基于深度递归级联卷积神经网络的并行磁共振成像方法

快速磁共振成像是磁共振研究领域重要的课题之一.随着大数据和深度学习的兴起,神经网络成为快速磁共振技术的重要方法.然而网络性能表现和网络参数量之间较难取得平衡,且对于多通道数据重建的并行成像问题,相关研究较少.本文构建了一种深度递归级联卷积神经网络结构,用于处理并行成像问题.这种网络结构在减少网络参数量的同时,能够尽可能地提高网络的表达能力,提高网络重建的精确度.实验结果表明,相较于传统并行成像方法,通过训练好的神经网络对欠采样磁共振数据进行重建,可以得到更准确的重建结果,且重建时间大大缩短.     

基于正则化迭代的并行磁共振图像重建算法

为了解决并行磁共振成像过程的病态性和图像信噪比下降问题,降低重建过程中噪声放大和异常值的干扰造成的图像信噪比的损失,提出了一种基于正则化共轭梯度迭代的并行磁共振成像重建算法;该算法基于最小二乘理论,引入正则化,优化方程,进而进行迭代重建;采用了不同加速因子的人脑磁共振K空间欠采样数据以验证该算法的重建性能,仿真结果表明了该算法相较于最小二乘法,能较大限度地降低噪声对重建结果的干扰,具有信噪比更高、误差更小、成像效果更好等特征;重建图像质量得到了较好的改善,对临床诊断更具有适用性。     

基于深层残差网络的加速图像超分辨率重建

针对目前卷积神经网络的超分辨率算法存在卷积层数少、模型简单、计算量大、收敛速度慢以及图像纹理模糊等问题,提出了一种基于深层残差网络的加速图像超分辨率重建方法,该方法在提高图像分辨率的同时加快收敛速度。设计更深的卷积神经网络模型来提高精确度,通过残差学习并且使用Adam优化方法使网络模型加速收敛。在原始低分辨率图像上直接进行特征映射,只在网络的末端引入子像素卷积层,将像素进行重新排列,得到高分辨率图像。实验结果表明,在set 5,set 14,BSD100测试集上,所提算法的峰值信噪比与结构相似性指数均高于现有的几种算法,能够恢复更多的图像细节,图像边缘也更加完整且收敛速度更快。     

基于深度神经网络的时空编码磁共振成像超分辨率重建方法

单扫描时空编码磁共振成像是一种新型超快速磁共振成像技术,它对磁场不均匀和化学位移伪影有较强的抵抗性,但是其固有的空间分辨率较低,因此通常需要进行超分辨率重建,以在不增加采样点数的情况下提高时空编码磁共振图像的空间分辨率.然而,现有的重建方法存在迭代求解时间长、重建结果有混叠伪影残留等问题.为此,本文提出了一种基于深度神经网络的单扫描时空编码磁共振成像超分辨率重建方法.该方法采用模拟样本训练深度神经网络,再利用训练好的网络模型对实际采样信号进行重建.数值模拟、水模和活体鼠脑的实验结果表明,该方法能快速重建出无残留混叠伪影、纹理信息清楚的超分辨率时空编码磁共振图像.适当增加训练样本数量以及在训练样本中加入适当的随机噪声水平,有助于改善重建效果.     

宽带电磁图像卷积神经网络盲恢复方法研究

在利用抛物反射面对电磁干扰源成像过程中,由于系统衍射受限及成像频带较宽,导致干扰源成像模糊,分辨率低,难以分辨,不同频率不同区域干扰源所成图像分辨率不同,采用已有超分辨算法难以提高分辨率。为了实现宽带电磁图像的盲复原, 应用卷积神经网络的方法。网络训练是直接输入模糊图像,不假设任何特定的模糊和噪声模型情况下,重建出高质量图像。实验和仿真结果证明了卷积神经网络盲恢复方法在宽频带不同成像区域下表现了优于其他盲恢复算法的优势。     

多局部残差连接注意网络的图像去模糊

针对现有的基于卷积神经网络的图像去模糊算法存在图像纹理细节恢复不清晰的问题,提出了一种基于多局部残差连接注意网络的图像去模糊算法。首先,采用一个卷积层进行浅层特征提取;其次,设计了一种新的基于残差连接和并行注意机制的多局部残差连接注意模块,用于消除图像模糊并提取上下文信息;再次,采用一个基于扩张卷积的成对连接模块进行细节恢复;最后,利用一个卷积层重建清晰图像。实验结果表明：在GoPro数据集上的PSNR (peak signal to noise ratio)和SSIM (structure similarity)分别为31.83 dB、0.927 5,在定性和定量两方面都表明所提方法能够有效地恢复模糊图像的纹理细节,网络性能优于对比方法。     

