Interpretable deep learning for multicenter gastric cancer T staging from CT images

Sep 29, 2026

Preoperative T staging of gastric cancer is essential for appropriate therapeutic stratification, yet contrast-enhanced CT interpretation remains subjective with inconsistent reliability, especially for subtle serosal invasion and T2–T3 separation. GTRNet is an interpretable end-to-end deep-learning framework that classifies T1–T4 from routine CT without manual segmentation or annotation. In a retrospective multicenter cohort of 1792 patients and two external cohorts, GTRNet showed strong discrimination and accuracy and generated Grad-CAM heatmaps. A rad-score combined with clinical factors produced a well-calibrated nomogram with greater net clinical benefit.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Gastric cancer T staging challenges

Limitations of conventional contrast-enhanced CT

Role and limits of endoscopic ultrasound (EUS) and other imaging

Interpretable deep learning approach

GTRNet framework and training strategy

Evaluation across internal and external cohorts

Model interpretability with Grad-CAM

Clinical decision support

Deep-learning rad-score and nomogram construction

Calibration and net clinical benefit

npj | digital medicine Article | Published in partnership with Seoul National University Bundang Hospital | | | | --- | --- | --- | | [https://doi.org/10.1038/s41746-025-02002-5](https://doi.org/10.1038/s41746-025-02002-5) | | | | Interpretable deep learning for multicenter gastric cancer T staging from CT images
Check for updates | | | | Guoliang Zheng1,12, Huan Wang2,12, Xiaomiao Chai3, Xin Xin4, Fuze Li1, Hongfei Li5, Yaoyang Ban1, Jinshi Wang1, Xinhui Qi6, Yingjie Li7, Zishuo Yan3, Fangning Guo8, Zhixue Jiang9, Dantong Zhu6, Yanqiang Zhang10, Zhendong Zheng6 , Xin Zhang11 , Jing Zhang3 & Yan Zhao 1 | | | | Preoperative T staging of gastric cancer is critical for therapeutic stratiﬁcation, yet conventional contrast-enhanced CT interpretation shows subjectivity and inconsistent reliability. We developed GTRNet, an interpretable end-to-end deep-learning framework that classiﬁes T1–T4 from routine CT without manual segmentation or annotation. In a retrospective multicenter study of 1792 patients, CT images underwent standardized preprocessing and the largest axial tumor slice was used for training; performance was then tested in two independent external cohorts. GTRNet achieved high discrimination (AUC 0.86–0.95) and accuracy (81–85%) in internal and external tests, surpassing radiologists. Grad-CAM heatmaps localized attention to the gastric wall and serosa. Combining a deep-learning rad-score with tumor size, differentiation and Lauren subtype, we constructed anomogram with good calibration and higher net clinical beneﬁt than conventional approaches. This automated and interpretable pipeline may standardize CT-based staging and support preoperative decision-making and neoadjuvant-therapy selection. | | | | Gastric cancer is a major global health burden and remains among the top causes of cancer-related deaths worldwide1. Despite advances, many patients still present with advanced disease due to limited early detection, complicating treatment2. Accurate preoperative T staging guides decisions from endoscopic resection for T1 to multimodal therapy for T3/T4 disease3–5. Accurate T staging thus enables more tailored interventions, potentially improving survival and reducing unnecessary morbidity.
Contrast-enhanced CT is the standard for preoperative evaluation, but conventional interpretation has known limitations. The overall accuracy of CT-based gastric T staging is often reported around 65–75%, with particular difﬁculty in discriminating T2 from T3 tumors and in recognizing subtle serosal invasion6. Endoscopic ultrasound (EUS) can visualize distinct layers of the gastric wall, thereby helping to distinguish | T1 disease from deeper invasion; however, its accuracy declines in advanced tumors and proximal lesions, and the technique is highly operator-dependent7,8. Other adjunct imaging methods, including double contrast-enhanced ultrasound, may improve diagnostic performance in experienced hands but are not universally available. Consequently, there is a clear clinical need for new tools to enhance the objectivity and accuracy ofT staging.
Artiﬁcial intelligence (AI), especially deep learning using convolutional neural networks (CNNs), has emerged as a powerful approach to analyzing medical images9. CNNs can learn multi-scale textural and morphological features from large volumes of data, surpassing traditional machine learning or radiologist interpretation in tasks such as tumor segmentation, disease classiﬁcation, and prognostic modeling. Notably, deep learning models have achieved high diagnostic | | 1Department of Gastric Surgery, Cancer Hospital of China Medical University, Liaoning Cancer Hospital & Institute, Cancer Hospital of Dalian University of Technology, Shenyang, Liaoning, China. 2Department of Radiotherapy, Cancer Hospital of China Medical University, Liaoning Cancer Hospital & Institute, Cancer Hospital of Dalian University of Technology, Shenyang, Liaoning, China. 3School of Pharmacy,

FAQs

What problem does GTRNet address in gastric cancer care?

It targets the subjectivity and inconsistent reliability of preoperative gastric cancer T staging using contrast-enhanced CT, especially in distinguishing similar categories such as T2 versus T3.

How does GTRNet perform T staging from CT images?

GTRNet provides fully automated four-class classification (T1–T4) from standard portal venous phase CT, using a single axial slice of the largest tumor cross-section and a modified ResNet-152 backbone.

How is the model made interpretable and used clinically?

Grad-CAM heatmaps localize attention to relevant gastric wall and serosa regions. A combined deep-learning rad-score and clinical variables form a nomogram with good calibration and improved net clinical benefit over conventional approaches.

Sep 29, 2026

Uploaded by Maya Linwood