Related Experiment Video
Updated: Nov 15, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.8K
Determining the invasiveness of ground-glass nodules using a 3D multi-task network
Ye Yu1, Na Wang2, Ning Huang2
1Department of Radiology, Renji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, 200127, China.
European Radiology
|March 5, 2021
Summary
A novel 3D multi-task deep learning model accurately determines the invasiveness of ground-glass nodules (GGNs). This AI tool aids in selecting patients with invasive lung lesions who require surgery and guides appropriate treatment methods.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Oncology
Background:
- Ground-glass nodules (GGNs) on lung CT scans require accurate invasiveness assessment for optimal patient management.
- Differentiating between pre-invasive and invasive pulmonary adenocarcinomas is crucial for surgical intervention decisions.
Purpose of the Study:
- To develop and evaluate a 3D multi-task deep learning network for determining the invasiveness of GGNs.
- To compare the deep learning model's performance against thoracic radiologists in classifying GGN invasiveness.
Main Methods:
- A novel 3D multi-task deep learning architecture was proposed.
- 770 patients with 909 GGNs were included, with data split into training (n=626) and testing (n=144) sets.
- The model performed three-category (AAH/AIS, MIA, IA) and binary (AAH/AIS/MIA vs. IA) classifications, compared with radiologist assessments.
Main Results:
- The deep learning model achieved 64.9% accuracy in three-category classification and 87.42% accuracy with an AUC of 0.89 in binary classification.
- For binary classification, the model's sensitivity and specificity were 69.57% and 95.24%, respectively.
- The model outperformed senior and junior radiologists in sensitivity, specificity, and accuracy for binary GGN invasiveness classification.
Conclusions:
- The 3D multi-task deep learning model demonstrates strong performance in classifying GGN invasiveness.
- The network's dual classification and segmentation branches effectively learn global and regional features.
- This AI tool can assist clinicians in identifying patients with invasive lung lesions needing surgery and selecting appropriate surgical strategies.
