WO2021159948A1 - 一种基于深度学习的低剂量pet三维重建方法 - Google Patents

一种基于深度学习的低剂量pet三维重建方法 Download PDF

Info

Publication number
WO2021159948A1
WO2021159948A1 PCT/CN2021/073462 CN2021073462W WO2021159948A1 WO 2021159948 A1 WO2021159948 A1 WO 2021159948A1 CN 2021073462 W CN2021073462 W CN 2021073462W WO 2021159948 A1 WO2021159948 A1 WO 2021159948A1
Authority
WO
WIPO (PCT)
Prior art keywords
pet
low
dose
image
projection
Prior art date
Application number
PCT/CN2021/073462
Other languages
English (en)
French (fr)
Inventor
朱闻韬
杨宝
周龙
叶宏伟
陈凌
饶璠
王瑶法
Original Assignee
之江实验室
明峰医疗系统股份有限公司
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 之江实验室, 明峰医疗系统股份有限公司 filed Critical 之江实验室
Priority to US17/616,161 priority Critical patent/US12118649B2/en
Publication of WO2021159948A1 publication Critical patent/WO2021159948A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10104—Positron emission tomography [PET]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/421—Filtered back projection [FBP]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/441—AI-based methods, deep learning or artificial neural networks

Definitions

  • the invention relates to the field of medical imaging, in particular to a low-dose PET three-dimensional reconstruction method based on deep learning.
  • PET Positron Emission Tomography
  • the imaging process of PET includes injecting a radioactive tracer into the patient before scanning. The tracer decays to produce positrons when it participates in physiological metabolism. The positrons and neighboring electrons have an annihilation effect to produce 511keV photon pairs that move in the opposite direction. The photon pair hits the PET scanner receiver to form a certain number of line of response (LOR), and save it as a three-dimensional PET raw data sinogram. After attenuation, randomness, and scatter correction are performed on the sinogram, the three-dimensional iterative reconstruction of it with statistical methods can be used to obtain a three-dimensional PET image that characterizes the metabolic intensity of various tissues in the human body.
  • LOR line of response
  • the purpose of the present invention is to break through the performance bottleneck of the existing low-dose PET reconstruction method based on image post-processing technology, and provide a PET image three-dimensional reconstruction method based on deep learning starting from PET raw data.
  • This method non-destructively back-projects the low-dose PET raw data to the image domain to obtain a highly blurred laminogram, uses a three-dimensional U-Net to fit the image domain deconvolution operation, recovers a rough estimate of the PET image from the laminogram, and connects the three-dimensional residual
  • the difference unit (Residual block) further refines the PET reconstructed image, uses the prior knowledge of training samples to learn and fix the network parameters, and finally applies it to the low-dose PET image three-dimensional reconstruction to obtain and reduce the traditional reconstruction results.
  • the low-dose PET reconstructed image with lower noise and higher resolution.
  • a low-dose PET 3D reconstruction method based on deep learning including the following steps:
  • the low-dose PET raw data is processed by attenuation correction, and then the system matrix transposition of the PET scanner is used to obtain the back-projection l pp_ac of the low-dose PET data after the attenuation correction.
  • step (1.4) in step (1.1) PET data acquisition in low dose backprojection l pp_ac step (1.2) after obtaining the attenuation correction backprojection l pp_ac random scattering data and subtraction, random scatter correction, divided by the step of ( 1.3) obtaining a PET image full backprojection 1 1 L, obtaining the correction, the low dose PET three-dimensional back projection regularization l bp:
  • step (1.4) Use the corrected and regularized low-dose PET three-dimensional backprojection l bp obtained in step (1.4) as the input of the deep neural network, use the standard-dose PET reconstructed image as the network label, and update the deep neural network through the Adam optimization algorithm Parameters to minimize the objective loss function and complete the training of the deep neural network.
  • the target loss function of the deep neural network training is:
  • N x , N y , and N z represent the total number of pixels in the horizontal, vertical, and axial directions of the low-dose PET back-projection or standard-dose PET image, respectively, and C( ⁇ ) represents the low-dose fitting of the three-dimensional deep neural network.
  • the mapping from the dose PET back-projection l bp to the standard dose PET reconstructed image f full , (i, j, k) represents the pixel in the image.
  • the present invention has the beneficial effect that the present invention starts from the low-dose PET raw data and suppresses the generation of artifacts and quantitative errors caused by the insufficient generalization ability of the image domain mapping neural network.
  • the present invention proposes a method of non-destructively back-projecting the low-dose PET raw data to the image domain, which reduces the computational complexity of the neural network fitting low-dose to standard dose mapping, improves the network training efficiency, and reduces the test network
