WO2022027327A1 - 一种图像重建方法及其应用 - Google Patents

一种图像重建方法及其应用 Download PDF

Info

Publication number
WO2022027327A1
WO2022027327A1 PCT/CN2020/107143 CN2020107143W WO2022027327A1 WO 2022027327 A1 WO2022027327 A1 WO 2022027327A1 CN 2020107143 W CN2020107143 W CN 2020107143W WO 2022027327 A1 WO2022027327 A1 WO 2022027327A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
sequence
feature vector
dynamic
pixel
Prior art date
Application number
PCT/CN2020/107143
Other languages
English (en)
French (fr)
Inventor
李彦明
郑海荣
江洪伟
万丽雯
Original Assignee
深圳高性能医疗器械国家研究院有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 深圳高性能医疗器械国家研究院有限公司 filed Critical 深圳高性能医疗器械国家研究院有限公司
Priority to PCT/CN2020/107143 priority Critical patent/WO2022027327A1/zh
Publication of WO2022027327A1 publication Critical patent/WO2022027327A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation

Definitions

  • the present application belongs to the technical field of image scanning, and in particular relates to an image reconstruction method and application thereof.
  • Image reconstruction is a technology that obtains shape information of three-dimensional objects through digital processing through data measured outside the object.
  • Image reconstruction technology was initially used in radiological medical equipment to display images of various parts of the human body, that is, computed tomography technology, or CT technology for short, and it was gradually applied in many fields. There are mainly projection reconstruction, shading recovery shape, stereo vision reconstruction and laser ranging reconstruction.
  • Dynamic myocardial perfusion CT imaging is an effective method for clinical diagnosis of coronary artery disease.
  • Existing clinical DMP-CT imaging requires repeated CT scans of the patient after the contrast-enhanced imaging agent reaches each tissue of the patient's heart to obtain dynamic time evolution information of the imaging agent.
  • Image reconstruction generally uses a commercial filtered back projection algorithm (FBP) to reconstruct the image of each frame separately.
  • FBP filtered back projection algorithm
  • the existing image reconstruction requires repeated CT scans of the same dose for the patient, which will bring a large dose of X-ray radiation to the patient, and there are potential Danger of causing radiation lesions in the scanned site.
  • the present application provides an image reconstruction method and its application.
  • the present application provides an image reconstruction method, the method includes the following steps:
  • the baseline image sequence is a scanning image performed before the imaging agent reaches the human tissue
  • the dynamic perfusion image sequence is a continuous image sequence performed after the imaging agent reaches the human tissue. Scan the image.
  • step 1) the dynamic perfusion image is obtained by scanning with a sparse down-sampling scanning scheme; the baseline image is obtained by scanning with a non-sparse full-sampling scanning scheme.
  • Another embodiment provided by the present application is: in the step 3), the maximum likelihood expectation maximization algorithm of total variation regularization is used to perform pre-reconstruction.
  • the pixel feature vector is related to the time evolution of the imaging agent.
  • the method for extracting the pixel feature vector in the step 4) is: using the corresponding pixel value of each frame image in the pre-reconstructed contrast-enhanced image sequence to The pixel feature vector that makes up the image pixels.
  • constructing a priori matrix of time series information includes: for each pixel feature vector feature set in pixels Determine k similar pixel feature vectors associated with it; The k-pixel eigenvectors in the k-nearest neighbor clusters are the same as The connection is established through a pre-designed Gaussian connection function, not in The k-nearest neighbor eigenvectors of other pixels in the cluster are the same as The connection function value of is defined as 0; will be based on the pixel feature vector All computed connection function values ⁇ ((i,j),(p,q)) are rearranged into a column vector of length 512 ⁇ 512.
  • the method for determining the feature vector of similar pixels is: calculating the feature vector of each pixel with the pixel feature vector under consideration Euclidean distance between , and The nearest k pixel feature vector of Euclidean distance is composed of The k-nearest neighbor clusters of .
  • the baseline image in step 7) is a normal dose baseline image reconstructed from the fully sampled baseline projection data through a filtered back-projection algorithm.
  • the present application also provides an application of an image reconstruction method, which is applied to dynamic myocardial perfusion scanning CT images, brain dynamic perfusion CT imaging and magnetic resonance perfusion imaging.
  • the image reconstruction method provided in this application is aimed at medical images.
  • the image reconstruction method provided by the present application is a dynamic perfusion CT image reconstruction method assisted by time series information.
  • the image reconstruction method provided by the present application is a sparse low-dose dynamic CT image reconstruction method of myocardial perfusion (DMP-CT) assisted by a time series in which perfusion-enhanced image information and baseline image information are separated. .
  • DMP-CT myocardial perfusion
  • the image reconstruction method provided in the present application is used to accurately reconstruct the dynamic perfusion CT image obtained by the sparse low-dose scanning scheme, and the reconstructed image meets the diagnostic criteria of clinical coronary artery disease (CAD) under the condition of reducing the radiation dose of X-ray scanning.
  • CAD clinical coronary artery disease
