NIH: Rapid whole-brain high-resolution myelin water fraction mapping from extremely under-sampled MRI data using deep neural network, 2023

[SNT Gatchaman]

Now published - link here

Preprint
Rapid whole-brain high-resolution myelin water fraction mapping from extremely under-sampled magnetic resonance imaging data using deep neural network
Zhaoyuan Gong, Nikkita Khattar, Matthew Kiely, Curtis Triebswetter, Mustapha Bouhrara

Changes in myelination are a cardinal feature of brain development and the pathophysiology of several cerebral diseases, including multiple sclerosis and dementias. Advanced magnetic resonance imaging (MRI) methods have been developed to probe myelin content through the measurement of myelin water fraction (MWF). However, the prolonged data acquisition and post-processing times of current MWF mapping methods pose substantial hurdles to their clinical implementation.

Recently, fast steady-state MRI sequences have been implemented to produce high spatial resolution whole-brain MWF mapping within 20 min. Despite the subsequent significant advances in the inversion algorithm to derive MWF maps from steady-state MRI, the high-dimensional nature of such inversion does not permit further reduction of the acquisition time by data under-sampling.

In this work, we present an unprecedented reduction in the computation (~30 s) and the acquisition time (~ 7 min) required for whole-brain high-resolution MWF mapping through a new Neural Network (NN)-based approach, named: Relaxometry of Extremely Under-SamplEd Data (NN-REUSED). Our analyses demonstrate virtually similar accuracy and precision in derived MWF values using the NN-REUSED approach as compared to results derived from the fully-sampled reference method.

The reduction in the acquisition and computation times represents a breakthrough toward clinically practical MWF mapping.

Link

 

[SNT Gatchaman]

As an electrical insulator essential for action potential conduction and trophic support to the neuronal axons of the CNS, myelin is crucial to higher-order integrative functions of the brain.

[...] probing myelin content and its integrity is critical to understanding cerebral development, maturation, and degeneration, as well as enhancing our capacity to identify novel therapeutics for myelin repair.

Clinical MRI of the brain basically looks at signal from the hydrogen atoms of water molecules, and how this is affected by the relevant tissue structure and composition.

[...] myelin water fraction (MWF), a direct proxy of myelin content, has been developed based on the multicomponent nature of water pools within image voxels [...] These water pools exhibit differential nuclear magnetic properties, including relaxation times. The fast relaxing component corresponds to the water trapped within the myelin sheets, while the moderately and slowly relaxing components are attributed to intra/extra cellular and cerebrospinal fluid (CSF) water compartments, respectively. The signal fraction of the fast relaxing component is defined as the MWF, which has been histologically validated as a specific measure of myelin content.

The multi-echo spin-echo (MESE) sequence was originally used to measure in vivo MWF. [...] Despite being the gold standard method for MWF mapping, MESE suffers the utmost slow data acquisition and fitting instability.

MRI loves its acronyms and I think many of them are backronyms, where the "cool" sounding term was coined and then an explanation hammered into it. I would like to suggest that medical physicists and radiologists are much better at this than certain well known BPS researchers.

multicomponent driven equilibrium steady-state observation of T1 and T2 (mcDESPOT) has been introduced for whole-brain high spatial resolution MWF determination within a clinically feasible acquisition time (∼20 min). [...] To improve the accuracy and precision in derived MWF values, the Bayesian Monte Carlo (BMC) analysis of mcDESPOT (BMC-mcDESPOT) was proposed ...

And yes there is indeed a rapper called Despot that should really style himself as "MC Despot" in my view. Anyhow

Despite these critical advances for clinically feasible and accurate MWF imaging, the acquisition time remains relatively long, particularly infeasible in investigations involving participants with limited cooperability [...] Further, due to its underlying complex mathematical modeling, BMC-mcDESPOT re- quires extensive computational power, with several hours needed to generate a single whole-brain MWF map; this limits real-time evaluation.

In recent work, Liu et al. have proposed deep learning [Neural Network] models for rapid computation of MWF maps from conventional multiple echo-time images [...] most of these methods focused on fast computation or providing MWF with a low spatial or temporal resolution. [...] we will provide details on our NN-based approach for whole-brain high-resolution MWF mapping from steady-state imaging data [...] within drastically reduced acquisition times as compared to the state-of-the-art method, BMC-mcDESPOT.

 

[SNT Gatchaman]

Here they show the dramatic reduction in white matter myelin content in a patient with Alzheimer disease vs HC. Top row is the conventional technique (BMC-mcDESPOT). Bottom row is the authors' new technique. The acquisition time (patient on magnet time) is the same - 21 minutes, but note the processing time: 30 seconds vs 30 hours. As proposed, this technique would be feasible for clinical practice and large-n research.

 

[SNT Gatchaman]

For more background, see Magnetic Resonance of Myelin Water: An in vivo Marker for Myelin (2016, Brain Plasticity). A few quotes —

Myelin is critical for healthy brain function. It plays a fundamental role in determining the speed of action potentials; when myelin integrity is compromised, brain function is affected. [...] It is an interesting time to review myelin water imaging because, while considerable progress has been made over the last 2 decades, the field is still not mature and fundamental questions remain to be explored.

