mcLARO: Multi-Contrast Learned Acquisition and Reconstruction Optimization for simultaneous quantitative multi-parametric mapping

Jinwei Zhang, Thanh D. Nguyen, Eddy Solomon, Chao Li, Qihao Zhang, Jiahao Li, Hang Zhang, Pascal Spincemaille, Yi Wang

Abstract

Purpose: To develop a method for rapid sub-millimeter T1, T2, T2*, and QSM mapping in a single scan using multi-contrast Learned Acquisition and Reconstruction Optimization (mcLARO).

Methods: A pulse sequence was developed by interleaving inversion recovery and T2 magnetization preparations with single-echo and multi-echo gradient echo acquisitions, sensitizing k-space data to T1, T2, T2*, and magnetic susceptibility. The proposed mcLARO framework uses deep learning to jointly optimize the multi-contrast k-space under-sampling pattern and the image reconstruction based on image feature fusion. The method was validated at R=8 under-sampling through retrospective ablation and prospective studies compared to conventional reference standards.

Results: The prospective study demonstrated that mcLARO (5:39 mins) achieved results comparable to reference scans (40:03 mins) with negligible bias and narrow 95% limits of agreement on regional quantitative values.

Conclusion: mcLARO enables fast, high-resolution multi-parametric mapping in a single scan, significantly reducing the time barrier for clinical quantitative MRI.

Introduction

Quantitative MRI (qMRI) represents a shift from qualitative "weighted" imaging to objective physical measurements of tissue properties, such as T1 and T2 relaxation times and magnetic susceptibility. These metrics are vital for tracking the progression of neurodegenerative conditions like Multiple Sclerosis, Parkinson’s, and Alzheimer’s disease.

However, the primary obstacle to the clinical adoption of qMRI is acquisition time. Obtaining a full suite of parameters typically requires multiple separate scans, leading to total exam times exceeding 40 minutes. This duration is not only costly but increases the likelihood of patient motion artifacts. The mcLARO framework addresses this by asking: can we use artificial intelligence to simultaneously optimize how data is sampled and how images are reconstructed to make "all-in-one" qMRI a clinical reality?

The mcLARO Framework

The mcLARO approach (Multi-Contrast Learned Acquisition and Reconstruction Optimization) consists of two integrated components: a specialized pulse sequence and a deep learning optimization framework.

1. Pulse Sequence Design

The researchers developed a custom MRI sequence that interleaves different magnetization preparations—Inversion Recovery for T1 sensitivity and T2-prep for T2 sensitivity—with single and multi-echo Gradient Recalled Echo (GRE) readouts. This ensures that the raw k-space data contains the necessary "fingerprints" for all four target parameters (T1, T2, T2*, and QSM) within a single acquisition window.

2. Joint AI Optimization

Unlike traditional methods that optimize sampling and reconstruction in isolation, mcLARO treats them as a single pipeline:

  • Learned Sampling: Utilizing a "LOUPE-style" (Learnable Optimal Under-sampling Pattern) module, the network identifies which parts of k-space are most informative for the specific multi-contrast reconstruction task. This allows for an 8x acceleration factor (R=8).
  • Deep Unrolled Reconstruction: The model employs a deep ADMM (Alternating Direction Method of Multipliers) network. This "unrolled" architecture incorporates the physics of MRI (data consistency) directly into the neural network layers, ensuring the output remains faithful to the acquired raw data.
  • Multi-Contrast Feature Fusion: Because T1, T2, and susceptibility maps of the same brain share the same underlying anatomy, the network uses a recurrent convolutional module to "fuse" features across different echoes. This allows the network to use structural information from one contrast to fill in missing details in another, maintaining sharpness despite heavy under-sampling.

Experimental Validation

The study involved 13 healthy subjects scanned on a 3T system. The model was trained using fully sampled data (34-minute scans) as the ground truth.

In the retrospective study, the model was tested by artificially removing data to simulate an 8x speedup. In the prospective study, the optimized 5:39-minute sequence was actually deployed on the scanner. Results were compared against a "gold standard" suite of conventional scans that took over 40 minutes to complete.

Key Results and Insights

  • Significant Acceleration: mcLARO reduced the total scan time from 40:03 minutes to 5:39 minutes while maintaining sub-millimeter resolution (0.75 x 0.75 x 1 mm³).
  • High Quantitative Fidelity: Bland-Altman analyses showed negligible bias and narrow limits of agreement for T1, T2, T2*, and QSM values compared to the reference standards.
  • The Power of Joint Optimization: Ablation studies revealed that removing either the learned sampling or the feature fusion module resulted in significant blurring, particularly in deep gray matter structures like the putamen and thalamus. This confirms that the synergy between the "camera" (sampling) and the "developer" (reconstruction) is essential for high-quality qMRI.
  • Denoising Capabilities: Interestingly, the AI-reconstructed maps often appeared cleaner than the fully sampled references. This suggests the network effectively learned to distinguish underlying signals from random thermal noise during the training process.

Conclusion and Future Directions

mcLARO demonstrates that deep learning can successfully bridge the gap between high-speed acquisition and high-fidelity quantitative mapping. By reducing scan times by nearly 85%, this method makes comprehensive brain mapping feasible for routine clinical use. Future work will focus on validating the model in patient populations with diverse pathologies and adapting the framework for motion-sensitive applications in the heart and liver.

Summary updated

Abstract

Purpose: To develop a method for rapid sub-millimeter T1, T2, T2* and QSM mapping in a single scan using multi-contrast Learned Acquisition and Reconstruction Optimization (mcLARO). Methods: A pulse sequence was developed by interleaving inversion recovery and T2 magnetization preparations and single-echo and multi-echo gradient echo acquisitions, which sensitized k-space data to T1, T2, T2* and magnetic susceptibility. The proposed mcLARO used a deep learning framework to optimize both the multi-contrast k-space under-sampling pattern and the image reconstruction based on image feature fusion. The proposed mcLARO method with R=8 under-sampling was validated in a retrospective ablation study using fully sampled data as reference and evaluated in a prospective study using separately acquired conventionally sampled quantitative maps as reference standard. Results: The retrospective ablation study showed improved image sharpness of mcLARO compared to the baseline network without multi-contrast sampling pattern optimization or image feature fusion, and negligible bias and narrow 95% limits of agreement on regional T1, T2, T2* and QSM values were obtained by the under-sampled reconstructions compared to the fully sampled reconstruction. The prospective study showed small or negligible bias and narrow 95% limits of agreement on regional T1, T2, T2* and QSM values by mcLARO (5:39 mins) compared to reference scans (40:03 mins in total). Conclusion: mcLARO enabled fast sub-millimeter T1, T2, T2* and QSM mapping in a single scan.
BibTeX
@inproceedings{zhang2023mclaro,
  author = {Jinwei Zhang and Thanh D. Nguyen and Eddy Solomon and Chao Li and Qihao Zhang and Jiahao Li and Hang Zhang and Pascal Spincemaille and Yi Wang},
  title = {{mcLARO: Multi-Contrast Learned Acquisition and Reconstruction Optimization for simultaneous quantitative multi-parametric mapping}},
  booktitle = {Magnetic},
  year = {2023},
  doi = {10.1002/mrm.29854},
}
Figure