WO2020177126A1 - 图像处理方法、***、计算设备及存储介质 - Google Patents

图像处理方法、***、计算设备及存储介质 Download PDF

Info

Publication number
WO2020177126A1
WO2020177126A1 PCT/CN2019/077344 CN2019077344W WO2020177126A1 WO 2020177126 A1 WO2020177126 A1 WO 2020177126A1 CN 2019077344 W CN2019077344 W CN 2019077344W WO 2020177126 A1 WO2020177126 A1 WO 2020177126A1
Authority
WO
WIPO (PCT)
Prior art keywords
class
blood vessel
original image
voxel
probability
Prior art date
Application number
PCT/CN2019/077344
Other languages
English (en)
French (fr)
Inventor
张保昌
周寿军
李娜
吴宗翰
李迟迟
王澄
Original Assignee
深圳先进技术研究院
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 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Priority to PCT/CN2019/077344 priority Critical patent/WO2020177126A1/zh
Publication of WO2020177126A1 publication Critical patent/WO2020177126A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • 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
    • G06T2207/10081—Computed x-ray tomography [CT]
    • 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
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • 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
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular

Definitions

  • the present invention belongs to the technical field of image processing, and particularly relates to an image processing method, system, computing device and storage medium.
  • the prior Chinese patent (application publication number CN109102511A) relates to a method, system and electronic device for cerebrovascular segmentation. Its main realization is to perform multi-scale filtering and enhancement processing on the original image containing brain tissue to obtain the enhanced vascular feature image and corresponding The direction vector field of ; establish a finite mixture model and estimate the parameters of the finite mixture model to obtain the class conditional probability; calculate the initial mark field of the original image, and form the Markov random field from the initial mark field and the direction vector field; and then obtain the class prior Probability: Based on the class prior probability and class conditional probability, by maximizing the posterior probability and iterative conditional mode, the cerebrovascular segmentation result is obtained.
  • the purpose of the present invention is to provide an image processing method, system, computing device and storage medium, aiming to solve the problem that the accuracy of cerebrovascular segmentation cannot be effectively improved due to the inaccurate fitting of the vascular tissue distribution interval in the prior art The problem.
  • the present invention provides an image processing method, which includes the following steps:
  • the fitting model is used to fit the gray histogram, and the fitting model is used to simulate the cerebrovascular Distribution or construction of the distribution function of the background distribution;
  • the parameters are iteratively updated.
  • the parameters that are the current iterative update target consist of: the label information corresponding to the voxel that has been first marked and the The unlabeled information corresponding to the first labeled voxel is constructed, the unlabeled information is constructed from the class posterior probability as the update result of the previous iteration, and the class posterior probability is constructed from the distribution function and corresponds to Background class and cerebrovascular class;
  • the class posterior probability corresponding to the voxel perform a second mark on the voxel to indicate that the voxel belongs to the background class or the cerebrovascular class, and obtain the class conditional probability;
  • the marker field is updated by maximizing the corresponding posterior probability to obtain the cerebrovascular segmentation result.
  • obtaining a blood vessel feature map from the original image specifically includes:
  • the multi-scale blood vessel enhancement value is converted into a blood vessel feature value with a meaning of blood vessel prediction probability, and the blood vessel feature map is composed of the blood vessel feature value.
  • determining the brain tissue area from the original image specifically includes:
  • the signal-to-noise ratio enhancement processing is performed on the original image to obtain the brain tissue region, and the signal-to-noise ratio enhancement processing includes decranial processing.
  • the clustering algorithm is a K-means clustering algorithm.
  • the fitting model is a Gaussian mixture model
  • the distribution function is a Gaussian distribution function
  • the two-point potential clump function is used to obtain the energy representation of the voxel to construct the Markov random field.
  • the two-point potential clump function is derived from the mark field obtained by the second mark and the result The construction of the vascular feature map.
  • the marker field is updated to obtain the cerebrovascular segmentation result, specifically:
  • the Bayesian criterion is used Calculate the posterior probability and maximize the posterior probability to update the second marking result of the voxel, thereby updating the marking field, and obtaining the cerebrovascular segmentation result, where N is an integer.
  • the present invention provides an image processing system, which includes:
  • a preprocessing unit for obtaining an original image containing brain tissue; determining a brain tissue area from the original image;
  • the initialization unit is used to use a clustering algorithm to process the gray histogram corresponding to the brain tissue region to obtain a preliminary classification result for preliminarily distinguishing the cerebrovascular and background in the brain tissue region;
  • Preliminary classification results initialize the parameters of the preset fitting model, the fitting model is used to fit the gray histogram, the fitting model is used to simulate the distribution of the cerebral blood vessels or Construction of the distribution function of the background distribution; window width and window level transformation analysis is performed on the original image to obtain voxels in the original image that meet the preset threshold requirements, and perform first marking on these voxels, so The threshold is used to distinguish the background and the cerebrovascular;
  • a feature map calculation unit for obtaining a blood vessel feature map from the original image
  • the segmentation unit is configured to iteratively update the parameters based on a preset iterative update model.
  • the parameter that is the current iterative update target corresponds to the voxel that has been first marked
  • the unlabeled information is constructed from the class posterior probability as the update result of the previous iteration
  • the class posterior probability is constructed from the distribution Function is constructed and corresponds to the background class and the cerebrovascular class; according to the posterior probability of the class corresponding to the voxel, a second label is performed on the voxel to indicate that the voxel belongs to the background class or the Cerebrovascular class, and obtain class conditional probability; combine the marker field obtained from the second label and the blood vessel feature map to construct a Markov random field; construct class prior probability from the Markov random field; based on For the class conditional probability and the class prior probability, in the iterative conditional mode, by
  • the present invention also provides a computing device including a memory and a processor, and the processor implements the steps in the above method when the processor executes the computer program stored in the memory.
  • the present invention also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above method are realized.
  • the present invention mainly uses the fitting model to fit the brain tissue area in the original image, and uses the iterative update model to iteratively update the parameters of the fitting model.
  • the marked information and unmarked information generated by the first mark are fully utilized Information, use semi-supervised parameter update to learn the parameters of the fitting model, so that the distribution curve of the fitting model is constantly approaching the gray histogram of the brain tissue area, so that the blood vessel tissue distribution interval can be accurately fitted, thereby improving the brain The accuracy of blood vessel segmentation.
  • FIG. 1 is an implementation flowchart of an image processing method provided by Embodiment 1 of the present invention.
  • Figure 2 is a schematic diagram of FSL-BET processing in the first embodiment of the present invention.
