Liu H, et al. Constructing a prediction model for acute pancreatitis severity based on liquid neural network[J]. Sci Rep, 2025, 15(1): 16655.
[9]	Zhang R F, Yin M Y, Jiang A Q, et al. Application value of the automated machine learning model based on modified computed tomography severity index combined with serological indicators in the early prediction of severe acute pancreatitis[J]. J Clin Gastroenterol, 2024, 58(7): 692-701.
[10]	Liu L J, Zhang R F, Shi Y, et al. Automated machine learning for predicting liver metastasis in patients with gastrointestinal stromal tumor: a SEER-based analysis[J]. Sci Rep, 2024, 14(1): 12415.
[11]	Russell B C, Torralba A, Murphy K P, et al. LabelMe: a database and web-based tool for image annotation[J]. Int J Comput Vis, 2008, 77(1): 157-173.
[12]	Shin H C, Roth H R, Gao M C, et al. Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning[J]. IEEE Trans Med Imaging, 2016, 35(5): 1285-1298.
[13]	Handelman G S, Kok H K, Chandra R V, et al. Peering into the black box of artificial intelligence: evaluation metrics of machine learning methods[J]. AJR Am J Roentgenol, 2019, 212(1): 38-43.
[14]	Nohara Y, Matsumoto K, Soejima H, et al. Explanation of machine learning models using shapley additive explanation and application for real data in hospital[J]. Comput Meth Programs Biomed, 2022, 214: 106584.
[15]	Fahmy A S, Csecs I, Arafati A, et al. An explainable machine learning approach reveals prognostic significance of right ventricular dysfunction in nonischemic cardiomyopathy[J]. JACC Cardiovasc Imaging, 2022, 15(5): 766-779.
[16]	Bendall O, James J, Pawlak K M, et al. Delayed bleeding after endoscopic resection of colorectal polyps: identifying high-risk patients[J]. Clin Exp Gastroenterol, 2021, 14: 477-492.
[17]	Zhang X Z, Jiang X X, Shi L. Risk factors for delayed colorectal postpolypectomy bleeding: a meta-analysis[J]. BMC Gastroenterol, 2024, 24(1): 162.
[18]	Albouys J, Montori Pina S, Boukechiche S, et al. Risk of delayed bleeding after colorectal endoscopic submucosal dissection: the Limoges Bleeding Score[J]. Endoscopy, 2024, 56(2): 110-118.
[19]	Lu Y, Zhou X Y, Chen H, et al. Establishment of a model for predicting delayed post-polypectomy bleeding: a real-world retrospective study[J]. Front Med, 2022, 9: 1035646.
[20]	王敏, 韦雪连, 李常伟, 等. 内镜下结直肠腺瘤性息肉的特征及经切除术后迟发出血的危险因素分析[J]. 胃肠病学和肝病学杂志, 2024, 33(6): 714-718.
	Wang M, Wei X L, Li C W, et al. Characteristics of colorectal adenomatous polyps under endoscopy and analysis of risk factors for delayed bleeding after resection[J]. Chinese Journal of Gastroenterology and Hepatology, 2024, 33(6): 714-718.
[21]	陈健, 王珍妮, 夏开建, 等. 基于深度学习的结直肠息肉内镜图像分割和分类方法比较[J]. 上海交通大学学报(医学版), 2024, 44(6): 762-772.
	Chen J, Wang Z N, Xia K J, et al. Comparative study on methods for colon polyp endoscopic image segmentation and classification based on deep learning[J]. Journal of Shanghai Jiao Tong University (Medical Science), 2024, 44(6): 762-772.

