What has happened in machine learning lately, and what does it mean for the future of medical image analysis? Machine learning has witnessed a tremendous amount of attention over the last few years. The current boom started around 2009 when so-called deep artificial neural networks began outperforming other established models on a number of important benchmarks. Deep neural networks are now the state-of-the-art machine learning models across a variety of areas, from image analysis to natural language processing, and widely deployed in academia and industry. These developments have a huge potential for medical imaging technology, medical data analysis, medical diagnostics and healthcare in general, slowly being realized. We provide a short overview of recent advances and some associated challenges in machine learning applied to medical image processing and image analysis. As this has become a very broad and fast expanding field we will not survey the entire landscape of applications, but put particular focus on deep learning in MRI. 
Our aim is threefold: (i) give a brief introduction to deep learning with pointers to core references; (ii) indicate how deep learning has been applied to the entire MRI processing chain, from acquisition to image retrieval, from segmentation to disease prediction; (iii) provide a starting point for people interested in experimenting and perhaps contributing to the field of machine learning for medical imaging by pointing out good educational resources, state-of-the-art open-source code, and interesting sources of data and problems related medical imaging.

An overview of deep learning in medical imaging focusing on MRI

What has happened in machine learning lately, and what does it mean for the future of medical image analysis? Machine learning has witnessed a tremendous amount of attention over the last few years. The current boom started around 2009 when so-called deep artificial neural networks began outperforming other established models on a number of important benchmarks. Deep neural networks are now the state-of-the-art machine learning models across a variety of areas, from image analysis to natural language processing, and widely deployed in academia and industry. These developments have a huge potential for medical imaging technology, medical data analysis, medical diagnostics and healthcare in general, slowly being realized. We provide a short overview of recent advances and some associated challenges in machine learning applied to medical image processing and image analysis. As this has become a very broad and fast expanding field we will not survey the entire landscape of applications, but put particular focus on deep learning in MRI. Our aim is threefold: (i) give a brief introduction to deep learning with pointers to core references; (ii) indicate how deep learning has been applied to the entire MRI processing chain, from acquisition to image retrieval, from segmentation to disease prediction; (iii) provide a starting point for people interested in experimenting and perhaps contributing to the field of deep learning for medical imaging by pointing out good educational resources, state-of-the-art open-source code, and interesting sources of data and problems related medical imaging.

Historically, clinical MRI started with main magnetic field strengths in the ∼0.05-0.35T range. In the past 40 years there have been considerable developments in MRI hardware, with one of the primary ones being the trend to higher magnetic fields. While resulting in large improvements in data quality and diagnostic value, such developments have meant that conventional systems at 1.5 and 3T remain relatively expensive pieces of medical imaging equipment, and are out of the financial reach for much of the world. In this review we describe the current state-of-the-art of low-field systems (defined as 0.25-1T), both with respect to its low cost, low foot-print, and subject accessibility. Furthermore, we discuss how low field could potentially benefit from many of the developments that have occurred in higher-field MRI. In the first section, the signal-to-noise ratio (SNR) dependence on the static magnetic field and its impact on the achievable contrast, resolution, and acquisition times are discussed from a theoretical perspective. In the second section, developments in hardware (eg, magnet, gradient, and RF coils) used both in experimental low-field scanners and also those that are currently in the market are reviewed. In the final section the potential roles of new acquisition readouts, motion tracking, and image reconstruction strategies, currently being developed primarily at higher fields, are presented. Level of Evidence: 5 Technical Efficacy Stage: 1 J. Magn. Reson. Imaging 2019.