  • the image reconstruction method provided in the present application solves the technical problem of reconstructing the myocardial perfusion dynamic CT image from the sparse projection angle data, and supports the clinical application of the myocardial perfusion dynamic CT scanning scheme with the sparse projection angle from the reconstruction algorithm level.
  • the image reconstruction method provided in the present application can obtain reconstructed images that meet clinical diagnostic standards while reducing the number of X-ray scans in the process of myocardial dynamic perfusion CT scanning, thereby significantly reducing the radiation dose of myocardial dynamic perfusion CT imaging to patients .
  • the image reconstruction method provided by the present application in terms of algorithm performance, the algorithm of the present application can effectively avoid the problem of reconstructed image artifacts and incorrect information caused by sparse sampling, and in the scene of sparse angle CT, the imaging quality is significantly higher than the existing commercial algorithms. Rebuild the result.
  • 1 is a schematic diagram of the overall framework of the low-dose DMP-CT image reconstruction algorithm assisted by time series information of the present application;
  • FIG. 2 is a schematic diagram of the reconstruction result of the low-dose DMP-CT image assisted by the time series information of the present application.
  • X-raying a patient with a sparse angle is an effective way to reduce the radiation dose of CT scans, especially for DMP-CT, a medical imaging examination that requires multiple scans of a patient.
  • existing reconstruction algorithms filtered back-projection, traditional algebraic iterative reconstruction
  • reconstruct CT images from sparsely collected data will result in a large number of artifacts in the reconstructed images due to the incompleteness and inconsistency of the data. seriously affect the diagnosis.
  • CT imaging process can be mathematically described by a simple matrix equation:
  • y is the projection data obtained by CT scanning (after data calibration and logarithmic transformation)
  • A is the system matrix corresponding to the scanning system, which represents the orthographic projection process from the scanned object to the projection data
  • x is the scanned object (to be reconstructed). image).
  • the image reconstruction task of CT imaging is to estimate the X-ray attenuation coefficient value x of the scanned object based on the obtained projection data y and the known system matrix A corresponding to the imaging system.
  • DMP-CT dynamic myocardial perfusion CT imaging
  • X ⁇ x 1 , x 2 ,..., x T ⁇ is the image sequence to be reconstructed.
  • the DMP-CT imaging task can be understood as estimating the dynamic image time series X according to the projection data time series Y based on the imaging system matrix A.
  • the present application provides an image reconstruction method, the method includes the following steps:
  • the baseline image sequence is the scanning image performed before the imaging agent reaches the human tissue
  • the dynamic perfusion image sequence is the continuous scanning image performed after the imaging agent reaches the human tissue.
  • the dynamic perfusion image is obtained by scanning with a sparse down-sampling scanning scheme; the baseline image is obtained by scanning with a non-sparse full-sampling scanning scheme.
  • the maximum likelihood expectation maximization algorithm of total variation regularization is used for pre-reconstruction.
  • the pixel feature vector in the step 4) is related to the time evolution of the imaging agent.
  • the extraction method of the pixel feature vector described in the step 4) is: the pixel value of the corresponding pixel point of each frame of the image in the pre-reconstructed contrast-enhanced image sequence is used to form the pixel feature vector of the image pixel point. .
  • constructing a priori matrix of time series information in the step 5) includes: for each pixel feature vector feature set in pixels Determine k similar pixel feature vectors associated with it; The k-pixel eigenvectors in the k-nearest neighbor clusters are the same as The connection is established through a pre-designed Gaussian connection function, not in The k-nearest neighbor eigenvectors of other pixels in the cluster are the same as The connection function value of is defined as 0; will be based on the pixel feature vector All computed connection function values ⁇ ((i,j),(p,q)) are rearranged into a column vector of length 512 ⁇ 512.
  • the method for determining the feature vector of the similar pixels is: calculating the feature vector of each pixel with the pixel feature vector under consideration Euclidean distance between , and The nearest k pixel feature vector of Euclidean distance is composed of The k-nearest neighbor clusters of .
  • the baseline image in the step 7) is a normal dose baseline image reconstructed from the fully sampled baseline projection data through a filtered back-projection algorithm.
  • the present application also provides an application of an image reconstruction method, which is applied to dynamic myocardial perfusion scanning CT images, brain dynamic perfusion CT imaging and magnetic resonance perfusion imaging.
  • image reconstruction method which is applied to dynamic myocardial perfusion scanning CT images, brain dynamic perfusion CT imaging and magnetic resonance perfusion imaging.
  • this application is just an example, not limited to the reconstruction of other types of images.
  • the contrast-enhanced images of cardiac tissue caused by imaging agents are reconstructed with the help of the time evolution information of imaging agents naturally contained in dynamic image sequences.
  • the reconstructed enhanced image is superimposed on the baseline CT image obtained by the normal dose acquisition scheme to obtain the final dynamic myocardial perfusion scan CT image sequence.
  • Image data acquisition baseline image acquisition for myocardial dynamic perfusion CT imaging, that is, the CT scan performed before the imaging agent reaches the human heart tissue, and dynamic perfusion image sequence acquisition, that is, after the imaging agent reaches the human heart tissue.
