CT Perfusion Map Generation from Multiphase CTA Using a Generative Adversarial Model for Acute Ischemic Stroke.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Multiphase CT Angiography (mCTA) shows promise for acute ischemic stroke but has limitations in assessing brain tissue perfusion.
• Current limitations in mCTA reduce its clinical interpretability for stroke diagnosis.
• Developing advanced imaging techniques is crucial for improving stroke assessment.

Purpose of the Study:

• To develop a generative adversarial network (GAN) capable of generating CT Perfusion (CTP)-like maps from mCTA data.
• To enhance the clinical interpretability of mCTA by simulating CTP parameters.
• To improve diagnostic capabilities for acute ischemic stroke using AI-driven image synthesis.

Main Methods:

• A GAN was trained on 714 cases with NCCT, CTP, and mCTA data.
• The GAN generated multiparametric CTP maps (Tmax, CBF, CBV) from mCTA.
• Model performance was evaluated using SSIM, PSNR, and FID; clinical utility was assessed by predicting infarct core and penumbra.

Main Results:

• The GAN achieved superior performance (SSIM, PSNR, FID) compared to CycleGAN and Pix2Pix on internal datasets.
• Predicted penumbra and infarct core showed Dice coefficients of 0.672 and 0.468, respectively.
• External data validation demonstrated promising Dice coefficients for penumbra (0.481) and core (0.301).

Conclusions:

• The developed GAN effectively generates CTP-like maps from mCTA, significantly improving image interpretability.
• The approach shows promising diagnostic performance for acute ischemic stroke.
• This AI-driven method is particularly valuable for resource-limited settings requiring efficient stroke assessment.