Related Experiment Video
Updated: Jun 24, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.1K
Res-TransNet: A Hybrid deep Learning Network for Predicting Pathological Subtypes of lung Adenocarcinoma in CT Images
Yue Su1, Xianwu Xia2, Rong Sun1
1School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China.
Journal of Imaging Informatics in Medicine
|June 11, 2024
Summary
A novel hybrid deep learning model, Res-TransNet, accurately predicts lung adenocarcinoma subtypes from CT scans. This advanced network integrates Residual Network (ResNet) and Vision Transformer (ViT) for improved diagnostic accuracy in early-stage lung cancer.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Oncology
- Radiology
Background:
- Early-stage lung adenocarcinoma diagnosis relies on accurate pathological subtyping.
- Deep learning models like Residual Network (ResNet) and Vision Transformer (ViT) show promise in medical image analysis.
- Integrating different deep learning architectures may enhance classification performance.
Purpose of the Study:
- To develop and evaluate a CT-based hybrid deep learning network, Res-TransNet, for predicting pathological subtypes of early-stage lung adenocarcinoma.
- To compare the performance of the hybrid Res-TransNet model against standalone ResNet and ViT models.
- To assess the potential of Res-TransNet in assisting radiologists with precision diagnosis.
Main Methods:
- A hybrid deep learning network (Res-TransNet) was developed by integrating 3D ResNet and ViT.
- The model was trained and validated on 1411 pathologically confirmed ground-glass nodules (GGNs) from two centers.
- Two classification tasks were performed: predicting invasive adenocarcinoma (IAC) from Non-IAC (Task 1) and classifying three subtypes (Task 2).
Main Results:
- The optimal Res-TransNet model achieved high AUC values of 0.986 (internal) and 0.933 (external) for Task 1, significantly outperforming ResNet and ViT.
- For Task 2, the fusion model achieved 68.3% accuracy and 66.1% weighted F1 score on the external validation set.
- Res-TransNet demonstrated significantly improved classification performance compared to the individual ResNet and ViT models.
Conclusions:
- The developed Res-TransNet model shows significant potential for improving the classification performance of lung adenocarcinoma subtypes.
- This hybrid deep learning approach can assist radiologists in achieving more precise diagnoses of early-stage lung cancer.
- Further validation and clinical implementation of Res-TransNet could enhance early lung cancer detection and management.