  • Fig. 3 is a statistical histogram of midbrain tissue regions in Example 1 of the present invention.
  • step S108 is a detailed flowchart of step S108 in the second embodiment of the present invention.
  • FIG. 5 is a schematic structural diagram of an image processing system provided by Embodiment 3 of the present invention.
  • FIG. 6 is a schematic structural diagram of a computing device provided in Embodiment 4 of the present invention.
  • Figure 1 shows the implementation flow of the image processing method provided in the first embodiment of the present invention, which can mainly realize the accurate segmentation of cerebrovascular tissue from the image. For ease of description, only the parts related to the embodiment of the present invention are shown. As follows:
  • step S101 an original image containing brain tissue is obtained.
  • the original image may be a Time of Flight-Magnetic Resonance Angiography (TOF-MRA) image, or other imaging images, such as a computer tomography (Computed Tomography, CT) image Or Positron Emission Computed Tomography (PET) images, etc.
  • TOF-MRA Time of Flight-Magnetic Resonance Angiography
  • CT computer tomography
  • PET Positron Emission Computed Tomography
  • step S102 the brain tissue area is determined from the original image.
  • the original image contains not only the brain tissue area, but also skull, eye tissue, background noise, etc.
  • the subsequent processing is still performed on the original image, the signal-to-noise ratio of the blood vessel is reduced in disguised form, which is not conducive to improvement. Accuracy and precision of subsequent processing.
  • the brain extraction tool (FSL Brain Extraction Tool, FSL-BET) in the medical processing tool FSL can be used to increase the signal-to-noise ratio of the original image to obtain the brain tissue area and the signal-to-noise ratio increase processing It includes decranial processing, eye tissue removal, background noise removal, etc., so as to improve the signal-to-noise ratio of blood vessels, reduce calculation costs, and facilitate more accurate extraction of cerebral blood vessels.
  • step S103 a clustering algorithm is used to process the gray histogram corresponding to the brain tissue region to obtain a preliminary classification result for preliminarily distinguishing the cerebral blood vessels and background in the brain tissue region.
  • the determined brain tissue area can be roughly divided into three gray scale steps: one is cerebrospinal fluid and lateral ventricle, the second is gray matter and white matter, and the third is cerebrovascular.
  • the cerebrospinal fluid and lateral ventricles are all called background.
  • Figure 3 can clearly indicate the approximate distribution range of gray matter and white matter.
  • the left and right sides of the peak are the cerebrospinal fluid and lateral ventricle areas, as well as the cerebrovascular area.
  • the K-MEANS clustering algorithm can be used, the number of clusters is set to 3, and the initial cluster centers are 1/4 point of the peak and valley, peak and valley point, and twice the point of the peak and valley point.
  • clustering algorithms such as: K-MEDOIDS algorithm, Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm using hierarchical methods Wait.
  • step S104 the parameters of the preset fitting model are initialized according to the preliminary classification results.
  • the fitting model is used to fit the gray histogram, and the fitting model is used to simulate the distribution of cerebral blood vessels or the background distribution. Distribution function construction.
  • the preset fitting model is a finite mixture model. It is found by testing the fitting effect of various probability density functions in the brain tissue region: Gaussian Mixture Model (GMM) composed of three Gaussian distributions ) It is better to fit the gray distribution of the brain tissue area, so the preferred fitting model is GMM, that is, three Gaussian distributions are used to respectively fit the regions corresponding to the above three gray-scale steps, that is, the first Gaussian distribution simulation Cerebrospinal fluid and lateral ventricle areas, the second Gaussian distribution simulates the gray matter and white matter areas, the third Gaussian distribution simulates the cerebrovascular area, and the proportion of each type of data in the three types of data after clustering w, mean u and The variance ⁇ is used as the initial parameter of GMM.
  • GMM Gaussian Mixture Model
  • I is the gray value
  • Gi is the Gaussian distribution indicator information
  • f Gi is the Gaussian distribution function
  • f G3 (I
  • step S105 perform window width and window level transformation analysis on the original image to obtain voxels in the original image that meet the preset threshold requirements, and perform first labeling on these voxels.
  • the threshold is used to distinguish background and cerebrovascular .
  • the imtool in the Matlab tool can be called to adjust the window width and window level of the original image, and set a tentative threshold for segmentation.
  • the voxels in the original image meet a certain threshold requirement, the voxels can be first marked to indicate that they are initially judged as background. If they meet another threshold requirement, the voxels can be first marked. A marker, which indicates that it was initially judged to be a cerebral blood vessel. There are still voxels that are not first marked in the original image.
  • the parameters are iteratively updated based on the preset iterative update model.
  • the parameters that are the current iterative update target are: the marking information corresponding to the voxel that has been first marked and the first marking has not been performed.
  • the unlabeled information corresponding to the labeled voxel is constructed.
  • the unlabeled information is constructed by the class posterior probability as the update result of the previous iteration.
  • the class posterior probability is constructed by the distribution function and corresponds to the background class and the cerebrovascular class.
  • the iterative update model may be as follows:
  • I j is the gray value of the j-th pixel
  • N( ⁇ ) is the number of voxels in the corresponding area
  • D li is the data belonging to the i-th distribution component
  • k is the number of iterations
  • I j ] k is the class posterior probability of the previous iteration update result.
  • the parameter as the update target of the current iteration is constructed by: the labeled information corresponding to the voxels that have been first labeled and the unlabeled information corresponding to the voxels that have not been first labeled.
  • the construction of probability probability includes:
  • [u i] k + 1 is constructed of unlabeled information flag information, the flag information comprising: I j, I j ⁇ D li and N (D li), without tag information includes: p [G i
  • labeled information includes: I j ,I j ⁇ D li and N(D li ), while unlabeled information includes: p[G i
  • [w i] k + 1 is constructed of unlabeled information flag information, the flag information comprising: N (D li), without tag information includes: p [G i
  • I j ) is constructed by the distribution function, specifically:
  • * may take the value of G 1 , G 2 , G 3 , which is one of the three categories, and ⁇ is the general term for the parameters in each Gaussian distribution, which includes u i , ⁇ i .
  • step S107 according to the class posterior probability corresponding to the voxel, a second mark is performed on the voxel to indicate that the voxel belongs to the background class or the cerebrovascular class, and the class conditional probability is obtained.
  • the body can be The voxel is judged to be the cerebrovascular class (L v ), otherwise it is the background class (L B ), so that the voxel can be second marked to indicate whether the voxel belongs to the cerebrovascular class or the background class, forming an initial label field L 0 .
  • l i ) is f Gi (I
  • the blood vessel feature map is obtained from the original image.
  • the blood vessel feature map can be formed by the gray value of each point of the original image.
  • step S109 the marker field obtained from the second marker and the blood vessel feature map are combined to construct a Markov random field.
  • the domain system of voxels can be defined in the brain tissue region, and the domain system can be a 6-neighbor system;
  • the two-point potential clique function is constructed from the marker field obtained by the second marker and the blood vessel feature map, where:
  • N i 6 neighborhood system of voxels in the i point.