  • Continuous CT scans with different doses of CT image acquisition schemes Baseline images were scanned using a normal-dose, ie, non-sparse, full-sampling scan scheme, and dynamic perfusion image sequences were scanned using a low-dose, ie, sparse, down-sampling scan scheme. This step can be described by the following formula:
  • Baseline projection data separation The image information corresponding to non-imaging agent introduction is separated from the acquired sparse dynamic projection data sequence.
  • the specific implementation method is: down-sampling the collected baseline projection data y baseline, full sampling , to ensure that the obtained sparse baseline projection data and dynamic perfusion projection data Y perfusion, each sparse frame corresponds to the same image acquisition geometric conditions (ie System Matrix A).
  • the downsampled sparse baseline projection data is subtracted from the dynamic perfusion projection data for each frame to obtain the enhanced projection data time series. This step can be described by the following formula:
  • contrast-enhanced projection data Y obtained from the above formula (4) is contrast- enhanced and sparsely pre-reconstructed using the Maximum Likelihood Expectation Maximization Algorithm (MLEM-TV) with total variation regularization, and the pre-reconstruction is obtained.
  • MLEM-TV Maximum Likelihood Expectation Maximization Algorithm
  • Time evolution feature extraction of imaging agent For each CT image pixel point (i, j), a corresponding pixel feature vector related to the time evolution of imaging agent is extracted.
  • the extraction method of the pixel feature vector is as follows: the pixel value x 1 ⁇ T (i,j) of the corresponding pixel point of each frame of the image in the pre-reconstructed contrast-enhanced image sequence obtained in the above step 3) is used to form the pixel point (i, j) ) pixel feature vector, which can be specifically described by the following formula:
  • T represents the frame number of the dynamic contrast-enhanced image.
  • Construction of a priori matrix of time series information based on the extracted feature vector of each pixel Construct a prior matrix of time series information.
  • the specific construction method is:
  • the method for determining the feature vector of similar pixels is: establish a feature vector about the pixel The k-nearest neighbor clusters, i.e. compute the feature vector for each pixel with the pixel feature vector under consideration Euclidean distance between , and The nearest k pixel feature vector of Euclidean distance is composed of The k-nearest neighbor clusters of .
  • Step b in The k-pixel eigenvectors in the k-nearest neighbor clusters are the same as The connection is established through a pre-designed Gaussian connection function, not in The k-nearest neighbor eigenvectors of other pixels in the cluster are the same as The connection function value of is defined as 0. Step b can be described by the following formula:
  • connection function values ⁇ ((i,j),(p,q)) are rearranged into a column vector of length 512 ⁇ 512. Therefore, for each pixel feature vector A column vector composed of the connection function value of the pixel eigenvector and all other pixel eigenvectors can be calculated, and all the column vectors form the time series information prior matrix ⁇ . This step can be described by the following formula:
  • the time series information prior matrix constructed in step 5) above is introduced into the reconstruction iteration of the contrast-enhanced image sequence in the form of an "image mask".
  • Contrast-enhanced CT images It is expressed as the product of an image mask and a coefficient image ⁇ , and the image mask is the time series prior matrix ⁇ constructed in the aforementioned step 5), namely
  • the contrast-enhanced image sequence X reconstructed according to the above steps is contrast-enhanced, and the reconstruction is added frame by frame to the normal dose baseline image reconstructed from the fully sampled baseline projection data through the filtered back projection algorithm (FBP). , namely obtain the reconstructed dynamic perfusion myocardial CT image time series X dynamic perfusion, and reconstruct .
  • FBP filtered back projection algorithm
  • Clinically used cardiac dynamic perfusion CT examination scan the patient with a normal dose before the imaging agent reaches the human tissue to achieve contrast enhancement effect, and perform a low-dose sparse CT dynamic scan on the patient after the imaging agent reaches the human tissue to achieve contrast enhancement .
  • the original image data obtained by this image acquisition scheme can be reconstructed using the method of the present application.
  • FIG. 2 shows the reconstruction result of the myocardial dynamic perfusion CT image reconstructed by the method of the present application.
  • the experiment was performed with a sparse magnification of 12 (984 projections/360° for normal dose full sampling and 82 projections/360° for sparse low dose sampling). It can be seen from the figure that the method of the present application has superior performance in the reconstruction of myocardial dynamic perfusion CT image with sparse angle sampling.
  • the high-low dose matching scanning scheme for myocardial dynamic perfusion CT scan that is, the normal dose scanning scheme is used before the imaging agent arrives and the sparse low-dose scanning scheme after the imaging agent arrives is switched.
  • the present application extracts a pixel feature vector from a pre-reconstructed contrast-enhanced image sequence, and constructs a time series prior matrix based on the pixel feature vector;
  • the present application transforms the CT image reconstruction task into a method for coefficient image reconstruction under image shading, and the application of the maximum likelihood expectation maximization algorithm in coefficient image reconstruction;

