Robust Myocardial Perfusion MRI Quantification With DeepFermi.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Physics

Background:

• Stress perfusion cardiac magnetic resonance (CMR) is crucial for assessing myocardial blood supply.
• Current visual assessment is subjective; quantitative methods are slow and sensitive to artifacts.
• Existing quantitative methods like deconvolution analysis are time-consuming and prone to motion-related outliers.

Purpose of the Study:

• To introduce a novel deep learning method for fast, accurate, and robust myocardial perfusion quantification.
• To develop a user-independent assessment of myocardial perfusion.
• To improve robustness against motion artifacts and data outliers in perfusion analysis.

Main Methods:

• Integration of the Fermi function with a neural network architecture for perfusion quantification.
• Utilizing a 3D convolutional neural network prior for spatio-temporal generalization.
• Employing a self-supervised learning framework and outlier-resistant training methodology.

Main Results:

• Demonstrated improved accuracy and robustness in simulation experiments compared to traditional deconvolution analysis.
• Achieved consistent performance across varying Signal-to-Noise Ratio scenarios with data outliers.
• Generated robust in vivo perfusion estimates aligning with clinical diagnoses, approximately five times faster than conventional algorithms.

Conclusions:

• The proposed deep learning method provides a significant advancement in myocardial perfusion quantification.
• This approach offers a faster, more accurate, and robust alternative to existing quantitative techniques.
• The self-supervised and outlier-resistant framework enhances clinical applicability of CMR perfusion analysis.