  • ⁇ 1, ⁇ 2 are proportional coefficients
  • V f (i) is the blood vessel score at the i-th voxel in the blood vessel feature map.
  • step S110 the class prior probability P(l i ) is constructed from the Markov random field, which can be specifically:
  • k is the traversal index
  • l k is the label of the k-th random voxel.
  • step S111 based on the class conditional probability and the class prior probability, in the iterative conditional mode, the marker field is updated by maximizing the corresponding posterior probability to obtain the cerebrovascular segmentation result.
  • l i ) can be used, which is equivalent to the aforementioned P(I
  • the fitting model is mainly used to fit the brain tissue area in the original image, and the iterative update model is used to iteratively update the parameters of the fitting model.
  • the label information generated by the first label is fully utilized
  • unlabeled information use semi-supervised parameter update to learn the parameters of the fitting model, so that the distribution curve of the fitting model is constantly approaching the gray histogram of the brain tissue area, so that the distribution interval of the blood vessel tissue can be accurately fitted, thereby Improved the accuracy of cerebrovascular segmentation.
  • the calculation process is performed on the brain tissue area after the skull is removed, which greatly eliminates many irrelevant tissues, improves the signal-to-noise ratio of blood vessels, and reduces the computational cost.
  • this embodiment further provides the following content:
  • step S108 specifically includes:
  • step S401 a multi-scale filter enhancement process is performed on the original image to obtain a primary feature map composed of multi-scale blood vessel enhancement values.
  • the multi-scale filtering technique based on the Hessian matrix can be used to enhance the tubular target in the data.
  • the original image data I and the multi-scale Gaussian kernel are convolved.
  • the point i with coordinates (x, y, z) has a gray value of I ⁇ (i )
  • the corresponding Hessian matrix is calculated as follows:
  • v 3 is redefined at each convolution scale as:
  • ⁇ is a threshold between 0 and 1
  • ⁇ is the filter scale.
  • the enhanced response is calculated by the following vessel response function:
  • step S402 under the threshold constraint condition constructed by the intracranial proportion of blood vessels, the multi-scale blood vessel enhancement value is converted into blood vessel feature values with the significance of blood vessel prediction probability, and the blood vessel feature map is composed of blood vessel feature values.
  • the multi-scale vessel enhancement value v can be transformed as follows:
  • V p ′ represents the collection of vascular enhancement values in the brain tissue region
  • ⁇ is the intracranial proportion of blood vessels (this value is equal to the Gaussian distribution weight w 3 corresponding to the blood vessels in GMM
  • ⁇ ( ⁇ ) is determined by the intracranial proportion of blood vessels.
  • the constructed threshold, the blood vessel characteristic value is V f .
  • the result of multi-scale blood vessel enhancement can be transformed to obtain the characteristic value of the blood vessel so that it has the significance of blood vessel prediction probability and embed it in the Markov random field, which is beneficial to better optimize the segmentation of GMM As a result, high-quality cerebrovascular segmentation is achieved.
  • FIG. 5 shows the structure of the image processing system provided by the third embodiment of the present invention. For ease of description, only the parts related to the embodiment of the present invention are shown, including:
  • the preprocessing unit 501 is configured to obtain an original image containing brain tissue; determine the brain tissue area from the original image;
  • the initialization unit 502 is configured to use a clustering algorithm to process the gray histogram corresponding to the brain tissue area to obtain a preliminary classification result for preliminarily distinguishing the cerebrovascular and background in the brain tissue area; According to the preliminary classification result, the parameters of the preset fitting model are initialized, the fitting model is used to fit the gray histogram, and the fitting model is used to simulate the distribution of the cerebral blood vessels or Constructing the distribution function of the background distribution; performing window width and window level transformation analysis on the original image to obtain voxels in the original image that meet the preset threshold requirements, and perform first marking on these voxels, The threshold is used to distinguish the background and the cerebral blood vessel;
  • the feature map calculation unit 503 is configured to obtain a blood vessel feature map from the original image.
  • the segmentation unit 504 is configured to iteratively update the parameters based on a preset iterative update model.
  • the parameter that is the current iterative update target is composed of: the voxel that has been first marked The corresponding labeled information and the unlabeled information corresponding to the voxels that are not first labeled are constructed.
  • the unlabeled information is constructed from the class posterior probability as the update result of the previous iteration, and the class posterior probability is constructed by the
  • the distribution function is constructed and corresponds to the background class and the cerebrovascular class; according to the posterior probability of the class corresponding to the voxel, a second label is performed on the voxel to indicate that the voxel belongs to the background class or Describe the cerebrovascular class and obtain the class conditional probability; combine the marker field obtained by the second marker and the blood vessel feature map to construct a Markov random field; construct a class prior probability from the Markov random field; Based on the class conditional probability and the class prior probability, in the iterative conditional mode, the marker field is updated by maximizing the corresponding posterior probability to obtain the cerebrovascular segmentation result.
  • each unit of the image processing system can be implemented by a corresponding hardware or software unit.
  • Each unit can be an independent software and hardware unit, or can be integrated into a software and hardware unit.
  • the present invention is not limited here. .
  • FIG. 6 shows the structure of the computing device provided in the fourth embodiment of the present invention. For ease of description, only the parts related to the embodiment of the present invention are shown.
  • the computing device in the embodiment of the present invention includes a processor 601 and a memory 602.
  • the processor 601 implements the steps in the foregoing method embodiments when executing the computer program 603 stored in the memory 602, such as steps S101 to S111 shown in FIG.
  • the processor 601 executes the computer program 603
  • the functions of the units in the foregoing device embodiments such as the functions of the units 501 to 504 shown in FIG. 5, are implemented.
  • the computing device in the embodiment of the present invention may be a single computer, or a computer network, or a single processing chip, or a chipset.
  • the processor 601 in the computing device executes the computer program 603 to implement the foregoing methods, reference may be made to the description of the foregoing method embodiments, which will not be repeated here.
  • a computer-readable storage medium stores a computer program, and the computer program implements the steps in the foregoing method embodiments when executed by a processor, for example, as shown in FIG. 1 Steps S101 to S111 are shown. Or, when the computer program is executed by the processor, the functions of the units in the foregoing system embodiments, such as the functions of the units 501 to 504 shown in FIG. 5, are realized.
  • the computer-readable storage medium in the embodiment of the present invention may include any entity or device or recording medium capable of carrying computer program code, such as ROM/RAM, magnetic disk, optical disk, flash memory and other memories.
  • the present invention has been verified on the public data set MIDAS, which contains 109 sets of TOF-MRA clinical data.
  • MIDAS public data set
  • Table 1 The results of the four evaluation measures of the three methods are given.
  • TP, FP, TN, and FN are true cases, false positive cases, true negative cases, and false negative cases, respectively.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)
  • Image Analysis (AREA)

