CN114119791A - MRI (magnetic resonance imaging) undersampled image reconstruction method based on cross-domain iterative network - Google Patents
MRI (magnetic resonance imaging) undersampled image reconstruction method based on cross-domain iterative networkDownload PDFInfo
- Publication number
- CN114119791A CN114119791ACN202010881742.5ACN202010881742ACN114119791ACN 114119791 ACN114119791 ACN 114119791ACN 202010881742 ACN202010881742 ACN 202010881742ACN 114119791 ACN114119791 ACN 114119791A
- Authority
- CN
- China
- Prior art keywords
- network
- data
- image
- space
- mri
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/41—Medical
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本发明涉及磁共振成像(Magnetic Resonance Imaging，MRI)、深度学习、欠采样重建等技术领域，具体涉及一种基于K空间域和图像域交叉迭代网络的MRI欠采样图像重建方法。The invention relates to the technical fields of magnetic resonance imaging (Magnetic Resonance Imaging, MRI), deep learning, undersampling reconstruction and the like, in particular to an MRI undersampling image reconstruction method based on a K-space domain and an image domain cross iterative network.
背景技术Background technique
MRI是一种非侵入、无电离辐射、任意切面方向、多参数影像学技术，软组织分辨率高，可以为临床诊断提供高分辨率的结构和功能信息。传统MRI需要在笛卡尔坐标系下采集2D或3D的数据矩阵，即k空间域数据，逆Fourier(傅里叶)变换后可以获得对应的图像域信息。k空间数据的每一行对应一步相位编码，不同相位编码数据的采集需要等待纵向磁化的恢复，即需要等待时间TR，根据不同的MRI成像方法，TR从数百毫秒到几秒不等，因此MRI是一种比较耗时的成像技术。常规MRI，理论前提是成像部位静止不动，否则会引入运动伪影，因此MRI检查过程中被检查者需要保持一个体位，坚持一到几分钟不动。腹部检查，需要多次屏气采集；心脏等具有自然生理运动的部位需要开发特殊的成像方法；同时MRI数据采集，病人需要处在一个半封闭的磁体空间内，这对大多病人长时间的配合是一种挑战，因此MRI迫切需要发展快速成像技术。MRI is a non-invasive, non-ionizing radiation, arbitrary section direction, multi-parameter imaging technology with high soft tissue resolution, which can provide high-resolution structural and functional information for clinical diagnosis. Traditional MRI needs to acquire a 2D or 3D data matrix in a Cartesian coordinate system, that is, k-space domain data, and corresponding image domain information can be obtained after inverse Fourier (Fourier) transformation. Each row of k-space data corresponds to one step of phase encoding. The acquisition of data with different phase encoding needs to wait for the recovery of longitudinal magnetization, that is, waiting time TR. According to different MRI imaging methods, TR ranges from hundreds of milliseconds to several seconds. Therefore, MRI It is a relatively time-consuming imaging technique. In conventional MRI, the theoretical premise is that the imaging site is still, otherwise motion artifacts will be introduced. Therefore, during the MRI examination, the examinee needs to maintain a position for one to several minutes. Abdominal examination requires multiple breath-holding acquisitions; parts with natural physiological motion such as the heart need to develop special imaging methods; meanwhile, for MRI data acquisition, the patient needs to be in a semi-enclosed magnet space, which is a long-term cooperation for most patients. A challenge, so MRI urgently needs to develop fast imaging techniques.
