Accelerating multipool CEST MRI of Parkinson's disease using deep learning-based Z-spectral compressed sensing.

Area of Science:

• Biomedical Imaging
• Machine Learning
• Neuroscience

Background:

• Parkinson's disease (PD) diagnosis relies on accurate imaging biomarkers.
• Multipool Chemical Exchange Saturation Transfer (CEST) MRI offers metabolic insights but suffers from long scan times.
• Reducing scan time is crucial for clinical applicability and patient comfort.

Purpose of the Study:

• To develop a deep learning approach to shorten multipool CEST MRI scan times for Parkinson's disease (PD) detection.
• To maintain high prediction accuracy despite reduced acquisition duration.
• To validate the proposed method against existing techniques.

Main Methods:

• A modified 1D U-Net deep learning model, termed Z-spectral compressed sensing (CS), was developed to reconstruct dense Z-spectra from sparse data.
• The model was trained using simulated Z-spectra derived from Bloch equations and validated with in vivo rat brain experiments.
• The method was applied to a 6-hydroxydopamine induced PD rat model, analyzing various CEST contrasts (APT, CEST@2ppm, NOE, DS, MT).

Main Results:

• Z-spectral CS demonstrated superior fidelity in Z-spectrum recovery compared to linear, pchip, and Lorentzian interpolation.
• Significant metabolic differences (APT, CEST@2ppm, NOE, DS) were detected between wild-type and PD rat striata.
• Multipool CEST MRI with Z-spectral CS achieved a 33% scan time reduction with maintained prediction accuracy, outperforming individual CEST contrasts.

Conclusions:

• Z-spectral CS effectively accelerates multipool CEST MRI acquisition.
• The integrated approach enhances diagnostic accuracy for Parkinson's disease.
• This method offers a viable solution for clinical translation of advanced MRI techniques.