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

本申请属于图像扫描技术领域，特别是涉及一种图像重建方法及其应用。现有的图像重建想要重建出达到临床诊断标准的动态图像序列，就必须对病人进行重复多次相同剂量CT扫描，这会给病人带来很大的X‐射线辐射剂量，并存在潜在的引发被扫描部位辐射病变的危险。本申请基于动态灌注增强图像的分离重建方法和动态图像序列的稀疏角度图像采集方案，对由显像剂引起的心脏组织对比增强图像借助动态图像序列中自然包含的显像剂时间演变信息进行重建，并将重建增强图像叠加到正常剂量采集方案得到的基线图像，获得最终的动态灌注扫描图像序列。显著降低动态灌注成像对病人的辐射剂量。

Description

一种图像重建方法及其应用 技术领域
本申请属于图像扫描技术领域，特别是涉及一种图像重建方法及其应用。
背景技术
图像重建是通过物体外部测量的数据，经数字处理获得三维物体的形状信息的技术。图像重建技术开始是在放射医疗设备中应用，显示人体各部分的图像，即计算机断层摄影技术，简称CT技术，后逐渐在许多领域获得应用。主要有投影重建、明暗恢复形状、立体视觉重建和激光测距重建。
心肌灌注动态CT成像(DMP-CT)是临床上诊断冠状动脉疾病的有效检查方法。现有临床DMP-CT成像需要再对比增强显像剂到达病人心脏各组织后对病人进行重复多次的CT扫描以获取动态的显像剂时间演变信息。图像重建一般借助于商用滤波反投影算法(FBP)对每一帧的图像分别进行重建。
现有的图像重建想要重建出达到临床诊断标准的动态图像序列，就必须对病人进行重复多次相同剂量CT扫描，这会给病人带来很大的X-射线辐射剂量，并存在潜在的引发被扫描部位辐射病变的危险。
发明内容
1.要解决的技术问题
基于现有的图像重建想要重建出达到临床诊断标准的动态图像序列，就必须对病人进行重复多次相同剂量CT扫描，这会给病人带来很大的X-射线辐射剂量，并存在潜在的引发被扫描部位辐射病变的危险的问题，本申请提供了一种图像重建方法及其应用。
2.技术方案
为了达到上述的目的，本申请提供了一种图像重建方法，所述方法包括如下步骤：
1)采集动态灌注图像序列和基线图像序列；
2)对基线图像投影数据序列进行降采样，在每一帧动态灌注图像序列中减去降采样的基线图像投影数据序列，得到增强投影数据时间序列；
3)对所述增强投影数据时间序列进行迭代预重建，得到预重建对比增强图像序列；
4)对每一个图像像素点均提取一个与之对应的像素特征向量；
5)基于所述像素特征向量构建时间序列信息先验矩阵；
6)将所述时间序列信息先验矩阵以“图像遮罩”的形式引入对比增强图像序列的重建 迭代中，得到重建的对比增强图像；
7)将所述对比增强图像投影数据序列逐帧与所述基线图像投影数据相加，得到重建的动态灌注图像时间序列。
本申请提供的另一种实施方式为：所述步骤1)中基线图像序列为显像剂到达人体组织之前进行的扫描图像，所述动态灌注图像序列为显像剂到达人体组织之后进行的连续扫描图像。
本申请提供的另一种实施方式为：所述步骤1)中所述动态灌注图像采用稀疏的降采样扫描方案进行扫描得到；所述基线图像采用非稀疏的全采样扫描方案进行扫描得到。
本申请提供的另一种实施方式为：所述步骤3)中采用全变分正则化的最大似然期望最大化算法进行预重建。
本申请提供的另一种实施方式为：所述步骤4)中所述像素特征向量与显像剂时间演变相关。
本申请提供的另一种实施方式为：所述步骤4)中所述像素特征向量的提取方法为：将所述预重建对比增强图像序列中的每一帧图像的对应像素点像素值用于组成图像像素点的像素特征向量。
本申请提供的另一种实施方式为：所述步骤5)中构建时间序列信息先验矩阵包括：对于每一个像素特征向量
Figure PCTCN2020107143-appb-000001
在像素特征集合
Figure PCTCN2020107143-appb-000002
中确定k个与之建立联系的相近像素特征向量；对处在
Figure PCTCN2020107143-appb-000003
的k-最邻近簇中的k个像素特征向量与
Figure PCTCN2020107143-appb-000004
通过一个预先设计的高斯连接函数建立联系，不处在
Figure PCTCN2020107143-appb-000005
的k-最邻近簇中的其他像素特征向量与
Figure PCTCN2020107143-appb-000006
的连接函数值定义为0；将基于像素特征向量
Figure PCTCN2020107143-appb-000007
计算的所有连接函数值ψ((i,j),(p,q))重排为一个长度为512×512的列向量。
本申请提供的另一种实施方式为：相近像素特征向量的确定方法为：计算每一个像素特征向量
Figure PCTCN2020107143-appb-000008
与被考虑的像素特征向量
Figure PCTCN2020107143-appb-000009
之间的欧几里得距离，与
Figure PCTCN2020107143-appb-000010
的欧几里得距离最近的k个像素特征向量即构成
Figure PCTCN2020107143-appb-000011
的k-最邻近簇。
本申请提供的另一种实施方式为：所述步骤7)中基线图像为从全采样基线投影数据经过滤波反投影算法重建的正常剂量基线图像。
本申请还提供一种图像重建方法的应用，将所述的图像重建方法应用于动态心肌灌注扫描CT图像、脑部动态灌注CT成像与磁共振灌注成像。
3.有益效果
与现有技术相比，本申请提供的一种图像重建方法的有益效果在于：
本申请提供的图像重建方法，针对医学图像。
本申请提供的图像重建方法，为一种时间序列信息协助的动态灌注CT图像重建方法。
本申请提供的图像重建方法，为一种基于灌注增强图像信息与基线图像信息分离的时间序列协助下的稀疏低剂量心肌灌注动态CT图像(DMP-CT)重建方法。。
本申请提供的图像重建方法，用于精确重建通过稀疏低剂量扫描方案获取的动态灌注CT图像，在降低X射线扫描辐射剂量的条件下，重建图像达到临床冠状动脉疾病(CAD)诊断标准。
本申请提供的图像重建方法，解决从稀疏投影角度数据重建心肌灌注动态CT图像的技术难题，从重建算法层面支撑稀疏投影角度的心肌灌注动态CT扫描方案的临床应用。
本申请提供的图像重建方法，可以在减小心肌动态灌注CT扫描过程中的X射线扫描幅数的情况下获得达到临床诊断标准的重建图像，从而显著降低心肌动态灌注CT成像对病人的辐射剂量。