  • the time required to generate low-dose PET reconstruction images is much lower than that of traditional iterative reconstruction algorithms, and PET images with a higher signal-to-noise ratio than traditional reconstruction algorithms can be obtained at the same dose.
  • Figure 1 is a flow chart of the back-projection method for low-dose PET raw data
  • Figure 2 is a comparison diagram of PET reconstructed images between the traditional algorithm and the algorithm of the present invention.
  • the low-dose raw data received by the PET scanner is a three-dimensional sinogram formed by the projection of the PET image through the three-dimensional X-ray transformation. Since the three-dimensional sinogram contains not only the axial plane projection, but also the inclined plane projection passing through the axial plane, it has the characteristics of large data volume and high redundancy of information.
  • the mapping between sinogram and PET image is directly used to fit the neural network. The limitations of computer computing and storage capacity are difficult to achieve.
  • the present invention proposes a lossless back projection method for three-dimensional sinogram, and its flow chart is shown in Fig. 1:
  • the low-dose PET raw data is processed by attenuation correction, and then the system matrix transposition of the PET scanner is used to obtain the back-projection l pp_ac of the low-dose PET data after attenuation correction, which is a highly blurred PET image.
  • the present invention proposes a regularization method for PET back-projection to generate a laminogram of all 1 simulated PET images, and Through the effect of the system matrix of the PET scanner, the projection result is back-projected to the image domain to obtain the back-projection l 1 of the PET image of all 1s.
  • step (1.4) in step (1.1) PET data acquisition in low dose backprojection l pp_ac step (1.2) after obtaining the attenuation correction backprojection l pp_ac random scattering data and subtraction, random scatter correction, divided by the step of ( 1.3) obtaining a PET image full backprojection 1 1 L, obtaining the correction, the low dose PET three-dimensional back projection regularization l bp:
  • f (x, y, z) and l bp (x, y, z) represent the activity value of the three-dimensional PET image and the backprojection after correction and regularization at a certain point (x, y, z)
  • H ( ⁇ , ⁇ ) is the three-dimensional Fourier transform of the rotationally symmetric PSF (point spread function) expressed in spherical coordinates, defined as:
  • the present invention proposes to use a three-dimensional deep neural network to fit
  • the deep neural network is composed of two parts.
  • the first part is a U-Net composed of a 3D convolutional layer, a 3D deconvolutional layer, and jump connections (shortcuts) between them.
  • the convolutional layer is used to encode the PET backprojection, extract high-level features, and then the deconvolution layer encodes the features to obtain a rough estimate of the PET image.
  • the shortcuts in the network combine the output of the convolution layer with the corresponding deconvolution layer. The output is superimposed to improve the network training effect and effectively prevent the degradation problem without increasing the network parameters.
  • the second part of the deep neural network is composed of multiple residual blocks in series, which is used to further refine the high-frequency details in the rough estimation of the PET image. Since the low-frequency information contained in the rough estimation of the PET image is similar to the low-frequency information of the standard dose PET image, the residual unit can only learn the high-frequency residual part between the two to improve the efficiency of network training.
  • the corrected and regularized low-dose PET three-dimensional backprojection l bp obtained in step (1.4) is used as the input of the deep neural network, and the standard-dose PET reconstructed image is used as the label of the deep neural network, which is updated by the Adam optimization algorithm
  • the parameters of the deep neural network minimize the objective loss function, complete the training of the deep neural network, and obtain an estimate of the mapping network
  • the prior knowledge of the standard dose learned by training the mapping network C( ⁇ ) can realize the compensation for the low-dose PET back projection during the test.
  • the target loss function of the deep neural network training is:
  • N x , N y , and N z represent the total number of pixels in the horizontal, vertical, and axial directions of the low-dose PET back-projection or standard-dose PET image, respectively, and C( ⁇ ) represents the low-dose fitting of the three-dimensional deep neural network.
  • the mapping from the dose PET back-projection l bp to the standard dose PET reconstructed image f full , (i, j, k) represents the pixel in the image.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Algebra (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Nuclear Medicine (AREA)

