Related Experiment Video
Updated: Jul 1, 2026
Author Spotlight: Standardization and Best Practices for Advancing Lung Imaging Using 129Xe MRI
Published on: November 21, 2023
Deep learning-based approach for acquisition time reduction in ventilation SPECT in patients after lung
Masahiro Nakashima1, Ryohei Fukui2, Seiichiro Sugimoto3
1Division of Radiological Technology, Okayama University Hospital, 2-5-1 Shikatacho, Kitaku, Okayama, 700-8558, Japan. nakas-m@cc.okayama-u.ac.jp.
A deep-learning approach using convolutional neural networks (CNNs) can significantly reduce lung ventilation single-photon emission computed tomography (SPECT) acquisition times. This method preserves image quality and diagnostic accuracy for chronic lung allograft dysfunction (CLAD) detection in lung transplant recipients.
Area of Science:
- Radiology and Imaging
- Pulmonary Medicine
- Transplantation Medicine
- Artificial Intelligence in Medicine
Background:
- Chronic lung allograft dysfunction (CLAD) is a major complication after lung transplantation.
- Accurate assessment of CLAD often relies on lung ventilation imaging, such as single-photon emission computed tomography (SPECT).
- Traditional SPECT acquisition times can be lengthy, potentially impacting patient comfort and throughput.
Purpose of the Study:
- To evaluate the image quality and diagnostic performance of abbreviated lung ventilation SPECT using a convolutional neural network (CNN).
- To assess the feasibility of significantly reducing SPECT acquisition times in lung transplant recipients.
- To determine if reduced acquisition times impact the diagnostic accuracy for CLAD.
Main Methods:
- Retrospective analysis of 93 lung transplant recipients who underwent ventilation SPECT/CT.
- A CNN was developed and trained to distinguish full-time from short-time SPECT acquisitions.
- Image quality was assessed using Structural Similarity Index (SSIM) and Normalized Mean Square Error (NMSE); diagnostic performance was evaluated using Area Under the Curve (AUC).
Main Results:
- CNN-predicted SPECT images showed significantly improved image quality (lower NMSE, higher SSIM) compared to short-time images.
- A strong correlation (r=0.955) was observed between the functional/morphological (F/M) volume ratios of full-time and predicted SPECT images.
- Diagnostic performance (AUC) for CLAD was high for full-time (0.942) and predicted SPECT (0.934) images, and lower for short-time images (0.872).
Conclusions:
- Deep learning-based approaches, utilizing CNNs, can effectively reduce lung ventilation SPECT acquisition times.
- This abbreviated approach maintains high image quality and diagnostic accuracy for CLAD detection.
- Shortened SPECT acquisition is feasible, offering potential benefits for lung transplant patient management.
