This notebook provides an intuitive, flexible and fully customizable MRI simulation tool for educational purposes. This version implements a spoiled GRE sequence applied to a numerical cardiac phantom with blood flow and simulates flow artefacts, as well as the effects of flow compensation.

Convolutional Neural Networks (CNNs) are highly effective tools for image reconstruction problems. Typically, CNNs are trained on large amounts of images, but, perhaps surprisingly, even without any training data, CNNs such as the Deep Image Prior and Deep Decoder achieve excellent imaging performance. Here, we build on those works by proposing an un-trained CNN for accelerated MRI along with performance-enhancing steps including enforcing data-consistency and combining multiple reconstructions

The rapid emergence of standardized robust non-invasive imaging methods and infrastructure for big data analysis has promoted the advent of a variety of large-scale neuroimaging studies. As a central component, versatile MR imaging offers the potential to investigate in-vivo the variability, development and anatomical layout of the human brain in health and disease. Increasing our knowledge of pre-symptomatic neuroanatomical changes supports research into disease etiology, risk and preserving factors, as well as potential intervention paradigms. Consequently, there is a need for efficient, scalable, and sensitive software tools to automatically extract clinically-relevant imaging markers. Traditional neuroimage analysis pipelines, however, often involve computationally intensive, time-consuming optimization steps, and thus, do not scale well to large cohort studies.

This tutorial provides an introduction to DOSMA, a Python-based deep-learning (DL), open-source MR image analysis framework. Topics include: 1) streamlining vendor-agnostic, multi-format data I/O, 2) using and deploying DL tools, 3) parallelizing compute intensive routines (e.g. registration, curve fitting), and 4) building scalable, reproducible analysis workflows.

Open-source Prototype of Myocardial T1 mapping (ProMyoT1) using pyPulseq which includes an inversion recovery T1 mapping sequence with a triggering scheme.

To utilize full 3D information in advanced reconstruction and fitting algorithms on memory limited GPUs, special solutions strategies are necessary to leverage the speed advantage, e.g., hiding memory latency of repeated transfers to/from the GPU to host memory. This can be achieved using asynchronous execution strategies. However, correct synchronization of critical operations can be error prone. To this end, we propose PyQMRI, a simple to use Python toolbox for quantitative MRI.

The repository hosts some example codes to perform MR image reconstruction with deep learning architectures. We showcase a comparison of denoising networks (denoising) and unrolled reconstruction networks (physics-based). Pure real-valued processing (real) is compared against complex-valued processing as complex-valued operations (complex) or 2-channel real-valued operations (2chreal). Respective complex-valued operations and data consistency layers are provided. A deep plug-and-play prior example is illustrated for cardiovascular MRI.

This educational lecture will give a broad overview of Quantitative Susceptibility Mapping. After explaining what magnetic susceptibility is and how we can measure it, I cover the data acquisition aspects, coil combination, unwrapping, masking, background field correction and dipole inversion.

This interactive Jupyter Book is comprised of 3 notebooks, outlining an overarching standard operating procedure to work with 1D ( audio files of MRI pulse sequences, guitar melodies and vocal tracks), 2D (BIDS formatted reconstructed images) and 3D (multi-channel k-space data in ISMRM-RD format) data: i) obtain meta-information about the data and use community-developed readers wherever possible, ii) use NumPy to prepare the data for further processing and iii) use Scipy modules to perform fundamental signal and image processing tasks.

Deep learning offers powerful tools for enhancing image quality and acquisition speed of MR images. Standard frameworks such as TensorFlow and PyTorch provide simple access to deep learning methods. However, they lack MRI specific operations and make reproducible research and code reuse more difficult due to fast changing APIs and complicated dependencies. In this tutorial, the user learns how to train a neural network with BART on the MNIST dataset. Moreover, the BART implementations of MoDL and the Variational Network are used for reconstruction.

Advanced reconstruction algorithms based on deep learning have recently drawn a lot of interest as they tend to outperform state-of-the-art methods. BART is a versatile framework for image reconstruction. In this work, we demonstrate how neural networks trained and tested with TensorFlow [5] can be integrated into BART. As an example, we discuss non-Cartesian parallel imaging using the SENSE model regularized by a log-likelihood image prior. The image prior is based on an autoregressive generative network pixel-cnn++. Furthermore, we validated the reconstruction pipeline using radial brain scans.

Today, most clinical scans are performed with Cartesian k-space sampling due to its robustness and ease to implement acquisition and reconstruction. However, there are numerous reasons to use non-Cartesian sampling methods, reasons that range from robustness to motion and flow, to less intrusive undersampling artifacts and more beneficial properties for advanced image reconstruction methods such as compressed sensing. This lecture covers the basics of image reconstruction with the non-uniform FFT.

This tutorial provides an introduction to quantitative T1 mapping, from an MRI physics perspective. Two widely used techniques are covered in-depth, Inversion Recovery and Variable Flip Angle (VFA), along with some discussions of cutting-edge variants of these techniques.

To shorten 4D flow acquisitions by shortening TRs with fast RF pulses and gradient waveforms. Real‐time convex optimization is used to generate these gradients waveforms on the scanner.

Quantitative analysis of dynamic contrast‐enhanced MRI (DCE‐MRI) requires an arterial input function (AIF) which is difficult to measure. We propose the reference region and input function tail (RRIFT) approach which uses a reference tissue and the washout portion of the AIF.

Inhomogeneous magnetization transfer (ihMT) is an emerging form of MRI contrast that may offer high specificity for myelinated tissue. Existing ihMT and pulsed MT sequences often use separate radiofrequency pulses for saturation and signal excitation. This study investigates the use of nonselective multiband radiofrequency pulses for simultaneous off‐resonance saturation and on‐resonance excitation specifically for generation of ihMT contrast within rapid steady‐state pulse sequences.

This notebook presents an example analysis pipeline using the Spinal Cord Toolbox (SCT), a suite of tools specialized for analysis of spinal cord MRI images of the spinal. Topics covered include: segmentation, masking, registration, warping, and quantitative metric computation. This tutorial was generated in a Jupyter Notebook and coded in Python.

We report a new postprocessing procedure that uses Fourier‐domain data analyses to improve the accuracy and reliability of phase unwrapping for MRI data of low SNR.

To introduce and demonstrate a software library for time‐optimal gradient waveform optimization with a wide range of applications. The software enables direct on‐the‐fly gradient waveform design on the scanner hardware for multiple vendors.

This online course is divided in 4 chapters (for more info see 📕 Course chapters ) that demonstrate different MRI techniques and display the results in Jupyer notebooks. The notebooks are written in Julia (1.4.1) and Python (3.7).