Mathieu Boudreau

Projects

I started my career in 2004, with an undergraduate physics degree at the Université de Moncton. My first research was in thin-film optics. I moved to hyperpolarized gas MRI for my master’s, on a low-field animal scanner. Since then I’ve worked mostly at clinical field strengths, on quantitative MRI and the software that supports it.

This page collects projects from across those years. Some finished a long time ago. Some are still going.

A gloved hand holds a glass slide coated in a magenta thin film. The film reflects the person photographing it, in a mask and safety glasses.
A single-material Bragg mirror I made at the Université de Moncton in 2009 — reflecting the person who made it.

Anatomy-based B0 field map simulation

B0 inhomogeneity is worst where tissue meets air — the sinuses, the ear canals, the trachea and the lungs. Shim coils can correct it, but designing one for a population means knowing how the field varies across a range of body shapes and postures, and no dataset covered the brain and the spine together. So we built one: 60 volunteers scanned head-to-torso at two 3 T sites, segmented into tissue labels, assigned susceptibility values from the literature, and converted to ΔB0 field maps by convolution with the dipole kernel. It is open on OpenNeuro, alongside the labels and susceptibility maps the field maps came from.

I set up the processing pipeline, organised the data into the BIDS structure and helped prepare it for open release, and worked on the susceptibility-to-field-map code — fixing bugs in the FFT and validating the output against the analytical solution for a sphere, so that the simulated fields were physically correct and not merely plausible.

One subject’s segmentation in three dimensions, rotating as it cycles through the labelled structures — skull and vertebrae, brain and spinal cord, airway and lungs. The body is cropped at the shoulders by the acquisition FOV.
Four-stage pipeline. A sagittal T1-weighted MRI of the head and torso is segmented into coloured anatomical labels covering skull, brain, vertebrae and soft tissue; the labels are converted to a susceptibility map; the susceptibility map is converted to a delta B-zero field map shown in red and blue.
The full pipeline, from a T1-weighted scan to a simulated ΔB0 field map: segmentation into anatomical labels, conversion to a tissue susceptibility map, then to the field map used for shim-coil development.
  • Abstract — ISMRM Annual Meeting, 2025 M. Boudreau, C. Pageot, N. Molinier, N. Laines-Medina, S. Rios, A. D’Astous, V. Callot, E. Alonso-Ortiz, J. Cohen-Adad, “Anatomy-Based Simulations of B0 Field Maps: A Large Database for Head and Spine Shim Coil Development”, digital poster. View poster Watch recording
  • Manuscript — in progress First author. The dataset, the simulation code and the processing pipeline are already public, on OpenNeuro and GitHub.

International reproducibility challenge on quantitative MRI

I led an international quantitative MRI acquisition challenge, run jointly by the ISMRM’s Reproducible Research and Quantitative MR study groups. Eighteen research groups across eight countries submitted data from 27 imaging sites on scanners from all three major vendors: 39 T1 maps of the ISMRM/NIST system phantom and 56 of healthy human brains, all at 3 T bar one at 0.35 T.

The question was deliberately narrow. Is the protocol described in a seminal T1 mapping paper enough for another group to reproduce it? It was not. Variability between submissions came out at twice the variability within a single one — 6.1% against 2.9% in the phantom, and 5.9% against 3.2% in the genu. Acquisition details communicated through a paper are not sufficient to reproduce a quantitative MRI protocol.

Everything was open from the start: data submission and quality control through GitHub issues, the fitting pipeline, the registration code, the analysis notebooks and the containers that run them, all in one organization. The results are browsable in an interactive dashboard.

Inter-sites Intra-sites

Every T1 measurement behind the result, one dot each, with the group mean marked. Hover a dot for the site and vendor it came from. The inter-site points scatter wider than the intra-site ones, most plainly in the brain regions. Phantom values are version 1 only, so they are comparable with the travelling-phantom group.
The live dashboard. Pick phantom or in vivo, then a region, to see the spread across submissions. Open it full size.

Longitudinal quantitative MRI
Study design: six participants scanned at ten time points on a 3T scanner between 2019 and 2022. Brain acquisitions include T1-weighted, MP2RAGE, MT saturation and diffusion; spine acquisitions include T1-weighted, T2-weighted, multi-echo gradient echo, MT saturation and diffusion-weighted imaging.
Six participants, up to ten sessions each, over three years — brain and cervical spinal cord, with the quantitative contrasts each measurement targets. The question was how much a qMRI biomarker drifts when nothing about the subject has changed. First author, Imaging Neuroscience, 2025; 10.1162/imag_a_00409.

B1 mapping for T1 bias correction
Axial brain images comparing five approaches across two acquisitions: nominal, reference double angle, Bloch-Siegert, actual flip angle imaging, and echo-planar double angle. Below, the corresponding B1 maps in a rainbow colour scale and the resulting T1 maps.
Five ways of measuring the transmit field, and what each one does to the T1 map derived from it. Choosing wrongly biases every downstream measurement, which is the thread running through most of my work. First author, Journal of Magnetic Resonance Imaging, 2017 — my most-cited first-author paper.

Hyperpolarized gas MRI for emphysema quantification

My master’s work. I ran lung imaging studies in animals on two machines: a 3 T GE scanner fitted with a high-powered gradient insert, and a custom-built low-field system at 73.5 mT. Both used hyperpolarized noble gas as the signal source — 3He and 129Xe — since neither lung tissue nor air gives you much to image with conventional MRI.

I was responsible for animal preparation, gas production, scanner operation, study design and the analyses. The measurement is the apparent diffusion coefficient of the gas: emphysema destroys the walls between airspaces, so the gas wanders further in the same amount of time, and the signal falls away faster as diffusion weighting is added.

Coronal MRI of a rat lung filled with hyperpolarized helium-3. Both lungs appear bright against a black chest, with the airway visible down the midline.
A rat lung from an emphysema cohort, imaged in vivo at 3 T with hyperpolarized 3He. One of the very first MRI images I ever took. The slider adds diffusion weighting; how quickly the signal fades is the measurement. Reconstructed from the original raw k-space.

Single-material Bragg mirrors by glancing-angle deposition

This was a fourth-year project. After a summer working on MoO3 and the GLAD deposition technique with a postdoc, G. Beydaghyan, I had a eureka moment: I realized I could use the increased porosity at high deposition angles to lower the index of refraction, enough to make an entire Bragg mirror out of a single material.

I found out shortly after that I was a few years late to that idea. It still led to the poster at CAP, and it was the catalyst for the next summer’s project, combining these mirrors with a UV photochromic reaction. Jason Riordon, an MSc student, was already working on MoO3 and photochromics; I proposed seeing how the mirrors would react. That led to my second co-authored paper.

The coated slide held in a gloved hand, reflecting orange.
A sample of the mirror, held in the cleanroom, seen in reflection.
Spectrophotometer Ellipsometer, 10° Optical model
400500600700020406080 Wavelength (nm) Reflectance (%)