WO2018176319A1 - 超声图像的分析方法及装置 - Google Patents

超声图像的分析方法及装置 Download PDF

Info

Publication number
WO2018176319A1
WO2018176319A1 PCT/CN2017/078782 CN2017078782W WO2018176319A1 WO 2018176319 A1 WO2018176319 A1 WO 2018176319A1 CN 2017078782 W CN2017078782 W CN 2017078782W WO 2018176319 A1 WO2018176319 A1 WO 2018176319A1
Authority
WO
WIPO (PCT)
Prior art keywords
ultrasound image
area
image
target
initial contour
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2017/078782
Other languages
English (en)
French (fr)
Inventor
刘磊
秦文健
温铁祥
辜嘉
李凌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
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 Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Publication of WO2018176319A1 publication Critical patent/WO2018176319A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/155—Segmentation; Edge detection involving morphological operators
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10132—Ultrasound image

Definitions

  • the present invention relates to the field of image processing technologies, and in particular, to a method and an apparatus for analyzing an ultrasound image.
  • CAD Computer-aided Diagnosis
  • the computer-aided diagnosis system is increasingly Used in clinical diagnosis.
  • diagnostic ultrasound results are typically provided by analyzing medical ultrasound images; for example, for ultrasound images of breast tumors, the ultrasound image is analyzed using a CAD system to provide diagnostic results for breast tumors, tumor location, or size.
  • CAD Computer-aided diagnosis systems
  • how to segment the lesion area (such as the tumor area) from the ultrasound image is a key link.
  • the current segmentation algorithm for ultrasound images is generally semi-automatic, that is, the clinician needs to manually select a representative point of the region of interest or the region of interest, and then perform segmentation by computer.
  • the semi-automatic segmentation method reduces the automation performance of the CAD system and cannot meet the needs of the massive medical image processing that is increasingly occurring in the clinic.
  • most of the automatic segmentation algorithms introduce some a priori constraint information such as shape, texture and spatial relative position to achieve fully automatic segmentation.
  • the results may be different, so it is difficult to accurately extract the prior constraint information of the breast tumor, thus affecting the analysis. The accuracy of the results.
  • the embodiment of the invention provides a method and a device for analyzing an ultrasonic image, which can improve the automation level and speed of the segmentation while ensuring the accuracy of the segmentation result.
  • An embodiment of the present invention provides an analysis method of an ultrasound image for segmenting a target region from the ultrasound image, the analysis method comprising: preprocessing the ultrasound image; and the pre-processed ultrasound An initial contour of the target area is extracted from the image; an initial contour of the target area is evolved to obtain an accurate boundary of the target area.
  • the step of pre-processing the ultrasound image includes: selecting an effective region of the ultrasound image; and performing denoising processing on the selected effective region.
  • the step of extracting the initial contour of the target region from the pre-processed ultrasound image comprises: pre-segmenting the pre-processed ultrasound image to obtain a pre-segmented image;
  • the pre-segmented image sequentially performs a series of processes, wherein the series of processes includes at least one of: morphological processing, void filling, and removing an area connected to the boundary; extracting through the series of processed images An enclosed area; determining an initial contour of the target area based on the size of the extracted closed area.
  • the step of pre-segmenting the pre-processed ultrasound image includes: processing the pre-processed ultrasound image by using a maximum inter-class variance method to obtain a first threshold; a threshold value, a foreground image is segmented from the pre-processed ultrasound image; the foreground image is processed by the maximum inter-class variance method to obtain a second threshold; and the second threshold is used
  • the pre-processed ultrasound image is pre-segmented to obtain the pre-segmented image.
  • the step of pre-segmenting the pre-processed ultrasound image includes: processing the pre-processed ultrasound image by using a maximum inter-class variance method to obtain a first threshold; a threshold value for pre-segmenting the pre-segmented ultrasound image to obtain the pre-segmented image; the step of sequentially performing a series of processing on the pre-segmented image includes: The shape-processed processing is performed a plurality of times on the pre-segmented image.
  • the step of determining an initial contour of the target region according to the size of the extracted closed region includes: for the extracted closed region, retaining n closed regions having the largest area, where n is an integer greater than 0 And determining an initial contour of the target area according to the size relationship of the n closed areas.
