Related Experiment Video
Updated: Jan 4, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Temporally downsampled cerebral CT perfusion image restoration using deep residual learning
Haichen Zhu1, Dan Tong2, Lu Zhang3
1Lab of Image Science and Technology, Key Laboratory of Computer Network and Information Integration (Ministry of Education), Southeast University, Nanjing, 210096, China.
International Journal of Computer Assisted Radiology and Surgery
|November 2, 2019
Summary
A novel deep convolutional neural network (CNN) reduces CT radiation dose by 50% for CT perfusion (CTP) imaging. This method effectively restores image quality for accurate acute ischemic stroke diagnosis.
Area of Science:
- Medical Imaging
- Radiology
- Artificial Intelligence
Background:
- Acute ischemic stroke is a leading global cause of death.
- CT perfusion (CTP) imaging is crucial for acute stroke diagnosis, but high radiation dose is a concern.
- Reducing radiation exposure during CTP is essential for patient safety.
Purpose of the Study:
- To develop a method for reducing radiation dose in CTP imaging.
- To investigate the efficacy of a deep convolutional neural network (DCNN) for restoring CTP image quality after dose reduction.
- To enable accurate stroke diagnosis with lower radiation exposure.
Main Methods:
- Downsampled original 30-pass CTP images to 15 passes, achieving a 50% radiation dose reduction.
- Developed and trained a residual DCNN model with 16 convolutional layers to restore 15-pass CTP images to 30 passes.
- Utilized 18 patients' CTP images for training and 6 for testing.
Main Results:
- The DCNN model successfully restored downsampled CTP images with high fidelity, achieving average SSIM of 0.981 and PSNR of 56.25.
- Restored images yielded perfusion maps comparable to original images, with average perfusion results being extremely close.
- Radiologists could detect hypoperfusion areas with comparable accuracy using restored images.
Conclusions:
- The proposed CNN model effectively restores temporally downsampled CTP images, enabling significant radiation dose reduction.
- This DCNN approach outperforms simple interpolation and generative adversarial networks in restoring essential CTP information.
- This method offers a viable option for reducing radiation dose in CTP imaging for acute stroke diagnosis.
