System Identification I

Analysis of big data by machine learning offers considerable advantages for assimilation and evaluation of large amounts of complex health-care data. However, to effectively use machine learning tools in health care, several limitations must be addressed and key issues considered, such as its clinical implementation and ethics in health-care delivery. Advantages of machine learning include flexibility and scalability compared with traditional biostatistical methods, which makes it deployable for many tasks, such as risk stratification, diagnosis and classification, and survival predictions. Another advantage of machine learning algorithms is the ability to analyse diverse data types (eg, demographic data, laboratory findings, imaging data, and doctors' free-text notes) and incorporate them into predictions for disease risk, diagnosis, prognosis, and appropriate treatments. Despite these advantages, the application of machine learning in health-care delivery also presents unique challenges that require data pre-processing, model training, and refinement of the system with respect to the actual clinical problem. Also crucial are ethical considerations, which include medico-legal implications, doctors' understanding of machine learning tools, and data privacy and security. In this Review, we discuss some of the benefits and challenges of big data and machine learning in health care.

Big data and machine learning algorithms for health-care delivery

Magnetic resonance image reconstruction from undersampled measurements using a patch-based nonlocal operator

Compressed sensing (CS) has the potential to reduce magnetic resonance (MR) data acquisition time. In order for CS-based imaging schemes to be effective, the signal of interest should be sparse or compressible in a known representation, and the measurement scheme should have good mathematical properties with respect to this representation. While MR images are often compressible, the second requirement is often only weakly satisfied with respect to commonly used Fourier encoding schemes. This paper investigates the use of random encoding for CS-MRI, in an effort to emulate the “universal” encoding schemes suggested by the theoretical CS literature. This random encoding is achieved experimentally with tailored spatially-selective radio-frequency (RF) pulses. Both simulation and experimental studies were conducted to investigate the imaging properties of this new scheme with respect to Fourier schemes. Results indicate that random encoding has the potential to outperform conventional encoding in certain scenarios. However, our study also indicates that random encoding fails to satisfy theoretical sufficient conditions for stable and accurate CS reconstruction in many scenarios of interest. Therefore, there is still no general theoretical performance guarantee for CS-MRI, with or without random encoding, and CS-based methods should be developed and validated carefully in the context of specific applications.

/pdf/compressed-sensing-mri-with-random-encoding-4s3g5574xo.pdf

Compressed-Sensing MRI With Random Encoding

Image reconstruction from undersampled k-space data has been playing an important role in fast magnetic resonance imaging (MRI). Recently, deep learning has demonstrated tremendous success in various fields and also shown potential in significantly accelerating MRI reconstruction with fewer measurements. This article provides an overview of deep-learning-based image reconstruction methods for MRI. Two types of deep-learningbased approaches are reviewed, those that are based on unrolled algorithms and those that are not, and the main structures of both are explained. Several signal processing issues for maximizing the potential of deep reconstruction in fast MRI are discussed, which may facilitate further development of the networks and performance analysis from a theoretical point of view.

Deep Magnetic Resonance Image Reconstruction: Inverse Problems Meet Neural Networks

In sensitivity encoding reconstruction, the issue of ill conditioning
becomes serious and thus the signal-to-noise ratio
becomes poor when a large acceleration factor is employed.
Total variation (TV) regularization has been used to address
this issue and shown to better preserve sharp edges than
Tikhonov regularization but may cause blocky effect. In this article,
we study nonlocal TV regularization for noise suppression
in sensitivity encoding reconstruction. The nonlocal TV regularization
method extends the conventional TV norm to a nonlocal
version by introducing a weighted nonlocal gradient
function calculated from the weighted difference between the
target pixel and its generalized neighbors, where the weights
incorporate the prior information of the image structure. The
method not only inherits the edge-preserving advantage of TV
regularization but also overcomes the blocky effect. The experimental
results from in vivo data show that nonlocal TV regularization
is superior to the existing competing methods in
preserving fine details and reducing noise and artifacts.

https://onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.22736

Sensitivity encoding reconstruction with nonlocal total variation regularization.

