A preoperative predictive model for lymphovascular invasion in poorly differentiated gastric adenocarcinoma utilizing EUS-based deep learning and radiomics
Aiming to create and validate a nomogram using endoscopic ultrasound (EUS) images to predict lymphovascular invasion (LVI) in poorly differentiated gastric adenocarcinoma (GC). A retrospective analysis was performed on a sample of 252 patients, which was divided into a training cohort ( n = 176) and a test cohort ( n = 76). Radiomics and deep learning (DL) features were extracted from EUS images and selected via LASSO. Various machine learning algorithms were used to build DL and radiomics models. The best DL and radiomics models were combined into a nomogram. This model’s performance was assessed. The integrated nomogram, incorporating the optimal DL and radiomics signatures, exhibits area under the curve (AUC) values of 0.953 in the training cohort and 0.878 in the test cohort. Calibration curves and the Hosmer-Lemeshow test showed no significant difference between predicted and observed probabilities. DCA indicated a substantial net benefit. This study developed and validated a novel nomogram using EUS imaging and machine learning, enhancing the clinical utility of EUS for assessing LVI status in poorly differentiated GC.
Authors
- Shanyu Qin (ORCID: https://orcid.org/0000-0002-3162-3070)
- Shuangyang Mo
- Guoqing Ouyang (ORCID: https://orcid.org/0000-0002-0130-7958)
- Huaiyang Cai
- Nan Yi (ORCID: https://orcid.org/0009-0001-1355-6505)
- Qiuju Huang
- Rili Jiang
- Binbin Huang
- Yingwei Wang
- Yanchang Lei
- Yan Zhang
- Ning Liu
Institutions
- Guangxi Medical University (CN)
- The People's Hospital of Guangxi Zhuang Autonomous Region (CN)
- First Affiliated Hospital of GuangXi Medical University (CN)
- Shandong Provincial QianFoShan Hospital (CN)
- Shandong First Medical University (CN)
Publication Details
- Journal
- BMC Medical Informatics and Decision Making
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1186/s12911-026-03895-2
- Primary Topic
- Radiomics and Machine Learning in Medical Imaging
- Type
- article
- Field-Weighted Citation Impact
- 0.00