Development of a Deep Learning Model for T1N0 Gastric Cancer Diagnosis Using 2.5D Radiomic Data in Preoperative CT Images

Abstract

Background: Early detection and precise preoperative staging of early gastric cancer (EGC) are critical, this study aims to develop a deep learning model using portal venous phase CT images to accurately distinguish EGC without lymph node metastasis (LNM).

Methods: This study included 3164 patients with gastric cancer (GC) who underwent radical surgery at two medical centers in China from 2006 to 2019. 2.5D radiomic data and multi instance learning (MIL) were novel approaches applied in this study. The features selected from MIL were then apply to six machine learning models to build a pT1N0 predicting model. Area under the curve (AUC) of the receiver operator characteristic (ROC) curves and decision curve analysis (DCA) were used to evaluate the diagnostic effectiveness and the clinical utility.

Findings: Basing 2.5D radiomic data and MIL to select features, the ResNet101 model combined with the XGBoost model represented a satisfactory performance for diagnosing pT1N0 GC with AUC of 0.920 (95% CI: 0.907-0.933), 0.774 (95% CI: 0.738-0.810) and 0.744 (95% CI: 0.686-0.803) in the training, internal and external validation cohorts, respectively. Furthermore, the training cohort had an accuracy of 86.3%, a sensitivity of 84.9%, and a specificity of 89.6%, the internal validation cohort had an accuracy of 76.7%, a sensitivity of 76.5%, and a specificity of 77.4%, meanwhile, the external validation cohort had an accuracy of 74.9%, a sensitivity of 72.0%, and a specificity of 83.7%. The 2.5D MIL-based model demonstrated a markedly superior predictive performance compared to traditional radiomics models and clinical models.

Interpretation: We firstly constructed a deep learning prediction model based on 2.5D radiomics and MIL for effectively diagnosing pT1N0 GC patients, which provides valuable information for the individualized treatment selection.

Funding: This study was supported by The National Key Research and Development Program of China (2021YFA0910100), Healthy Zhejiang One Million People Cohort (K-20230085), Post-doctoral Innovative Talent Support Program (BX2023375), National Natural Science Foundation of China (82304946, 82473489, 82403546), Natural Science Foundation of Zhejiang Province (LR21H280001, MS25H160146, TGY23H160038), and The Medicine and Health Science Fund of Zhejiang Province (2025KY047, 2023KY073).

Declaration of Interest: All coauthors report no conflict of interest.

Ethical Approval:  The study has been approved by the Ethics Committees of all participating centers (IRB-2024- 220).

Keywords: Early gastric cancer, Lymph node metastasis, Deep Learning, Multi instance learning, Radiomics