Related Experiment Video
Updated: Aug 22, 2025
Author Spotlight: A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
Pulmonary nodule volumetric accuracy of a deep learning-based reconstruction algorithm in low-dose computed
Shota Watanabe1, Kenta Sakaguchi2, Shigetoshi Kitaguchi2
1Division of Positron Emission Tomography, Institute of Advanced Clinical Medicine, Kindai University Hospital, 377-2 Ohno-Higashi, Osakasayama, Osaka 589-8511, Japan; Radiology Center, Kindai University Hospital, 377-2 Ohno-Higashi, Osakasayama, Osaka 589-8511, Japan.
Deep learning-based reconstruction (DLR) offers superior noise reduction in low-dose CT (LDCT) compared to hybrid iterative reconstruction (hybrid IR). DLR maintains pulmonary nodule volumetric accuracy, even with increased noise non-stationarity.
Area of Science:
- Medical Imaging
- Radiology
- Artificial Intelligence in Medicine
Background:
- Low-dose computed tomography (LDCT) is crucial for lung cancer screening.
- Image reconstruction techniques significantly impact image quality and diagnostic accuracy in LDCT.
- Pulmonary nodule detection and characterization rely on precise volumetric measurements.
Purpose of the Study:
- To compare image properties and pulmonary nodule volumetric accuracy among deep learning-based reconstruction (DLR), filtered back projection (FBP), and hybrid iterative reconstruction (hybrid IR) in LDCT.
- To evaluate the noise reduction performance and stationarity of different reconstruction methods.
- To assess the impact of reconstruction techniques on the volumetric accuracy of pulmonary nodules at varying dose levels.
Main Methods:
- A chest phantom with artificial pulmonary nodules was scanned at various low doses.
- Standard deviations of pixel values and volumetric percentage errors were calculated for FBP, hybrid IR, and DLR.
- Noise non-stationarity index (NNSI) was computed to assess noise stationarity across reconstruction methods.
Main Results:
- DLR achieved higher standard deviation reduction rates (79%-90%) compared to hybrid IR (62%-85%).
- Volumetric percentage errors for nodules were clinically acceptable across all methods, with hybrid IR and DLR showing equivalent or lower errors than FBP for hypoattenuating nodules.
- The noise non-stationarity index (NNSI) was significantly higher for DLR than for FBP and hybrid IR (p < 0.01).
Conclusions:
- Deep learning-based reconstruction (DLR) provides superior noise suppression in LDCT compared to hybrid iterative reconstruction (hybrid IR).
- DLR maintains pulmonary nodule volumetric accuracy without compromising diagnostic performance, despite increased noise non-stationarity.
- These findings support the potential of DLR as an advanced reconstruction technique for LDCT imaging.
