Truace
Subscribe

Multimodal Transformer Integrating CT and Pathology for Postoperative Risk Stratification in Stage I Non-Small Cell Lung Cancer

Purpose To develop and evaluate an interpretable, multimodal deep learning model to improve postoperative risk stratification and predict adjuvant therapy benefit in stage I non-small cell lung cancer (NSCLC). Materials and Methods In this multicenter retrospective study of patients with stage I NSCLC, a multimodal model called the Deep Learning-Radiomic-Text Transformer (DLRTransformer) was developed by integrating multiview 2.5-dimensional deep learning features from CT scans, radiomics, clinical variables, an…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1589· cite & verify
Multimodal Transformer Integrating CT and Pathology for Postoperative Risk Stratification in Stage I Non-Small Cell Lung Cancer
Pancoast Tumor 2 by Jmarchn. CC BY-SA 3.0 · https://creativecommons.org/licenses/by-sa/3.0

In brief

Researchers built the Deep Learning-Radiomic-Text Transformer, which combines CT-derived deep learning features, radiomics, clinical variables, and pathology report embeddings from a large language model to predict disease-free survival after surgery for stage I non-small cell lung cancer. It was developed and tested in a retrospective study of 1073 patients from five centers.

On independent external datasets the model scored a concordance index of 0.77-0.78 versus 0.60-0.65 for guideline-based stratification, and it separated patients who gained from adjuvant therapy from those who did not. This could help target treatment in early-stage disease, but the evidence is retrospective, so prospective confirmation is still needed.

Main points

  1. The model fuses multiview 2.5-dimensional CT deep learning features, radiomics, clinical variables, and LLM-extracted pathology report embeddings using a central transformer.
  2. External concordance index was 0.77-0.78, compared with 0.60-0.65 for guideline-based risk stratification.
  3. Risk scores were linked to tumor microenvironment features, including increased neoplastic cellularity and reduced tertiary lymphoid structures.
  4. A treatment-by-risk interaction was observed (interaction HR 0.14; 95% CI 0.06, 0.32; P < .01).

The gain

On external datasets, the DLRTransformer predicted disease-free survival better than guideline-based risk stratification and flagged high-risk stage I NSCLC patients who benefited from adjuvant therapy.

The rundown

The cohort comprised 1073 patients enrolled from five centers, with a mean age of 60.8 years and 550 men. The study was multicenter and retrospective, and performance was assessed on independent external datasets.

In propensity score-matched subcohorts, adjuvant therapy improved disease-free survival in high-risk patients (HR 0.30; 95% CI 0.11, 0.84; P = .02). In low-risk patients no clear benefit was seen (HR 1.70; 95% CI 0.97, 2.99; P = .06).

The authors analyzed associations between model-derived risk scores and tumor microenvironment phenotypes to assess biologic interpretability, finding links to higher neoplastic cellularity and fewer tertiary lymphoid structures.

Sources

  1. Peer-reviewedRadiology: Imaging Cancer2026-11-01

The debate