基于去模糊空间变换RCNN的毫米波图像目标检测（英文）

提出一种包含去模糊的空间变换区域卷积神经网络的目标检测算法.首先,基于主动毫米波圆柱扫描成像原理对人体进行三维成像(频率24~30 GHz),建立毫米波图像数据集.然后,估计毫米波图像的模糊核,通过卷积去噪网络获得图像先验知识,将其集成到半二次分裂的优化方法中,以实现非盲目去模糊.最后,由定位网络、网格生成器和采样网络三部分组成空间变换网络,将它融入到特征提取网络中,在去模糊后实现目标检测.通过该非盲目去模糊算法得到的图像的峰值信噪比可达27.49 dB,目标检测算法的平均精度可达80.9%.实验结果表明,与现有的先进方法相比,该方法可以有效地提高图像质量和检测精度,为毫米波图像中隐藏危险品的目标检测提供了新的技术支持.     

基于去模糊空间变换RCNN的毫米波图像目标检测（英文）

提出一种包含去模糊的空间变换区域卷积神经网络的目标检测算法.首先,基于主动毫米波圆柱扫描成像原理对人体进行三维成像(频率24～30GHz),建立毫米波图像数据集.然后,估计毫米波图像的模糊核,通过卷积去噪网络获得图像先验知识,将其集成到半二次分裂的优化方法中,以实现非盲目去模糊.最后,由定位网络、网格生成器和采样网络三部分组成空间变换网络,将它融入到特征提取网络中,在去模糊后实现目标检测.通过该非盲目去模糊算法得到的图像的峰值信噪比可达27.49dB,目标检测算法的平均精度可达80.9%.实验结果表明,与现有的先进方法相比,该方法可以有效地提高图像质量和检测精度,为毫米波图像中隐藏危险品的目标检测提供了新的技术支持     

一种基于时空变换和压缩感知的磁共振螺旋采样的图像重建算法

螺旋采样磁共振快速成像在功能性成像、并行成像和动态成像等领域发挥着越来越重要的作用.螺旋采样图像重建的传统算法是用核函数将螺旋状分布的k空间数据插值到均匀网格中,再利用傅里叶变换和最小二乘法进行重建.但是基于网格化的算法对核函数过于依赖,在网格化过程中产生难以避免的误差.该文提出了基于时空变换和压缩感知的l1范数的最优化模型和重建算法.时空变换矩阵描述了空间上的磁共振图像与采集到的时域信号间的关系,使得算法直接使用采集到的数据作为保真约束项,避免了网格化过程产生的误差.此外,基于图像处理单元的并行计算被用来提高时空变换矩阵的运算速度,使得算法具有较强的应用价值.     

基于深度学习的粒子场数字全息成像研究进展

粒子场的数字全息成像中，由一幅粒子场全息图重建出高精度的三维粒子场分布，是数字全息技术领域的经典问题之一。相比于传统反向重建算法，深度学习算法可以从单个全息图直接重建出三维粒子场来简化算法复杂度，提高计算效率和准确率。介绍国内外研究团队将深度学习算法结合数字全息技术实现粒子场数字全息成像的研究进展，从不同粒子表征方法入手，叙述了支持向量机、全连接神经网络、全卷积网络、U-Net网络、深度神经网络在粒子场数字全息成像中粒子表征及粒子场反向重建过程中的应用原理、实现途径和准确率。最后指出了深度学习算法在这一研究领域的优势及目前基于深度学习算法的不足，并对如何进一步提高该方法的准确率进行了展望。     

DeepcomplexMRI: Exploiting deep residual network for fast parallel MR imaging with complex convolution

This paper proposes a multi-channel image reconstruction method, named DeepcomplexMRI, to accelerate parallel MR imaging with residual complex convolutional neural network. Different from most existing works which rely on the utilization of the coil sensitivities or prior information of predefined transforms, DeepcomplexMRI takes advantage of the availability of a large number of existing multi-channel groudtruth images and uses them as target data to train the deep residual convolutional neural network offline. In particular, a complex convolutional network is proposed to take into account the correlation between the real and imaginary parts of MR images. In addition, the k-space data consistency is further enforced repeatedly in between layers of the network. The evaluations on in vivo datasets show that the proposed method has the capability to recover the desired multi-channel images. Its comparison with state-of-the-art methods also demonstrates that the proposed method can reconstruct the desired MR images more accurately.     