Abstract

本发明公开了一种基于深度学习的低剂量PET三维重建方法，该方法包括将低剂量PET原始数据无损反投影至图像域，选取适当的三维深度神经网络结构以拟合低剂量PET反投影到标准剂量PET图像之间的映射，通过训练样本学习并固定网络参数后，实现从低剂量PET原始数据出发的PET图像三维重建，以获取比传统重建算法与图像域降噪处理噪声更低，分辨率更高的低剂量PET重建图像。

Description

一种基于深度学习的低剂量PET三维重建方法 技术领域
本发明涉及医学影像领域，具体地涉及一种基于深度学习的低剂量PET三维重建方法。
背景技术
正电子发射断层扫描(Positron Emission Tomography，PET)是一种能提供活体生化和定量生理信息的医学影像，在肿瘤学、心脏学、神经学，以及精神疾病等方面具有重要应用，PET/CT也已经成为国际公认的肿瘤检测的金标准。PET的成像过程包括扫描前对病人注入放射性示踪剂，示踪剂在参与生理代谢时发生衰变产生正电子，正电子和邻近的电子发生湮灭效应，产生逆向运动的511keV光子对。光子对打到PET扫描仪接收器上形成一定数量的符合响应线(Line of Response，LOR)，并将其保存成三维PET原始数据sinogram。对sinogram进行衰减、随机和散射校正后再用统计学方法对其进行三维迭代重建，即可获得表征人体内各组织代谢强度的三维PET图像。
通过减少注入病人体内的放射性示踪剂剂量，可以降低对病人及医护人员的辐射，缩短扫描时间，降低造影成本，因此低剂量PET重建已经成为核医学领域最热门的课题之一，尤其在婴幼儿癫痫以及青少年抑郁等应用领域获得广泛关注。然而注射示踪剂的减少导致了PET扫描仪接收到的真符合事件(true coincidence)受随机(random coincidence)、散射符合事件(scattered coincidence)干扰的增加，使得PET原始数据信噪比的降低，从而进一步导致由传统重建方法获取的PET图像被大量噪声污染，严重影响了临床医生的读片与诊断。
近年来，随着计算机技术的迅速发展，深度学习方法在自然图像处理、语音识别等问题上取得了重大突破，同时在医学影像上的应用也越来越广泛。为了抑制传统重建算法获取的低剂量PET重建图像中的噪声并保留有用病灶信息，基于深度学习方法的降噪处理已经被引入以改善低剂量PET成像。目前，此类方法主要集中在图像后处理范畴，即在图像域建立高噪声低剂量PET重建图像到标准剂量PET图像的映射。由于传统的低剂量PET重建过程并不能完全保留原始数据中的有效高频信息，用于拟合映射的神经网络的泛化能力不足导致图像伪影和定量误差成为了此类技术路线的性能瓶颈。
发明内容
本发明的目的在于突破现有基于图像后处理技术的低剂量PET重建方法的性能瓶颈，提供一种从PET原始数据出发的，基于深度学习的PET图像三维重建方法。该方法将低剂量PET原始数据无损反投影到图像域获取高度模糊的laminogram，使用三维U-Net拟合图像域 去卷积操作，从laminogram中恢复出PET图像的粗估计，并通过串联三维残差单元(Residual block)进一步细化PET重建图像，利用训练样本先验知识学习并固定网络参数，最后将其应用于低剂量PET图像三维重建，以获取与传统重建结果以及对传统重建结果进行降噪后处理所得的PET图像相比，噪声更低，分辨率更高的低剂量PET重建图像。
本发明的目的是通过以下技术方案实现的：一种基于深度学习的低剂量PET三维重建方法，包括以下步骤：
(1)执行低剂量PET原始数据反投影方法，具体包括以下子步骤：
(1.1)将低剂量PET原始数据经过衰减校正处理，再经PET扫描仪的系统矩阵转置作用，获取衰减校正后低剂量PET数据反投影l pp_ac。
(1.2)将低剂量PET原始数据中的随机与散射事件数据经过PET扫描仪的系统矩阵转置的作用，获取随机与散射数据的反投影l rs。
(1.3)生成全1仿真PET图像，并将其经过PET扫描仪的系统矩阵作用，获得其三维投影，再将三维投影的结果反投影至图像域，获得全1的PET图像的反投影l 1。
(1.4)将步骤(1.1)中获取衰减校正后低剂量PET数据反投影l pp_ac与步骤(1.2)中获取随机与散射数据的反投影l pp_ac相减，进行随机散射校正，再除以步骤(1.3)获得全1的PET图像的反投影l 1，获得校正、正则化后的低剂量PET三维反投影l bp：
Figure PCTCN2021073462-appb-000001
(2)将步骤(1.4)获取的校正、正则化后的低剂量PET三维反投影l bp作为深度神经网络的输入，将标准剂量PET重建图像作为网络的标签，通过Adam优化算法更新深度神经网络参数，使目标损失函数最小化，完成对深度神经网络的训练。所述深度神经网络训练的目标损失函数为：
Figure PCTCN2021073462-appb-000002
其中N x、N y、N z分别表示低剂量PET反投影或标准剂量PET图像在水平、竖直和轴向的像素点总个数，C(·)表示三维深度神经网络拟合的从低剂量PET反投影l bp到标准剂量PET重建图像f full的映射，(i,j,k)表示图像中的像素点。
(3)对新采集的低剂量PET原始数据，执行步骤一的反投影方法，再输入步骤二训练好的深度神经网络中，获得对应低剂量PET重建图像。