https://onlinelibrary.wiley.com/doi/pdf/10.1002/jmri.26637

Low‐field MRI: An MR physics perspective

Quantitative magnetic resonance imaging (qMRI) goes beyond conventional MRI, which aims primarily at local image contrast. It provides specific physical parameters related to the nuclear spin of protons in water, such as relaxation times. These parameters carry information about the local microstructural environment of the protons (such as myelin in the brain). Non-invasive in vivo histology using MRI (hMRI) aims to use this information to directly characterize biological tissue microstructure, partially replacing or complementing classical invasive histology. The understanding of MRI tissue contrast provided by hMRI is, in turn, crucial for further improvements of qMRI, and they should be considered closely interlinked. We discuss concepts, models and validation approaches, pointing out challenges and the latest advances in this field. Further, we point out links to physics, including computational and analytical approaches and developments in materials science and photonics, that aid in reference data acquisition and model validation. Quantitative magnetic resonance imaging and in vivo histology go beyond standard magnetic resonance imaging, aiming at characterizing tissue microstructure of the living brain. This Technical Review discusses advances in concepts, instrumentation, biophysical models and validation approaches facilitating this rapidly developing field.

Quantitative magnetic resonance imaging of brain anatomy and in vivo histology

Magnetic resonance fingerprinting (MRF) is a quantitative imaging technique that can simultaneously measure multiple important tissue properties of human body. Although MRF has demonstrated improved scan efficiency as compared to conventional techniques, further acceleration is still desired for translation into routine clinical practice. The purpose of this paper is to accelerate MRF acquisition by developing a new tissue quantification method for MRF that allows accurate quantification with fewer sampling data. Most of the existing approaches use the MRF signal evolution at each individual pixel to estimate tissue properties, without considering the spatial association among neighboring pixels. In this paper, we propose a spatially constrained quantification method that uses the signals at multiple neighboring pixels to better estimate tissue properties at the central pixel. Specifically, we design a unique two-step deep learning model that learns the mapping from the observed signals to the desired properties for tissue quantification, i.e.: 1) with a feature extraction module for reducing the dimension of signals by extracting a low-dimensional feature vector from the high-dimensional signal evolution and 2) a spatially constrained quantification module for exploiting the spatial information from the extracted feature maps to generate the final tissue property map. A corresponding two-step training strategy is developed for network training. The proposed method is tested on highly undersampled MRF data acquired from human brains. Experimental results demonstrate that our method can achieve accurate quantification for T1 and T2 relaxation times by using only 1/4 time points of the original sequence (i.e., four times of acceleration for MRF acquisition).

/pdf/erratum-to-deep-learning-for-fast-and-spatially-constrained-586vfabv7a.pdf

Erratum to “Deep Learning for Fast and Spatially Constrained Tissue Quantification From Highly Accelerated Data in Magnetic Resonance Fingerprinting”

Purpose The purpose of this study was to accelerate the acquisition and reconstruction time of 3D magnetic resonance fingerprinting scans. Methods A 3D magnetic resonance fingerprinting scan was accelerated by using a single-shot spiral trajectory with an undersampling factor of 48 in the x-y plane, and an interleaved sampling pattern with an undersampling factor of 3 through plane. Further acceleration came from reducing the waiting time between neighboring partitions. The reconstruction time was accelerated by applying singular value decomposition compression in k-space. Finally, a 3D premeasured B1 map was used to correct for the B1 inhomogeneity. Results The T1 and T2 values of the International Society for Magnetic Resonance in Medicine/National Institute of Standards and Technology MRI phantom showed a good agreement with the standard values, with an average concordance correlation coefficient of 0.99, and coefficient of variation of 7% in the repeatability scans. The results from in vivo scans also showed high image quality in both transverse and coronal views. Conclusions This study applied a fast acquisition scheme for a fully quantitative 3D magnetic resonance fingerprinting scan with a total acceleration factor of 144 as compared with the Nyquist rate, such that 3D T1 , T2 , and proton density maps can be acquired with whole-brain coverage at clinical resolution in less than 5 min. Magn Reson Med 79:2190-2197, 2018. © 2017 International Society for Magnetic Resonance in Medicine.

Fast 3D magnetic resonance fingerprinting for a whole-brain coverage.