目前，结合多线圈单元并行成像的方法加快了成像速度，加速因子在2上下，但是这些方法依赖特殊的硬件或序列，且进一步提高采样速度的空间有限。基于压缩感知的MRI(Compressed Sensing-M,CS-MRI)是一种能够以远低于奈奎斯特采样定律下欠采样k空间数据从而加快成像速度的技术，无需特定硬件和序列。但是CS-MRI技术也存在一些限制，比如CS-MRI采样轨迹需要满足不相关性准则，典型的加速因子在2.5到3之间；CS-MRI在全局系数变换中利用到的稀疏系数较少，具有潜在隐藏图像细节结构甚至引入模糊伪影的风险，不能准确描述复杂的生物组织结构。At present, methods that combine parallel imaging with multi-coil units accelerate the imaging speed by a factor of around 2, but these methods rely on special hardware or sequences, and there is limited space for further improving the sampling speed. Compressed Sensing-M (CS-MRI) is a technique that can undersample k-space data much lower than Nyquist's sampling law to speed up imaging without specific hardware and sequences. However, CS-MRI technology also has some limitations. For example, the CS-MRI sampling trajectory needs to meet the irrelevance criterion, and the typical acceleration factor is between 2.5 and 3; CS-MRI uses fewer sparse coefficients in the global coefficient transformation, It has the risk of potentially hiding the detailed structure of the image or even introducing blurring artifacts, and cannot accurately describe the complex biological tissue structure.
卷积神经网络(CNN)把局部感受野、权值共享以及时间或空间亚采样这三种思想结合起来，可获得某种程度的位移、尺度、形变不变性，更类似于生物神经网络，降低了网络模型的复杂度，减少了权值的数量。近年来被用于MRI欠采样重建领域，通过训练网络获得欠采样图像和参考图像(全采样图像)之间的非线性关系，基本没有充分利用采集到的k空间数据信息，且加速因子较高时不能重建出图像的细节信息。Convolutional Neural Network (CNN) combines the three ideas of local receptive field, weight sharing, and temporal or spatial subsampling to obtain a certain degree of displacement, scale, and deformation invariance, which is more similar to biological neural network, reducing The complexity of the network model is reduced and the number of weights is reduced. In recent years, it has been used in the field of MRI undersampling reconstruction. The nonlinear relationship between the undersampling image and the reference image (full sampling image) is obtained by training the network. The acquired k-space data information is basically not fully utilized, and the acceleration factor is high. The details of the image cannot be reconstructed.
发明内容SUMMARY OF THE INVENTION
基于加速成像需求和现有的技术问题，本发明提出了一种基于k空间域和图像域交叉迭代网络的MRI欠采样图像重建方法。Based on the requirement of accelerated imaging and the existing technical problems, the present invention proposes an MRI undersampling image reconstruction method based on a k-space domain and an image domain cross iterative network.
本发明的技术方案是这样实现的：一种基于交叉域迭代网络的MRI欠采样图像重建方法，包括以下步骤：The technical scheme of the present invention is achieved as follows: an MRI undersampling image reconstruction method based on a cross-domain iterative network, comprising the following steps:
步骤1：回顾性的对全采样的k空间数据进行预处理，获得训练集和验证集数据；Step 1: Retrospectively preprocess the fully sampled k-space data to obtain training set and validation set data;
步骤2：构建交叉域迭代神经网络(KI网络)，KI网络由N个模块级联构成，每个模块由k空间域CNN网络Kcnn和图像域CNN网络Icnn组成，两个网络之间通过FT(傅里叶变换)或者逆fourier变换IFT(逆傅里叶变换)连接；预定义KI网络的损失函数loss；Step 2: Build a cross-domain iterative neural network (KI network). The KI network is composed of N modules cascaded. Each module is composed of a k-space domain CNN network Kcnn and an image domain CNN network Icnn. Fourier transform) or inverse fourier transform IFT (inverse Fourier transform) connection; predefined loss function loss of KI network;
步骤3：训练KI网络，将训练集输入KI网络进行训练，若损失函数loss最小，则作为训练好的中间网络；若损失函数loss不是最小，则反向传播更新网络参数，返回KI网络进行训练；Step 3: Train the KI network and input the training set into the KI network for training. If the loss function loss is the smallest, it will be used as a trained intermediate network; if the loss function loss is not the smallest, then backpropagation updates the network parameters and returns to the KI network for training ;
步骤4：检测KI网络，将验证集输入步骤3训练好的中间网络，获得重建图像，检测神经网络的泛化能力和过拟合情况，满足评判标准就输出该网络及其参数作为图像重建网络使用；不满足评判标准，就调整网络参数和层数、扩大训练集返回步骤2；Step 4: Detect the KI network, input the verification set into the intermediate network trained in step 3, obtain the reconstructed image, detect the generalization ability and overfitting of the neural network, and output the network and its parameters as the image reconstruction network if the criteria are met. Use; if the evaluation criteria are not met, adjust the network parameters and layers, expand the training set and return to step 2;
步骤5：重建k空间欠采样的图像，将欠采样的k空间数据输入步骤4获得的的图像重建网络，重建高质量的图像。Step 5: Reconstruct the k-space undersampled image, and input the undersampled k-space data into the image reconstruction network obtained in step 4 to reconstruct a high-quality image.
