Time domain principal component analysis for rapid, real-time 2D MRI reconstruction from undersampled data.

Area of Science:

• Medical Imaging
• Biophysics
• Computer-Aided Diagnosis

Background:

• Real-time magnetic resonance imaging (MRI) is crucial for dynamic applications like MR-guided radiotherapy.
• Current methods face challenges with latency and reconstruction speed.
• Principal Component Analysis (PCA) offers potential for accelerated MRI reconstruction.

Purpose of the Study:

• To develop and evaluate a rapid, real-time 2D accelerated MRI method using temporal domain PCA.
• To assess the method's performance in terms of speed, robustness, and image quality.
• To determine the suitability of this technique for real-time tracking applications.

Main Methods:

• A novel 2D accelerated MRI reconstruction method utilizing PCA in the temporal domain was developed.
• The technique employs a moving window of previous frames to reconstruct the current frame.
• Retrospective testing on 15 lung patient datasets with varying acceleration factors (3x-8x) and added noise levels.

Main Results:

• The method demonstrated good temporal robustness with consistent metrics throughout imaging sessions.
• Optimal number of Principal Components (PCs) for temporal fitting was dependent on acceleration rate, decreasing with higher accelerations.
• Reconstruction achieved approximately 50 ms per frame on a standard CPU, with flexibility in hardware requirements.

Conclusions:

• The proposed 2D real-time MR acceleration method shows superior robustness compared to previous real-time PCA techniques up to 8x acceleration.
• Improved temporal robustness, image structure contourability, and accurate definition were achieved.
• The method is flexible, hardware-independent, and suitable for real-time applications requiring low latency.