Multiparametric quantitative imaging is gaining increasing interest due to its widespread advantages in clinical applications. Magnetic resonance fingerprinting is a recently introduced approach of fast multiparametric quantitative imaging. In this article, magnetic resonance fingerprinting acquisition, dictionary generation, reconstruction, and validation are reviewed.

https://onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.27403

Magnetic resonance fingerprinting: a technical review.

Magnetic Resonance Fingerprinting-An Overview.

Purpose This study explores the possibility of using a gradient moment balanced sequence with a quadratically varied RF excitation phase in the magnetic resonance fingerprinting (MRF) framework to quantify T2 * in addition to δ f , T1 , and T2 tissue properties. Methods The proposed quadratic RF phase-based MRF method (qRF-MRF) combined a varied RF excitation phase with the existing balanced SSFP (bSSFP)-based MRF method to generate signals that were uniquely sensitive to δ f , T1 , T2 , as well as the distribution width of intravoxel frequency dispersion, Γ . A dictionary, generated through Bloch simulation, containing possible signal evolutions within the physiological range of δ f , T1 , T2 , and Γ , was used to perform parameter estimation. The estimated T2 and Γ were subsequently used to estimate T2 * . The proposed method was evaluated in phantom experiments and healthy volunteers (N = 5). Results The T1 and T2 values from the phantom by qRF-MRF demonstrated good agreement with values obtained by traditional gold standard methods (r2 = 0.995 and 0.997, respectively; concordance correlation coefficient = 0.978 and 0.995, respectively). The T2 * values from the phantom demonstrated good agreement with values obtained through the multi-echo gradient-echo method (r2 = 0.972, concordance correlation coefficient = 0.983). In vivo qRF-MRF-measured T1 , T2 , and T2 * values were compared with measurements by existing methods and literature values. Conclusion The proposed qRF-MRF method demonstrated the potential for simultaneous quantification of δ f , T1 , T2 , and T2 * tissue properties.

/pdf/magnetic-resonance-fingerprinting-with-quadratic-rf-phase-y41w7j4l3j.pdf

Magnetic resonance fingerprinting with quadratic RF phase for measurement of T2* simultaneously with δf, T1, and T2

Purpose The purpose of this study is to increase the robustness of MR fingerprinting (MRF) toward subject motion. Methods A novel reconstruction algorithm, MOtion insensitive MRF (MORF), was developed, which uses an iterative reconstruction based retrospective motion correction approach. Each iteration loops through the following steps: pattern recognition, metric based identification of motion corrupted frames, registration based motion estimation, and motion compensated data consistency verification. The proposed algorithm was validated using in vivo 2D brain MRF data with retrospective in-plane motion introduced at different stages of the acquisition. The validation was performed using qualitative and quantitative comparisons between results from MORF, the iterative multi-scale (IMS) algorithm, and with the IMS results using data without motion for a ground truth comparison. Additionally, the MORF algorithm was evaluated in prospectively motion corrupted in vivo 2D brain MRF datasets. Results For datasets corrupted by in-plane motion both prospectively and retrospectively, MORF noticeably reduced motion artifacts compared with iterative multi-scale and closely resembled the results from data without motion, even when ∼54% of data was motion corrupted during different parts of the acquisition. Conclusions MORF improves the insensitivity of MRF toward rigid-body motion occurring during any part of the MRF acquisition.

/pdf/image-reconstruction-algorithm-for-motion-insensitive-mr-1qidoz5xql.pdf

Bhairav Bipin Mehta

Papers

Fast 3D magnetic resonance fingerprinting for a whole-brain coverage.

Magnetic resonance fingerprinting: a technical review.

Magnetic Resonance Fingerprinting-An Overview.

Magnetic resonance fingerprinting with quadratic RF phase for measurement of T2* simultaneously with δf, T1, and T2

Image reconstruction algorithm for motion insensitive MR Fingerprinting (MRF): MORF