进一步地，步骤1中，对全采样的k空间数据进行随机欠采样，获得欠采样的k空间数据ku，对全采样的k空间数据进行逆fourier变换IFT，获得目标图像数据x，训练集由多组欠采样的k空间数据ku、全采样的k空间数据和目标图像数据x构成；验证集由多组欠采样的k空间数据ku和目标图像数据x构成。Further, in
进一步地，步骤2中，Kcnn和Icnn网络都采用残差神经网络，均由一个输入层、多个隐藏层和一个输出层构成，隐藏层的激活函数为线性整流函数(Rectified Linear Unit,ReLU)，输出层的激活函数为softmax函数。Further, in step 2, both Kcnn and Icnn networks use residual neural networks, which are composed of an input layer, multiple hidden layers and an output layer, and the activation function of the hidden layer is a linear rectification function (Rectified Linear Unit, ReLU) , the activation function of the output layer is the softmax function.
进一步地，Icnn网络隐藏层包含了通道注意力层CA(channel-wise attentionlayer)，用于调整各个通道的权重；Icnn网络输出前添加数据保真层DC层(Dataconsistency layer)。Further, the hidden layer of the Icnn network includes a channel-wise attention layer CA (channel-wise attention layer), which is used to adjust the weight of each channel; a data fidelity layer DC (Dataconsistency layer) is added before the output of the Icnn network.
进一步地，Kcnn网络的目标数据集是全采样的k空间数据，Icnn网络的目标数据是目标图像数据x；第一个模块Kcnn网络的输入层数据是训练集中的欠采样的k空间数据ku，将复数k空间数据分成实部和虚部分别输入网络；第n个模块中Kcnn网络的输入层数据是第n-1个模块中Icnn网路输出的图像数据xn-1经过FT后的数据FT(xn-1)和前边所有模块中Kcnn网络的输出的k空间数据kn-1，kn-2…k1。Further, the target data set of the Kcnn network is the fully sampled k-space data, and the target data of the Icnn network is the target image data x; the input layer data of the first module Kcnn network is the under-sampled k- space data ku in the training set. , divide the complex k-space data into real and imaginary parts and input them into the network respectively; the input layer data of the Kcnn network in the nth module is the image data xn-1 output by the Icnn network in the n-1th module after FT. Data FT(xn-1 ) and k-space data kn-1 , kn-2 . . . k1 of the output of the Kcnn network in all previous modules.
进一步地，第n个模块中Icnn网络的输入数据是第n个模块中Kcnn网路输出的k空间数据kn经过IFT后的数据IFT(kn)和前边所有模块中Icnn网络输出的图像数据xn-1，xn-2…x1。Further, the input data of the Icnn network in thenth module is the k-space data kn output by the Kcnn network in thenth module after IFT The data IFT(kn) and the image data output by the Icnn network in all previous modules xn-1 , xn-2 . . . x1 .
进一步地，步骤2中，预定义KI网络的损失函数loss为：Further, in step 2, the loss function loss of the predefined KI network is:
其中，in,
其中，Wk和WI是权重因子，K是训练集中全采样的k空间数据，
进一步地，步骤3中，将训练集中欠采样的k空间数据ku输入KI网络，将全采样的k空间数据和目标图像数据x作为目标数据训练KI网络。Further, in step 3, the undersampled k- space data ku in the training set is input into the KI network, and the fully-sampled k-space data and target image data x are used as target data to train the KI network.
