Performance of a fusion model based on multi-parametric MRI deep learning algorithm and clinicopathological features predicting lymph node metastasis in rectal cancer
10.12025/j.issn.1008-6358.2026.20260189
- VernacularTitle:多参数磁共振成像深度学习算法与临床病理特征融合模型预测直肠癌淋巴结转移的效能
- Author:
Songbin LIN
1
;
Yang LÜ
2
;
Peng ZHENG
2
Author Information
1. Department of General Surgery, Zhongshan Hospital (Xiamen), Fudan University, Xiamen 361015, Fujian, China.
2. Department of Colorectal Surgery, Zhongshan Hospital, Fudan University, Shanghai 200032, China.
- Publication Type:Shortarticle
- Keywords:
rectal cancer;
lymph node metastasis;
MRI;
deep learning;
clinicopathological features
- From:
Chinese Journal of Clinical Medicine
2026;33(4):668-674
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop and validate a fusion model that integrates a multi-parametric magnetic resonance imaging deep learning algorithm with clinicopathological factors to improve the accuracy of preoperative prediction for lymph node metastasis (LNM) in rectal cancer. Methods A total of 191 rectal cancer patients who received treatment at Zhongshan Hospital (Xiamen), Fudan University from January 2022 to December 2024 (modeling group, n＝141) and at Zhongshan Hospital, Fudan University from January to June 2021 (external validation group, n＝50) were retrospectively included. Magnetic resonance imaging and clinical data of all patients were collected to compare the diagnostic performance of the previously developed algorithm with that of two radiologists. Logistic regression was further employed to identify independent risk factors for LNM, and an integrated model combining ‘‘algorithm + clinicopathological features’’ was constructed. The clinical value of this model was evaluated using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). Results In the modeling group, the algorithm’s predictive performance for LNM (AUC＝0.763) was higher than those of two physicians (AUC＝0.675, 0.655). Multivariate analysis results indicated that carcinoembryonic antigen ＞5 μg/L and low differentiation were independent risk factors for LNM (P＜0.01). The fusion model improved predictive performance, with an AUC of 0.872 in the modeling group and 0.830 in the external validation group. DCA analysis showed that the fusion model provided higher net benefits across the clinically common threshold range. Conclusions The fusion model based on multi-parametric magnetic resonance image deep learning algorithm and clinical features shows good predictive performance for LNM of rectal cancer, which is superior to simple algorithms and manual interpretation. It is expected to become an auxiliary tool for preoperative risk stratification and treatment strategy selection of rectal cancer.
