Ran Yang

IIR GRAPPA for parallel MR image reconstruction

Accelerated parallel MRI has advantage in imaging speed, and its image quality has been improved continuously in recent years. This paper introduces a two‐dimensional infinite impulse response model of inverse filter to replace the finite impulse response model currently used in generalized autocalibrating partially parallel acquisitions class image reconstruction methods. The infinite impulse response model better characterizes the correlation of k‐space data points and better approximates the perfect inversion of parallel imaging process, resulting in a novel generalized image reconstruction method for accelerated parallel MRI. This k‐space‐based reconstruction method includes the conventional generalized autocalibrating partially parallel acquisitions class methods as special cases and has a new infinite impulse response data estimation mechanism for effective improvement of image quality. The experiments on in vivo MRI data show that the proposed method significantly reduces reconstruction errors compared with the conventional two‐dimensional generalized autocalibrating partially parallel acquisitions method, particularly at the high acceleration rates. Magn Reson Med, 2010.

Computer-aided diagnosis of early knee osteoarthritis based on MRI T2 mapping.

This work was aimed at studying the method of computer-aided diagnosis of early knee OA (OA: osteoarthritis). Based on the technique of MRI (MRI: Magnetic Resonance Imaging) T2 Mapping, through computer image processing, feature extraction, calculation and analysis via constructing a classifier, an effective computer-aided diagnosis method for knee OA was created to assist doctors in their accurate, timely and convenient detection of potential risk of OA. In order to evaluate this method, a total of 1380 data from the MRI images of 46 samples of knee joints were collected. These data were then modeled through linear regression on an offline general platform by the use of the ImageJ software, and a map of the physical parameter T2 was reconstructed. After the image processing, the T2 values of ten regions in the WORMS (WORMS: Whole-organ Magnetic Resonance Imaging Score) areas of the articular cartilage were extracted to be used as the eigenvalues in data mining. Then,a RBF (RBF: Radical Basis Function) network classifier was built to classify and identify the collected data. The classifier exhibited a final identification accuracy of 75%, indicating a good result of assisting diagnosis. Since the knee OA classifier constituted by a weights-directly-determined RBF neural network didnt require any iteration, our results demonstrated that the optimal weights, appropriate center and variance could be yielded through simple procedures. Furthermore, the accuracy for both the training samples and the testing samples from the normal group could reach 100%. Finally, the classifier was superior both in time efficiency and classification performance to the frequently used classifiers based on iterative learning. Thus it was suitable to be used as an aid to computer-aided diagnosis of early knee OA.

Compressed sensing MRI by two-dimensional wavelet filter banks

How to speed up the scanning process is the bottleneck problem of magnetic resonance imaging (MRI). As a newly developed mathematical framework of signal sampling and recovery, compressed sensing (CS) provides a solution to this problem because of its potential of reconstructing MR images from fewer samples. Recent work has demonstrated successful application of CS to MRI. However, the frequently used sparsifying transform is the traditional discrete wavelet transform, which has shortcomings, such as oscillations, lack of directionality and shift variance. This paper implements compressed sensing MRI reconstruction based on a new kind of two-dimensional wavelet filter banks which has improved directional selectivity and approximate shift invariance. Our experiments show that the method can significantly reduce aliasing and achieve higher peak signal to noise ratio (PSNR).

Analysis of generalized rosette trajectory for compressed sensing MRI

PURPOSEnThe application of compressed sensing (CS) technology in magnetic resonance imaging (MRI) is to accelerate the MRI scan speed by incoherent undersampling of k-space data and nonlinear iterative reconstruction of MRI images. This paper generalizes the existing rosette trajectories to configure the sampling patterns for undersampled k-space data acquisition in MRI scans. The arch and curvature characteristics of the generalized rosette trajectories are analyzed to explore their feasibility and advantages for CS reconstruction of MRI images.nnnMETHODSnTwo key properties crucial to the CS MRI application, the scan speed and sampling incoherence of the generalized rosette trajectories, are analyzed. The analysis on the scan speed of generalized rosette trajectories is based on the transversal time derived from the curvature of the trajectories, and the sampling incoherence is based on the evaluation of the point spread function for the measurement matrix. The results of analysis are supported by extensive simulations where the performances of rosette, spiral, and radial sampling patterns at different acceleration factors are compared.nnnRESULTSnIt is shown that compared with spiral trajectories, the arch and curvature characteristics of the generalized rosette trajectories are more feasible to meet the physical requirements of undersampled k-space data acquisition in terms of time shortness and scan area. It is further shown that the sampling pattern of the rosette trajectory has higher incoherence than that of the other trajectories and can thus achieve higher reconstruction performance. Reconstruction performances illustrate that the rosette trajectory can achieve about 10% higher peak signal-to-noise ratio than radial and spiral trajectories under the high acceleration factor R = 10.nnnCONCLUSIONSnThe generalized rosette trajectories can be a desirable candidate for CS reconstruction of MRI.

