Related Experiment Video
Updated: Jul 12, 2026
10:26
A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
Deep learning-based malignancy probability estimation of pulmonary nodules in PET/CT imaging
Lars Leijten1, Erik H J G Aarntzen2,3, Roel L J Verhoeven4
1Department of Medical Imaging, Radboud University Medical Center, Nijmegen, The Netherlands. lars.leijten@radboudumc.nl.
European Radiology
|July 10, 2026
Summary
A new deep learning model (AITO-PETCT-MP) for pulmonary nodule malignancy estimation on [18F]FDG-PET/CT shows non-inferior performance to the Herder model. This imaging-only model performs comparably to expert clinicians, offering a potential improvement in diagnostic accuracy.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Current British Thoracic Society (BTS) guidelines recommend [18F]FDG-PET/CT followed by Herder model risk stratification for suspicious pulmonary nodules.
- The Herder model's reliance on limited imaging features may impact diagnostic accuracy.
Purpose of the Study:
- To develop and evaluate a deep learning (DL) model, AITO-PETCT-MP, for malignancy probability estimation in pulmonary nodules using PET/CT imaging.
- To compare the diagnostic performance of AITO-PETCT-MP against the Herder model and expert clinicians.
Main Methods:
- A retrospective study of 533 indeterminate pulmonary nodules (268 malignant) from 436 patients.
- Histopathology or a 2-year benign registry follow-up served as the reference standard.
- Diagnostic performance was assessed on a test set of 161 nodules (80 malignant) in a reader study comparing AITO-PETCT-MP, the Herder model, and seven clinicians.
Main Results:
- AITO-PETCT-MP achieved an AUC of 0.78, demonstrating non-inferiority to the Herder model (AUC 0.73, p=0.005).
- The average AUC for clinicians was 0.80.
- The Herder model referred more benign nodules for treatment (26/81) compared to AITO-PETCT-MP and clinicians (3/81).
- AITO-PETCT-MP and clinicians assigned more malignant cases to CT surveillance than the Herder model.
Conclusions:
- The PET/CT-based deep learning model (AITO-PETCT-MP) shows comparable performance to expert clinicians and is non-inferior to the Herder model for pulmonary nodule malignancy estimation.
- The study highlights potential differences in patient management recommendations between the Herder model and current clinical practice based on BTS follow-up categories.
