Related Experiment Video
Updated: Jul 11, 2025
Author Spotlight: A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
Diagnostic performance of a deep learning-based method in differentiating malignant from benign subcentimeter (≤10
Jianing Liu1, Linlin Qi1, Yawen Wang1
1Department of Diagnostic Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
A deep learning (DL) model significantly outperformed experienced radiologists in diagnosing malignant subcentimeter solid pulmonary nodules (SSPNs) on CT scans. This AI tool shows promise for reducing diagnostic uncertainty and improving accuracy, particularly for smaller nodules.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging Analysis
Background:
- Subcentimeter solid pulmonary nodules (SSPNs) pose diagnostic challenges.
- Differentiating malignant from benign SSPNs is crucial for patient management.
- Deep learning (DL) models offer potential for enhanced image analysis.
Purpose of the Study:
- To evaluate the diagnostic performance of a DL model for SSPNs.
- To compare the DL model's accuracy against experienced radiologists.
- To assess the DL model's effectiveness across different nodule size subgroups.
Main Methods:
- Retrospective collection of 200 SSPNs (100 benign, 100 malignant).
- Utilized a DL model to predict malignancy probability from CT images.
- Compared DL model's diagnostic accuracy and indeterminate results with radiologists using McNemar-Bowker test.
Main Results:
- DL model achieved significantly higher accuracy (71.5%) than radiologists (38.5%) (P<0.001).
- DL model reported fewer indeterminate results across all nodule sizes (≤10 mm).
- Superior performance observed in subgroups: 3-6 mm (75.5% vs 28.3%), 6-8 mm (62.0% vs 28.2%), and 8-10 mm (77.6% vs 55.3%).
Conclusions:
- The DL-based method demonstrates superior performance in differentiating malignant and benign SSPNs.
- This DL model can potentially reduce diagnostic uncertainty and improve accuracy.
- The model is particularly effective for SSPNs smaller than 8 mm.