An improved GRAPPA algorithm based on sensitivity estimation

This paper analyzes the famous GRAPPA algorithm, which is one of most widely used image reconstruction algorithms for parallel magnetic resonance imaging (pMRI). Inherently the existing GRAPPA type algorithms ignore the physical background of k-space data and treat the image reconstruction problem as a pure data interpolation problem which is solved based on an assumption that the k-space data are shift-invariant autoregressive process. Based on physical principles of MRI, this paper reveals the difficulty of such assumption. New GRAPPA algorithm is developed where the above assumption is relaxed and the missing k-space data are reconstructed based on physical properties of k-space data and coil sensitivity profiles, which can be estimated using Auto-Calibrating Signal (ACS) lines. This proposed algorithm can greatly improve the image quality even at very high acceleration factor. The in vivo examples demonstrate the overwhelming advantages of the proposed algorithm.

Super-resolution reconstruction of dynamic MRI by patch learning

Achieving both high spatial and temporal resolution is desired in dynamic Magnetic Resonance Imaging (MRI), however, it is difficult to satisfy because of the slow scanning speed of MRI caused by physical and physiological limits. In order to guarantee the temporal resolution, the amount of acquired k-space data is usually reduced. In this paper, a novel method-patch learning-based dynamic MRI super-resolution reconstruction, is proposed, where high resolution dynamic images are reconstructed based on the sampled low frequency k-space data together with a small amount of fully sampled frames as training data. The proposed method is also demonstrated by in-vivo dynamic MRI data.

On analysis of k-t BLAST for dynamic MRI

k-t BLAST (Broad-use Linear Acquisition Speed-up Technique) is a popular means of improving imaging speed for dynamic MRI. The impact of several factors such as dealing DC(direct current) separately, the width of the temporal frequency band and the signal estimation capacity of the unaliasing formalisms are investigated. It is shown that the compactness of the temporal frequency representation is crucial to the performance of k-t BLAST. The accuracy of the priori information retrieved from the training data has only diminishing impact on reconstruction quality due to the inaccurate unaliasing equation. The estimation capacity of the unaliasing formalism is insufficient when the signal overlap is severe, and changing the unaliasing formalism accordingly may improve the reconstruction quality.

2D IIR filter for parallel magnetic resonance image reconstruction

Accelerated parallel MRI has been widely used in medical research and clinical diagnoses. This paper presents a two dimensional (2D) infinite impulse response (IIR) model of inverse filter to replace the finite impulse response model currently used in GRAPPA class image reconstruction methods. The IIR model better characterizes the correlation of k-space data points and better approximates the inversion of parallel imaging process. The experiments on in vivo accelerated cardiac imaging show that the proposed method significantly reduces reconstruction errors compared with conventional 2D GRAPPA method, particularly at the high acceleration rates.

GESPIRiT: ESPIRiT combined with GRAPPA while autocalibration data is insufficient

Parallel magnetic resonance imaging (pMRI) which utilizes the redundant information from different coil sensitivities has been widely used. As the number of coils increases, algorithms that reconstruct single combined image are more suitable for pMRI. Most single combined image reconstruction need explicit functions of coil sensitivity. As a result, estimation of the coil sensitivities is important in pMRI. Nowadays, ESPIRiT(Iterative self-consistent parallel imaging reconstruction using eigenvector maps) based on the estimation of coil sensitivities is an outstanding algorithm. However, ESPIRiT has its limitation, which estimates the coil sensitivities not very correctly, while the autocalibration data is insufficient. In this paper, we focus on the pMRI algorithms based on improving the estimation of coil sensitivities and propose a new algorithm called GESPIRiT (ESPIRiT combined with GRAPPA), which is based on combining ESPIRiT with GRAPPA to obtain better estimation of coil sensitivities and final reconstructed image. Experiments are applied to demonstrate its feasibility and efficiency.

Image reconstruction of dynamic MRI based on adaptive motion estimation

In recent years, dynamic MRI plays a crucial role in dynamic clinical studies. Since there are significant correlations in temporal dimension, the methods of making full use of the redundancy between frames have been widely concerned. In order to improve under-sampled rate in k-space, an effective way is to borrow the thoughts of motion estimation from video compression coding. This paper aims at improving under-sampled rate and image quality through the technique of motion estimation and motion compensation. Based on the conception of motion vector prediction and multiple reference frames, a technique of adaptive size for block matching is proposed. Then the non-sampled frames can be reconstructed through interpolation. For the sake of clarity, this paper only focus on the cases with single receiver coil, but it can also be applied to the cases with multiple receiver coils.