Dual-domain cascade of U-nets for multi-channel magnetic resonance image reconstruction

The U-net is a deep-learning network model that has been used to solve a number of inverse problems. In this work, the concatenation of two-element U-nets, termed the W-net, operating in k-space (K) and image (I) domains, were evaluated for multi-channel magnetic resonance (MR) image reconstruction. The two-element network combinations were evaluated for the four possible image-k-space domain configurations: a) W-net II, b) W-net KK, c) W-net IK, and d) W-net KI. Selected four element (WW-nets) and six element (WWW-nets) networks were also examined. Two configurations of each network were compared: 1) each coil channel was processed independently, and 2) all channels were processed simultaneously. One hundred and eleven volumetric, T1-weighted, 12-channel coil k-space datasets were used in the experiments. Normalized root mean squared error, peak signal-to-noise ratio and visual information fidelity were used to assess the reconstructed images against the fully sampled reference images. Our results indicated that networks that operate solely in the image domain were better when independently processing individual channels of multi-channel data. Dual-domain methods were better when simultaneously reconstructing all channels of multi-channel data. In addition, the best cascade of U-nets performed better (p < 0.01) than the previously published, state-of-the-art Deep Cascade and Hybrid Cascade models in three out of four experiments.     

IKWI-net: A cross-domain convolutional neural network for undersampled magnetic resonance image reconstruction

Magnetic resonance imaging (MRI) is widely used to get the information of anatomical structure and physiological function with the advantages of high resolution and non-invasive scanning. But the long acquisition time limits its application. To reduce the time consumption of MRI, compressed sensing (CS) theory has been proposed to reconstruct MRI images from undersampled k-space data. But conventional CS methods mostly use iterative methods that take lots of time. Recently, deep learning methods are proposed to achieve faster reconstruction, but most of them only pay attention to a single domain, such as the image domain or k-space. To take advantage of the feature representation in different domains, we propose a cross-domain method based on deep learning, which first uses convolutional neural networks (CNNs) in the image domain, k-space and wavelet domain simultaneously. The combined order of the three domains is also first studied in this work, which has a significant effect on reconstruction. The proposed IKWI-net achieves the best performance in various combinations, which utilizes CNNs in the image domain, k-space, wavelet domain and image domain sequentially. Compared with several deep learning methods, experiments show it also achieves mean improvements of 0.91 dB in peak signal-to-noise ratio (PSNR) and 0.005 in structural similarity (SSIM).     

基于支撑先验与深度图像先验的无预训练磁共振图像重建方法

基于深度学习的磁共振成像(magnetic resonance imaging, MRI)方法需要大规模、高质量的病患数据样本集进行预训练.然而,由于病患隐私及设备等因素限制,获取大规模、高质量的磁共振数据集在实际临床应用中面临挑战.本文提出一种新的基于深度学习的欠采样磁共振图像重建方法,该方法无需预训练、不依赖训练数据集,而是充分利用待重建的目标MR图像的结构先验和支撑先验,并将其引入深度图像先验(deep image prior, DIP)框架,从而削减对训练数据集的依赖,提升学习效率.基于参考图像与目标图像的相似性,采用高分辨率参考图像作为深度网络输入,将结构先验信息引入网络;将参考图像在小波域中幅值大的系数索引集作为目标图像的已知支撑集,构造正则化约束项,将网络训练转化为网络参数的最优化求解过程.实验结果表明,本文方法可由欠采样k空间数据重建得到更精确的磁共振图像,且在保留组织特征、细节纹理方面具有明显优势.     

Imaging of stroke: a comparison between X-ray fluorescence and magnetic resonance imaging methods

A dual imaging approach, combining magnetic resonance imaging to localize lesions and synchrotron rapid scanning X-ray fluorescence (XRF) mapping to localize and quantify calcium, iron and zinc was used to examine one case of recent stroke with hemorrhage and two cases of ischemia 3 and 7 years before death with the latter showing superficial necrosis. In hemorrhagic lesions, more Fe is found accompanied with less Zn. In chronic ischemic lesions, Fe, Zn and Ca are lower indicating that these elements are removed as the normal tissue dies and scar tissue forms. Both susceptibility and T2* maps were calculated to visualize iron in hemorrhages and validated by XRF Ca and Fe maps. The former was superior for imaging iron in hemorrhagic transformation and necrosis but did not capture ischemic lesions. In contrast, T2* could not differentiate Ca from Fe in necrotic tissue but did capture ischemic lesions, complementing the susceptibility mapping. The spatial localization, accurate quantitative data and elemental differentiation shown here could also be valuable for imaging other brain tissue damage with abnormal Ca and Fe content.     

