Back to all papers

Multimodal multitask deep learning for predicting tertiary lymphoid structures and peritoneal recurrence in gastric cancer: a multicentre development, validation, and testing study.

November 19, 2026pubmed logopapers

Authors

Li Z,Zhang T,Wang W,Luo X,Zheng W,Han Z,Ahmad MU,Yuan Q,Chen C,Xi S,Wu L,Li W,Xie J,Feng W,Yu J,Kinahan PE,Li Z,Li G,Jiang Y

Affiliations (14)

  • Department of Bioengineering, University of Washington, Seattle, WA, USA; Department of Radiology, University of Washington, Seattle, WA, USA.
  • Department of General Surgery & Guangdong Provincial Key Laboratory of Precision Medicine for Gastrointestinal Tumor, Nanfang Hospital, Southern Medical University, Guangzhou, China.
  • Department of Gastric Surgery, Sun Yat-sen University Cancer Center, Guangzhou, China.
  • Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA, USA.
  • Department of Surgery, Stanford University School of Medicine, Stanford, CA, USA.
  • Department of Medical Imaging Center, Nanfang Hospital, Southern Medical University, Guangzhou, China.
  • The Reproductive Medical Center, The Seventh Affiliated Hospital of Sun Yat-sen University, Shenzhen, China.
  • Department of Pathology, the Third Affiliated Hospital of Kunming Medical University, Yunnan Cancer Hospital, Yunnan Cancer Center, Kunming, China.
  • Department of Pathology, Wake Forest University School of Medicine, Winston Salem, NC, USA.
  • Graduate Group of Epidemiology, University of California Davis, Davis, CA, USA.
  • Department of Pathology, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China.
  • Department of Radiology, the Third Affiliated Hospital of Kunming Medical University, Yunnan Cancer Hospital, Yunnan Cancer Center, Kunming, China.
  • Department of Gastrointestinal Surgery, School of Clinical Medicine, Tsinghua University, Beijing Tsinghua Changgung Hospital, Beijing, China.
  • Department of Radiation Oncology, Wake Forest University School of Medicine, Winston Salem, North Carolina, USA. Electronic address: [email protected].

Abstract

Peritoneal recurrence remains a common and fatal complication following curative gastric cancer surgery. Tertiary lymphoid structures are crucial indicators of immune response in cancer, and their maturation is associated with better prognosis and therapy outcomes. Evaluation of these structures could improve patient stratification for treatment of cancer but is hindered by the invasive and biased nature of histological methods. This study aimed to explore a non-invasive radiological method of assessment in patients with gastric cancer, using a novel multitask hybrid model to predict the presence and maturation status of tertiary lymphoid structures and peritoneal recurrence-free survival. We developed CTransNet, a multitask hybrid model combining a convolutional neural network (ResNet50) and a Transformer model to predict tertiary lymphoid structure maturation and peritoneal recurrence-free survival in patients with gastric cancer. The study used retrospective data from eight Chinese cohorts of patients treated for gastric cancer at Nanfang Hospital, Sun Yat-sen University Cancer Center, Yunnan Cancer Hospital, and Guangdong Provincial Hospital of Chinese Medicine (one training cohort, two internal validation cohorts, three external validation cohorts, and two independent cohorts of patients treated with anti-PD-1 or anti-PD-L1 inhibitors for at least two cycles). Patients were classified into three categories based on the presence and maturation of tertiary lymphoid structures (identified using haematoxylin-eosin and immunohistochemistry staining): TLS0 (absent tertiary lymphoid structures), TLS1 (primary follicle-like tertiary lymphoid structures without germinal centres), and TLS2 (secondary follicle-like tertiary lymphoid structures with germinal centres [ie, more mature structures]). Peritoneal recurrence was assessed using imaging (ultrasonography, CT, or PET-CT), clinical signs, or reoperation. We evaluated the accuracy of the multitask hybrid model in classifying patient-level tertiary lymphoid structure status using receiver operating characteristic analysis. Patients were stratified into risk groups using the median predicted score from the training cohort (high vs low), which was then applied to the validation cohorts. Survival outcomes (peritoneal recurrence-free survival, disease-free survival, and overall survival) were analysed using Kaplan-Meier methods and Cox proportional hazards models. We also analysed the relationships between predicted scores and responses to chemotherapy and immunotherapy. Patients in the Chinese cohorts were recruited between Jan 1, 2005, and Feb 28, 2022. In the training cohort, for classifying patient-level tertiary lymphoid structure status, values for area under the receiver operating characteristic curve were 0·910 (95% CI 0·882-0·938) for TLS0, 0·925 (0·899-0·951) for TLS1, and 0·953 (0·933-0·974) for TLS2. For predicting peritoneal recurrence-free survival, the model achieved a concordance index of 0·654 (95% CI 0·620-0·688) in the training cohort. Across all cohorts, patients with low model scores had substantially higher 5-year peritoneal recurrence rates than those with high scores (eg, 66·69%, 95% CI 60·11-73·27 vs 26·28%, 20·05-32·51; p<0·0001 in the training cohort). Multivariate analysis confirmed risk stratification based on the multitask hybrid model score as an independent predictor of peritoneal recurrence-free survival, disease-free survival, and overall survival. Analysis of patients with stage II or stage III disease in the training cohort and internal and external validation cohorts showed that in those with low predicted scores versus high predicted scores, adjuvant chemotherapy was associated with improved disease-free survival and overall survival in patients with low scores but not those with high scores. In the immunotherapy cohorts, objective response rates were higher for patients with high predicted scores than for those with low predicted scores (54·5% vs 5·6%; p<0·0001 in immunotherapy cohort 1 and 53·1% vs 9·7%; p=0·0010 in immunotherapy cohort 2). By accurately predicting tertiary lymphoid structure status, peritoneal recurrence, and survival, the multitask hybrid model shows strong potential as a non-invasive tool for personalised gastric cancer care. The model's potential to identify patients who might benefit from adjuvant chemotherapy and are more likely to respond to immunotherapy could help clinicians to optimise treatment strategies. None.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.