本申请提供的图像重建方法，在算法性能方面，本申请算法能够有效避免由于稀疏采样导致的重建图像伪影与错误信息问题，在稀疏角度CT的场景下，成像质量明显高于现有商用算法重建结果。
附图说明
图1是本申请的时间序列信息协助的低剂量DMP-CT图像重建算法总体框架示意图；
图2是本申请的时间序列信息协助的低剂量DMP-CT图像重建结果示意图。
具体实施方式
在下文中，将参考附图对本申请的具体实施例进行详细地描述，依照这些详细的描述，所属领域技术人员能够清楚地理解本申请，并能够实施本申请。在不违背本申请原理的情况下，各个不同的实施例中的特征可以进行组合以获得新的实施方式，或者替代某些实施例中的某些特征，获得其它优选的实施方式。
使用稀疏角度对病人进行X射线是有效的降低CT扫描辐射剂量的方法，特别对于DMP-CT这种需要对病人进行多次扫描的医学影像检查。使用现有重建算法(滤波反投影、传统代数迭代重建)从稀疏采集的数据重建CT图像会由于数据的不完整性(incompleteness)和不一致性(inconsistency)，导致重建图像中存在大量的伪影，严重影响诊断。
不失一般性地，CT成像过程在数学上可以用一个简单的矩阵方程描述：
y＝Ax，       (式1)
其中y为CT扫描得到的投影数据(数据校准与对数变换之后)，A为与扫描系统对应的系统 矩阵，表征从被扫描物体到投影数据的正投影过程，x为被扫描物(待重建图像)。CT成像的图像重建任务即是根据所获得的投影数据y，基于与成像系统对应的已知系统矩阵A，估算被扫描物体的X射线衰减系数值x。
对于心肌灌注动态CT成像(DMP-CT)而言，若对于每一帧图像的投影几何(投影角度，投影幅数，投影角度间隔，管电压与管电流等)是完全相同的，那么每一帧图像的成像过程均可用上述(式1)表示，则DMP-CT成像可用下式描述：
Y＝AX，          (式2)
其中，Y＝{y 1,y 2,…,y T}为投影数据时间序列，X＝{x 1,x 2,…,x T}为待重建图像序列。则DMP-CT成像任务可理解为基于成像系统矩阵A，根据投影数据时间序列Y，估计动态图像时间序列X。
参见图1～2，本申请提供一种图像重建方法，所述方法包括如下步骤：
1)采集动态灌注图像序列和基线图像序列；
2)对基线图像投影数据序列进行降采样，在每一帧动态灌注图像序列中减去降采样的基线图像投影数据序列，得到增强投影数据时间序列；
3)对所述增强投影数据时间序列进行迭代预重建，得到预重建对比增强图像序列；
4)对每一个图像像素点均提取一个与之对应的像素特征向量；
5)基于所述像素特征向量构建时间序列信息先验矩阵；
6)将所述时间序列信息先验矩阵以“图像遮罩”的形式引入对比增强图像序列的重建迭代中，得到重建的对比增强图像；
7)将所述对比增强图像投影数据序列逐帧与所述基线图像投影数据相加，得到重建的动态灌注图像时间序列。
进一步地，所述步骤1)中基线图像序列为显像剂到达人体组织之前进行的扫描图像，所述动态灌注图像序列为显像剂到达人体组织之后进行的连续扫描图像。
进一步地，所述步骤1)中所述动态灌注图像采用稀疏的降采样扫描方案进行扫描得到；所述基线图像采用非稀疏的全采样扫描方案进行扫描得到。
进一步地，所述步骤3)中采用全变分正则化的最大似然期望最大化算法进行预重建。
进一步地，所述步骤4)中所述像素特征向量与显像剂时间演变相关。
进一步地，所述步骤4)中所述像素特征向量的提取方法为：将所述预重建对比增强图像序列中的每一帧图像的对应像素点像素值用于组成图像像素点的像素特征向量。
进一步地，所述步骤5)中构建时间序列信息先验矩阵包括：对于每一个像素特征向量
Figure PCTCN2020107143-appb-000012
在像素特征集合
Figure PCTCN2020107143-appb-000013
中确定k个与之建立联系的相近像素特征向量；对处在
Figure PCTCN2020107143-appb-000014
的k-最邻近簇中的k个像素特征向量与
Figure PCTCN2020107143-appb-000015
通过一个预先设计的高斯连接函数建立联系，不处在
Figure PCTCN2020107143-appb-000016
的k-最邻近簇中的其他像素特征向量与
Figure PCTCN2020107143-appb-000017
的连接函数值定义为0；将基于像素特征向量
Figure PCTCN2020107143-appb-000018
计算的所有连接函数值ψ((i,j),(p,q))重排为一个长度为512×512的列向量。
进一步地，所述相近像素特征向量的确定方法为：计算每一个像素特征向量
Figure PCTCN2020107143-appb-000019
与被考虑的像素特征向量
Figure PCTCN2020107143-appb-000020
之间的欧几里得距离，与
Figure PCTCN2020107143-appb-000021
的欧几里得距离最近的k个像素特征向量即构成
Figure PCTCN2020107143-appb-000022
的k-最邻近簇。
进一步地，所述步骤7)中基线图像为从全采样基线投影数据经过滤波反投影算法重建的正常剂量基线图像。
本申请还提供一种图像重建方法的应用，将所述的图像重建方法应用于动态心肌灌注扫描CT图像、脑部动态灌注CT成像与磁共振灌注成像。当然，本申请中只是列举，不限于其他类型图像的重建。
实施例
基于动态灌注增强CT图像的分离重建方法和动态图像序列的稀疏角度图像采集方案，对由显像剂引起的心脏组织对比增强图像借助动态图像序列中自然包含的显像剂时间演变信息进行重建，并将重建增强图像叠加到正常剂量采集方案得到的基线CT图像，获得最终的动态心肌灌注扫描CT图像序列。
图像数据采集：对心肌动态灌注CT成像的基线(baseline)图像采集，即显像剂到达人体心脏组织之前进行的CT扫描，和动态灌注图像序列采集，即显像剂到达人体心脏组织之后进行的连续CT扫描，采取不同剂量的CT图像采集方案。基线图像使用正常剂量，即非稀疏的全采样扫描方案进行扫描，动态灌注图像序列使用低剂量，即稀疏的降采样扫描方案进行扫描。该步骤可用下述式子描述：
Figure PCTCN2020107143-appb-000023
