68/100

Moderate evidence

Cite
More
EmailLinkedInX

AI model reading scans sorted gastric cancer patients by recurrence risk, study finds

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

Evidence type
Controlled or validated comparison
Sources
1 sourcePeer-reviewed
Published
2 min read

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

Researchers built CTransNet, a model combining a convolutional neural network (ResNet50) and a Transformer, to estimate from radiological imaging whether gastric cancer patients have tertiary lymphoid structures, immune formations linked to prognosis, and how mature they are. It also predicts peritoneal recurrence-free survival. In the training cohort, the area under the curve was 0.910 for absent structures (TLS0), 0.925 for primary follicle-like structures (TLS1) and 0.953 for mature structures (TLS2).

Read the study 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 pati…

Multimodal multitask deep learning for predicting tertiary lymphoid structures and peritoneal recurrence in gastric cancer: a multicentre development, validation, and testing study
"374 MDG enhances operational readiness at Yokota" by Airman 1st Class David S. Calcote, U.S. Air Force, public domain.

The gain

In a retrospective study of eight Chinese gastric cancer cohorts, the CTransNet model split patients by risk: in the training cohort, 5-year peritoneal recurrence was 66.69% with low scores versus 26.28% with high scores.

In brief

Patients with low model scores had much higher 5-year peritoneal recurrence rates than those with high scores: 66.69% versus 26.28% in the training cohort. The score remained an independent predictor of peritoneal recurrence-free, disease-free and overall survival in multivariate analysis. Tissue-based assessment of these structures is invasive and biased, which is the gap a scan-based method would fill.

The score also tracked treatment response. Among stage II or III patients, adjuvant chemotherapy was linked to better disease-free and overall survival in those with low scores but not those with high scores. In two immunotherapy cohorts, objective response rates were 54.5% versus 5.6% and 53.1% versus 9.7% for high versus low scores. The authors describe the model as a potential tool for personalised care, not one proven in clinical use.

Main points

  1. CTransNet pairs ResNet50 with a Transformer and predicts tertiary lymphoid structure status (TLS0, TLS1, TLS2) and peritoneal recurrence-free survival from imaging.
  2. Training-cohort AUCs for classifying structure status were 0.910 (TLS0), 0.925 (TLS1) and 0.953 (TLS2).
  3. Five-year peritoneal recurrence in the training cohort was 66.69% for low-score patients versus 26.28% for high-score patients (p<0.0001).
  4. Risk stratification by model score was an independent predictor of peritoneal recurrence-free, disease-free and overall survival.
  5. Among two immunotherapy cohorts treated with anti-PD-1 or anti-PD-L1 drugs, response rates were 54.5% vs 5.6% and 53.1% vs 9.7% for high vs low scores.

The full story

The study used retrospective data from patients treated at Nanfang Hospital, Sun Yat-sen University Cancer Center, Yunnan Cancer Hospital and Guangdong Provincial Hospital of Chinese Medicine. The eight cohorts comprised one training cohort, two internal validation cohorts, three external validation cohorts and two independent immunotherapy cohorts. Patients were recruited between 1 January 2005 and 28 February 2022.

Structure status was established with haematoxylin-eosin and immunohistochemistry staining, and peritoneal recurrence was assessed by imaging, clinical signs or reoperation. Patients were split into high and low risk groups using the median predicted score from the training cohort, a cut-off then applied to the validation cohorts. Survival was analysed with Kaplan-Meier methods and Cox models.

For predicting peritoneal recurrence-free survival, the model reached a concordance index of 0.654 in the training cohort. In stage II and III patients, low scores identified those in whom adjuvant chemotherapy was associated with better survival, while high-score patients showed no such association. The authors suggest this could help clinicians optimise treatment strategies.

Sources

1 on file
  1. The Lancet Digital Healthdoi:10.1016/j.landig.2026.101075
    Peer-reviewed2026-11-19

Talk it over

Three questions to start on, two more if you have time
In the training cohort, 5-year peritoneal recurrence was 66.69% for low-score patients versus 26.28% for high-score patients. How would a patient and a doctor use a score like this differently?

Start with a question

AI model reading scans sorted gastric cancer patients by recurrence risk, study findshttps://truvace.com/item/multimodal-multitask-deep-learning-for-predicting-tertiary-lymphoid-structures-a

  1. In the training cohort, 5-year peritoneal recurrence was 66.69% for low-score patients versus 26.28% for high-score patients. How would a patient and a doctor use a score like this differently?

    Answer in the thread
  2. The model was built and tested on retrospective data from eight cohorts at Chinese hospitals, and its concordance index was 0.654 in the training cohort. What can this design tell us about patients treated elsewhere?

    Answer in the thread
  3. In the immunotherapy cohorts, response rates were 54.5% versus 5.6% for high versus low scores. What finding would convince you that the score should guide treatment choices?

    Answer in the thread
If you have more time
  • Adjuvant chemotherapy was linked to better survival in stage II or III patients with low scores but not high scores. What kind of study would you want before anyone acted on that link?

    Answer in the thread
  • Tissue-based assessment of these structures is invasive, while the model works from radiological imaging. How would you weigh a scan-based estimate against a tissue sample in a clinic you know?

    Answer in the thread

What readers say