Related Experiment Video
Updated: Jun 25, 2025
Author Spotlight: A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
Value of CT-Based Deep Learning Model in Differentiating Benign and Malignant Solid Pulmonary Nodules ≤ 8 mm
Yuan Li1, Xing-Tao Huang2, Yi-Bo Feng3
1Department of Thoracic Surgery, the First Affiliated Hospital of Chongqing Medical University, No.1 Youyi Road, Yuzhong District, Chongqing, China (Y.L.); Department of Thoracic Surgery, National Cancer Center/ National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China (Y.L.).
Deep learning models using computed tomography (CT) effectively distinguish small benign and malignant pulmonary nodules. These artificial intelligence tools also accurately differentiate benign tumors from inflammatory nodules.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Solid pulmonary nodules (SPNs) ≤ 8 mm pose diagnostic challenges.
- Accurate differentiation of benign vs. malignant SPNs is crucial for patient management.
- Distinguishing benign tumors from inflammatory nodules is also clinically important.
Purpose of the Study:
- To evaluate the effectiveness of deep learning (DL) models based on computed tomography (CT) for differentiating benign and malignant SPNs ≤ 8 mm.
- To compare the performance of DL models against conventional algorithms.
- To assess the ability of DL models to distinguish benign tumors from inflammatory nodules.
Main Methods:
- Development of five DL models using the Multiscale Dual Attention Network (MDANet) incorporating nodule and peri-nodular features.
- Internal and external validation of models using CT scans from 719 patients with resected SPNs.
- Comparison of the best-performing DL model against four conventional algorithms (VGG19, ResNet50, ResNeXt50, DenseNet121).
Main Results:
- The best DL model (Model 4) incorporating the nodule and a 15 mm peri-nodular region achieved an AUC of 0.730 for differentiating benign and malignant SPNs in the external validation cohort.
- This DL model outperformed four conventional algorithms.
- Another DL model (Model 8) achieved an AUC of 0.871 for distinguishing benign tumors from inflammatory nodules.
Conclusions:
- CT-based DL models, specifically those built with MDANet, demonstrate high accuracy in discriminating between small benign and malignant SPNs.
- These models are also effective in differentiating benign tumors from inflammatory nodules.
- DL offers a promising tool for improving the diagnostic accuracy of pulmonary nodule classification.