基线投影数据分离：从所采集的稀疏的动态投影数据序列分理出对应于非显像剂引入的图像信息。具体实现方法为：将所采集基线投影数据y 基线,全采样进行降采样，保证所得稀疏基线投影数据与动态灌注投影数据Y 灌注,稀疏的每一帧都对应与相同的图像采集几何条件(即系统矩阵A)。在每一帧的动态灌注投影数据中减去降采样的稀疏基线投影数据，得到增强投影数 据时间序列。该步骤可用下述式子描述：
Figure PCTCN2020107143-appb-000024
对比增强投影数据的迭代预重建：对上述式(4)所得对比增强投影数据Y 对比增强,稀疏使用全变分正则化的最大似然期望最大化算法(MLEM-TV)进行预重建，得到预重建的对比增强图像序列。该步骤可用下述式子描述：
Figure PCTCN2020107143-appb-000025
显像剂时间演变特征提取：对于每一个CT图像像素点(i,j),均提取一个与之对应的与显像剂时间演变相关的像素特征向量
Figure PCTCN2020107143-appb-000026
像素特征向量的提取方法为：将上述步骤3)所得预重建对比增强图像序列中的每一帧图像的对应像素点像素值x 1～T(i,j)用于组成像素点(i,j)的像素特征向量，具体可用下述式子描述：
Figure PCTCN2020107143-appb-000027
其中，T表示动态对比增强图像的帧数。
时间序列信息先验矩阵的构建：基于所提取的关于每一个像素点像素特征向量
Figure PCTCN2020107143-appb-000028
构建时间序列信息先验矩阵。具体的构建方法为：
a.对于每一个像素特征向量
Figure PCTCN2020107143-appb-000029
在像素特征集合
Figure PCTCN2020107143-appb-000030
中确定k个与之建立联系的相近像素特征向量。相近像素特征向量的确定方法为：建立关于像素特征向量
Figure PCTCN2020107143-appb-000031
的k-最邻近簇，即计算每一个像素特征向量
Figure PCTCN2020107143-appb-000032
与被考虑的像素特征向量
Figure PCTCN2020107143-appb-000033
之间的欧几里得距离，与
Figure PCTCN2020107143-appb-000034
的欧几里得距离最近的k个像素特征向量即构成
Figure PCTCN2020107143-appb-000035
的k-最邻近簇。
b.处在
Figure PCTCN2020107143-appb-000036
的k-最邻近簇中的k个像素特征向量与
Figure PCTCN2020107143-appb-000037
通过一个预先设计的高斯连接函数建立联系，不处在
Figure PCTCN2020107143-appb-000038
的k-最邻近簇中的其他像素特征向量与
Figure PCTCN2020107143-appb-000039
的连接函数值定义为0。步骤b可用下述式描述：
Figure PCTCN2020107143-appb-000040
c.将基于像素特征向量
Figure PCTCN2020107143-appb-000041
计算的所有连接函数值ψ((i,j),(p,q))重排为一个长度为 512×512的列向量。因此，对于每一个像素特征向量
Figure PCTCN2020107143-appb-000042
都可以计算得一个由该像素特征向量与其他所有像素特征向量的连接函数值组成的列向量，所有的列向量即组成时间序列信息先验矩阵Ψ。该步骤可用下述式描述：
Figure PCTCN2020107143-appb-000043
时间序列信息先验矩阵协助的增强图像重建：将上述步骤5)构建的时间序列信息先验矩阵以“图像遮罩”的形式引入对比增强图像序列的重建迭代中。将对比增强CT图像
Figure PCTCN2020107143-appb-000044
表示为一个图像遮罩与一个系数图像ζ的乘积，该图像遮罩即为前述步骤5)所构建的时间序列先验矩阵Ψ，即
Figure PCTCN2020107143-appb-000045
则对应于时间帧t的对比增强图像成像任务可描述为：
Figure PCTCN2020107143-appb-000046
将上述(式10)应用于极大似然期望值最大化迭代重建，则有关于系数图像ζ的迭代更新规则：
Figure PCTCN2020107143-appb-000047
根据上式迭代计算所得系数图像ζ可由上述(式9)恢复为重建的对比增强图像
Figure PCTCN2020107143-appb-000048
基线图像与对比增强图像的融合：将根据上述步骤重建得到的对比增强图像序列X 对比增强,重建逐帧与从全采样基线投影数据经过滤波反投影算法(FBP)重建的正常剂量基线图像相加，即得到重建的动态灌注心肌CT图像时间序列X 动态灌注,重建。该步骤可用下述时描述：
Figure PCTCN2020107143-appb-000049
Figure PCTCN2020107143-appb-000050
临床使用的心脏动态灌注CT检查：在显像剂到达人体组织达到对比增强效果之前对病人进行正常剂量的扫描，在显像剂到达人体组织形成对比增强后对病人实施低剂量的稀疏CT动态扫描。以此图像采集方案得到的原始图像数据即可使用本申请方法进行重建。
本申请方法经过猪心脏医学临床图像的仿真实验验证，效果明显：图2展示了使用本申请方法重建的心肌动态灌注CT图像重建结果。该实验在稀疏倍率为12的条件下进行(正常剂量全采样为984幅投影/360°，稀疏低剂量采样为82幅投影/360°)。由图可见本申请方法在稀疏角度采样心肌动态灌注CT图像重建中的优越性能。
本申请对于心肌动态灌注CT扫描的高低剂量配合扫描方案，即显像剂未到达之前使用 正常剂量扫描方案与显像剂到达后的稀疏低剂量扫描方案切换。
本申请从预重建对比增强图像序列中提取像素特征向量，并基于像素特征向量构建时间序列先验矩阵的方法；
本申请将CT图像重建任务转化为图像遮蔽下的系数图像重建问题的方法，及最大似然期望最大化算法在系数图像重建中的应用；
本申请中灌注CT图像的分离重建思想，即将灌注图像的基线部分与对比增强部分分开重建，重建后再融合相加得到最终动态灌注CT图像。
尽管在上文中参考特定的实施例对本申请进行了描述，但是所属领域技术人员应当理解，在本申请公开的原理和范围内，可以针对本申请公开的配置和细节做出许多修改。本申请的保护范围由所附的权利要求来确定，并且权利要求意在涵盖权利要求中技术特征的等同物文字意义或范围所包含的全部修改。