进一步地，步骤4中，选择重建图像和目标图像之间的峰值信噪比(PSNR)和结构相似度(SSIM)作为评判依据，评判标准为PSNR不小于35dB且结构相似度不低于0.95。Further, in step 4, the peak signal-to-noise ratio (PSNR) and the structural similarity (SSIM) between the reconstructed image and the target image are selected as the evaluation basis, and the evaluation criteria are that the PSNR is not less than 35dB and the structural similarity is not less than 0.95.
本发明的有益效果：Beneficial effects of the present invention:
本发明提出了一种基于k空间域和图像域交叉迭代网络的MRI欠采样图像重建方法，准备全采样、欠采样的MRI k空间数据和图像数据集合，形成训练集和验证集数据；构建k空间域和图像域交叉的CNN稠密连接网络；将训练集数据输入网络进行训练；利用验证集测试网络的性能和泛化能力，确定网络模型和参数；将欠采样的k空间数据输入训练得到的图像重建网络重建图像。The invention proposes an MRI under-sampling image reconstruction method based on a k-space domain and an image domain cross iterative network, prepares fully-sampled and under-sampled MRI k-space data and image data sets to form training set and verification set data; constructs k The CNN densely connected network that intersects the spatial domain and the image domain; input the training set data into the network for training; use the validation set to test the performance and generalization ability of the network, and determine the network model and parameters; The image reconstruction network reconstructs the image.
网络模块迭代过程中采用稠密连接，可有效利用中间层信息；在构建k空间域网络时考虑到MRI k空间数据的分步布特征，采用残差网络；构建图像域网络时引入通道注意力机制，有效调整各个通道的权重，实现特征的优化提取，可有效恢复图像的细节信息；在损失函数中既考虑图像域损失又考虑了k空间域的损失，使用多监督训练，有效避免梯度回传时的监督信息丢失。本发明采用的k空间域残差网络可缓和k空间的数据强度分布不均匀的特性，且可抑制图像中的高频震荡伪影，本发明在消除因欠采样引入的高频震荡伪影的同时可恢复图像的细节信息，从而实现欠采样加速MRI采样速度的同时获得高质量的图像。In the iterative process of the network module, dense connections are used, which can effectively utilize the information of the intermediate layer; when constructing the k-space domain network, considering the distributed distribution characteristics of MRI k-space data, the residual network is adopted; when constructing the image domain network, the channel attention mechanism is introduced , effectively adjust the weight of each channel, realize the optimal extraction of features, and effectively restore the detailed information of the image; in the loss function, both the loss of the image domain and the loss of the k-space domain are considered, and the multi-supervised training is used to effectively avoid the gradient return. When the supervision information is lost. The k-space domain residual network used in the present invention can alleviate the uneven distribution of data intensity in k-space, and can suppress high-frequency oscillation artifacts in the image. At the same time, the detailed information of the image can be recovered, so that the under-sampling can accelerate the MRI sampling speed and obtain high-quality images.
附图说明Description of drawings
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.
图1为本发明的流程图；Fig. 1 is the flow chart of the present invention;
图2为交叉域迭代网络的结构图；Figure 2 is a structural diagram of a cross-domain iterative network;
图3为k空间域网络Kcnn结构图；Figure 3 is a structural diagram of the k-space domain network Kcnn;
图4为图像域网络Icnn结构图；Fig. 4 is an image domain network Icnn structure diagram;
图5为k空间欠采样模板和重建出的图像。Figure 5 shows the k-space undersampling template and the reconstructed image.
具体实施方式Detailed ways
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有付出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
一种基于交叉域迭代网络的MRI欠采样图像重建方法，流程如图1所示，具体步骤如下：An MRI undersampling image reconstruction method based on cross-domain iterative network, the process is shown in Figure 1, and the specific steps are as follows:
步骤1：对k空间数据进行预处理，获得训练集和验证集数据。Step 1: Preprocess the k-space data to obtain training set and validation set data.
