A rapid and accurate guanidine CEST imaging in ischemic stroke using a machine learning approach

Area of Science:

• Biomedical Imaging
• Neuroimaging
• Medical Physics

Background:

• Accurate brain tissue pH mapping is vital for ischemic stroke diagnosis and management.
• Amide proton transfer (APT) imaging has limitations in lesion contrast and signal intensity.
• Guanidine chemical exchange saturation transfer (CEST) imaging offers improved contrast but faces quantification challenges.

Purpose of the Study:

• To develop a machine learning (ML) framework for accurate and robust quantification of guanidine CEST effects.
• To reduce scan time for pH-sensitive imaging.
• To improve lesion detection and characterization in ischemic stroke.

Main Methods:

• Training an ML model on partially synthetic data incorporating experimental line-shape information.
• Utilizing gradient-based feature selection to identify optimal frequency offsets.
• Systematically varying CEST parameters (solute fraction, exchange rate, relaxation) in simulations.

Main Results:

• The ML model demonstrated superior accuracy compared to conventional fitting methods.
• Gradient-based feature selection reduced acquisition points by 72%, significantly cutting scan time.
• The model produced clear hyperintense lesion maps in vivo, outperforming conventional methods.
• A strong negative correlation between guanidine and APT effects was observed, supporting physiological relevance.

Conclusions:

• Partially synthetic training data effectively combines realistic spectral features with ground truth.
• The ML framework enables robust quantification of guanidine CEST effects for pH imaging.
• This approach shows significant potential for rapid and accurate pH-sensitive neuroimaging.