Abstract

本发明适用计算机技术领域，提供了一种图像处理方法、***、计算设备及存储介质，该方法主要利用拟合模型对原始图像中脑组织区域进行拟合，利用迭代更新模型对拟合模型的参数进行迭代更新，迭代更新过程中，充分利用第一标记所产生的标记信息和未标记信息，使用半监督参数更新来学习拟合模型的参数，使得拟合模型的分布曲线不断逼近脑组织区域的灰度直方图，这样，能对血管组织分布区间进行精确拟合，从而提高了脑血管分割精度。

Description

图像处理方法、***、计算设备及存储介质 技术领域
本发明属于图像处理技术领域，尤其涉及一种图像处理方法、***、计算设备及存储介质。
背景技术
中国在先专利(申请公布号CN109102511A)涉及一种脑血管分割方法、***及电子设备，其主要实现：对包含脑组织的原始图像进行多尺度滤波增强处理，得到增强后的血管特征图像以及对应的方向向量场；建立有限混合模型并估计有限混合模型参数，得到类条件概率；计算原始图像的初始标记场，并由初始标记场与方向向量场构成马尔科夫随机场；进而得到类先验概率；基于类先验概率和类条件概率，通过最大化后验概率和迭代条件模式，得到脑血管分割结果。
上述在先专利虽然能够提取有效的血管候选空间，提取出低对比度下的脑血管结构，但由于在有限混合模型参数迭代过程中，采用的是传统无监督的期望最大化算法，从而不利于血管组织分布区间的精确拟合，最终使得脑血管分割精度无法得到有效提高。另外，直接对原始数据进行计算处理，无法剔除很多不相关的组织对血管的影响，同时也加大了计算代价。
发明内容
本发明的目的在于提供一种图像处理方法、***、计算设备及存储介质，旨在解决现有技术所存在的、因血管组织分布区间无法精确拟合而导致的脑血管分割精度无法得到有效提高的问题。
一方面，本发明提供了一种图像处理方法，所述方法包括下述步骤：
获得包含脑组织的原始图像；
从所述原始图像中确定脑组织区域；
采用聚类算法，对所述脑组织区域所对应的灰度直方图进行处理，得到用于初步区分所述脑组织区域中的脑血管及背景的初步分类结果；
根据所述初步分类结果，对预设的拟合模型的参数进行初始化，所述拟合模型用于对所述灰度直方图进行拟合，所述拟合模型由用于模拟所述脑血管分布或所述背景分布的分布函数构建；
对所述原始图像进行窗宽窗位变换分析，得到满足预设的阈值要求的所述原始图像中的体素，并对该些体素进行第一标记，所述阈值用于区分所述背景及所述脑血管；
基于预设的迭代更新模型，对所述参数进行迭代更新，所述迭代更新模型中，作为当前迭代更新目标的所述参数由：已进行第一标记的所述体素对应的标记信息以及未进行第一标记的所述体素对应的未标记信息构建，所述未标记信息由作为前一迭代更新结果的类后验概率构建，所述类后验概率由所述分布函数构建且对应于背景类及脑血管类；
根据所述体素所对应的所述类后验概率，对所述体素进行第二标记以指示所述体素属于所述背景类或所述脑血管类，并得到类条件概率；
从所述原始图像得到血管特征图；
将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场；
由所述马尔科夫随机场构建类先验概率；
基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果。
进一步的，从所述原始图像得到血管特征图，具体包括：
对所述原始图像进行多尺度滤波增强处理，得到由多尺度血管增强值构成的初级特征图；
在由血管颅内占比所构建的阈值约束条件下，将所述多尺度血管增强值转换为具有血管预测概率意义的血管特征值，所述血管特征图由所述血管特征值构成。
进一步的，从所述原始图像中确定脑组织区域具体为：
对所述原始图像进行信噪比提升处理，得到所述脑组织区域，所述信噪比提升处理包含去颅骨处理。
进一步的，所述聚类算法为K均值聚类算法。
进一步的，所述拟合模型为高斯混合模型，所述分布函数为高斯分布函数。
进一步的，将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场，具体为：
在所述脑组织区域中定义所述体素的领域***；
基于所述领域***，利用双点势团函数得到所述体素的能量表示，从而构建所述马尔科夫随机场，所述双点势团函数由所述第二标记所得的标记场及所述血管特征图构建。
进一步的，基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果，具体为：
利用所求得的所述类条件概率及所述类先验概率，利用预先构建的迭代条件模型，在第N次迭代时，在已知观测数据和标记场的条件下，利用贝叶斯准则计算所述后验概率，并最大化所述后验概率，以更新所述体素的第二标记结果，从而更新所述标记场，得到所述脑血管分割结果，其中，N为整数。
另一方面，本发明提供了一种图像处理***，所述***包括：
预处理单元，用于获得包含脑组织的原始图像；从所述原始图像中确定脑组织区域；
初始化单元，用于采用聚类算法，对所述脑组织区域所对应的灰度直方图进行处理，得到用于初步区分所述脑组织区域中的脑血管及背景的初步分类结果；根据所述初步分类结果，对预设的拟合模型的参数进行初始化，所述拟合 模型用于对所述灰度直方图进行拟合，所述拟合模型由用于模拟所述脑血管分布或所述背景分布的分布函数构建；对所述原始图像进行窗宽窗位变换分析，得到满足预设的阈值要求的所述原始图像中的体素，并对该些体素进行第一标记，所述阈值用于区分所述背景及所述脑血管；
特征图计算单元，用于从所述原始图像得到血管特征图；以及，
分割单元，用于基于预设的迭代更新模型，对所述参数进行迭代更新，所述迭代更新模型中，作为当前迭代更新目标的所述参数由：已进行第一标记的所述体素对应的标记信息以及未进行第一标记的所述体素对应的未标记信息构建，所述未标记信息由作为前一迭代更新结果的类后验概率构建，所述类后验概率由所述分布函数构建且对应于背景类及脑血管类；根据所述体素所对应的所述类后验概率，对所述体素进行第二标记以指示所述体素属于所述背景类或所述脑血管类，并得到类条件概率；将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场；由所述马尔科夫随机场构建类先验概率；基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果。
另一方面，本发明还提供了一种计算设备，包括存储器及处理器，所述处理器执行所述存储器中存储的计算机程序时实现如上述方法中的步骤。
另一方面，本发明还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现如上述方法中的步骤。
本发明主要利用拟合模型对原始图像中脑组织区域进行拟合，利用迭代更新模型对拟合模型的参数进行迭代更新，迭代更新过程中，充分利用第一标记所产生的标记信息和未标记信息，使用半监督参数更新来学习拟合模型的参数，使得拟合模型的分布曲线不断逼近脑组织区域的灰度直方图，这样，能对血管组织分布区间进行精确拟合，从而提高了脑血管分割精度。
附图说明
图1是本发明实施例一提供的图像处理方法的实现流程图；
图2是本发明实施例一中FSL-BET处理示意图；
图3是本发明实施例一中脑组织区域统计直方图；
图4是本发明实施例二中步骤S108的细化流程图；
图5是本发明实施例三提供的图像处理***的结构示意图；
图6是本发明实施例四提供的计算设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