  • the step of evolving the initial contour of the target area comprises: using an area-based active contour model to evolve an initial contour of the target area.
  • the edge of the target away from the evolution curve is detected by the edge indication function to guide the evolution curve to stop at the boundary of the target contour, thereby improving the convergence speed;
  • the edge indication function is:
  • ⁇ (0,1) is the control coefficient of the image boundary intensity field versus evolution velocity
  • ⁇ is the proportional constant
  • R is the edge strength obtained by the exponentially weighted average ratio operator.
  • An embodiment of the present invention provides an ultrasound image analysis apparatus for segmenting a target area from the ultrasound image, the analysis apparatus comprising: a preprocessing module for preprocessing the ultrasound image; an initial contour An extraction module, configured to extract an initial contour of the target region from the pre-processed ultrasound image; and an evolution module configured to evolve an initial contour of the target region to obtain an accuracy of the target region boundary.
  • the automation level and speed of the segmentation can be improved while ensuring the accuracy of the segmentation result.
  • FIG. 1 is a flow chart showing an embodiment of an analysis method of an ultrasonic image of the present invention
  • FIG. 2a is a schematic flow chart of an embodiment of step 101 in FIG. 1;
  • Figure 2b is a schematic illustration of an embodiment of an original ultrasound image and an active area
  • FIG. 3 is a schematic flow chart of an embodiment of step 102 in FIG. 1;
  • Figure 4a and Figure 4b are schematic diagrams showing the evolution results of the conventional CV model and the improved CV model, respectively;
  • Fig. 5 is a view showing the configuration of an embodiment of an ultrasonic image analyzing apparatus of the present invention.
  • FIG. 1 is a schematic flow chart of an embodiment of an analysis method of an ultrasonic image of the present invention.
  • the method for analyzing the ultrasound image can be integrated into the CAD system for segmenting the target region from the ultrasound image, thereby facilitating the CAD system to provide auxiliary diagnostic data.
  • the ultrasound image may be, for example, a breast tumor ultrasound image, but the invention is not limited thereto.
  • the analysis method of the ultrasonic image includes the following steps:
  • Step 101 Preprocess the ultrasound image.
  • the ultrasound image can be generated, for example, by an ultrasound device and then input to a CAD system for processing.
  • Step 102 Extract an initial contour of the target area from the pre-processed ultrasound image.
  • the target area may be, for example, a tumor area in an ultrasound image of a breast tumor.
  • Step 103 Evolve the initial contour of the target area to obtain an accurate boundary of the target area.
  • the automation level and speed of the segmentation can be improved while ensuring the accuracy of the segmentation result.
  • FIG. 2a it is a schematic flowchart of an embodiment of step 101 in FIG. It includes the following steps:
  • Step 201 Select an effective area of the ultrasound image.
  • the manual cutting method may be used to select the effective area.
  • the so-called manual cutting method manually uses the mouse to draw a box to trim the original image, thereby removing the surrounding irrelevant area and retaining the center of the image. region.
  • the program setting method can also be used to select the effective area; because the image size and the frame size collected by the ultrasound device of the same model are fixed, it is possible to reserve only the pixels of a specific size area in the image by program setting, thereby ignoring the surrounding area. Irrelevant information.
  • the original ultrasound image and the image of the effective region obtained after the processing in step 201 wherein the left image in FIG. 2b is the original ultrasound image, and the image on the right is the image of the effective region.
  • Step 202 Perform denoising processing on the selected effective area.
  • the image may be filtered by a SRAD (Speckle Reducing Anisotropic Diffusion) model to remove speckle noise.
  • SRAD Styreducing Anisotropic Diffusion
  • the SRAD model can effectively filter out the ultrasonic image speckle noise, simplify the image structure, improve the image quality, and maintain the detail information at the edge of the image.
  • FIG. 3 it is a schematic flowchart of an embodiment of step 102 in FIG. 1, which includes the following steps:
  • Step 301 Perform pre-segmentation on the pre-processed ultrasound image to obtain a pre-segmented image.
  • the pre-segmented ultrasound image is pre-segmented by using two maximum inter-class variance methods. It should be noted that the details of the maximum inter-class variance method are familiar to those skilled in the art, and therefore are not described herein.