Claims (10)

  1. 一种图像重建方法，其特征在于：所述方法包括如下步骤：
    1)采集动态灌注图像序列和基线图像序列；
    2)对基线图像投影数据序列进行降采样，在每一帧动态灌注图像序列中减去降采样的基线图像投影数据序列，得到增强投影数据时间序列；
    3)对所述增强投影数据时间序列进行迭代预重建，得到预重建对比增强图像序列；
    4)对每一个图像像素点均提取一个与之对应的像素特征向量；
    5)基于所述像素特征向量构建时间序列信息先验矩阵；
    6)将所述时间序列信息先验矩阵以“图像遮罩”的形式引入对比增强图像序列的重建迭代中，得到重建的对比增强图像；
    7)将所述对比增强图像投影数据序列逐帧与所述基线图像投影数据相加，得到重建的动态灌注图像时间序列。
  2. 如权利要求1所述的图像重建方法，其特征在于：所述步骤1)中基线图像序列为显像剂到达人体组织之前进行的扫描图像，所述动态灌注图像序列为显像剂到达人体组织之后进行的连续扫描图像。
  3. 如权利要求1所述的图像重建方法，其特征在于：所述步骤1)中所述动态灌注图像采用稀疏的降采样扫描方案进行扫描得到；所述基线图像采用非稀疏的全采样扫描方案进行扫描得到。
  4. 如权利要求3所述的图像重建方法，其特征在于：所述步骤3)中采用全变分正则化的最大似然期望最大化算法进行预重建。
  5. 如权利要求1所述的图像重建方法，其特征在于：所述步骤4)中所述像素特征向量与显像剂时间演变相关。
  6. 如权利要求1所述的图像重建方法，其特征在于：所述步骤4)中所述像素特征向量的提取方法为：将所述预重建对比增强图像序列中的每一帧图像的对应像素点像素值用于组成图像像素点的像素特征向量。
  7. 如权利要求1所述的图像重建方法，其特征在于：所述步骤5)中构建时间序列信息先验矩阵包括：对于每一个像素特征向量
    Figure PCTCN2020107143-appb-100001
    在像素特征集合
    Figure PCTCN2020107143-appb-100002
    中确定k个与之建立联系的相近像素特征向量；对处在
    Figure PCTCN2020107143-appb-100003
    的k-最邻近簇中的k个像素特征向量与
    Figure PCTCN2020107143-appb-100004
    通过一个预先设计的高斯连接函数建立联系，不处在
    Figure PCTCN2020107143-appb-100005
    的k-最邻近簇中的其他像素特征向量与
    Figure PCTCN2020107143-appb-100006
    的连接函数值定义为0；将基于像素特征向量
    Figure PCTCN2020107143-appb-100007
    计算的所有连接函数值ψ((i,j),(p,q))重排为一个长度为512×512的列 向量。
  8. 如权利要求7所述的图像重建方法，其特征在于：所述相近像素特征向量的确定方法为：计算每一个像素特征向量
    Figure PCTCN2020107143-appb-100008
    与被考虑的像素特征向量
    Figure PCTCN2020107143-appb-100009
    之间的欧几里得距离，与
    Figure PCTCN2020107143-appb-100010
    的欧几里得距离最近的k个像素特征向量即构成
    Figure PCTCN2020107143-appb-100011
    的k-最邻近簇。
  9. 如权利要求1所述的图像重建方法，其特征在于：所述步骤7)中基线图像为从全采样基线投影数据经过滤波反投影算法重建的正常剂量基线图像。
  10. 一种图像重建方法的应用，其特征在于：将权利要求1～9中任一项所述的图像重建方法应用于动态心肌灌注扫描CT图像、脑部动态灌注CT成像与磁共振灌注成像。
PCT/CN2020/107143 2020-08-05 2020-08-05 一种图像重建方法及其应用 WO2022027327A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/107143 WO2022027327A1 (zh) 2020-08-05 2020-08-05 一种图像重建方法及其应用

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/107143 WO2022027327A1 (zh) 2020-08-05 2020-08-05 一种图像重建方法及其应用

Publications (1)

Publication Number Publication Date
WO2022027327A1 true WO2022027327A1 (zh) 2022-02-10