对全采样的k空间数据进行随机欠采样，采样模板如图5所示，获得欠采样的k空间数据ku；对全采样的k空间数据进行逆fourier变换IFT，获得与其对应的图像x，简称为目标图像数据x，作为标记数据使用。Perform random under-sampling on the fully sampled k-space data, and the sampling template is shown in Figure 5 to obtain the under-sampled k- space data ku ; perform the inverse Fourier transform IFT on the fully-sampled k-space data to obtain the corresponding image x, It is abbreviated as target image data x, and is used as marker data.
训练集由多组欠采样的k空间数据、全采样的k空间数据及目标图像数据x构成。The training set consists of multiple sets of undersampled k-space data, fully sampled k-space data and target image data x.
验证集由多组欠采样的k空间数据和目标图像数据x构成。The validation set consists of multiple sets of undersampled k-space data and target image data x.
步骤2：构建交叉域迭代神经网络(KI网络)，如图2所示。Step 2: Construct a cross-domain iterative neural network (KI network), as shown in Figure 2.
1)KI网络由N个模块级联构成，每个模块由k空间域CNN网络Kcnn和图像域CNN网络Icnn组成，Kcnn网络与Icnn网络之间通过逆fourier变换IFT连接，Icnn网络与Kcnn网络之间通过fourier变换FT连接，每个模块中Kcnn网络与其前边的模块中Kcnn级联连接；每个模块中Icnn网络与其前边模块中的Icnn级联连接；1) KI network is composed of N modules cascaded, each module is composed of k-space domain CNN network Kcnn and image domain CNN network Icnn, Kcnn network and Icnn network are connected by inverse Fourier transform IFT, Icnn network and Kcnn network are connected. Through the Fourier transform FT connection, the Kcnn network in each module is connected in cascade with the Kcnn in the preceding module; the Icnn network in each module is connected in cascade with the Icnn in the preceding module;
2)如图3和图4所示，Kcnn和Icnn网络都是残差神经网络，均由一个输入层、多个隐藏层和一个输出层构成，隐藏层的激活函数为线性整流函数(Rectified Linear Unit,ReLU)，输出层的激活函数为softmax函数。2) As shown in Figure 3 and Figure 4, the Kcnn and Icnn networks are both residual neural networks, which are composed of an input layer, multiple hidden layers and an output layer. The activation function of the hidden layer is a linear rectification function (Rectified Linear function). Unit, ReLU), the activation function of the output layer is the softmax function.
3)如图4所示，Icnn网络隐藏层包含了通道注意力层(CA，channel-wiseattention layer)，用于调整各个通道的权重；Icnn网络输出前添加数据保真层DC层(Dataconsistency layer)。DC层计算如下：3) As shown in Figure 4, the hidden layer of the Icnn network includes a channel attention layer (CA, channel-wise attention layer), which is used to adjust the weight of each channel; the data fidelity layer DC layer (Dataconsistency layer) is added before the output of the Icnn network. . The DC layer is calculated as follows:
其中：
4)Kcnn、Icnn网络的目标数据集，即标记数据分别是全采样的k空间数据和其IFT变换后的图像数据x。4) The target datasets of Kcnn and Icnn networks, that is, the labeled data are the fully sampled k-space data and its IFT transformed image data x, respectively.
如图2和图3所示，第一个模块Kcnn网络的输入数据是训练集中的欠采样的k空间数据，将复数k空间数据分成实部和虚部分别输入网络；第n个模块中Kcnn网络的输入数据是第n-1个模块中Icnn网路输出的图像数据xn-1经过FT后的数据FT(xn-1)和前边所有模块中Kcnn网络的输出的k空间数据kn-1，kn-2…k1。As shown in Figure 2 and Figure 3, the input data of the first module Kcnn network is the undersampled k-space data in the training set, and the complex k-space data is divided into real and imaginary parts and input to the network respectively; The input data of the network is the image data xn-1 output by the Icnn network in the n-1th module, the data FT(xn-1 ) after FT and the k-space data kn of the output of the Kcnn network in all the previous modules.-1 , kn-2 ... k1 .