This paper introduces a modified principal dynamic modes (PDM) method, which is able to separate the dynamics of sympathetic and parasympathetic nervous activities. The PDM is based on the principle that among all possible choices of expansion bases, there are some that require the minimum number of basis functions to achieve a given mean-square approximation of the system output. Such a minimum set of basis functions is termed PDMs of the nonlinear system. We found that the first two dominant PDMs have similar frequency characteristics for parasympathetic and sympathetic activities, as reported in the literature. These results are consistent for all nine of our healthy human subjects using our modified PDM approach. Validation of the purported separation of parasympathetic and sympathetic activities was performed by the application of the autonomic nervous system blocking drugs atropine and propranolol. With separate applications of the respective drugs, we found a significant decrease in the amplitude of the waveforms that correspond to each nervous activity. Furthermore, we observed near complete elimination of these dynamics when both drugs were given to the subjects. Comparison of our method to the conventional low-frequency/high-frequency ratio shows that our proposed approach provides more accurate assessment of the autonomic nervous balance. Our nonlinear PDM approach allows a clear separation of the two autonomic nervous activities, the lack of which has been the main reason why heart rate variability analysis has not had wide clinical acceptance.

Nonlinear analysis of the separate contributions of autonomic nervous systems to heart rate variability using principal dynamic modes

Magnetic resonance imaging (MRI) is known to be a slow imaging modality and undersampling in k-space has been used to increase the imaging speed. However, image reconstruction from undersampled k-space data is an ill-posed inverse problem. Iterative algorithms based on compressed sensing have been used to address the issue. In this work, we unroll the iterations of the primal-dual hybrid gradient algorithm to a learnable deep network architecture, and gradually relax the constraints to reconstruct MR images from highly undersampled k-space data. The proposed method combines the theoretical convergence guarantee of optimization methods with the powerful learning capability of deep networks. As the constraints are gradually relaxed, the reconstruction model is finally learned from the training data by updating in k-space and image domain alternatively. Experiments on in vivo MR data demonstrate that the proposed method achieves superior MR reconstructions from highly undersampled k-space data over other state-of-the-art image reconstruction methods.

Model Learning: Primal Dual Networks for Fast MR Imaging

Image reconstruction in magnetic resonance imaging (MRI) clinical applications has become increasingly more complicated. However, diagnostic and treatment require very fast computational procedure. Modern competitive platforms of graphics processing unit (GPU) have been used to make high-performance parallel computations available, and attractive to common consumers for computing massively parallel reconstruction problems at commodity price. GPUs have also become more and more important for reconstruction computations, especially when deep learning starts to be applied into MRI reconstruction. The motivation of this survey is to review the image reconstruction schemes of GPU computing for MRI applications and provide a summary reference for researchers in MRI community.

A survey of GPU-based acceleration techniques in MRI reconstructions.

The paper presents a novel approach of pseudo 2D random sampling scheme for application of compressed sensing in Cartesian magnetic resonance imaging (MRI). The proposed scheme is realized by a pulse sequence program which switches the directions of phase encoding and frequency encoding during data acquisition such that both k x and k y directions can be undersampled randomly. The resulting random sampling pattern approximates the ideal but impractical 2D patterns. Both the simulation and experiment results show the proposed method is superior to the existing 1D random sampling and similar to the ideal 2D random sampling in terms of the reconstruction quality. This method can potentially improve the MR imaging speed through the application of compressed sensing in conventional MRI.

Haifeng Wang

Papers

Sensitivity encoding reconstruction with nonlocal total variation regularization.

Nonlinear analysis of the separate contributions of autonomic nervous systems to heart rate variability using principal dynamic modes

Model Learning: Primal Dual Networks for Fast MR Imaging

A survey of GPU-based acceleration techniques in MRI reconstructions.

Pseudo 2D random sampling for compressed sensing MRI