Related Experiment Video
Updated: Jun 11, 2025
07:30
Electromagnetic Navigation Transthoracic Nodule Localization for Minimally Invasive Thoracic Surgery
Published on: May 4, 2022
3.3K
A vision transformer-based deep transfer learning nomogram for predicting lymph node metastasis in lung
Chuanyu Chen1, Yi Luo1, Qiuyang Hou1
1Department of Radiology, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, Anhui, China.
Medical Physics
|September 28, 2024
Summary
A novel deep transfer learning nomogram using vision transformers accurately predicts lymph node metastasis in lung adenocarcinoma patients. This tool aids clinicians in treatment planning by improving diagnostic accuracy from chest CT scans.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Oncology
- Radiology
Background:
- Lymph node metastasis (LNM) is critical in lung cancer management, but chest computed tomography (CT) has limitations in detecting it.
- Accurate LNM status prediction is essential for effective treatment strategies in lung adenocarcinoma.
Purpose of the Study:
- To develop and validate a vision transformer (ViT)-based deep transfer learning nomogram (DTLN) for predicting LNM in lung adenocarcinoma.
- To assess the predictive performance of the ViT-based DTLN using preoperative unenhanced chest CT imaging.
Main Methods:
- A cohort of 528 lung adenocarcinoma patients was divided into training (70%) and validation (30%) sets.
- A pretrained ViT model extracted deep transfer learning features, and a logistic regression model constructed a ViT-based DTL model.
- A ViT-based DTLN was developed by integrating the DTL signature with clinical predictors (tumor size, location, density).
Main Results:
- The ViT-based DTL model achieved an AUC of 0.821 (training) and 0.825 (validation), comparable to classical CNN models.
- The DTLN demonstrated superior predictive performance with AUCs of 0.865 (training) and 0.894 (validation).
- The DTLN significantly outperformed both the clinical factor model and the ViT-based DTL model (p < 0.001).
Conclusions:
- A ViT-based DTL model is effective for predicting LNM in lung adenocarcinoma, showing viability for medical image deep learning tasks.
- The developed ViT-based DTLN offers excellent predictive performance, aiding clinicians and radiologists in accurate diagnosis and treatment planning.