如图2和图3所示，第n个模块中Icnn网络的输入数据是第n个模块中Kcnn网路输出的图像数据kn经过IFT后的数据IFT(xn)和前边所有模块中Icnn网络输出的图像数据xn-1，xn-2…x1。As shown in Figure 2 and Figure 3, the input data of the Icnn network in the nth module is the image data kn output by the Kcnn network in thenth module after IFT. IFT(xn) andIcnn in all previous modules The image data xn-1 , xn-2 ... x1 output by the network.
5)KI网络定义的损失函数loss如下：5) The loss function loss defined by the KI network is as follows:
其中，in,
其中，Wk取0.1，WI取0.99，K是训练集中全采样的k空间数据，
步骤3：训练KI网络。Step 3: Train the KI network.
如图1和图2所示，将训练集中欠采样的k空间数据ku输入KI网络，将全采样k空间和其经IFT变换后获得的图像数据x作为目标图像数据训练神经网络。As shown in Figures 1 and 2, the undersampled k- space data ku in the training set is input into the KI network, and the fully-sampled k-space and the image data x obtained after IFT transformation are used as the target image data to train the neural network.
步骤4：检测神经网络(KI网络)。Step 4: Detection Neural Network (KI Network).
如图1所示，将验证集的欠采样k空间数据和与其对应的目标图像数据x数据输入训练好的中间网络，获得重建图像。As shown in Figure 1, the undersampled k-space data of the validation set and the corresponding target image data x data are input into the trained intermediate network to obtain the reconstructed image.
选择重建图像和目标图像之间的峰值信噪比(PSNR)和结构相似度(SSIM)作为评判依据。The peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) between the reconstructed image and the target image were selected as the judging criteria.
评判标准为PSNR不小于35dB且结构相似度不低于0.95。The evaluation criteria are that the PSNR is not less than 35dB and the structural similarity is not less than 0.95.
满足评判标准就输出该网络及其参数作为图像重建网络使用；If the evaluation criteria are met, the network and its parameters are output as an image reconstruction network;
不满足评判标准，就调整网络参数和层数、扩大训练集返回第二步，从而确保该神经网络的泛化能力的和过拟合情况，扩大训练集指的是增大训练集数据组的数目。If the evaluation criteria are not met, adjust the network parameters and layers, expand the training set and return to the second step, so as to ensure the generalization ability and over-fitting of the neural network. Expanding the training set refers to increasing the size of the training set data set. number.
最终使用的网络模块数为6。The final number of network modules used is 6.
步骤5：重建k空间欠采样的图像。Step 5: Reconstruct the k-space undersampled image.
将欠采样的k空间数据输入上述图像重建网络，重建出高质量的图像，如图5所示，分别是k空间为欠采样的模板Mask和重建出的图像。Input the undersampled k-space data into the above image reconstruction network to reconstruct high-quality images, as shown in Figure 5, which are the undersampled template Mask and the reconstructed image in k-space.
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the present invention. within the scope of protection.
Claims (9)
Translated fromChinese
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010881742.5ACN114119791B (en) | 2020-08-28 | 2020-08-28 | A cross-domain iterative network-based MRI undersampling image reconstruction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010881742.5ACN114119791B (en) | 2020-08-28 | 2020-08-28 | A cross-domain iterative network-based MRI undersampling image reconstruction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114119791Atrue CN114119791A (en) | 2022-03-01 |
| CN114119791B CN114119791B (en) | 2025-02-25 |
Family
ID=80374862
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010881742.5AActiveCN114119791B (en) | 2020-08-28 | 2020-08-28 | A cross-domain iterative network-based MRI undersampling image reconstruction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114119791B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114693823A (en)* | 2022-03-09 | 2022-07-01 | 天津大学 | A method of MR image reconstruction based on space-frequency dual-domain parallel reconstruction |
| US11751825B2 (en) | 2019-03-27 | 2023-09-12 | Siemens Healthcare Gmbh | Devices, systems, and methods for controlling acquisition parameters when carrying out a medical x-ray examination |
| CN117576250A (en)* | 2024-01-19 | 2024-02-20 | 中国科学技术大学先进技术研究院 | Rapid reconstruction method and system for prospective undersampled MRI Dixon data |