A comparison of ferumoxytol with gadolinium as contrast agents for the diagnostic magnetic resonance imaging of osteomyelitis

BackgroundFerumoxytol, an FDA-approved superparamagnetic iron oxide nanoparticle (SPION) preparation used for the treatment of iron deficiency anemia, is also known to be taken up by macrophages in areas of infection or inflammation, where it produces negative contrast changes on T2-weighted MR images.PurposeWe sought to compare Ferumoxytol-induced MRI contrast changes with those observed using standard-of-care Gadolinium in patients presenting with symptoms suggestive of osteomyelitis.SubjectsOut of eighteen enrolled patients, 15 had MR imaging with both ferumoxytol and gadolinium. Based on clinical and/or pathologic criteria, 7 patients were diagnosed with osteomyelitis, 5 patients had osteomyelitis ruled out, and in 3 patients a definitive diagnosis could not be made.Field strength1.5 Tesla.SequencesUsed included STIR, T1-weighted and T2-weighted spin echo.AssessmentThe mean contrast changes upon ferumoxytol and gadolinium administration were measured from lesion regions of interest and compared with control regions.Statistical testsStudent's t-test, propagation of errors. Data are reported as means ± S.E.ResultsThe mean contrast changes, ΔC, associated with a diagnosis of osteomyelitis were found to be ΔCFe = −2.7 ± 0.7 when Ferumoxytol and T2w imaging sequences were used and ΔCGd = +3.1 ± 1.1 (P < 0.001) when Gadolinium and a T1w imaging sequence was used. The MRI contrast changes for both agents correlated with systemic markers of inflammation, such as the erythrocyte sedimentation rate. In patients without osteomyelitis, no significant contrast changes were observed in T2-weighted, Ferumoxytol-contrasted MRI. The macrophages in osteomyelitic lesions were found to take up at least 16 times as much iron as benign bone marrow.Data conclusionWe conclude that in terms of its MRI diagnostic accuracy for osteomyelitis Ferumoxytol-contrasted MRI is a promising approach for diagnosing osteomyelitis that merits further study.     

Bayesian image processing in magnetic resonance imaging.

In the past several years, image processing techniques based on Bayesian models have received considerable attention. In our earlier work, we developed a novel Bayesian approach which was primarily aimed at the processing and reconstruction of images in positron emission tomography. In this paper, we describe how the technique has been adopted to process magnetic resonance images in order to reduce noise and artifacts, thereby improving image quality. In this framework, the image is assumed to be a statistical variable whose posterior probability density conditional on the observed image is modeled by the product of the likelihood function of the observed data with a prior density based our prior knowledge. A Gibbs random field incorporating local continuity information and with edge-detection capability is used as the prior model. Based on the formalism of the posterior density, we can compute an estimate of the image using an iterative technique. We have implemented this technique and applied it to phantom and clinical images. Our results indicate that the approach works reasonably well for reducing noise, enhancing edges, and removing ringing artifact.     

Elementary single turn solenoids used as the transmitter and receiver in magnetic resonance imaging

A single turn solenoid, also called a loop-gap resonator, is a device that is efficient for radio frequency spectroscopy on relatively large samples. Thus, the device provides an effective means for magnetic imaging where the single turn solenoid may serve both as the transmitter and receiver coil. The device is readily constructed and provides very efficient use of radio frequency (RF) power for imaging extremities such as breasts, arms, feet, and hands. The resulting magnetic images are acquired in short times with good anatomical resolution and considerable reduction of the RF power delivered to the patient.     

SNR-optimality of sum-of-squares reconstruction for phased-array magnetic resonance imaging

We consider the commonly used "Sum-of-Squares" (SoS) reconstruction method for phased-array magnetic resonance imaging with unknown coil sensitivities. We show that the signal-to-noise ratio (SNR) in the image produced by SoS is asymptotically (as the input SNR--> infinity ) equal to that of maximum-ratio combining, which is the best unbiased reconstruction method when the coil sensitivities are known. Finally, we discuss the implications of this result.