Diagnostically Competitive Performance of a Physiology-Informed Generative Multi-Task Network for Contrast-Free CT...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed tomography perfusion (CTP) imaging is vital for stroke assessment but relies on contrast agents.
• Contrast agents in CTP can cause allergic reactions, adverse effects, and significant costs.
• There is a need for safer, more cost-effective perfusion imaging techniques.

Purpose of the Study:

• To introduce Multitask Automated Generation of Intermodal CT perfusion maps (MAGIC), a deep learning framework.
• To generate contrast-free CTP imaging maps from non-contrast CT scans.
• To improve image fidelity and diagnostic accuracy in perfusion imaging.

Main Methods:

• Developed a novel deep learning framework (MAGIC) using generative AI and physiological information.
• Mapped non-contrast CT imaging to multiple contrast-free CTP maps.
• Incorporated physiological characteristics into loss terms to enhance image fidelity.
• Trained and validated the network on stroke patient data from UF Health.

Main Results:

• Demonstrated robustness to brain perfusion abnormalities.
• A double-blinded study with neuroradiologists validated MAGIC's visual quality and diagnostic accuracy.
• MAGIC showed favorable performance compared to traditional contrast-enhanced CTP.

Conclusions:

• MAGIC offers a promising contrast-free, cost-effective, and rapid solution for perfusion imaging.
• This technology has the potential to revolutionize stroke assessment and treatment planning.
• The framework enhances healthcare by providing safer and more accessible perfusion diagnostics.