Related Experiment Video
Updated: Sep 8, 2025
10:26
Author Spotlight: A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
2.0K
Diagnostic performance of artificial intelligence models for pulmonary nodule classification: a multi-model
Sarah K Herber1, Lukas Müller2, Daniel Pinto Dos Santos1
1Department of Diagnostic and Interventional Radiology, University Medical Center of the Johannes Gutenberg-University Mainz, Mainz, Germany.
European Radiology
|July 27, 2025
Summary
Commercial artificial intelligence (AI) models show low accuracy in classifying pulmonary nodules, with high false-negative rates and many intermediate-risk results. These AI tools are not yet reliable for standalone clinical use in lung cancer diagnosis.
Area of Science:
- Pulmonary Medicine
- Radiology
- Artificial Intelligence in Healthcare
Background:
- Lung cancer is a leading cause of cancer mortality, and early detection of pulmonary nodules is crucial for improving survival rates.
- Distinguishing malignant from benign pulmonary nodules using imaging remains a diagnostic challenge.
- Artificial intelligence (AI) is being explored to enhance the accuracy of pulmonary nodule classification.
Purpose of the Study:
- To evaluate the diagnostic performance of commercially available AI software models in classifying pulmonary nodules.
- To compare AI model accuracy against histopathology as the gold standard.
- To identify potential limitations of AI in clinical application for pulmonary nodule diagnosis.
Main Methods:
- A retrospective analysis of 158 pulmonary nodules (4-30 mm) from CT scans was conducted.
- Three AI software models were used to classify nodules, with sensitivity, specificity, and false rates calculated.
- Diagnostic accuracy was assessed using the area under the receiver operating characteristic curve (AUC), with subgroup analyses performed based on nodule characteristics and CT scan parameters.
Main Results:
- One AI model classified a significant proportion of nodules as intermediate risk, hindering assessment.
- The other AI models exhibited moderate sensitivity (53.1-70.3%) but low specificity (46.7-66.7%), resulting in high false-positive rates (45.5-52.4%).
- Area under the ROC curve (AUC) values ranged from 0.5 to 0.6, indicating limited diagnostic capability. Up to 49% of nodules were classified as intermediate risk, and performance varied with CT scan type.
Conclusions:
- Current AI-based software models demonstrate insufficient specificity and high false-negative rates for pulmonary nodule classification.
- A substantial percentage of nodules were classified as intermediate risk, limiting clinical utility.
- The evaluated AI models are not yet suitable for standalone clinical application in pulmonary nodule diagnosis.