Feature	Training set (n=1 947)	Test set (n=835)	t/Z/χ²	P value
Age/year	58.50±11.30	58.19±10.93	0.668	0.507
DBIL/(μmol·L^-1)	5.01 (1.73, 8.98)	4.79 (1.94, 8.14)	1.801	0.214
TBIL/(μmol·L^-1)	29.01 (10.97, 48.23)	26.47 (11.87, 46.33)	1.771	0.218
Cholesterol/(mmol·L^-1 )	7.88 (6.76, 8.98)	7.75 (6.68, 8.80)	1.959	0.047
AST/(U·L^-1 )	100.59 (78.61, 124.08)	101.89 (79.45, 122.13)	0.293	0.946
Platelet count/(10⁹·L^-1)	246.20 (186.34, 311.28)	249.63 (180.30, 311.17)	-0.023	0.982
BUN/(mg·dL^-1 )	6.33 (4.76, 7.86)	6.29 (4.85, 7.96)	-0.512	0.545
ALP/(U·L^-1 )	110.63 (81.65, 140.05)	109.15 (80.61, 140.26)	0.523	0.531
BMI/(kg·m^-2)	24.10 (22.10, 26.20)	24.00 (22.10, 26.20)	0.316	0.667
PT/s	11.73±0.82	11.70±0.86	0.804	0.415
INR	1.02 (0.98, 1.06)	1.02 (0.98, 1.06)	1.391	0.110
Albumin/(g·L^-1)	41.67 (38.33, 44.34)	41.59 (38.52, 44.49)	-0.711	0.712
Maximum basal diameter/cm	1.76 (1.33, 2.27)	1.69 (1.29, 2.19)	2.023	0.056
E-Score	0 (0, 0.02)	0 (0, 0.03)	-1.512	0.042
Smoking/n(%)			2.496	0.114
No	1 801 (92.5)	787 (94.3)
Yes	146 (7.5)	48 (5.7)
Alcohol consumption/n(%)			0.477	0.490
No	1 588 (81.6)	671 (80.4)
Yes	359 (18.4)	164 (19.6)
Hypertension/n(%)	949 (48.7)	420 (50.3)	0.507	0.477
No
Yes	998 (51.3)	415 (49.7)
Coronary heart disease/n(%)			0.342	0.559
No	344 (17.7)	156 (18.7)
Yes	1 603 (82.3)	679 (81.3)
Gender/n(%)			1.384	0.239
Female	1 026 (52.7)	461 (55.2)
Male	921 (47.3)	374 (44.8)
Diabetes mellitus/n(%)	1 614 (82.9)	675 (80.8)	1.560	0.212
No
Yes	333 (17.1)	160 (19.2)
Pedunculated status/n(%)			0.761	0.383
No	1 370 (70.4)	573 (68.6)
Yes	577 (29.6)	262 (31.4)
Antithrombotic drug use history/n(%)			0.010	0.982
No	1 847 (94.9)	793 (95.0)
Yes	100 (5.1)	42 (5.0)
Polyp location/n(%)			5.139	0.077
Left colon	769 (39.5)	334 (40.0)
Right colon	571 (29.3)	273 (32.7)
Bilateral colon	607 (31.2)	228 (27.3)
Number of polyps/n(%)			0.153	0.696
<3	1 551 (79.7)	659 (78.9)
≥3	396 (20.3)	176 (21.1)

Feature	Training set (n=1 947)	Test set (n=835)	t/Z/χ²	P value
Age/year	58.50±11.30	58.19±10.93	0.668	0.507
DBIL/(μmol·L^-1)	5.01 (1.73, 8.98)	4.79 (1.94, 8.14)	1.801	0.214
TBIL/(μmol·L^-1)	29.01 (10.97, 48.23)	26.47 (11.87, 46.33)	1.771	0.218
Cholesterol/(mmol·L^-1 )	7.88 (6.76, 8.98)	7.75 (6.68, 8.80)	1.959	0.047
AST/(U·L^-1 )	100.59 (78.61, 124.08)	101.89 (79.45, 122.13)	0.293	0.946
