Interpretable deep learning for multicenter gastric cancer T staging from CT images.
Authors
Affiliations (14)
Affiliations (14)
- Department of Gastric Surgery, Cancer Hospital of China Medical University, Liaoning Cancer Hospital & Institute, Cancer Hospital of Dalian University of Technology, Shenyang, Liaoning, China.
- Department of Radiotherapy, Cancer Hospital of China Medical University, Liaoning Cancer Hospital & Institute, Cancer Hospital of Dalian University of Technology, Shenyang, Liaoning, China.
- School of Pharmacy, China Medical University, Shenyang, Liaoning, China.
- Shenyang Mental Health Center, Shenyang, Liaoning, China.
- Outpatient Clinic, 15th Retirement Cadre Sanatorium of Shenyang, Liaoning Military Region, Shenyang, Liaoning, China.
- Department of Oncology, General Hospital of Northern Theater Command, Shenyang, Liaoning, China.
- Department of Radiology, Hunnan Central Hospital, Shenyang, Liaoning, China.
- China-UK Joint College, China Medical University, Shenyang, Liaoning, China.
- School of Health Management, China Medical University, Shenyang, Liaoning, China.
- Department of Gastric Surgery, Zhejiang Cancer Hospital, Hangzhou, Zhejiang, China.
- Department of Oncology, General Hospital of Northern Theater Command, Shenyang, Liaoning, China. [email protected].
- Department of Nuclear Medicine, Shengjing Hospital of China Medical University, Shenyang, Liaoning, China. [email protected].
- School of Pharmacy, China Medical University, Shenyang, Liaoning, China. [email protected].
- Department of Gastric Surgery, Cancer Hospital of China Medical University, Liaoning Cancer Hospital & Institute, Cancer Hospital of Dalian University of Technology, Shenyang, Liaoning, China. [email protected].
Abstract
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 a nomogram 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.