Jonathan Rafael Patino Lopez

Related publications (15)

Estimation of numerical substrate properties with compartmentalized models from Monte-Carlo simulated DW-MRI signals

Erick Jorge Canales Rodriguez, Rémy Marc M Gardier, Gabriel Girard, Jonathan Rafael Patino Lopez, Jean-Philippe Thiran, Juan Luis Villarreal Haro

2022

, , , , ,

One main goal of Diffusion-Weighted Magnetic Resonance Imaging is to infer microscopic tissue properties from the measured signal. The use of Monte-Carlo Simulations for DW-MRI on realistic substrates will help the study of DWI-MRI signals in controlled environments and to investigate, extract, and validate approaches for the understanding of white matter features. In this work, we present a novel framework for creating complex white matter phantoms with unprecedented characteristics like high packing density greater than 90% intra-axonal volume fraction and voxel size of (140 um)3.

2022

A Signal Peak Separation Index for axisymmetric B-tensor encoding

Gabriel Girard, Benoît Macq, Jonathan Rafael Patino Lopez, Marco Pizzolato, Gaëtan Olivier D Rensonnet, Jean-Philippe Thiran

Diffusion-weighted MRI (DW-MRI) has recently seen a rising interest in planar, spherical and general B-tensor encodings. Some of these sequences have aided traditional linear encoding in the estimation of white matter microstructural features, generally by making DW-MRI less sensitive to the orientation of axon fascicles in a voxel. However, less is known about their potential to make the signal more sensitive to fascicle orientation, especially in crossing-fascicle voxels. Although planar encoding has been commended for the resemblance of its signal with the voxel's orientation distribution function (ODF), linear encoding remains the near undisputed method of choice for orientation estimation. This paper presents a theoretical framework to gauge the sensitivity of axisymmetric B-tensors to fascicle orientations. A signal peak separation index (SPSI) is proposed, motivated by theoretical considerations on a simple multi-tensor model of fascicle crossing. Theory and simulations confirm the intuition that linear encoding, because it maximizes B-tensor anisotropy, possesses an intrinsic advantage over all other axisymmetric B-tensors. At identical SPSI however, oblate B-tensors yield higher signal and may be more robust to acquisition noise than their prolate counterparts. The proposed index relates the properties of the B-tensor to those of the tissue microstructure in a straightforward way and can thus guide the design of diffusion sequences for improved orientation estimation and tractography.

Springer2021

Simulation-based Microstructure Imaging: From Forward Signal Modeling to Machine Learning

Jonathan Rafael Patino Lopez

This thesis focuses on the signal modelling of Diffusion-Weighted Magnetic Resonance Imaging (DW-MRI) in the brain's white matter (WM). DW-MRI a non-invasive technique with enormous potential for the study of the brain's microstructure by measuring the diffusion properties of biological tissue. In order to infer such microstructure properties from DW-MRI signals --- for example the axon's diameter distribution or the neurite dispersion, just to mention a few --- many models have been proposed in the past decades. However, such models have relied on disregarding several structural components of the WM, like the diameter and direction changes along the axons or the volumetric changes in the tissues' water compartments, which inherently affect the diffusion properties of the studied media. This large heterogeneity in the WM tissue, summed to the high structural variability between different brains regions or WM tracks, have prevented researchers from formulating accurate analytical models to this day.The following work proposes a paradigm change from the conventional analytic model-based approach to a simulated-based one, centered in the simulation of DW-MRI signals in realistic virtual tissue. Our work starts with the challenge of creating a robust framework for the simulation of DW-MRI as a forward modelling tool for microstructure estimation and continues exploding the simulator capabilities to create a simulation-based microstructure modelling strategy. Each chapter of this thesis presents several contributions to the field for the construction of realistic white matte numerical phantoms, which can later be used for the estimation of the axons' density and diameter distribution, or the design of novel microstructure-specific DW-MRI acquisition sequences. Thus, the contributions presented in this work pave the way for the modelling, study, and validation of complex microstructure models of brain white matter.

EPFL2021

, , , , , ,

Diffusion MRI (DW-MRI) allows for the detailed exploration of the brain white matter microstructure, with applications in both research and the clinic. However, state-of-the-art methods for microstructure estimation suffer from known limitations, such as the overestimation of the mean axon diameter, and the infeasibility of fitting diameter distributions. In this study, we propose to eschew current modeling based approaches in favor of a novel, simulation-assisted machine learning approach. In particular, we train machine learning (ML) algorithms on a large dataset of simulated diffusion MRI signals from white matter regions with different axon diameter distributions and packing densities. We show, on synthetic data, that the trained models in-silico and from DW-MRI data with moderately high b-values (4000 s/mm2). Further, we show, on in-vivo data, that the estimators trained from simulations can provide parameter estimates which are close to the values expected from histology.

