Spatio-temporal reconstruction of substance dynamics using compressed sensing in multi-spectral magnetic resonance spectroscopic imaging

Utako Yamamoto, Hirohiko Imai, Kei Sano, Masayuki Ohzeki, Tetsuya Matsuda, Toshiyuki Tanaka

TL;DR

Researchers have developed a new method to capture the "movie" of chemical changes inside a living body using multi-spectral magnetic resonance spectroscopic imaging (MRSI). By combining compressed sensing with a model that uses known chemical "fingerprints," they achieved a temporal resolution of just four seconds—orders of magnitude faster than traditional full-sampling methods that would typically take hours.

What Problem Does This Paper Address?

Magnetic Resonance Spectroscopic Imaging (MRSI) is a powerful tool because it doesn't just show anatomy; it maps the spatial distribution of specific chemicals (metabolites) like glucose or lactate. However, MRSI has a major "speed limit." To get high-quality spectral and spatial data, the scanner usually needs to collect a massive amount of data, which can take a very long time.

This prolonged acquisition time makes it nearly impossible to observe substance dynamics—how chemicals move and transform in real-time (e.g., a tumor metabolizing glucose). If the chemical distribution changes while the scanner is still mid-acquisition, the resulting image is blurred or inaccurate. The authors aim to break this speed limit to observe metabolic processes, such as the Warburg effect in tumors, with high temporal resolution.

Methods

The core innovation is a reconstruction method that shifts the goalpost: instead of trying to reconstruct the entire complex multi-dimensional spectrum first, it reconstructs the spatio-temporal distributions of the substances directly.

Key components include:

  • Partially Separable Function (PSF) Model: The authors model the data as a linear combination of "base spectra" (the known chemical signatures of substances like glucose, lactate, and fat) multiplied by their spatial distributions over time.
  • Compressed Sensing (CS): They use random undersampling (specifically via Sobol sequences) to collect only a small fraction of the data required for a full scan.
  • Regularization: To fill in the gaps from undersampling, they apply "prior knowledge" through a minimization problem:
  • Spatial Sparsity ($\ell_1$ regularization): Assuming substances are localized in specific areas.
  • Temporal Smoothness (Elastic Net): Assuming chemical changes happen gradually rather than in jagged jumps.
  • ADMM Algorithm: They solved this complex optimization problem using the Alternating Direction Method of Multipliers (ADMM), making the large-scale computation feasible.

Datasets

The method was evaluated using two primary setups:

  1. Phantom Data: Glass tubes filled with increasing amounts of glucose or lactate solutions over time to simulate controlled dynamic changes.
  2. Animal Data: A tumor-bearing mouse (Colon 26 carcinoma). The researchers injected glucose and used the system to monitor its uptake and subsequent conversion into lactate within the tumor—a hallmark of cancer metabolism known as the Warburg effect.
  3. Base Spectra: 2D MRS data was acquired separately to serve as the "fingerprints" for the substances of interest.

Key Insights and Ideas

  • Direct Mapping: The "aha moment" is realizing that if you already know what glucose and lactate look like spectrally, you don't need to "discover" their spectra from the raw data. You only need to find where they are and how much of them exist at any given time.
  • Independence of Dimensions: Because the method targets the substance maps directly, it is mathematically independent of the number of spectral or spatial dimensions, making it highly flexible for different types of MRSI sequences.
  • The 4-Second Breakthrough: By assigning undersampled data points to specific 4-second windows, they created a time-series of images that effectively functions as a metabolic movie, showing glucose flowing into the injection site and lactate appearing in the tumor.

Key Results

  • Temporal Resolution: Achieved a resolution of 4 seconds per frame. For comparison, a fully sampled 2D spatial/2D spectral movie frame would have taken roughly 2.3 to 4.5 hours to acquire.
  • Metabolic Observation: In the mouse study, the system successfully captured the rapid increase of glucose at the injection site, its spread through the body, and its gradual conversion to lactate inside the tumor.
  • Robustness: The phantom experiments confirmed that the reconstructed intensity of the chemicals matched the actual rates at which the solutions were instilled into the tubes.

Limitations and Future Work

  • Stair-step Artifacts: The authors noted some "stair-step" artifacts in the temporal profiles, likely caused by the regularization parameters. While they can be smoothed by adjusting the math, there is a trade-off between smoothness and the ability to detect rapid changes.
  • Requirement of Base Spectra: The current model requires pre-acquired "base spectra" for every substance of interest. If a substance is present but its spectrum isn't provided to the model, it won't be reconstructed.
  • Future AI Integration: The authors suggest that future versions could use machine learning to "learn" these base spectra or subspaces automatically, potentially allowing the discovery of unknown metabolic byproducts without prior templates.

Summary updated

Abstract

The objective of our study is to observe dynamics of multiple substances in vivo with high temporal resolution from multi-spectral magnetic resonance spectroscopic imaging (MRSI) data. The multi-spectral MRSI can effectively separate spectral peaks of multiple substances and is useful to measure spatial distributions of substances. However it is difficult to measure time-varying substance distributions directly by ordinary full sampling because the measurement requires a significantly long time. In this study, we propose a novel method to reconstruct the spatio-temporal distributions of substances from randomly undersampled multi-spectral MRSI data on the basis of compressed sensing (CS) and the partially separable function model with base spectra of substances. In our method, we have employed spatio-temporal sparsity and temporal smoothness of the substance distributions as prior knowledge to perform CS. The effectiveness of our method has been evaluated using phantom data sets of glass tubes filled with glucose or lactate solution in increasing amounts over time and animal data sets of a tumor-bearing mouse to observe the metabolic dynamics involved in the Warburg effect in vivo. The reconstructed results are consistent with the expected behaviors, showing that our method can reconstruct the spatio-temporal distribution of substances with a temporal resolution of four seconds which is extremely short time scale compared with that of full sampling. Since this method utilizes only prior knowledge naturally assumed for the spatio-temporal distributions of substances and is independent of the number of the spectral and spatial dimensions or the acquisition sequence of MRSI, it is expected to contribute to revealing the underlying substance dynamics in MRSI data already acquired or to be acquired in the future.
BibTeX
@article{yamamoto2023spatio,
  author = {Utako Yamamoto and Hirohiko Imai and Kei Sano and Masayuki Ohzeki and Tetsuya Matsuda and Toshiyuki Tanaka},
  title = {{Spatio-temporal reconstruction of substance dynamics using compressed sensing in multi-spectral magnetic resonance spectroscopic imaging}},
  journal = {Expert Systems with Applications},
  year = {2023},
  url = {https://openreview.net/forum?id=wCJevNHBIp},
}

Loading related content...

Figure