以下结合具体实施例对本发明的具体实现进行详细描述：
实施例一：
图1示出了本发明实施例一提供的图像处理方法的实现流程，主要能实现从图像中精确分割出脑血管组织，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S101中，获得包含脑组织的原始图像。
本实施例中，原始图像可以为时间飞跃磁共振血管成像(Time of Flight-Magnetic Resonance Angiography，TOF-MRA)图像，也可以为其他成像图像，例如：电子计算机断层扫描(Computed Tomography，CT)图像或正电子发射型计算机断层显像(Positron Emission Computed Tomography，PET)图像等。
在步骤S102中，从原始图像中确定脑组织区域。
本实施例中，由于原始图像中不仅包含脑组织区域，而且还包含颅骨、眼部组织、背景噪声等，如果后续处理仍针对原始图像进行，则变相降低了血管 的信噪比，不利于提升后续处理的准确度和精度。因此，如图2所示，可利用医学处理工具FSL中的脑提取工具(FSL Brain Extraction Tool，FSL-BET)，对原始图像进行信噪比提升处理，得到脑组织区域，信噪比提升处理中包含去颅骨处理，还可以包含去眼部组织、去背景噪声等处理，从而提高血管的信噪比，减少计算代价，有利于更加精准地提取脑血管。
在步骤S103中，采用聚类算法，对脑组织区域所对应的灰度直方图进行处理，得到用于初步区分脑组织区域中的脑血管及背景的初步分类结果。
本实施例中，针对所确定的脑组织区域，可以大致分为三个灰度阶梯：其一为脑脊液和侧脑室，其二为灰质和白质，其三为脑血管。在本申请中，由于目的是分割脑血管，那么，脑脊液和侧脑室，以及灰质和白质，均被称为背景。通过统计直方图分析，脑组织区域的统计直方图可如图3所示，由于灰质和白质占比最大，图3中可明显指示出灰质和白质大致的分布范围，此外，根据经验认为在其波峰左右两侧分别是脑脊液和侧脑室区域，以及脑血管区域。进而可采用K均值(K-MEANS)聚类算法，设置聚类个数为3，初始聚类中心分别为峰谷的1/4点、峰谷点，以及峰谷点的2倍点。
当然，在其他应用示例中，还可以采用其他聚类算法，例如：K中心点(K-MEDOIDS)算法、利用层次方法的平衡迭代规约和聚类(Balanced Iterative Reducing and Clustering using Hierarchies，BIRCH)算法等。
在步骤S104中，根据初步分类结果，对预设的拟合模型的参数进行初始化，拟合模型用于对灰度直方图进行拟合，拟合模型由用于模拟脑血管分布或背景分布的分布函数构建。
本实施例中，预设的拟合模型为有限混合模型，由于通过测试各种概率密度函数在脑组织区域的拟合效果发现：由三个高斯分布组成的高斯混合模型(Gaussian Mixture Model，GMM)对脑组织区域灰度分布拟合效果较佳，那么优选拟合模型为GMM，即采用三个高斯分布分别对上述三个灰度阶梯对应的区域进行拟合，即第一个高斯分布模拟脑脊液和侧脑室区域，第二个高斯分布 模拟灰质和白质区域，第三个高斯分布模拟脑血管区域，并利用聚类后的三类数据中的每一类数据的占比w、均值u和方差σ作为GMM的初始参数。
对应的GMM表达式如下：
Figure PCTCN2019077344-appb-000001
其中，I为灰度值，i为聚类数(i＝1,2,3)，Gi为高斯分布指示信息，f Gi(I|u i,σ i)为高斯分布函数，f G3(I|u 3,σ 3)对应于脑血管类别。
在步骤S105中，对原始图像进行窗宽窗位变换分析，得到满足预设的阈值要求的原始图像中的体素，并对该些体素进行第一标记，阈值用于区分背景及脑血管。
本实施例中，可调用Matlab工具中的imtool，对原始图像实现调节数据的窗宽窗位，并设置试探性的阈值进行分割。
原始图像中的体素若满足某一阈值要求，则可对该些体素进行第一标记，表征其被初判为背景，若满足某另一阈值要求，则可对该些体素进行第一标记，表征其被初判为脑血管。原始图像中还存在未被第一标记的体素。
相应的，脑组织区域的所有体素D可被划分为已进行第一标记的体素集合D l以及未进行第一标记的体素集合D u，即D＝D u∪D l。
在步骤S106中，基于预设的迭代更新模型，对参数进行迭代更新，迭代更新模型中，作为当前迭代更新目标的参数由：已进行第一标记的体素对应的标记信息以及未进行第一标记的体素对应的未标记信息构建，未标记信息由作为前一迭代更新结果的类后验概率构建，类后验概率由分布函数构建且对应于背景类及脑血管类。
本实施例中，迭代更新模型可如下：
Figure PCTCN2019077344-appb-000002
Figure PCTCN2019077344-appb-000003
Figure PCTCN2019077344-appb-000004
其中，I j为第j个像素的灰度值，N(·)为相应区域中体素的数量，D li为属于第i个分布成分的数据，k为迭代次数，p[G i|I j] k作为前一迭代更新结果的类后验概率。
作为当前迭代更新目标的参数由：已进行第一标记的体素对应的标记信息以及未进行第一标记的体素对应的未标记信息构建，未标记信息由作为前一迭代更新结果的类后验概率构建，具体包含：
[u i] k+1由标记信息与未标记信息构建，标记信息包含：I j,I j∈D li以及N(D li)，而未标记信息包含：p[G i|I j] k、I j,I j∈D u。
Figure PCTCN2019077344-appb-000005
由标记信息与未标记信息构建，标记信息包含：I j,I j∈D li以及N(D li)，而未标记信息包含：p[G i|I j] k、I j,I j∈D u。当然，
Figure PCTCN2019077344-appb-000006
的表达中，还包含相应的均值[u i] k。
[w i] k+1由标记信息与未标记信息构建，标记信息包含：N(D li)，而未标记信息包含：p[G i|I j] k,I j∈D u。
类后验概率p(*|I j)由分布函数构建，具体为：
Figure PCTCN2019077344-appb-000007
其中，*可能取值为G 1,G 2,G 3，即三个类别中的一个，φ是每个高斯分布中 参数的统称，即包含u i,σ i。
在步骤S107中，根据体素所对应的类后验概率，对体素进行第二标记以指示体素属于背景类或脑血管类，并得到类条件概率。
本实施例中，根据贝叶斯判别准则，对于TOF-MRA中的每一个体素，当且仅当脑血管类的类后验概率大于任意背景类的类后验概率时，可将该体素判断为脑血管类(L v)，否则为背景类(L B)，从而可以对体素进行第二标记以指示体素属于脑血管类，还是属于背景类，形成初始标记场L 0。
类条件概率P(I|l i)即f Gi(I|u i,σ i)，其中，l i为类标签。
在步骤S108中，从原始图像得到血管特征图，例如：可由原始图像各个点的灰度值构成血管特征图。
在步骤S109中，将第二标记所得的标记场及血管特征图结合，构建马尔科夫随机场。
本实施例中，可在脑组织区域中定义体素的领域***，该领域***可为6邻域***；
基于领域***，利用双点势团函数
Figure PCTCN2019077344-appb-000008
得到所述体素的能量表示U(l i)，从而构建马尔科夫随机场，双点势团函数由第二标记所得的标记场及血管特征图构建，其中：