Family

ID=80118656

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/107143 WO2022027327A1 (zh) 2020-08-05 2020-08-05 一种图像重建方法及其应用

Country Status (1)

Country Link
WO (1) WO2022027327A1 (zh)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102314698A (zh) * 2011-08-10 2012-01-11 南方医科大学 基于阿尔法散度约束的全变分最小化剂量ct重建方法
US20130342539A1 (en) * 2010-08-06 2013-12-26 Google Inc. Generating Simulated Eye Movement Traces For Visual Displays
CN111179366A (zh) * 2019-12-18 2020-05-19 深圳先进技术研究院 基于解剖结构差异先验的低剂量图像重建方法和系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130342539A1 (en) * 2010-08-06 2013-12-26 Google Inc. Generating Simulated Eye Movement Traces For Visual Displays
CN102314698A (zh) * 2011-08-10 2012-01-11 南方医科大学 基于阿尔法散度约束的全变分最小化剂量ct重建方法
CN111179366A (zh) * 2019-12-18 2020-05-19 深圳先进技术研究院 基于解剖结构差异先验的低剂量图像重建方法和系统

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
CHANGFEI GONG; CE HAN; GUANGHUI GAN; ZHENXIANG DENG; YONGQIANG ZHOU; JINLING YI; XIAOMIN ZHENG; CONGYING XIE; XIANCE JIN: "Low-dose dynamic myocardial perfusion CT image reconstruction using pre-contrast normal-dose CT scan induced structure tensor total variation regularization", PHYSICS IN MEDICINE AND BIOLOGY, INSTITUTE OF PHYSICS PUBLISHING, BRISTOL GB, vol. 62, no. 7, 1 March 2017 (2017-03-01), Bristol GB , pages 2612 - 2635, XP020315162, ISSN: 0031-9155, DOI: 10.1088/1361-6560/aa5d40 *
GONG CHANGFEI: "High Quality Low-dose Myocardial CT Perfusion Imaging Method", CHINESE MASTER’S THESES DATABASE, MEDICINE & HEALTH SCIENCES- SOUTHERN MEDICAL UNIVERSITY GUANGDONG PROVINCE, 24 May 2016 (2016-05-24), XP055895074 *
JIA XIAO: "Regional Feature Preserving Based Low-Dose CT Precision Imaging", CHINESE DOCTORAL DISSERTATIONS DATABASE, MEDICAL AND HEALTH SCIENCES-SOUTHERN MEDICAL UNIVERSITY GUANGDONG PROVINCE, 11 May 2019 (2019-05-11), pages 1 - 114, XP055895087 *
MA JIANHUA; HUANG JING; FENG QIANJIN; ZHANG HUA; LU HONGBING; LIANG ZHENGRONG; CHEN WUFAN: "Low-dose computed tomography image restoration using previous normal-dose scan", MEDICAL PHYSICS., AIP, MELVILLE, NY., US, vol. 38, no. 10, 1 October 2011 (2011-10-01), US , pages 5713 - 5731, XP012153023, ISSN: 0094-2405, DOI: 10.1118/1.3638125 *

Similar Documents

Publication Publication Date Title
Li et al. Investigation of low-dose CT image denoising using unpaired deep learning methods
CN105046675B (zh) 基于mvct图像消除kvct图像中金属伪影的方法
CN109961419B (zh) 对pet活度分布图像进行衰减校正的校正信息获取方法
Zhou et al. DuDoUFNet: dual-domain under-to-fully-complete progressive restoration network for simultaneous metal artifact reduction and low-dose CT reconstruction
CN112258642B (zh) 基于深度学习的低剂量pet数据三维迭代更新重建方法
CN111915696A (zh) 三维图像数据辅助的低剂量扫描数据重建方法及电子介质
WO2020214911A1 (en) Method and system for generating attenuation map from spect emission data based upon deep learning
TW201219013A (en) Method for generating bone mask
WO2022246677A1 (zh) 一种增强ct图像的重建方法
CN108038840B (zh) 一种图像处理方法、装置、图像处理设备及存储介质
Jung et al. Deep learning cross-phase style transfer for motion artifact correction in coronary computed tomography angiography
CN115984394A (zh) 一种联合先验图像及卷积稀疏网络的低剂量ct重建方法
Li et al. Learning non-local perfusion textures for high-quality computed tomography perfusion imaging
Isola et al. Cardiac motion-corrected iterative cone-beam CT reconstruction using a semi-automatic minimum cost path-based coronary centerline extraction
Du et al. X-ray CT image denoising with MINF: A modularized iterative network framework for data from multiple dose levels
Khodajou-Chokami et al. A deep learning method for high-quality ultra-fast CT image reconstruction from sparsely sampled projections
Gholizadeh-Ansari et al. Low-dose CT denoising using edge detection layer and perceptual loss
Liu et al. Motion estimation based on projective information disentanglement for 3D reconstruction of rotational coronary angiography
WO2022027327A1 (zh) 一种图像重建方法及其应用
Khodajou-Chokami et al. PARS-NET: a novel deep learning framework using parallel residual conventional neural networks for sparse-view CT reconstruction
Liang et al. A model-based unsupervised deep learning method for low-dose CT reconstruction
CN111862262B (zh) 一种图像重建方法及其应用
Fung et al. A multimodal approach to image-derived input functions for brain PET
Yan et al. Calcium removal from cardiac CT images using deep convolutional neural network
Makki et al. High-resolution temporal reconstruction of ankle joint from dynamic MRI

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20948184

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.07.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20948184

Country of ref document: EP

Kind code of ref document: A1