Using Deep Learning to Shorten MRI Acquisition Time for Stroke Diagnosis
Author Information
Author(s): Peng Yun, Wu Chunmiao, Sun Ke, Li Zihao, Xiong Liangxia, Sun Xiaoyu, Wan Min, Gong Lianggeng
Primary Institution: The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University
Hypothesis
Can deep learning generate synthetic T2-weighted images from diffusion-weighted imaging b0 images to shorten MRI acquisition time?
Conclusion
The Hi-Net can generate synthetic T2-weighted images from low-resolution b0 images, outperforming the pix2pix algorithm.
Supporting Evidence
- The Hi-Net algorithm showed higher PSNR and SSIM compared to the pix2pix algorithm.
- Radiologists identified the nature of synthetic images with accuracy rates of 69.90% and 71.20%.
- The overall quality score of synthetic images ranged from 1.63 to 4.45 on a five-point scale.
Takeaway
This study shows that computers can help create clearer brain images faster, which is important for quickly diagnosing strokes.
Methodology
The study used a retrospective design with 53 patients, employing deep learning algorithms to synthesize T2-weighted images from b0 images.
Limitations
The study is based on a small sample size and the quality of some regions in the generated images needs improvement.
Participant Demographics
Patients who underwent head MRI scans, aged 18 and older.
Statistical Information
P-Value
<0.001
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website