| WO2024052814A1 (en)* | 2022-09-05 | 2024-03-14 | B. G. Negev Technologies And Applications Ltd., At Ben-Gurion University | System and method for compressed sensing using partial ensemble measurements |
| CN119762608A (en)* | 2024-12-11 | 2025-04-04 | 中国科学院精密测量科学与技术创新研究院 | Human body 5.0T multi-core31P magnetic resonance spectrum imaging self-supervision reconstruction method |
| US12417535B2 (en) | 2022-03-25 | 2025-09-16 | Siemens Healthineers Ag | Trained function for providing magnetic field data, and the application of the trained function |
| CN119762608B (en)* | 2024-12-11 | 2025-10-10 | 中国科学院精密测量科学与技术创新研究院 | Self-supervised reconstruction method for 5.0T multi-nuclear 31P magnetic resonance spectroscopy imaging of human body |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101777720B1 (en)* | 2016-05-24 | 2017-09-12 | 가천대학교 산학협력단 | Image processing apparatus for mri system and method thereof |
| CN108535675A (en)* | 2018-04-08 | 2018-09-14 | 朱高杰 | A kind of magnetic resonance multichannel method for reconstructing being in harmony certainly based on deep learning and data |
| CN110151181A (en)* | 2019-04-16 | 2019-08-23 | 杭州电子科技大学 | A fast magnetic resonance imaging method based on recursive residual U-shaped network |
| CN110288672A (en)* | 2019-06-28 | 2019-09-27 | 闽江学院 | A kind of compressed sensing MR image reconstruction method based on the dense network of ultra-deep |
| US20200033431A1 (en)* | 2018-07-30 | 2020-01-30 | Jo Schlemper | Deep learning techniques for magnetic resonance image reconstruction |
| US20200065969A1 (en)* | 2018-08-27 | 2020-02-27 | Siemens Healthcare Gmbh | Medical image segmentation from raw data using a deep attention neural network |
| CN111436936A (en)* | 2020-04-29 | 2020-07-24 | 浙江大学 | CT image reconstruction method based on MRI |
- 2020
- 2020-08-28CNCN202010881742.5Apatent/CN114119791B/enactiveActive
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101777720B1 (en)* | 2016-05-24 | 2017-09-12 | 가천대학교 산학협력단 | Image processing apparatus for mri system and method thereof |
| CN108535675A (en)* | 2018-04-08 | 2018-09-14 | 朱高杰 | A kind of magnetic resonance multichannel method for reconstructing being in harmony certainly based on deep learning and data |
| US20200033431A1 (en)* | 2018-07-30 | 2020-01-30 | Jo Schlemper | Deep learning techniques for magnetic resonance image reconstruction |
| WO2020028257A2 (en)* | 2018-07-30 | 2020-02-06 | Hyperfine Research, Inc. | Deep learning techniques for magnetic resonance image reconstruction |
| US20200065969A1 (en)* | 2018-08-27 | 2020-02-27 | Siemens Healthcare Gmbh | Medical image segmentation from raw data using a deep attention neural network |
| CN110151181A (en)* | 2019-04-16 | 2019-08-23 | 杭州电子科技大学 | A fast magnetic resonance imaging method based on recursive residual U-shaped network |
| CN110288672A (en)* | 2019-06-28 | 2019-09-27 | 闽江学院 | A kind of compressed sensing MR image reconstruction method based on the dense network of ultra-deep |
| CN111436936A (en)* | 2020-04-29 | 2020-07-24 | 浙江大学 | CT image reconstruction method based on MRI |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11751825B2 (en) | 2019-03-27 | 2023-09-12 | Siemens Healthcare Gmbh | Devices, systems, and methods for controlling acquisition parameters when carrying out a medical x-ray examination |
| CN114693823A (en)* | 2022-03-09 | 2022-07-01 | 天津大学 | A method of MR image reconstruction based on space-frequency dual-domain parallel reconstruction |
| CN114693823B (en)* | 2022-03-09 | 2024-06-04 | 天津大学 | Magnetic resonance image reconstruction method based on space-frequency double-domain parallel reconstruction |
| US12417535B2 (en) | 2022-03-25 | 2025-09-16 | Siemens Healthineers Ag | Trained function for providing magnetic field data, and the application of the trained function |
| WO2024052814A1 (en)* | 2022-09-05 | 2024-03-14 | B. G. Negev Technologies And Applications Ltd., At Ben-Gurion University | System and method for compressed sensing using partial ensemble measurements |
