A data‐driven approach to optimising the encoding for multi‐shell diffusion MRI with application to neonatal imaging

Jacques‐Donald Tournier, Daan Christiaens, Jana Hutter, Anthony N. Price, Lucilio Cordero‐Grande, Emer Hughes, Matteo Bastiani, Stamatios N. Sotiropoulos, Stephen M. Smith, Daniel Rueckert, Serena J. Counsell, A. David Edwards, Joseph V. Hajnal

TL;DR

This paper introduces a novel data-driven method to optimize multi-shell diffusion MRI (dMRI) acquisition protocols, focusing on maximizing information content rather than relying on specific analysis models. Applied to neonatal imaging, their approach suggests a 3-shell acquisition strategy that balances sensitivity and practicality for this unique developmental stage.

What Problem Does This Paper Address?

Diffusion MRI (dMRI) is a powerful, non-invasive tool for understanding the brain's microstructure and connectivity. Multi-shell dMRI, which uses multiple b-values (representing diffusion weighting), offers richer information than single-shell approaches. However, there's no consensus on how to best choose the number of shells, their specific b-values, or the number of diffusion-weighted directions for each shell. This lack of standardization makes it challenging to compare studies and hinders the development of robust, universally applicable analysis pipelines. This is particularly problematic for specialized populations like neonates, whose developing brains have distinct diffusion properties compared to adults, requiring tailored acquisition strategies.

Methods

The authors propose a data-driven, information-theoretic approach to optimize multi-shell dMRI acquisition. Their method decouples the optimization of b-values from the optimization of diffusion directions.

  1. B-value Optimization: They use Singular Value Decomposition (SVD) on the diffusion-weighted signal data to identify a data-driven basis that represents the signal's dependence on b-values. This allows them to estimate the "effect size" of different signal components. The goal is to find a set of b-values that maximizes sensitivity to these components.
  2. Direction Optimization: For each shell, they determine the minimum number of diffusion directions required to avoid aliasing, based on the angular frequency content of the signal. This is done independently for each shell.
  3. Sensitivity Measure: They define a "sum of squared coefficients of variation" (SSCV) as a measure of sensitivity, aiming to minimize this metric to maximize the signal-to-noise ratio (CNR) of the estimated diffusion parameters.
  4. T2 Decay Consideration: The method also incorporates T2 decay effects, which are crucial for understanding signal attenuation at higher b-values, especially in neonatal imaging.

This approach is "agnostic" to specific diffusion models, meaning it aims to capture the most informative signal regardless of how it will be analyzed later.

Datasets

The study utilized data acquired from five neonates scanned at term-equivalent age. This data was collected on a Philips Achieva 3T magnet at the Evelina Newborn Imaging Center. The acquisition parameters included a DW pulsed gradient spin-echo EPI sequence with specific TE/TR, voxel size, and matrix dimensions. The data was preprocessed to correct for motion and eddy current-induced distortions.

Key Insights and Ideas

  • Model-Agnostic Optimization is Crucial: The paper strongly argues that optimizing acquisition parameters based on specific diffusion models (like DTI) leads to protocols that are not necessarily optimal for other models or even different outputs of the same model. A data-driven, model-agnostic approach is therefore essential for future-proofing acquisition protocols.
  • Neonatal Brains are Different: The angular frequency content of the dMRI signal in neonates is significantly lower than in adults. This implies that fewer diffusion directions are needed per shell compared to adult protocols, making it feasible to acquire more informative data within clinical time constraints.
  • A 3-Shell Strategy is Optimal for Neonates: Their data-driven approach converged on a 3-shell acquisition (plus b=0) as the optimal strategy for neonatal imaging, balancing information content with practical scan times.
  • Robustness of b-values: The identified optimal b-values (around 400 and 1000 s/mm²) were remarkably consistent across subjects and even across different numbers of shells, suggesting a stable underlying signal characteristic.
  • T2 Decay Matters, But Less Than Expected: While T2 decay influences the maximum b-value, its impact on the optimal b-value selection was relatively minor for neonates, likely due to their longer T2 relaxation times compared to adults.

Key Results

The developed data-driven method was used to design the acquisition scheme for the developing Human Connectome Project (dHCP) neonatal imaging. The final protocol, currently in use, consists of:

  • b=0 images: 20 volumes
  • Shell 1: b = 400 s/mm² with 64 directions
  • Shell 2: b = 1000 s/mm² with 88 directions
  • Shell 3: b = 2600 s/mm² with 128 directions

This protocol aims to acquire high-quality data suitable for a wide range of analysis methods. The study also demonstrated that optimizing for specific DTI metrics (like FA or MD) using different fitting algorithms (OLS, WLS, IWLS) yielded different "optimal" protocols, highlighting the problem of model-dependent optimization.

Limitations and Future Work

  • Angular Dependence in Adults: The authors acknowledge that their current approach, which focuses on the mean DW signal per shell, might not be fully optimal for adult brains where angular dependence is much stronger. They suggest that joint optimization of both angular and b-value dependences is an area for ongoing work.
  • Specific Applications: While the proposed method provides a generally suitable protocol, they note that for highly specific applications, a tailored acquisition might be necessary, potentially sacrificing some generality.
  • Neonatal Specificity: The identified parameters are optimal for the neonatal age range. Applying this method to other age groups or ex vivo scanning would require re-optimization.
  • Future Work: The authors mention that joint optimization of angular and b-value dependences is an area of ongoing research. They also suggest that their framework can be adapted for other specific applications or cohorts.

Summary updated

Abstract

Diffusion MRI has the potential to provide important information about the connectivity and microstructure of the human brain during normal and abnormal development, noninvasively and in vivo. Recent developments in MRI hardware and reconstruction methods now permit the acquisition of large amounts of data within relatively short scan times. This makes it possible to acquire more informative multi-shell data, with diffusion sensitisation applied along many directions over multiple b-value shells. Such schemes are characterised by the number of shells acquired, and the specific b-value and number of directions sampled for each shell. However, there is currently no clear consensus as to how to optimise these parameters. In this work, we propose a means of optimising multi-shell acquisition schemes by estimating the information content of the diffusion MRI signal, and optimising the acquisition parameters for sensitivity to the observed effects, in a manner agnostic to any particular diffusion analysis method that might subsequently be applied to the data. This method was used to design the acquisition scheme for the neonatal diffusion MRI sequence used in the developing Human Connectome Project (dHCP), which aims to acquire high quality data and make it freely available to the research community. The final protocol selected by the algorithm, and currently in use within the dHCP, consists of 20 b=0 images and diffusion-weighted images at b = 400, 1000 and 2600 s/mm<sup>2</sup> with 64, 88 and 128 directions per shell, respectively.
BibTeX
@inproceedings{tournier2020datadriven,
  author = {Jacques‐Donald Tournier and Daan Christiaens and Jana Hutter and Anthony N. Price and Lucilio Cordero‐Grande and Emer Hughes and Matteo Bastiani and Stamatios N. Sotiropoulos and Stephen M. Smith and Daniel Rueckert and Serena J. Counsell and A. David Edwards and Joseph V. Hajnal},
  title = {{A data‐driven approach to optimising the encoding for multi‐shell diffusion MRI with application to neonatal imaging}},
  booktitle = {NMR},
  year = {2020},
  doi = {10.1002/nbm.4348},
}
Figure