Figure PCTCN2019077344-appb-000009
其中，N i为体素i的6邻域***中的点。
Figure PCTCN2019077344-appb-000010
来源于两部分，一部分来源于第二标记所得的标记场L 0，另一部分来源于血管特征图V f：
Figure PCTCN2019077344-appb-000011
其中，α 1,α 2为比例系数，
Figure PCTCN2019077344-appb-000012
Figure PCTCN2019077344-appb-000013
其中，V f(i)为血管特征图中第i个体素处的血管得分。
在步骤S110中，由马尔科夫随机场构建类先验概率P(l i)，具体可为：
Figure PCTCN2019077344-appb-000014
Figure PCTCN2019077344-appb-000015
其中，k为遍历下标，l k是第k个随机体素的标签。
在步骤S111中，基于类条件概率及类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新标记场，得到脑血管分割结果。
在本实施例中，可利用所求得的类条件概率P(Y i|l i)即等价于前述P(I|l i)及类先验概率P(l i)，利用预先构建的迭代条件模型，在第N次迭代时，在已知观测数据Y和标记场L N的条件下，利用贝叶斯准则计算后验概率P(l i|Y i)，并最大化后验概率P(l i|Y i)，以更新体素的第二标记结果，从而更新标记场，得到脑血管分割结果，其中，N为整数，其中，最大化后验概率P(l i|Y i)，以更新体素的第二标记结果l new具体可为：
Figure PCTCN2019077344-appb-000016
实施本实施例，主要利用拟合模型对原始图像中脑组织区域进行拟合，利用迭代更新模型对拟合模型的参数进行迭代更新，迭代更新过程中，充分利用第一标记所产生的标记信息和未标记信息，使用半监督参数更新来学习拟合模型的参数，使得拟合模型的分布曲线不断逼近脑组织区域的灰度直方图，这样， 能对血管组织分布区间进行精确拟合，从而提高了脑血管分割精度。另外，是对去颅骨等之后的脑组织区域进行计算处理，大大剔除了很多不相关的组织，提高了血管的信噪比，也减少了计算代价。
实施例二：
本实施例在实施例一基础上，进一步提供了如下内容：
如图4所示，本实施例中，步骤S108具体包括：
在步骤S401中，对原始图像进行多尺度滤波增强处理，得到由多尺度血管增强值构成的初级特征图。
本实施例中，首先可基于Hessian矩阵的多尺度滤波技术，增强数据中的管状目标。为了得到不同尺度的血管特性，将原始图像数据I与多尺度高斯核进行卷积操作，在尺度σ下，坐标为(x,y,z)的点i，其灰度值为I σ(i)，对应的Hessian矩阵计算如下：
Figure PCTCN2019077344-appb-000017
对H(i,σ)进行特征值分解，可以得到3个特征值(v 1,v 2,v 3)，其中，|v 1|≤|v 2|≤|v 3|。一般情况下，血管点对应的Hessian矩阵的特征值满足以下关系式：
v 2≈v 3
|v 2,3|>>|v 1|……(13)
为了改善低对比度的情况，对v 3在每个卷积尺度下进行重新定义为：
Figure PCTCN2019077344-appb-000018
其中，τ是一个0到1之间的阈值，σ是滤波尺度。
增强响应如下血管响应函数计算得出：
Figure PCTCN2019077344-appb-000019
由此得到了多尺度血管增强后的结果，即多尺度血管增强值v，对应于式(15)中的V p。
在步骤S402中，在由血管颅内占比所构建的阈值约束条件下，将多尺度血管增强值转换为具有血管预测概率意义的血管特征值，血管特征图由血管特征值构成。
本实施例中，可对多尺度血管增强值v进行如下变换：
Figure PCTCN2019077344-appb-000020
Figure PCTCN2019077344-appb-000021
其中，V p′表示脑组织区域的血管增强值的集合，β是血管颅内占比(该值等于GMM中血管对应的高斯分布权重w 3，θ(β)为由血管颅内占比所构建的阈值，血管特征值为V f。
实施本实施例，可将多尺度血管增强后的结果，通过变换得到血管特征值，使其具有血管预测概率意义，并将其嵌入马尔科夫随机场，从而有利于更好地优化GMM的分割结果，实现优质的脑血管分割。
实施例三：
图5示出了本发明实施例三提供的图像处理***的结构，为了便于说明，仅示出了与本发明实施例相关的部分，其中包括：
预处理单元501，用于获得包含脑组织的原始图像；从所述原始图像中确定脑组织区域；
初始化单元502，用于采用聚类算法，对所述脑组织区域所对应的灰度直方图进行处理，得到用于初步区分所述脑组织区域中的脑血管及背景的初步分类结果；根据所述初步分类结果，对预设的拟合模型的参数进行初始化，所述拟合模型用于对所述灰度直方图进行拟合，所述拟合模型由用于模拟所述脑血管分布或所述背景分布的分布函数构建；对所述原始图像进行窗宽窗位变换分析，得到满足预设的阈值要求的所述原始图像中的体素，并对该些体素进行第一标记，所述阈值用于区分所述背景及所述脑血管；
特征图计算单元503，用于从所述原始图像得到血管特征图；以及，
分割单元504，用于基于预设的迭代更新模型，对所述参数进行迭代更新，所述迭代更新模型中，作为当前迭代更新目标的所述参数由：已进行第一标记的所述体素对应的标记信息以及未进行第一标记的所述体素对应的未标记信息构建，所述未标记信息由作为前一迭代更新结果的类后验概率构建，所述类后验概率由所述分布函数构建且对应于背景类及脑血管类；根据所述体素所对应的所述类后验概率，对所述体素进行第二标记以指示所述体素属于所述背景类或所述脑血管类，并得到类条件概率；将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场；由所述马尔科夫随机场构建类先验概率；基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果。
在本发明实施例中，图像处理***的各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。
图像处理***的各单元在实现上述图像处理方法时实现的步骤，可参考前 述方法实施例的描述，在此不再赘述。
实施例四：
图6示出了本发明实施例四提供的计算设备的结构，为了便于说明，仅示出了与本发明实施例相关的部分。
本发明实施例的计算设备包括处理器601及存储器602，处理器601执行存储器602中存储的计算机程序603时实现上述各个方法实施例中的步骤，例如图1所示的步骤S101至S111。或者，处理器601执行计算机程序603时实现上述各装置实施例中各单元的功能，例如图5所示单元501至504的功能。
本发明实施例的计算设备可以为单个计算机，也可以为计算机组网，也可以为单个处理芯片，也可以为芯片组等。该计算设备中处理器601执行计算机程序603时实现上述各方法时实现的步骤，可参考前述方法实施例的描述，在此不再赘述。
实施例五：
在本发明实施例中，提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序被处理器执行时实现上述各方法实施例中的步骤，例如，图1所示的步骤S101至S111。或者，该计算机程序被处理器执行时实现上述各***实施例中各单元的功能，例如图5所示单元501至504的功能。