与现有技术相比，本发明的有益效果是：本发明从低剂量PET原始数据出发，抑制了由于图像域映射神经网络的泛化能力不足所导致的伪影和定量误差的产生。另外，本发明提出将低剂量PET原始数据无损反投影至图像域的方法，减小了拟合低剂量到标准剂量映射的神经网络的计算复杂度，提高了网络训练效率，降低了测试网络即生成低剂量PET重建图像所需时间，使其远低于传统迭代重建算法耗时，并且在相同剂量下可获取比传统重建算法更高信噪比的PET图像。
附图说明
图1是低剂量PET原始数据反投影方法流程图；
图2是传统算法与本发明算法PET重建图像对比图。
具体实施方式
下面结合附图和实施例对本发明的内容做进一步阐述。
PET扫描仪接收到的低剂量原始数据是由PET图像经过三维X-ray变换所得投影构成的三维sinogram。由于三维sinogram中不仅包含轴向平面投影，还包含穿过轴向平面的倾斜平面投影，它具有数据量大和信息高度冗余的特点，直接用神经网络拟合sinogram到PET图像之间的映射受到计算机运算及存储能力的限制，很难实现。本发明提出针对三维sinogram的无损反投影方法，其流程图如图1所示：
(1.1)将低剂量PET原始数据经过衰减校正处理，再经PET扫描仪的系统矩阵转置作用，获取衰减校正后低剂量PET数据反投影l pp_ac，即高度模糊的PET图像。
(1.2)将低剂量PET原始数据中的随机与散射事件数据经过PET扫描仪的系统矩阵转置的作用，获取随机与散射数据的反投影l rs。
(1.3)为了削弱由PET扫描仪轴向范围有限所导致的三维投影与反投影过程的空间变化性，本发明提出对于PET反投影的正则化方法，生成全1仿真PET图像的laminogram，并将其经过PET扫描仪的系统矩阵的作用，再将投影的结果反投影至图像域，获得全1的PET图像的反投影l 1。
(1.4)将步骤(1.1)中获取衰减校正后低剂量PET数据反投影l pp_ac与步骤(1.2)中获取随机与散射数据的反投影l pp_ac相减，进行随机散射校正，再除以步骤(1.3)获得全1的PET图像的反投影l 1，获得校正、正则化后的低剂量PET三维反投影l bp：
Figure PCTCN2021073462-appb-000003
所述校正、正则化后的低剂量PET三维反投影l bp与PET重建图像之间的关系表示为：
Figure PCTCN2021073462-appb-000004
其中，f(x,y,z)和l bp(x,y,z)分别表示三维PET图像和校正、正则化后反投影在某一点(x,y,z)处的活度值，H(υ,ψ)是用球面坐标表示的旋转对称的PSF(point spread function)的三维傅里叶变换，定义为：
Figure PCTCN2021073462-appb-000005
Figure PCTCN2021073462-appb-000006
表示ramp-type三维图像域滤波器，PET反投影与之卷积可恢复出高分辨率PET图像。
本发明提出用三维深度神经网络拟合
Figure PCTCN2021073462-appb-000007
该深度神经网络由两部分构成，第一部分是由3D卷积层，3D反卷积层，以及它们之间的跳跃连接(shortcuts)构成的U-Net。其中卷积层用来对PET反投影进行编码，提取高层特征，反卷积层再对特征进行编码获取PET图像的粗估计，网络中的shortcuts将卷积层的输出与对应反卷积层的输出进行叠加，实现在不增加网络参数的基础上，改善网络训练效果并有效防止退化问题。
深度神经网络的第二部分由多个残差单元(Residual block)串联构成，用于进一步细化PET图像粗估计中的高频细节。由于PET图像的粗估计包含的低频信息与标准剂量PET图像的低频信息相近，使用残差单元可以只学习两者之间的高频残差部分，以提高网络训练效率。
(2)因此将步骤(1.4)获取的校正、正则化后的低剂量PET三维反投影l bp作为深度神经网络的输入，将标准剂量PET重建图像作为深度神经网络的标签，通过Adam优化算法更新所述深度神经网络参数，使目标损失函数最小化，完成对深度神经网络的训练，获得映射网络的估计
Figure PCTCN2021073462-appb-000008
训练映射网络C(·)所学习到的标准剂量先验知识可以在测试过程中实现对低 剂量PET反投影的补偿。所述深度神经网络训练的目标损失函数为：
Figure PCTCN2021073462-appb-000009
其中N x、N y、N z分别表示低剂量PET反投影或标准剂量PET图像在水平、竖直和轴向的像素点总个数，C(·)表示三维深度神经网络拟合的从低剂量PET反投影l bp到标准剂量PET重建图像f full的映射，(i,j,k)表示图像中的像素点。
(3)用训练好的映射
Figure PCTCN2021073462-appb-000010
对新采集的低剂量PET原始数据进行重建：首先执行步骤一的反投影方法，将处理后的反投影输入映射网络
Figure PCTCN2021073462-appb-000011
获得对应低剂量PET重建图像。
传统重建算法对低剂量PET数据的重建结果如图2(a)所示，重建图像噪声大，无法区分病灶与噪声；使用本发明所提出的重建算法可获取如图2(b)所示的低剂量PET重建图像，图像2(b)的信噪比明显高于图像2(a)。