  • the pre-processed ultrasound image is first processed using a maximum inter-class variance method to obtain a first threshold (ie, a global threshold). Then, according to the first threshold, the preprocessed ultrasound image is divided into a background image and a foreground image, wherein the foreground image may include a target region. Then, the foreground image is processed using the maximum inter-class variance method to obtain a second threshold (optimum threshold). Finally, the pre-segmented ultrasound image is pre-segmented by using the second threshold to obtain a pre-segmented image. It should be noted that the pre-segmented image may be a binarized image.
  • the maximum inter-class variance method may be performed only once, that is, the pre-processed ultrasound image is first processed by using a maximum inter-class variance method to obtain a global threshold. Then use this The global threshold is used to pre-segment the pre-processed ultrasound image to obtain a pre-segmented image.
  • step 301 when the maximum inter-class variance method is used twice, the mis-segmentation can be reduced.
  • the maximum inter-class variance method in order to improve the accuracy, the number of morphological processing can be increased in a series of subsequent processing.
  • Step 302 Perform a series of processing on the pre-segmented image, wherein the series of processing includes at least one of the following: morphological processing, hole filling, and removing the area connected to the boundary.
  • the morphological processing may mean that the binarized image obtained after the pre-segmentation often contains some noise fragments caused by speckle noise. Therefore, in order to filter out small burrs and isolated points in the image, cut off the elongated connection, and smooth the edge of the lesion (such as a tumor), the binarized image (ie, the pre-segmented image) is subjected to morphological processing. At the same time, in order to achieve better results, larger structural elements can be used for corrosion and smaller structural elements for expansion.
  • the area connected to the boundary in the binarized image is often mis-segmented by artifacts, not the real lesion area, so in order to eliminate the influence, the area connected to the boundary in the binarized image is removed.
  • Step 303 Extract a closed area in the series of processed images.
  • some closed connected regions (generally black) appear in the image, which are tumor candidate regions.
  • all closed areas are extracted and can be sorted according to the size of each closed area.
  • Step 304 Determine an initial contour of the target area according to the size of the extracted closed area.
  • n closed areas having the largest area are reserved, wherein n is an integer greater than 0; and the target area is determined according to the size relationship of the n closed areas The initial outline of the field. It should be noted that if there is only one closed area, the closed area is directly used as the initial outline of the target area.
  • the two closed areas with the largest area are extracted. If the difference between the area of the larger closed area and the area of the smaller closed area is less than a preset threshold, the two closed areas are simultaneously determined as targets. The initial contour of the region; if the difference between the area of the larger enclosed area and the area of the smaller enclosed area is greater than a predetermined threshold, only the larger enclosed area is determined as the initial contour of the target area.
  • the preset threshold may be half of the area of the larger closed area or other values.
  • the initial contour of the target area can be automatically extracted without manually setting the initial contour, so that the automation level of the CAD system can be improved.
  • the region-based active contour model is mainly used to evolve the initial contour of the target region, thereby obtaining an accurate boundary of the target region.
  • the region-based active contour model also known as CV model
  • CV model is an energy model based on image global region information proposed by Chan and Vese in 2001.
  • the core idea is to use segmentation constants to approximate various parts of the image.
  • the variational method is used to introduce the level set to establish the equation, and the differential method is used for numerical calculation.
  • the evolution problem of the boundary contour is transformed into the energy minimization problem.
  • the CV model is improved to improve the convergence speed.
  • the embodiment of the present invention detects the far-away evolution by adding an edge indication function (specifically, replacing the Dirac function in the traditional CV mode with an edge indication function).
  • the target edge of the curve is used to guide the evolution curve to stop at the boundary of the target contour, thereby increasing the convergence speed.
  • the edge indication function can be expressed as:
  • ⁇ (0,1) is the control coefficient of the image boundary intensity field versus evolution rate
  • ⁇ is the proportional constant
  • R is the edge intensity obtained by the ratio of exponentially weighted averages (ROEWA) operator.
  • the gradient value adopted by the ROEWA operator instead of the general image segmentation algorithm is that the speckle noise in the ultrasound image is multiplicative noise, and the ROEWA operator can better adapt to the image in which the noise is a multiplicative model.