本发明实施例的计算机可读存储介质可以包括能够携带计算机程序代码的任何实体或装置、记录介质，例如，ROM/RAM、磁盘、光盘、闪存等存储器。
本发明在公开数据集MIDAS上进行了验证，其中包含109套TOF-MRA临床数据，为了定量分析本算法的有效性，从中抽取了20套数据，合作医院的医生对其进行了标记，表1给出三种方法的四种评估测度的结果。
表1三种方法的比较
Figure PCTCN2019077344-appb-000022
Figure PCTCN2019077344-appb-000023
注：
Figure PCTCN2019077344-appb-000024
其中TP、FP、TN、FN分别是真正例、假正例、真反例、假反例。
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、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种图像处理方法，其特征在于，所述方法包括下述步骤：
    获得包含脑组织的原始图像；
    从所述原始图像中确定脑组织区域；
    采用聚类算法，对所述脑组织区域所对应的灰度直方图进行处理，得到用于初步区分所述脑组织区域中的脑血管及背景的初步分类结果；
    根据所述初步分类结果，对预设的拟合模型的参数进行初始化，所述拟合模型用于对所述灰度直方图进行拟合，所述拟合模型由用于模拟所述脑血管分布或所述背景分布的分布函数构建；
    对所述原始图像进行窗宽窗位变换分析，得到满足预设的阈值要求的所述原始图像中的体素，并对该些体素进行第一标记，所述阈值用于区分所述背景及所述脑血管；
    基于预设的迭代更新模型，对所述参数进行迭代更新，所述迭代更新模型中，作为当前迭代更新目标的所述参数由：已进行第一标记的所述体素对应的标记信息以及未进行第一标记的所述体素对应的未标记信息构建，所述未标记信息由作为前一迭代更新结果的类后验概率构建，所述类后验概率由所述分布函数构建且对应于背景类及脑血管类；
    根据所述体素所对应的所述类后验概率，对所述体素进行第二标记以指示所述体素属于所述背景类或所述脑血管类，并得到类条件概率；
    从所述原始图像得到血管特征图；
    将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场；
    由所述马尔科夫随机场构建类先验概率；
    基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果。
  2. 如权利要求1所述的方法，其特征在于，从所述原始图像得到血管特征 图，具体包括：
    对所述原始图像进行多尺度滤波增强处理，得到由多尺度血管增强值构成的初级特征图；
    在由血管颅内占比所构建的阈值约束条件下，将所述多尺度血管增强值转换为具有血管预测概率意义的血管特征值，所述血管特征图由所述血管特征值构成。
  3. 如权利要求1所述的方法，其特征在于，从所述原始图像中确定脑组织区域具体为：
    对所述原始图像进行信噪比提升处理，得到所述脑组织区域，所述信噪比提升处理包含去颅骨处理。
  4. 如权利要求1所述的方法，其特征在于，所述聚类算法为K均值聚类算法。
  5. 如权利要求1所述的方法，其特征在于，所述拟合模型为高斯混合模型，所述分布函数为高斯分布函数。
  6. 如权利要求2所述的方法，其特征在于，将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场，具体为：
    在所述脑组织区域中定义所述体素的领域***；
    基于所述领域***，利用双点势团函数得到所述体素的能量表示，从而构建所述马尔科夫随机场，所述双点势团函数由所述第二标记所得的标记场及所述血管特征图构建。
  7. 如权利要求1所述的方法，其特征在于，基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果，具体为：
    利用所求得的所述类条件概率及所述类先验概率，利用预先构建的迭代条件模型，在第N次迭代时，在已知观测数据和标记场的条件下，利用贝叶斯准则计算所述后验概率，并最大化所述后验概率，以更新所述体素的第二标记结 果，从而更新所述标记场，得到所述脑血管分割结果，其中，N为整数。
  8. 一种图像处理***，其特征在于，所述***包括：
    预处理单元，用于获得包含脑组织的原始图像；从所述原始图像中确定脑组织区域；
    初始化单元，用于采用聚类算法，对所述脑组织区域所对应的灰度直方图进行处理，得到用于初步区分所述脑组织区域中的脑血管及背景的初步分类结果；根据所述初步分类结果，对预设的拟合模型的参数进行初始化，所述拟合模型用于对所述灰度直方图进行拟合，所述拟合模型由用于模拟所述脑血管分布或所述背景分布的分布函数构建；对所述原始图像进行窗宽窗位变换分析，得到满足预设的阈值要求的所述原始图像中的体素，并对该些体素进行第一标记，所述阈值用于区分所述背景及所述脑血管；
    特征图计算单元，用于从所述原始图像得到血管特征图；以及，
    分割单元，用于基于预设的迭代更新模型，对所述参数进行迭代更新，所述迭代更新模型中，作为当前迭代更新目标的所述参数由：已进行第一标记的所述体素对应的标记信息以及未进行第一标记的所述体素对应的未标记信息构建，所述未标记信息由作为前一迭代更新结果的类后验概率构建，所述类后验概率由所述分布函数构建且对应于背景类及脑血管类；根据所述体素所对应的所述类后验概率，对所述体素进行第二标记以指示所述体素属于所述背景类或所述脑血管类，并得到类条件概率；将所述第二标记所得的标记场及所述血管特征图结合，构建马尔科夫随机场；由所述马尔科夫随机场构建类先验概率；基于所述类条件概率及所述类先验概率，在迭代条件模式下，通过最大化相应的后验概率，更新所述标记场，得到脑血管分割结果。
  9. 一种计算设备，包括存储器及处理器，其特征在于，所述处理器执行所述存储器中存储的计算机程序时实现如权利要求1至7任一项所述方法中的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程 序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至7任一项所述方法中的步骤。
PCT/CN2019/077344 2019-03-07 2019-03-07 图像处理方法、***、计算设备及存储介质 WO2020177126A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/077344 WO2020177126A1 (zh) 2019-03-07 2019-03-07 图像处理方法、***、计算设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/077344 WO2020177126A1 (zh) 2019-03-07 2019-03-07 图像处理方法、***、计算设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020177126A1 true WO2020177126A1 (zh) 2020-09-10