Claims (1)

  1. 一种基于深度学习的低剂量PET三维重建方法，其特征在于，包括以下步骤：
    (1)执行低剂量PET原始数据反投影方法，具体包括以下子步骤：
    (1.1)将低剂量PET原始数据经过衰减校正处理，再经PET扫描仪的系统矩阵转置作用，获取衰减校正后低剂量PET数据反投影l pp_ac。
    (1.2)将低剂量PET原始数据中的随机与散射事件数据经过PET扫描仪的系统矩阵转置的作用，获取随机与散射数据的反投影l rs。
    (1.3)生成全1仿真PET图像，并将其经过PET扫描仪的系统矩阵作用，获得其三维投影，再将三维投影的结果反投影至图像域，获得全1的PET图像的反投影l 1。
    (1.4)将步骤(1.1)中获取衰减校正后低剂量PET数据反投影l pp_ac与步骤(1.2)中获取随机与散射数据的反投影l pp_ac相减，进行随机散射校正，再除以步骤(1.3)获得全1的PET图像的反投影l 1，获得校正、正则化后的低剂量PET三维反投影l bp：
    Figure PCTCN2021073462-appb-100001
    (2)将步骤(1.4)获取的校正、正则化后的低剂量PET三维反投影l bp作为深度神经网络的输入，将标准剂量PET重建图像作为网络的标签，通过Adam优化算法更新深度神经网络参数，使目标损失函数最小化，完成对深度神经网络的训练。所述深度神经网络训练的目标损失函数为：
    Figure PCTCN2021073462-appb-100002
    其中N x、N y、N z分别表示低剂量PET反投影或标准剂量PET图像在水平、竖直和轴向的像素点总个数，C(·)表示三维深度神经网络拟合的从低剂量PET反投影l bp到标准剂量PET重建图像f full的映射，(i,j,k)表示图像中的像素点。
    (3)对新采集的低剂量PET原始数据，执行步骤一的反投影方法，再输入步骤二训练好的深度神经网络中，获得对应低剂量PET重建图像。
PCT/CN2021/073462 2020-02-11 2021-01-23 一种基于深度学习的低剂量pet三维重建方法 WO2021159948A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/616,161 US12118649B2 (en) 2020-02-11 2021-01-23 Deep learning based three-dimensional reconstruction method for low-dose PET imaging

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010087761.0 2020-02-11
CN202010087761.0A CN111325686B (zh) 2020-02-11 2020-02-11 一种基于深度学习的低剂量pet三维重建方法