用于SAR估计的基于U-Net网络的快速膝关节模型重建

膝关节高场磁共振成像（MRI）时，射频功率沉积（SAR）是一个关键的安全指标.目前对于局部SAR的准确估计只能通过电磁仿真实现，这就要求得到每一个个体的膝关节模型.本文提出一种针对低场磁共振图像的基于卷积神经网络的分割方法，以实现膝关节磁共振图像的快速重建.数据集来自于矢位T₁加权自旋回波图像，将膝关节组织按照"肌肉-脂肪-骨骼"模型进行简化，除脂肪与骨骼之外的其他组织归类为肌肉.采用一种全卷积的神经网络，即U-Net进行逐层的图像分割，卷积层数为4，训练采用交叉熵函数.本文对图像的自动分割结果与手动标注结果进行了定量的比较.此外，采用3 T正交鸟笼线圈进行了SAR仿真，结果验证了组织简化对于SAR估计的可行性，并且所提方法构建的模型可以得到较为精准的局部SAR分布.     

Transfer learning in deep neural network based under-sampled MR image reconstruction

In Magnetic Resonance Imaging (MRI), the success of deep learning-based under-sampled MR image reconstruction depends on: (i) size of the training dataset, (ii) generalization capabilities of the trained neural network. Whenever there is a mismatch between the training and testing data, there is a need to retrain the neural network from scratch with thousands of MR images obtained using the same protocol. This may not be possible in MRI as it is costly and time consuming to acquire data. In this research, a transfer learning approach i.e. end-to-end fine tuning is proposed for U-Net to address the data scarcity and generalization problems of deep learning-based MR image reconstruction. First the generalization capabilities of a pre-trained U-Net (initially trained on the human brain images of 1.5 T scanner) are assessed for: (a) MR images acquired from MRI scanners of different magnetic field strengths, (b) MR images of different anatomies and (c) MR images under-sampled by different acceleration factors. Later, end-to-end fine tuning of the pre-trained U-Net is proposed for the reconstruction of the above-mentioned MR images (i.e. (a), (b) and (c)). The results show successful reconstructions obtained from the proposed method as reflected by the Structural SIMilarity index, Root Mean Square Error, Peak Signal-to-Noise Ratio and central line profile of the reconstructed images.     

Deep neural network inspired by iterative shrinkage-thresholding algorithm with data consistency (NISTAD) for fast Undersampled MRI reconstruction

With the aim of developing a fast algorithm for high-quality MRI reconstruction from undersampled k-space data, we propose a novel deep neural Network, which is inspired by Iterative Shrinkage Thresholding Algorithm with Data consistency (NISTAD). NISTAD consists of three consecutive blocks: an encoding block, which models the flow graph of ISTA, a classical iteration-based compressed sensing magnetic resonance imaging (CS-MRI) method; a decoding block, which recovers the image from sparse representation; a data consistency block, which adaptively enforces consistency with the acquired raw data according to learned noise level. The ISTA method is thereby mapped to an end-to-end deep neural network, which greatly reduces the reconstruction time and simplifies the tuning of hyper-parameters, compared to conventional model-based CS-MRI methods. On the other hand, compared to general deep learning-based MRI reconstruction methods, the proposed method uses a simpler network architecture with fewer parameters. NISTAD has been validated on retrospectively undersampled diencephalon standard challenge data using different acceleration factors, and compared with DAGAN and Cascade CNN, two state-of-the-art deep neural network-based methods which outperformed many other state-of-the-art model-based and deep learning-based methods. Experimental results demonstrated that NISTAD reconstruction achieved comparable image quality with DAGAN and Cascade CNN reconstruction in terms of both PSNR and SSIM metrics, and subjective assessment, though with a simpler network structure.     