  • the constructed edge indication function is a monotonically decreasing function. When R ⁇ , the value of g(R) ⁇ 0, g tends to be 0, indicating that the closer to the true boundary contour of the lesion. Therefore, the edge indication function itself can also speed up the convergence of the CV model.
  • step 103 of FIG. 1 by improving the conventional CV model, the convergence speed of the CV model can be improved, thereby speeding up the processing.
  • the split time comparison table of the improved CV model and the conventional CV model at different iteration times It can be seen from the table that the segmentation time of the improved CV model is significantly faster than the traditional CV model.
  • the analysis device 500 can be used to segment a target region from an ultrasound image.
  • the analysis device 500 includes: a pre-processing module 501 for pre-processing the ultrasound image; and an initial contour extraction module 502, configured to extract an initial of the target region from the pre-processed ultrasound image An outline; and an evolution module 503, configured to evolve an initial contour of the target area to obtain an accurate boundary of the target area.
  • the ultrasonic image analyzing device of the embodiment of the invention can improve the automation level and speed of the segmentation while ensuring the accuracy of the segmentation result.
  • the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Ultra Sonic Daignosis Equipment (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种超声图像的分析方法及装置，用于从所述超声图像中分割出目标区域。其中，分析方法包括：对所述超声图像进行预处理（101）；从预处理后的所述超声图像中提取出所述目标区域的初始轮廓（102）；以及对所述目标区域的初始轮廓进行演化，从而获取所述目标区域的准确边界（103）。该方法及装置能够在保证分割结果准确性的同时，提高分割的自动化水平和速度。

Description

超声图像的分析方法及装置 技术领域
本发明涉及图像处理技术领域，尤其涉及一种超声图像的分析方法及装置。
背景技术
在临床上，借助于CAD(Computer-aided Diagnosis,计算机辅助诊断)系统来辅助病症的诊断，可以避免主观因素的影响以及提高诊断结果的准确性和客观性，因此计算机辅助诊断系统被越来越多地应用于临床诊断中。在计算机辅助诊断系统中，一般通过分析医学超声图像来提供诊断结果；例如，对于乳腺肿瘤超声图像，利用CAD系统来对该超声图像进行分析以提供是否患乳腺肿瘤、肿瘤位置或大小等诊断结果。而对于计算机辅助诊断系统而言，如何从超声图像中分割出病灶区域(如肿瘤区)是其中的关键一环。
