Validation of a Deep Learning-Based Method for Accelerating Susceptibility-Weighted Imaging in Clinical Settings
Xiao Wu1, Shan Xu1, Yao Zhang1
1Department of Radiology, The Second Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou, China.
NMR in Biomedicine
|January 8, 2025
Summary
Deep learning (DL) reconstruction accelerates susceptibility-weighted imaging (SWI) acquisition, reducing scan time by over 60% while maintaining diagnostic image quality. This DL-SWI method shows significant improvements in artifacts and noise, with potential for clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Radiology
Background:
- Susceptibility-weighted imaging (SWI) is crucial in clinical diagnostics but often limited by long acquisition times.
- Accelerating SWI acquisition is essential for improving patient throughput and reducing motion artifacts.
- Deep learning (DL) offers a promising approach for reconstructing accelerated imaging data.
Purpose of the Study:
- To evaluate the feasibility of a DL-based reconstruction method for accelerating SWI acquisition in clinical settings.
- To compare the image quality and diagnostic performance of DL-accelerated SWI (DL-SWI) with conventional parallel imaging (PI) accelerated SWI.
- To assess the potential of DL-SWI for reducing scan time without compromising diagnostic accuracy.
Main Methods:
- A DL-based reconstruction method (ReconNet3D) was used to accelerate SWI acquisition with an acceleration factor of 5 (1:46 acquisition time).
- Prospective under-sampling and PI (acceleration factor of 2, 4:45 acquisition time) were used as a comparison.
- Quantitative metrics (SSIM, PSNR) and qualitative assessments by two raters (artifacts, noise, sharpness, lesion conspicuity, overall quality) were performed on 61 subjects.
- Microbleed counts and non-inferiority assessments were conducted.
Main Results:
- DL-SWI achieved comparable SSIM (0.89 ± 0.02) and PSNR (36.91 ± 2.41) to PI-SWI.
- DL-SWI demonstrated significantly superior scores for artifacts, noise, and overall image quality (p < 0.001).
- While DL-SWI showed slightly reduced sharpness (p = 0.031), lesion conspicuity and microbleed detection were not significantly different, with no false positives or negatives.
Conclusions:
- DL-based SWI reconstruction effectively accelerates acquisition, reducing scan time by over 60% compared to PI.
- The DL method maintains high image quality and diagnostic performance, comparable to conventional PI-SWI.
- DL-SWI presents a viable and promising tool for efficient clinical neuroimaging, enhancing workflow and patient experience.