Springer Nature2020

HOTmix: characterizing hindered diffusion using a mixture of generalized higher order tensors

Muhamed Barakovic, Erick Jorge Canales Rodriguez, Alessandro Daducci, Tim Bjørn Dyrby, Gabriel Girard, Jonathan Rafael Patino Lopez, Marco Pizzolato, Gaëtan Olivier D Rensonnet, David Paul Roger Romascano, Jean-Philippe Thiran

We present HOTmix, a new model to describe the diffusion MRI signal for molecules undergoing hindered diffusion. HOTmix is based on a mixture of generalized higher order tensors, explicitly incorporating the diffusion sequence’s time-dependent parameters. The method was evaluated on simulated diffusion MRI signals obtained through Monte Carlo simulations, using intermediate diffusion times, mimicking both ex-vivo and in-vivo conditions. HOTmix provided better reconstructions compared to the standard diffusion tensor, the kurtosis tensor, and a single generalized higher order tensor. In future work, we will explore whether modelling the hindered compartment using HOTmix improves microstructural features estimated using dMRI.

2019

Learning global brain microstructure maps using trainable sparse encoders

Muhamed Barakovic, Alessandro Daducci, Gabriel Girard, Jonathan Rafael Patino Lopez, Jean-Philippe Thiran

Diffusion-Weighted Magnetic Resonance Imaging is the only non-invasive technique available to infer the underlying brain tissue microstructure. Currently, one of the promising methods for microstructure imaging is signal modelling using convex formulation, e.g. using the COMMIT framework. Despite the benefits introduced with such a framework, an important limitation is the long convergence time, making the method unappealing for clinical applications. In order to address this limitation, we propose to use a neural network to learn the sparse encode representation of the data and performed an end-to-end reconstruction of the microstructure estimates directly from the diffusion-weighted data. Our results show that the neural network can accurately estimate the microstructure maps, 4 orders of magnitude faster than the convex formulation.

2019

, , , , ,

Probabilistic parameter estimation in model fitting runs the gamut from maximum likelihood or maximum a posteriori point estimates from optimization to Markov Chain Monte Carlo (MCMC) sampling. The latter, while more computationally intensive, generally provides a better characterization of the underlying parameter distribution than that of point estimates. However, in order to efficiently explore distributions, MCMC methods ideally require generating uncorrelated samples while also preserving reasonable acceptance probabilities; this becomes particularly important in problematic regions of parameter space. In this paper, we extend a recently proposed Hamiltonian MCMC sampler parametrized by neural networks (L2HMC) by modifying the loss function to jointly optimize the distance between samples and the acceptance probability such that it is stable and efficient. We apply this enhanced sampler to parameter estimation in a recently proposed MRI model, the multi-echo spherical mean technique. We show that it generally outperforms the state of the art Hamiltonian No-U-Turn (NUTS) sampler, L2HMC, and a least squares fitting in terms of accuracy and precision, also enabling the generation of more informative parameter posterior distributions. This illustrates the potential of machine learning enhanced samplers for improving probabilistic parameter estimation for medical imaging applications.

SPRINGER INTERNATIONAL PUBLISHING AG2019

Sparse wars: A survey and comparative study of spherical deconvolution algorithms for diffusion MRI

Muhamed Barakovic, Erick Jorge Canales Rodriguez, Alessandro Daducci, Gabriel Girard, Jonathan Rafael Patino Lopez, Marco Pizzolato, Gaëtan Olivier D Rensonnet, David Paul Roger Romascano, Jean-Philippe Thiran

Spherical deconvolution methods are widely used to estimate the brain's white-matter fiber orientations from diffusion MRI data. In this study, eight spherical deconvolution algorithms were implemented and evaluated. These included two model selection techniques based on the extended Bayesian information criterion (i.e., best subset selection and the least absolute shrinkage and selection operator), iteratively reweighted l2- and l1-norm approaches to approximate the l0-norm, sparse Bayesian learning, Cauchy deconvolution, and two accelerated Richardson-Lucy algorithms. Results from our exhaustive evaluation show that there is no single optimal method for all different fiber configurations, suggesting that further studies should be conducted to find the optimal way of combining solutions from different methods. We found l0-norm regularization algorithms to resolve more accurately fiber crossings with small inter-fiber angles. However, in voxels with very dominant fibers, algorithms promoting more sparsity are less accurate in detecting smaller fibers. In most cases, the best algorithm to reconstruct fiber crossings with two fibers did not perform optimally in voxels with one or three fibers. Therefore, simplified validation systems as employed in a number of previous studies, where only two fibers with similar volume fractions were tested, should be avoided as they provide incomplete information. Future studies proposing new reconstruction methods based on high angular resolution diffusion imaging data should validate their results by considering, at least, voxels with one, two, and three fibers, as well as voxels with dominant fibers and different diffusion anisotropies.

2019

Tractography Reproducibility Challenge With Empirical Data (TraCED): The 2017 ISMRM Diffusion Study Group Challenge

Muhamed Barakovic, Alessandro Daducci, Maxime Descoteaux, Tim Bjørn Dyrby, Matteo Frigo, Gabriel Girard, Jonathan Rafael Patino Lopez, David Paul Roger Romascano, Wei Sun, Jean-Philippe Thiran

2019