然而，目前对于超声图像的分割算法一般是半自动的，即需要临床医生手工选取感兴趣区域或感兴趣区域的代表点，然后通过计算机进行区域分割。半自动的分割方法降低了CAD系统的自动化性能，无法满足临床上日渐产生的海量医学图像处理的需要。而目前的全自动分割算法，大多是引入了一些形状、纹理和空间相对位置等先验约束信息来实现全自动分割。但是，由于病灶的大小、形状千差万别以及边界不清晰等问题，而且即使同一患者由不同的超声设备进行扫描所得的结果也可能不相同，因此难以准确提取乳腺肿瘤的先验约束信息，从而影响分析结果的准确性。
因此，在现有的CAD系统中，其针对医学超声图像所采取的分割方式存在 效率与准确性难以同时兼顾的问题。
发明内容
本发明实施例提供了一种超声图像的分析方法及装置，能够在保证分割结果准确性的同时，提高分割的自动化水平和速度。
本发明实施例提供了一种超声图像的分析方法，用于从所述超声图像中分割出目标区域，所述分析方法包括：对所述超声图像进行预处理；从预处理后的所述超声图像中提取出所述目标区域的初始轮廓；对所述目标区域的初始轮廓进行演化，从而获取所述目标区域的准确边界。
其中，对所述超声图像进行预处理的步骤包括：选取所述超声图像的有效区域；以及对选取的所述有效区域进行去噪处理。
其中，所述从预处理后的所述超声图像中提取出所述目标区域的初始轮廓的步骤包括：对所述预处理后的所述超声图像进行预分割，得到预分割后的图像；对所述预分割后的图像依次进行一系列处理，其中该一系列处理包括如下至少一项：形态学处理、空洞填充、以及去除与边界相连的区域；提取经过所述一系列处理后的图像中的封闭区域；根据所述提取到的封闭区域的大小，确定所述目标区域的初始轮廓。
其中，对所述预处理后的所述超声图像进行预分割的步骤包括：采用最大类间方差法，来处理所述预处理后的所述超声图像，从而得到第一阈值；根据所述第一阈值，从所述预处理后的所述超声图像中划分出前景图像；采用所述最大类间方差法，来处理所述前景图像，以得到第二阈值；利用所述第二阈值来对所述预处理后的所述超声图像进行预分割，以得到所述预分割后的图像。
其中，对所述预处理后的所述超声图像进行预分割的步骤包括：采用最大类间方差法，来处理所述预处理后的所述超声图像，从而得到第一阈值；根据所述第一阈值，来对所述预处理后的所述超声图像进行预分割，以得到所述预分割后的图像；所述对所述预分割后的图像依次进行一系列处理的步骤包括：对所述预分割后的图像多次执行所述形状学处理。
其中，根据所述提取到的封闭区域的大小，确定所述目标区域的初始轮廓的步骤包括：对于所述提取到的封闭区域，保留面积最大的n个封闭区域，其中n为大于0的整数；根据该n个封闭区域的大小关系，确定所述目标区域的初始轮廓。
其中，所述根据该n个封闭区域的大小关系，确定所述目标区域的初始轮廓的步骤包括：当n＝2时，若该2个封闭区域中，较大封闭区域的面积与较小封闭区域的面积之差小于预设阈值，则将该2个封闭区域同时确定为所述目标区域的初始轮廓；若所述较大封闭区域的面积与所述较小封闭区域的面积之差大于预设阈值，则将所述较大封闭区域确定为所述目标区域的初始轮廓。
其中，对所述目标区域的初始轮廓进行演化的步骤包括：采用基于区域的活动轮廓模型，来对所述目标区域的初始轮廓进行演化。
其中，在所述基于区域的活动轮廓模型中，通过边缘指示函数检测远离演化曲线的目标边缘，以引导所述演化曲线停止在目标轮廓的边界处，从而提高收敛速度；
其中边缘指示函数为：
Figure PCTCN2017078782-appb-000001
或者，
Figure PCTCN2017078782-appb-000002
其中，α∈(0,1)，为图像边界强度场对演化速度的控制系数，β为比例常数， R为指数加权平均比率算子得到的边缘强度。
本发明实施例提供了一种超声图像的分析装置，用于从所述超声图像中分割出目标区域，所述分析装置包括：预处理模块，用于对所述超声图像进行预处理；初始轮廓提取模块，用于从预处理后的所述超声图像中提取出所述目标区域的初始轮廓；以及演化模块，用于对所述目标区域的初始轮廓进行演化，从而获取所述目标区域的准确边界。
本发明实施例的有益效果是：
本发明实施例，能够在保证分割结果准确性的同时，提高分割的自动化水平和速度。
附图说明
图1是是本发明的超声图像的分析方法的实施例的流程示意图；
图2a是图1中的步骤101的实施例的流程示意图；
图2b是原始超声图像和有效区域的实施例的示意图；
图3是图1中的步骤102的实施例的流程示意图；
图4a和图4b分别为传统CV模型和改进的CV模型的演化结果示意图；
图5是本发明的超声图像的分析装置的实施例的结构示意图。