Joint Reconstruction of Multi-contrast Images and Multi-channel Coil Sensitivities

Magnetic resonance imaging (MRI) has an important feature that it provides multiple images with different contrasts for complementary diagnostic information. However, a large amount of data is needed for multi-contrast images depiction, and thus, the scan is time-consuming. Many methods based on parallel magnetic resonance imaging (pMRI) and compressed sensing (CS) are applied to accelerate multi-contrast MR imaging. Nevertheless, the image reconstructed by sophisticated pMRI methods contains residual aliasing artifact that degrades the quality of the image when the acceleration factor is high. Other methods based on CS always suffer the regularization parameter-selecting problem. To address these issues, a new method is presented for joint multi-contrast image reconstruction and coil sensitivity estimation. The coil sensitivities can be shared during the reconstruction due to the identity of coil sensitivity profiles of different contrast images for imaging stationary tissues. The proposed method uses the coil sensitivities as sharable information during the reconstruction to improve the reconstruction quality. As a result, the residual aliasing artifact can be effectively removed in the reconstructed multi-contrast images even if the acceleration factor is high. Besides, as there is no regularization term in the proposed method, the troublesome regularization parameter selection in the CS can also be avoided. Results from multi-contrast in vivo experiments demonstrated that multi-contrast images can be jointly reconstructed by the proposed method with effective removal of the residual aliasing artifact at a high acceleration factor.     

基于Gadgetron平台的多GPU分布式磁共振图像重建

为了满足磁共振成像（MRI）临床扫描的需求,磁共振图像重建算法的开发一直在不断进行.目前广泛使用的算法实现方式是利用中央处理器（CPU）对磁共振扫描数据进行数学变换得到图像,随着算法复杂度的提升,计算性能问题逐渐显露.利用CPU在大数据量下执行复杂算法时,计算并行性的缺失以及运算中产生的海量数据的存储负荷会导致计算变得极为缓慢,使得一些算法因为重建时间过长,在临床上面临难以推广的问题,也制约了基础研究中新算法的研发.本文设计并实现了一种新的重建算法执行方式,利用Gadgetron磁共振软件重建平台在多核CPU基础上搭载多块图形处理器（GPU）,将磁共振图像重建以分布式并行计算方式实现,并以重建耗时较长的3D径向数据采集Stack of Star（SOS）的图像重建为实例,展示这种重建的实现方法能以相对低廉的硬件成本极大提升重建的速度.     

Parallel variable-density spiral imaging using nonlocal total variation reconstruction

The relatively long scan time is still a bottleneck for both clinical applications and research of magnetic resonance imaging. To reduce the data acquisition time, we propose a novel fast magnetic resonance imaging method based on parallel variable-density spiral acquisition, which combines undersampling optimization and nonlocal total variation reconstruction.The undersampling optimization promotes the incoherence of resultant aliasing artifact via the "worst-case" residual error metric, and thus accelerates the data acquisition. Moreover, nonlocal total variation reconstruction is utilized to remove such an incoherent aliasing artifact and so improve image quality. The feasibility of the proposed method is demonstrated by both numerical phantom simulation and in vivo experiment. The experimental results show that the proposed method can achieve high acceleration factor and effectively remove an aliasing artifact from data undersampling with well-preserved image details. The image quality is better than that achieved with the total variation method.     

基于级联网络的膝关节图像分割与模型构建

采用膝关节模型进行电磁仿真是计算膝关节局部射频功率沉积（SAR）的主要方法,为了构建膝关节模型,本文提出了一种包含两个卷积神经网络——U-Net的级联网络结构,用于膝关节磁共振图像的分割.第一个网络在整幅图上分割肌肉、脂肪等占比较大的组织,并从分割结果中预测软骨与半月板的大致位置信息,第二个网络基于该信息在一个更小的子图上分割小组织以提高分割精度.两个网络均采用焦点损失函数,它们的分割结果合并在一起构成膝关节模型.我们将该方法与其它4种方法的分割结果进行了定量指标的对比研究,并分别构建膝关节模型,计算局部SAR值.结果表明本文提出的级联网络结构可以更精确的构建用于SAR仿真的膝关节模型.     

A novel acquisition-reconstruction algorithm for surface magnetic resonance imaging

In U-shaped, hand-size magnetic resonance surface scanners, imaging is performed along only one spatial direction, with the application of just one gradient (one-dimensional imaging). Lateral spatial resolution can be obtained by magnet displacement, but, in this case, resolution is very poor (on the order of some millimeters) and cannot be useful for high-resolution imaging applications. In this article, an innovative technique for acquisition and reconstruction of images produced by U-shaped, hand-size MRI surface scanners is presented. The proposed method is based on the acquisition of overlapping strips and an analytical reconstruction technique; it is capable of arbitrarily improving spatial lateral resolution without either using a second magnetic field gradient or making any assumptions about the imaged sample extension. Numerical simulations on synthetic images are reported demonstrating the method functionalities. The presented method also makes it possible to use U-shaped, hand-size MRI surface scanners for high-resolution biomedical applications, such as the imaging of skin lesions.