Publications (1)

Publication Number Publication Date
WO2021159948A1 true WO2021159948A1 (zh) 2021-08-19

Family

ID=71172639

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/073462 WO2021159948A1 (zh) 2020-02-11 2021-01-23 一种基于深度学习的低剂量pet三维重建方法

Country Status (2)

Country Link
CN (1) CN111325686B (zh)
WO (1) WO2021159948A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117201691A (zh) * 2023-11-02 2023-12-08 湘江实验室 一种基于深度学习的全景扫描关联成像方法

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325686B (zh) * 2020-02-11 2021-03-30 之江实验室 一种基于深度学习的低剂量pet三维重建方法
CN111915696B (zh) * 2020-08-07 2023-09-22 南京安科医疗科技有限公司 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
CN112053412B (zh) * 2020-08-31 2022-04-29 浙江大学 基于教师-学生生成器的低剂量Sinogram去噪与PET图像重建方法
CN111862320B (zh) * 2020-09-21 2020-12-11 之江实验室 一种spect三维重建图像到标准视图的自动转向方法
CN112541871B (zh) * 2020-12-07 2024-07-23 深圳先进技术研究院 低剂量图像去噪网络的训练方法、低剂量图像的去噪方法
CN112258642B (zh) * 2020-12-21 2021-04-09 之江实验室 基于深度学习的低剂量pet数据三维迭代更新重建方法
CN112489158B (zh) * 2021-01-13 2023-05-12 河北大学 基于cGAN的自适应网络用于低剂量PET图像的增强方法
CN112927132B (zh) * 2021-01-25 2022-07-19 浙江大学 提升pet系统空间分辨率均匀性的pet图像重建方法
CN112991477B (zh) * 2021-01-28 2023-04-18 明峰医疗系统股份有限公司 基于深度学习的pet图像处理方法
CN113057667B (zh) * 2021-03-26 2023-08-15 上海联影医疗科技股份有限公司 Pet探测器信号采样方法、装置、电子装置和存储介质
CN113256753B (zh) * 2021-06-30 2021-10-29 之江实验室 基于多任务学习约束的pet图像感兴趣区域增强重建方法
CN113838161B (zh) * 2021-11-25 2022-02-22 之江实验室 一种基于图学习的稀疏投影重建方法
US12008689B2 (en) 2021-12-03 2024-06-11 Canon Medical Systems Corporation Devices, systems, and methods for deep-learning kernel-based scatter estimation and correction
CN114332287B (zh) * 2022-03-11 2022-07-15 之江实验室 基于transformer特征共享的PET图像重建方法、装置、设备及介质
CN114758031B (zh) * 2022-06-15 2022-10-21 之江实验室 电容耦合电阻抗层析成像图像重建方法和装置
CN116503506B (zh) * 2023-06-25 2024-02-06 南方医科大学 一种图像重建方法、系统、装置及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102737392A (zh) * 2012-06-07 2012-10-17 南方医科大学 一种低剂量x线ct图像的非局部正则化先验重建方法
EP2711738A1 (en) * 2012-09-21 2014-03-26 Technische Universität München A method and a device to generate virtual X-ray computed tomographic image data
CN103810735A (zh) * 2014-02-28 2014-05-21 南方医科大学 一种低剂量x射线ct图像统计迭代重建方法
CN111325686A (zh) * 2020-02-11 2020-06-23 之江实验室 一种基于深度学习的低剂量pet三维重建方法