From the point of view of an MR imaging scientist, CNS tissue contains two fundamentally different kinds of cells (myelinated and unmyelinated) and four different locations for water (intracellular, extracellular, between myelin bilayers and cerebrospinal fluid). While the non-aqueous components of CNS tissue, i.e. lipids, proteins and nucleic acids, have complex dynamic structures, 70–85% of brain mass is simply H2O. Unmyelinated neurons and glial cells have single bilayer membranes while myelinated neurons contain multi-bilayer membranes for which approximately 40% of the mass is compartmentalized water.

Myelination in humans begins in the fifth fetal month. Rapid myelin growth occurs during the first two years and continues throughout life until the 6th decade. The onset of myelination is marked by a rapid increase in brain lipid and protein content and a corresponding decrease in brain water content.

Using the mcDESPOT technique, several studies have reported the pattern of normal myelin development from 2.5 months to 5.5 years of age. These studies demonstrate a spatial temporal pattern of myelination beginning in the cerebellum, pons, and internal capsule; proceeding caudocranially from the splenium of the corpus callosum and optic radiations (3–4 months) to the occipital and parietal lobes (4–6 months) and then to the genu of the corpus callosum and frontal and temporal lobes (6 – 8 months). This pattern of myelination is consistent with reports of myelination begining in posterior regions before anterior regions of the brain. Distinct male and female developmental patterns were observed ...

An aside —

A second factor which may be linked to cognitive performance is the presence of the apolipoprotein E (APOE) e4 allele a major susceptibility gene for late-onset Alzheimer’s disease. The presence of this gene may influence brain development already in infancy, as demonstrated by a recent myelin water MRI study where it was found that infant e4 carriers had a lower myelin water fraction than noncarriers in areas of the brain preferentially affected in Alzheimer’s disease.

And in relation to the "all tests are normal" and in particular normal clinical MRI brain imaging for ME, note the following in relation to the normal appearing white matter in multiple sclerosis —

Normal appearing white matter (NAWM) are areas of brain and spinal cord which appear ‘normal’ on standard clinical imaging. Much effort has also been put into studying NAWM in MS, which is known from post-mortem histological studies to demonstrate myelin abnormalities demyelination. MWF of NAWM was found to be diffusely reduced in both brain (by 6–37%) and spinal cord (by 11–25%) when compared to healthy controls. NAWM MWF can distinguish between different subtypes of MS and reduced brain MWF is also related to increased disability, suggested that more global disease process can influence clinical status.

Definitely want to see the fast technique in the thread's preprint evaluating ME/LC. Finally this comment on concussion, noting the neurological symptom overlap with ME —

Traumatic brain injury and concussion are a major public health concern but continue to be poorly understood. Evidence for myelin damage in mild traumatic brain injury (TBI) comes from one prospective study that examined two varsity hockey teams (45 players) over one season of athletic competition. 11 players sustained a concussion, and were scanned at 72 hours, 2 weeks, and 2 months post-injury. At 2 weeks post-injury, MWF was reduced in several brain areas relative to preseason; values recovered to pre-season values by 2 months post-injury. These results could indicate a transient myelin disruption following a single mTBI, with subsequent remyelination of affected neurons.

 

[SNT Gatchaman]

For current state-of-the-art see —
So You Want to Image Myelin Using MRI: An Overview and Practical Guide for Myelin Water Imaging (2020, paywall, sci-hub)

Also —
Using myelin water imaging to link underlying pathology to clinical function in multiple sclerosis: A scoping review (2022)
Macromolecular Proton Fraction as a Myelin Biomarker: Principles, Validation, and Applications (2022)
Mechanisms of myelin repair, MRI techniques and therapeutic opportunities in multiple sclerosis (2021)

 

[SNT Gatchaman]

Published as —

REUSED: A deep neural network method for rapid whole-brain high-resolution myelin water fraction mapping from extremely under-sampled MRI (2023)
Gong; Khattar; Kiely; Triebswetter; Bouhrara

Changes in myelination are a cardinal feature of brain development and the pathophysiology of several central nervous system diseases, including multiple sclerosis and dementias. Advanced magnetic resonance imaging (MRI) methods have been developed to probe myelin content through the measurement of myelin water fraction (MWF).

However, the prolonged data acquisition and post-processing times of current MWF mapping methods pose substantial hurdles to their clinical implementation. Recently, fast steady-state MRI sequences have been implemented to produce high-spatial resolution whole-brain MWF mapping within ~20 min. Despite the subsequent significant advances in the inversion algorithm to derive MWF maps from steady-state MRI, the highdimensional nature of such inversion does not permit further reduction of the acquisition time by data undersampling.

In this work, we present an unprecedented reduction in the computation (~30 s) and the acquisition time (~7 min) required for whole-brain high-resolution MWF mapping through a new Neural Network (NN)based approach, named NN-Relaxometry of Extremely Under-SamplEd Data (NN-REUSED). Our analyses demonstrate virtually similar accuracy and precision in derived MWF values using NN-REUSED compared to results derived from the fully sampled reference method. The reduction in the acquisition and computation times represents a breakthrough toward clinically practical MWF mapping.

Link (Computerized Medical Imaging and Graphics)