Platelet count/(10⁹·L^-1)	246.20 (186.34, 311.28)	249.63 (180.30, 311.17)	-0.023	0.982
BUN/(mg·dL^-1 )	6.33 (4.76, 7.86)	6.29 (4.85, 7.96)	-0.512	0.545
ALP/(U·L^-1 )	110.63 (81.65, 140.05)	109.15 (80.61, 140.26)	0.523	0.531
BMI/(kg·m^-2)	24.10 (22.10, 26.20)	24.00 (22.10, 26.20)	0.316	0.667
PT/s	11.73±0.82	11.70±0.86	0.804	0.415
INR	1.02 (0.98, 1.06)	1.02 (0.98, 1.06)	1.391	0.110
Albumin/(g·L^-1)	41.67 (38.33, 44.34)	41.59 (38.52, 44.49)	-0.711	0.712
Maximum basal diameter/cm	1.76 (1.33, 2.27)	1.69 (1.29, 2.19)	2.023	0.056
E-Score	0 (0, 0.02)	0 (0, 0.03)	-1.512	0.042
Smoking/n(%)			2.496	0.114
No	1 801 (92.5)	787 (94.3)
Yes	146 (7.5)	48 (5.7)
Alcohol consumption/n(%)			0.477	0.490
No	1 588 (81.6)	671 (80.4)
Yes	359 (18.4)	164 (19.6)
Hypertension/n(%)	949 (48.7)	420 (50.3)	0.507	0.477
No
Yes	998 (51.3)	415 (49.7)
Coronary heart disease/n(%)			0.342	0.559
No	344 (17.7)	156 (18.7)
Yes	1 603 (82.3)	679 (81.3)
Gender/n(%)			1.384	0.239
Female	1 026 (52.7)	461 (55.2)
Male	921 (47.3)	374 (44.8)
Diabetes mellitus/n(%)	1 614 (82.9)	675 (80.8)	1.560	0.212
No
Yes	333 (17.1)	160 (19.2)
Pedunculated status/n(%)			0.761	0.383
No	1 370 (70.4)	573 (68.6)
Yes	577 (29.6)	262 (31.4)
Antithrombotic drug use history/n(%)			0.010	0.982
No	1 847 (94.9)	793 (95.0)
Yes	100 (5.1)	42 (5.0)
Polyp location/n(%)			5.139	0.077
Left colon	769 (39.5)	334 (40.0)
Right colon	571 (29.3)	273 (32.7)
Bilateral colon	607 (31.2)	228 (27.3)
Number of polyps/n(%)			0.153	0.696
<3	1 551 (79.7)	659 (78.9)
≥3	396 (20.3)	176 (21.1)

Model	Sensitivity/%	Specificity/%	Accuracy/%	PPV/%	NPV/%	F1 score/%	AUC (95%CI)
LR	71.80	76.14	73.97	75.03	73.00	73.38	0.809 (0.791‒0.833)
DCT	62.53	80.57	71.56	76.27	68.29	68.72	0.771 (0.752‒0.790)
RF	66.06	81.10	73.58	77.73	70.52	71.42	0.804 (0.781‒0.821)
XGBoost	77.28^①	82.27	79.78^①	81.32^①	78.39^①	79.25^①	0.831 (0.814‒0.852)
LNN	63.32	82.53^①	72.93	78.35	69.26	70.04	0.807 (0.790‒0.833)

Model	Sensitivity/%	Specificity/%	Accuracy/%	PPV/%	NPV/%	F1 score/%	AUC (95%CI)
LR	71.80	76.14	73.97	75.03	73.00	73.38	0.809 (0.791‒0.833)
DCT	62.53	80.57	71.56	76.27	68.29	68.72	0.771 (0.752‒0.790)
RF	66.06	81.10	73.58	77.73	70.52	71.42	0.804 (0.781‒0.821)
XGBoost	77.28^①	82.27	79.78^①	81.32^①	78.39^①	79.25^①	0.831 (0.814‒0.852)
LNN	63.32	82.53^①	72.93	78.35	69.26	70.04	0.807 (0.790‒0.833)