| CN117576250A (en)* | 2024-01-19 | 2024-02-20 | 中国科学技术大学先进技术研究院 | Rapid reconstruction method and system for prospective undersampled MRI Dixon data |
| CN117576250B (en)* | 2024-01-19 | 2024-04-02 | 中国科学技术大学先进技术研究院 | Rapid reconstruction method and system for prospective undersampled MRI Dixon data |
| CN119762608A (en)* | 2024-12-11 | 2025-04-04 | 中国科学院精密测量科学与技术创新研究院 | Human body 5.0T multi-core31P magnetic resonance spectrum imaging self-supervision reconstruction method |
| CN119762608B (en)* | 2024-12-11 | 2025-10-10 | 中国科学院精密测量科学与技术创新研究院 | Self-supervised reconstruction method for 5.0T multi-nuclear 31P magnetic resonance spectroscopy imaging of human body |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114119791B (en) | 2025-02-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114119791A (en) | MRI (magnetic resonance imaging) undersampled image reconstruction method based on cross-domain iterative network | |
| CN108335339B (en) | Magnetic resonance reconstruction method based on deep learning and convex set projection | |
| Sandino et al. | Compressed sensing: From research to clinical practice with deep neural networks: Shortening scan times for magnetic resonance imaging | |
| US11170543B2 (en) | MRI image reconstruction from undersampled data using adversarially trained generative neural network | |
| CN108535675A (en) | A kind of magnetic resonance multichannel method for reconstructing being in harmony certainly based on deep learning and data | |
| CN101975936A (en) | Rapid magnetic resonance imaging (MRI) method based on CS ( compressed sensing ) technique | |
| US20240290011A1 (en) | Dual-domain self-supervised learning for accelerated non-cartesian magnetic resonance imaging reconstruction | |
| CN116597037B (en) | Physical generation data-driven rapid magnetic resonance intelligent imaging method | |
| CN104434108B (en) | Multi-layer magnetic resonance fast imaging method based on adjacent layer information and undersampling | |
| CN103323805A (en) | Method for SPEED rapid magnetic resonance imaging based on wavelet domain sparse representation | |
| CN113487507A (en) | Dual-domain recursive network MR reconstruction method based on multi-module feature aggregation | |
| CN104013403A (en) | Three-dimensional heart magnetic resonance imaging method based on tensor composition sparse bound | |
| Pezzotti et al. | An adaptive intelligence algorithm for undersampled knee mri reconstruction: Application to the 2019 fastmri challenge | |
| CN120279134A (en) | Non-Cartesian magnetic resonance rapid intelligent imaging method for k-space direct learning | |
| He et al. | Low-rank and framelet based sparsity decomposition for interventional MRI reconstruction | |
| CN115471580B (en) | Physical intelligent high-definition magnetic resonance diffusion imaging method | |
| Lv et al. | Parallel imaging with a combination of sensitivity encoding and generative adversarial networks | |
| CN113920211B (en) | Quick magnetic sensitivity weighted imaging method based on deep learning | |
| Shangguan et al. | Multi-slice compressed sensing MRI reconstruction based on deep fusion connection network | |
| Li et al. | Deep residual network for highly accelerated fMRI reconstruction using variable density spiral trajectory | |
| CN113866694B (en) | Rapid three-dimensional magnetic resonance T1 quantitative imaging method, system and medium | |
| CN119379832A (en) | Anatomical structure-aware low-field multi-contrast fast MRI joint reconstruction method | |
| CN106093814A (en) | A kind of cardiac magnetic resonance imaging method based on multiple dimensioned low-rank model | |
| CN116725515B (en) | Magnetic resonance rapid imaging method | |
| CN112489150A (en) | Deep neural network multi-scale sequential training method for rapid MRI |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |