Related Experiment Videos
Multimodal multitask deep learning for grading management system in non-small cell lung cancer.
Xinyue Liu1,2, Fang Dai3, Jiawei Dai4
1Department of Medical Oncology, Shanghai Pulmonary Hospital, Tongji University Medical School Cancer Institute, Tongji University School of Medicine, Shanghai, China.
Nature Communications
|June 15, 2026
Summary
A new deep learning system accurately predicts non-small cell lung cancer (NSCLC) subtype, stage, and survival risk using PET/CT scans and clinical data, improving patient management.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate histologic subtyping, TNM staging, and prognostic assessment are crucial for non-small cell lung cancer (NSCLC) management.
- Challenges in NSCLC management include tumor heterogeneity, limited biomarker performance, and diagnostic uncertainty.
Purpose of the Study:
- To develop and validate a multimodal, multi-task deep learning scoring system (MM-DLS) for non-invasive prediction of NSCLC subtype, stage, and survival risk.
- To integrate pretreatment PET/CT images with clinical variables for enhanced NSCLC assessment.
Main Methods:
- Development and validation of a multimodal, multi-task deep learning scoring system (MM-DLS).
- Integration of pretreatment PET/CT images and clinical variables.
- Validation in a cohort of 4,164 patients from multiple centers.
Main Results:
- MM-DLS achieved an area under the ROC curve of 0.86 for histologic subtype classification.
- MM-DLS demonstrated high accuracy for TNM staging: 0.86 for stages I-II, 0.86 for stage III, and 0.88 for stage IV.
- The model showed strong discrimination for 1-, 3-, and 5-year survival across various treatment regimens.
Conclusions:
- MM-DLS provides an effective, non-invasive framework for NSCLC subtype prediction, staging, and prognostic stratification.
- This deep learning approach can significantly aid in the clinical management of non-small cell lung cancer.