Family

ID=72338413

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/077344 WO2020177126A1 (zh) 2019-03-07 2019-03-07 图像处理方法、***、计算设备及存储介质

Country Status (1)

Country Link
WO (1) WO2020177126A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112508888A (zh) * 2020-11-26 2021-03-16 中国科学院苏州生物医学工程技术研究所 面向医学影像的脑动脉快速自动分割方法及***
CN112733825A (zh) * 2020-12-18 2021-04-30 北京城市排水集团有限责任公司 一种摆臂雷达管道图像病害空间提取方法
CN113936012A (zh) * 2021-10-21 2022-01-14 东南大学 一种多尺度超像素空间共生混合模型的图像分割方法
CN114119626A (zh) * 2021-10-18 2022-03-01 北京理工大学 一种基于统计学模型和多尺度滤波的脑血管图像分割方法
CN114742784A (zh) * 2022-03-31 2022-07-12 精诚工坊电子集成技术（北京）有限公司 皮肤图像红血丝标记方法、评价方法及其***
CN115661467A (zh) * 2022-12-28 2023-01-31 中国科学院自动化研究所 脑血管图像分割方法、装置、电子设备及存储介质
CN116246779A (zh) * 2023-05-10 2023-06-09 潍坊护理职业学院 基于用户影像数据的牙科诊疗方案生成方法及***
CN116342588B (zh) * 2023-05-22 2023-08-11 徕兄健康科技(威海)有限责任公司 一种脑血管图像增强方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009052562A1 (en) * 2007-10-23 2009-04-30 Commonwealth Scientific And Industrial Research Organisation Automatic segmentation of articular cartilage in mr images
CN102763135A (zh) * 2009-12-23 2012-10-31 通用电气公司 用于自动分割和时间跟踪的方法
CN102800088A (zh) * 2012-06-28 2012-11-28 华中科技大学 超声颈动脉斑块自动分割方法
CN105551041A (zh) * 2015-12-15 2016-05-04 中国科学院深圳先进技术研究院 普适的血管分割方法及***
US20180322369A1 (en) * 2016-06-08 2018-11-08 Gopro, Inc. Combining independent solutions to an image or video processing task
CN109102511A (zh) * 2018-07-06 2018-12-28 深圳先进技术研究院 一种脑血管分割方法、***及电子设备
CN109300137A (zh) * 2018-09-20 2019-02-01 北京航空航天大学 一种多曲面估计区间二型模糊聚类磁共振脑图像分割方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009052562A1 (en) * 2007-10-23 2009-04-30 Commonwealth Scientific And Industrial Research Organisation Automatic segmentation of articular cartilage in mr images
CN102763135A (zh) * 2009-12-23 2012-10-31 通用电气公司 用于自动分割和时间跟踪的方法
CN102800088A (zh) * 2012-06-28 2012-11-28 华中科技大学 超声颈动脉斑块自动分割方法
CN105551041A (zh) * 2015-12-15 2016-05-04 中国科学院深圳先进技术研究院 普适的血管分割方法及***
US20180322369A1 (en) * 2016-06-08 2018-11-08 Gopro, Inc. Combining independent solutions to an image or video processing task
CN109102511A (zh) * 2018-07-06 2018-12-28 深圳先进技术研究院 一种脑血管分割方法、***及电子设备
CN109300137A (zh) * 2018-09-20 2019-02-01 北京航空航天大学 一种多曲面估计区间二型模糊聚类磁共振脑图像分割方法

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112508888A (zh) * 2020-11-26 2021-03-16 中国科学院苏州生物医学工程技术研究所 面向医学影像的脑动脉快速自动分割方法及***
CN112508888B (zh) * 2020-11-26 2023-12-19 中国科学院苏州生物医学工程技术研究所 面向医学影像的脑动脉快速自动分割方法及***
CN112733825A (zh) * 2020-12-18 2021-04-30 北京城市排水集团有限责任公司 一种摆臂雷达管道图像病害空间提取方法
CN114119626A (zh) * 2021-10-18 2022-03-01 北京理工大学 一种基于统计学模型和多尺度滤波的脑血管图像分割方法
CN113936012A (zh) * 2021-10-21 2022-01-14 东南大学 一种多尺度超像素空间共生混合模型的图像分割方法
CN114742784A (zh) * 2022-03-31 2022-07-12 精诚工坊电子集成技术（北京）有限公司 皮肤图像红血丝标记方法、评价方法及其***
CN115661467A (zh) * 2022-12-28 2023-01-31 中国科学院自动化研究所 脑血管图像分割方法、装置、电子设备及存储介质
CN115661467B (zh) * 2022-12-28 2023-05-05 中国科学院自动化研究所 脑血管图像分割方法、装置、电子设备及存储介质
CN116246779A (zh) * 2023-05-10 2023-06-09 潍坊护理职业学院 基于用户影像数据的牙科诊疗方案生成方法及***
CN116342588B (zh) * 2023-05-22 2023-08-11 徕兄健康科技(威海)有限责任公司 一种脑血管图像增强方法

Similar Documents

Publication Publication Date Title
WO2020177126A1 (zh) 图像处理方法、***、计算设备及存储介质
Tang et al. A two-stage approach for automatic liver segmentation with Faster R-CNN and DeepLab
Feng et al. Image segmentation and bias correction using local inhomogeneous iNtensity clustering (LINC): A region-based level set method
US9984283B2 (en) Methods, systems, and computer readable media for automated detection of abnormalities in medical images
Yang et al. Automatic segmentation model combining U-Net and level set method for medical images
Mahapatra Analyzing training information from random forests for improved image segmentation
Qazi et al. Auto‐segmentation of normal and target structures in head and neck CT images: a feature‐driven model‐based approach
Cordero-Grande et al. Unsupervised 4D myocardium segmentation with a Markov Random Field based deformable model
Lu et al. Fully automatic liver segmentation combining multi-dimensional graph cut with shape information in 3D CT images
US8724866B2 (en) Multi-level contextual learning of data
CN109102511B (zh) 一种脑血管分割方法、***及电子设备
Li et al. Fully automatic myocardial segmentation of contrast echocardiography sequence using random forests guided by shape model
US7773789B2 (en) Probabilistic minimal path for automated esophagus segmentation
CN104766340B (zh) 一种图像分割方法
Lei et al. Automatic segmentation of organs-at-risk from head-and-neck CT using separable convolutional neural network with hard-region-weighted loss
Lu et al. Precise segmentation of multiple organs in CT volumes using learning-based approach and information theory
Huang et al. Region-based nasopharyngeal carcinoma lesion segmentation from MRI using clustering-and classification-based methods with learning
He et al. Fast automatic 3D liver segmentation based on a three‐level AdaBoost‐guided active shape model
Zhu et al. Automatic delineation of the myocardial wall from CT images via shape segmentation and variational region growing
Song et al. Kidney segmentation in CT sequences using SKFCM and improved GrowCut algorithm
CN115661467A (zh) 脑血管图像分割方法、装置、电子设备及存储介质
Kisilev et al. Semantic description of medical image findings: structured learning approach.
Francis Automatic lesion identification in MRI of multiple sclerosis patients
Lee et al. Unsupervised segmentation of lung fields in chest radiographs using multiresolution fractal feature vector and deformable models
CN110046646B (zh) 图像处理方法、***、计算设备及存储介质

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: 19918060

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: 19918060

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19918060

Country of ref document: EP

Kind code of ref document: A1

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 23.02.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19918060

Country of ref document: EP

Kind code of ref document: A1