Table of Contents
Fetching ...

Quantitative multi-metabolite imaging of Parkinson's disease using AI boosted molecular MRI

Hagar Shmuely, Michal Rivlin, Or Perlman

TL;DR

A rapid molecular MRI acquisition paradigm with deep learning based reconstruction for multi-metabolite quantification of glutamate, mobile proteins, semisolid, and mobile macromolecules in an acute MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) mouse model emerges as promising PD biomarkers.

Abstract

Traditional approaches for molecular imaging of Parkinson's disease (PD) in vivo require radioactive isotopes, lengthy scan times, or deliver only low spatial resolution. Recent advances in saturation transfer-based PD magnetic resonance imaging (MRI) have provided biochemical insights, although the image contrast is semi-quantitative and nonspecific. Here, we combined a rapid molecular MRI acquisition paradigm with deep learning based reconstruction for multi-metabolite quantification of glutamate, mobile proteins, semisolid, and mobile macromolecules in an acute MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) mouse model. The quantitative parameter maps are in general agreement with the histology and MR spectroscopy, and demonstrate that semisolid magnetization transfer (MT), amide, and aliphatic relayed nuclear Overhauser effect (rNOE) proton volume fractions may serve as PD biomarkers.

Quantitative multi-metabolite imaging of Parkinson's disease using AI boosted molecular MRI

TL;DR

A rapid molecular MRI acquisition paradigm with deep learning based reconstruction for multi-metabolite quantification of glutamate, mobile proteins, semisolid, and mobile macromolecules in an acute MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) mouse model emerges as promising PD biomarkers.

Abstract

Traditional approaches for molecular imaging of Parkinson's disease (PD) in vivo require radioactive isotopes, lengthy scan times, or deliver only low spatial resolution. Recent advances in saturation transfer-based PD magnetic resonance imaging (MRI) have provided biochemical insights, although the image contrast is semi-quantitative and nonspecific. Here, we combined a rapid molecular MRI acquisition paradigm with deep learning based reconstruction for multi-metabolite quantification of glutamate, mobile proteins, semisolid, and mobile macromolecules in an acute MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) mouse model. The quantitative parameter maps are in general agreement with the histology and MR spectroscopy, and demonstrate that semisolid magnetization transfer (MT), amide, and aliphatic relayed nuclear Overhauser effect (rNOE) proton volume fractions may serve as PD biomarkers.

Paper Structure

This paper contains 24 sections, 6 figures.

Figures (6)

  • Figure 1: A deep learning pipeline for multi-metabolite chemical exchange saturation transfer magnetic resonance fingerprinting (CEST MRF). Four rapid pulse sequences are implemented (Supplementary Fig. 1), where each yields 31 raw molecular-feature-embedded images for the semisolid magnetization transfer (MRF$_{MT}$), aliphatic relayed nuclear Overhauser effect (MRF$_{rNOE}$), amide (MRF$_{amide}$) and glutamate (MRF$_{Glu}$) proton pools (shown in gray-scale). These images, together with the water T$_1$ and T$_2$ maps (color coded brains, left), then serve as input for three fully connected neural networks, applied sequentially and in a pixelwise manner. Relevant information is shared across the pipeline (squared input neurons) to improve the quantification accuracy. The ultimate pipeline output are eight proton volume fraction (f$_{ss}$ / f$_{s}$) and exchange rate (k$_{ssw}$ / k$_{sw}$) maps for the semisolid MT or rNOE, amide, and glutamate proton pools, respectively.
  • Figure 2: In vitro quantification of glutamate proton exchange parameters.a. MRF-based glutamate concentration (top) and proton exchange rate (bottom) maps, acquired under physiological conditions (pH = 7.0, T = 37 °C) at 7T. b. MRF-determined glutamate concentrations in phantoms correlated significantly with the known concentrations (Pearson’s r = 0.9646, p < 0.0001, n = 9 vials). c. The MRF determined proton exchange rates for all phantoms were successfully decoupled from the concentration dynamics. Data are presented as mean $\pm$ standard deviation.
  • Figure 3: Quantitative molecular images and relaxometry of a representative mouse before and after MPTP administration.a. T$_1$ and T$_2$ relaxometry maps. b. Semisolid MT proton volume fraction (f$_{ss}$, top) and exchange rate (k$_{ssw}$, bottom) maps. c. rNOE proton volume fraction ($f_s$, top) and exchange rate ($k_{sw}$, bottom) maps. d. Amide proton volume fraction ($f_s$, top) and exchange rate ($k_{sw}$, bottom) maps. e. Glutamate concentration (top) and amine proton exchange rate ($k_{sw}$, bottom) maps. White squares represent the striatal region of interest (ROI). Note the T$_1$ relaxation shortening and elevated semisolid MT, amide, and rNOE proton volume fractions post MPTP administration.
  • Figure 4: Proton volume fraction maps from three additional representative mice, before and after MPTP administration. Note the increased semisolid MT, amide, and rNOE proton volume fraction in the striatum (white squared region of interest).
  • Figure 5: Statistical analysis of the CEST MRF-based parameters, before and after MPTP administration. The analysis was performed on the striatum region of interest (ROI), as delineated in Fig.\ref{['fig:hero']} and \ref{['fig:3reps']}. a. Water T$_1$ and T$_2$ relaxation times. b,c,d. Proton volume fraction (top) and exchange rate (bottom) for the semisolid MT, rNOE and amide proton pools, respectively. e. Glutamate concentration (top) and amine proton exchange rate (bottom). The central horizontal lines in all box plots mark the median, the box limits represent the upper (third) and lower (first) quartiles, the whiskers represent 1.5 $\times$ the interquartile range above and below the upper and lower quartiles, respectively, and all data points are plotted.
  • ...and 1 more figures