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104183012B (zh) * 2013-10-31 2015-12-09 上海联影医疗科技有限公司 一种pet三维图像重建方法和装置
US20180197317A1 (en) * 2017-01-06 2018-07-12 General Electric Company Deep learning based acceleration for iterative tomographic reconstruction
US10475214B2 (en) * 2017-04-05 2019-11-12 General Electric Company Tomographic reconstruction based on deep learning
CN107133997B (zh) * 2017-04-11 2019-10-15 浙江大学 一种基于深度神经网络的双示踪剂pet重建方法
US11234666B2 (en) * 2018-05-31 2022-02-01 Canon Medical Systems Corporation Apparatus and method for medical image reconstruction using deep learning to improve image quality in position emission tomography (PET)
CN109102550B (zh) * 2018-06-08 2023-03-31 东南大学 基于卷积残差网络的全网络低剂量ct成像方法及装置
CN109166161B (zh) * 2018-07-04 2023-06-30 东南大学 一种基于噪声伪影抑制卷积神经网络的低剂量ct图像处理系统
CN109559360B (zh) * 2018-09-29 2020-08-28 深圳先进技术研究院 一种pet图像重建方法、装置和计算设备
CN109712213B (zh) * 2018-12-24 2023-10-20 上海联影医疗科技股份有限公司 Pet图像重建方法、系统、可读存储介质和设备
CN110009613A (zh) * 2019-03-28 2019-07-12 东南大学 基于深度稠密网络的低剂量ct成像方法、装置及系统
CN110599420B (zh) * 2019-05-27 2022-03-15 首都师范大学 基于深度学习的ct图像分块重建方法及系统
CN110415311B (zh) * 2019-07-29 2024-04-16 上海联影医疗科技股份有限公司 Pet图像重建方法、系统、可读存储介质和设备
CN110559008B (zh) * 2019-09-03 2022-10-25 明峰医疗系统股份有限公司 一种正电子断层扫描及重建方法
CN110717951B (zh) * 2019-09-12 2021-08-03 浙江大学 一种基于cGANs的PET图像直接重建方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102737392A (zh) * 2012-06-07 2012-10-17 南方医科大学 一种低剂量x线ct图像的非局部正则化先验重建方法
EP2711738A1 (en) * 2012-09-21 2014-03-26 Technische Universität München A method and a device to generate virtual X-ray computed tomographic image data
CN103810735A (zh) * 2014-02-28 2014-05-21 南方医科大学 一种低剂量x射线ct图像统计迭代重建方法
CN111325686A (zh) * 2020-02-11 2020-06-23 之江实验室 一种基于深度学习的低剂量pet三维重建方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117201691A (zh) * 2023-11-02 2023-12-08 湘江实验室 一种基于深度学习的全景扫描关联成像方法
CN117201691B (zh) * 2023-11-02 2024-01-09 湘江实验室 一种基于深度学习的全景扫描关联成像方法

Also Published As

Publication number Publication date
CN111325686B (zh) 2021-03-30
US20220383565A1 (en) 2022-12-01
CN111325686A (zh) 2020-06-23

Similar Documents

Publication Publication Date Title
WO2021159948A1 (zh) 一种基于深度学习的低剂量pet三维重建方法
JP7539954B2 (ja) ディープ畳み込みニューラルネットワークを使用した医用画像化のための線量低減
CN111627082B (zh) 基于滤波反投影算法和神经网络的pet图像重建方法
CN107481297B (zh) 一种基于卷积神经网络的ct图像重建方法
CN113256753B (zh) 基于多任务学习约束的pet图像感兴趣区域增强重建方法
WO2020118844A1 (zh) Pet图像的重建方法、计算机存储介质、计算机设备
CN107123095A (zh) 一种pet图像重建方法、成像系统
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
Ma et al. An encoder-decoder network for direct image reconstruction on sinograms of a long axial field of view PET
Shao et al. A learned reconstruction network for SPECT imaging
Song et al. Low-dose cardiac-gated spect studies using a residual convolutional neural network
CN109903356A (zh) 基于深度多重解析网络的缺失ct投影数据估计方法
He et al. Downsampled imaging geometric modeling for accurate CT reconstruction via deep learning
CN113870138A (zh) 基于三维U-net的低剂量CT图像去噪方法及系统
CN111325695A (zh) 基于多剂量等级的低剂量图像增强方法、系统及存储介质
Feng et al. Dual residual convolutional neural network (DRCNN) for low-dose CT imaging
Jang et al. Head motion correction based on filtered backprojection for x‐ray CT imaging
Li et al. Learning non-local perfusion textures for high-quality computed tomography perfusion imaging
Liu et al. MRCON-Net: Multiscale reweighted convolutional coding neural network for low-dose CT imaging
Zhang et al. Hformer: highly efficient vision transformer for low-dose CT denoising
Li et al. Enhanced PET imaging using progressive conditional deep image prior
CN116245969A (zh) 一种基于深度神经网络的低剂量pet图像重建方法
Kim et al. CNN-based CT denoising with an accurate image domain noise insertion technique
Jiang et al. GDAFormer: Gradient-guided Dual Attention Transformer for Low-Dose CT image denoising
US12118649B2 (en) Deep learning based three-dimensional reconstruction method for low-dose PET imaging

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: 21754019

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21754019

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 21754019

Country of ref document: EP

Kind code of ref document: A1