TL;DR
This paper introduces a novel method for rapidly acquiring multiple T1 maps simultaneously from the human brain and liver. By combining a specialized radial MRI acquisition with a sophisticated model-based reconstruction, they achieve high-resolution T1 maps in just 4 seconds, overcoming previous limitations in speed and multislice capability.
What Problem Does This Paper Address?
Quantitative T1 mapping is crucial for medical diagnostics and research, providing insights into tissue properties. However, traditional methods are often slow, requiring minutes to acquire multiple slices, which is impractical for many clinical scenarios, especially for abdominal imaging requiring breath-holds. Existing fast methods often sacrifice resolution or accuracy, or are limited to single-slice acquisitions. This paper aims to bridge this gap by developing a technique that can acquire multiple T1 maps simultaneously at high resolution and with excellent accuracy and precision within a clinically relevant timeframe.
Methods
The core of this work lies in the combination of two key innovations:
- Single-Shot Inversion-Recovery (IR) Radial FLASH Acquisition: Instead of acquiring data slice-by-slice, the method uses a "spoke-interleaved" radial acquisition. This means that data from multiple slices are acquired concurrently in a single, continuous scan after a single inversion pulse. Radial sampling, combined with a golden-angle scheme, allows for efficient k-space coverage.
- Model-Based Nonlinear Inverse Reconstruction: This is the crucial component for handling the highly undersampled data. Unlike traditional image-space reconstructions, this method directly reconstructs the desired parameter maps (equilibrium magnetization, effective flip angle, and T1) from the raw k-space data. It leverages a "nonlinear forward model" that describes the physics of the MRI signal and incorporates "joint sparsity constraints" to exploit redundancies in the data. This approach is more efficient than image-space methods and allows for simultaneous estimation of all parameters and coil sensitivities. The reconstruction is performed using an iteratively regularized Gauss-Newton method (IRGNM).
Datasets
The authors evaluated their method using:
- Numerical Phantom: A simulated phantom designed in a previous study [8] was used to validate the accuracy and precision of the method under controlled conditions.
- Experimental Phantom: A commercial phantom with defined T1 values was used for experimental validation.
- Human Subjects: Data were acquired from healthy volunteers for brain and liver imaging.
The paper does not introduce a new benchmark dataset but rather validates its method against existing standards and previous work.
Key Insights and Ideas
- Exploiting Data Redundancy: The model-based reconstruction is key to unlocking the potential of highly undersampled radial data. By directly modeling the signal generation process, it can extract more information from fewer measurements than image-space methods.
- Simultaneous Multislice Acquisition is Feasible and Effective: The "spoke-interleaved" approach demonstrates that acquiring data from multiple slices concurrently is not only possible but also maintains high accuracy and precision, even with significant undersampling.
- Radial Sampling's Advantage: Radial trajectories, when combined with model-based reconstruction, are well-suited for fast, undersampled acquisitions, offering flexibility and robustness.
- Trade-offs in Flip Angle Selection: The study highlights the need to carefully select flip angles for multislice acquisitions to balance signal-to-noise ratio (SNR) with T1 accuracy, as higher flip angles can impact accuracy.
Key Results
- Speed: The method achieves 7 simultaneous T1 maps of the brain at 0.5 × 0.5 × 4 mm³ resolution within a single 4-second scan. For abdominal applications, it yields 3 simultaneous T1 maps at 1.25 × 1.25 × 6 mm³ resolution within a 4-second breath hold.
- Accuracy and Precision: Phantom studies showed comparable accuracy and precision to single-slice acquisitions. For brain and liver imaging, the quantitative results were in good agreement with literature values.
- Visual Quality: Visual inspection of the T1 maps from multislice acquisitions showed no discernible differences compared to single-slice acquisitions, indicating excellent image quality.
Limitations and Future Work
- Computational Time: The authors acknowledge that the offline computational time for the model-based reconstruction is a significant limitation, although they suggest acceleration techniques like multi-GPU implementations.
- Abdominal Artifacts: For abdominal applications, the presence of off-resonance effects and the need for more spokes to avoid artifacts can limit the number of simultaneous slices achievable compared to brain imaging.
- Future Directions: The authors suggest exploring combinations with Simultaneous Multislice (SMS) techniques to further increase the number of T1 maps or achieve even higher spatial resolution. They also propose extending the method to other applications beyond T1 mapping.
Summary updated