Related Experiment Video
Updated: Sep 11, 2025
12:15
Optimized System for Cerebral Perfusion Monitoring in the Rat Stroke Model of Intraluminal Middle Cerebral Artery Occlusion
Published on: February 17, 2013
23.3K
Generative Artificial Intelligence to Automate Cerebral Perfusion Mapping in Acute Ischemic Stroke from Non-contrast
Nicholas J Primiano1, Abhinav R Changa2, Shaun Kohli3
1Department of Radiology, Mount Sinai West, One Gustave Levy Place, Box 1137, New York, NY, USA. nicholasjprimiano@gmail.com.
Journal of Imaging Informatics in Medicine
|August 11, 2025
Summary
Generative AI can now predict critical stroke perfusion parameters from non-contrast head CT scans, improving acute ischemic stroke (AIS) diagnosis. This approach streamlines workflows and expands access to vital hemodynamic data, aiding timely treatment decisions.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Acute ischemic stroke (AIS) requires rapid reperfusion, but CT perfusion (CTP) has limitations like processing time and radiation.
- Non-contrast head CT (NCHCT) is widely available in acute stroke settings.
Purpose of the Study:
- To explore a generative artificial intelligence (AI) approach to predict key perfusion parameters (relative cerebral blood flow [rCBF] and time-to-maximum [Tmax]) directly from NCHCT.
- To streamline stroke imaging workflows and expand access to critical perfusion data.
Main Methods:
- A modified pix2pix-turbo generative adversarial network (GAN) was developed to translate NCHCT images into perfusion maps.
- The GAN was trained on paired NCHCT-CTP data from 99 AIS patients.
- Performance was assessed using SSIM, PSNR, and FID metrics.
Main Results:
- GAN-generated Tmax maps achieved an SSIM of 0.827 and rCBF maps achieved an SSIM of 0.79.
- The model demonstrated moderate approximation to ground truth perfusion maps and captured key hemodynamic features.
- The AI approach successfully generated functional perfusion maps from NCHCT images.
Conclusions:
- Generating perfusion maps from NCHCT using a modified GAN is feasible.
- This cross-modality approach can be a valuable adjunct for AIS evaluation, especially in resource-limited settings.
- Further studies with larger datasets and model refinement are needed to enhance clinical utility.