具体实施方式
为了使本发明所解决的技术问题、技术方案及有益效果更加清楚明白，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。
如图1所示，是本发明的超声图像的分析方法的实施例的流程示意图。其中该超声图像的分析方法可以集成于CAD系统中，以用来从超声图像中分割出目标区域，从而便于CAD系统提供辅助诊断数据。其中，该超声图像例如可以是乳腺肿瘤超声图像，但是本发明不限制于此。如图1所示，该超声图像的分析方法包括如下步骤：
步骤101：对超声图像进行预处理。其中，超声图像例如可以由超声设备产生，然后输入至CAD系统进行处理。
步骤102：从预处理后的超声图像中提取出目标区域的初始轮廓。其中，目标区域例如可以是乳腺肿瘤超声图像中的肿瘤区域。
步骤103：对目标区域的初始轮廓进行演化，从而获取目标区域的准确边界。
本实施例，通过对超声图像进行预处理、目标区域的初始轮郭的提取，以及初始轮郭的演化，从而能够在保证分割结果准确性的同时，提高分割的自动化水平和速度。
下面将依次对图1的各个步骤进行说明。
如图2a所示，是图1中的步骤101的实施例的流程示意图。其包括如下步骤：
步骤201：选取所述超声图像的有效区域。
其中，在步骤201中，可以采用手动剪切法来选取有效区域，所谓手动剪切法即手动利用鼠标划取一个方框对原始图像进行剪裁，从而去掉周边的无关区域而保留图像的中心大区域。另外，也可以采取程序设置法来选取有效区域；因为同一型号的超声设备采集到的图像大小及边框大小是固定的，因此可以通过程序设置只保留图像中特定大小区域的像素，从而忽略掉周边的无关信息。
举例而言，如图2b所示，是原始的超声图像和经过步骤201处理后，得到的有效区域的图像；其中图2b中的左图为原始的超声图像，而右图为有效区域的图像。
步骤202：对选取的有效区域进行去噪处理。
其中，在步骤202中，可以采用SRAD(Speckle Reducing Anisotropic Diffusion,各向异性扩散)模型对图像进行滤波，以去除斑点噪声。其中，采用SRAD模型既能有效滤除超声图像斑点噪声、简化图像结构、提高图像质量，又能保持图像边缘处细节信息。
如图3所示，是图1中的步骤102的实施例的流程示意图，其包括如下步骤：
步骤301：对预处理后的超声图像进行预分割，得到预分割后的图像。
其中，在步骤301中，采用两次最大类间方差法来对预处理后的超声图像进行预分割。需要说明的是，最大类间方差法的详细内容对于本领域技术人员而言是熟悉的，因此不赘述。
具体而言，在步骤301中，首先采用最大类间方差法，来处理预处理后的超声图像，以得到第一阈值(即全局阈值)。接着，根据该第一阈值，将预处理后的超声图像划分为背景图像和前景图像，其中前景图像会包含目标区域。然后，继续采用最大类间方差法来处理前景图像，以得到第二阈值(最优阈值)。最后，利用该第二阈值来对预处理后的超声图像进行预分割，以得到预分割后的图像。需要说明的是，该预分割后的图像可以为一个二值化图像。
可选地，在步骤301中，可以仅执行一次最大类间方差法，即首先采用最大类间方差法，来处理预处理后的超声图像，以得到全局阈值。接着，利用该 全局阈值来对预处理后的超声图像进行预分割，以得到预分割后的图像。
需要说明的是，在步骤301中，当采用两次最大类间方差法时，可以减小误分割。而当采用一次最大类间方差法时，为了提高精度，可以在后续的一系列处理中，增加形态学处理的次数。
步骤302：对预分割后的图像依次进行一系列处理，其中该一系列处理包括如下至少一项：形态学处理、空洞填充、以及去除与边界相连的区域。
其中，形态学处理可以是指：在经过预分割后得到的二值化图像中，往往含有一些因斑点噪声引起的噪声碎片。因此为了滤除图像中的小毛刺和孤立点、切断细长连接、平滑病灶(如肿瘤)边缘，对二值化图像(即预分割后的图像)进行形态学处理。同时，为了取得更好的效果，可以在腐蚀时采用较大的结构元素，膨胀时采用较小的结构元素。
其中，由于残余噪声、病灶内部灰度不均等影响，二值化后的病灶内部会产生“空洞”，故需对封闭区域内的小孔洞进行填充操作。
其中，二值化图像中与边界相连的区域往往是由伪影引起的误分割，不是真实的病灶区域，因此为了消除其影响，去除二值化图像中与边界相连的区域。
步骤303：提取经过一系列处理后的图像中的封闭区域。
其中，经过前述步骤的处理，图像中会出现一些封闭的连通区域(一般为黑色)，这些即是肿瘤候选区域。在步骤中，将所有的封闭区域提取出来，并且可以按照各封闭区域的面积大小对其进行排序。
步骤304：根据提取到的封闭区域的大小，确定目标区域的初始轮廓。
其中，在步骤304中，对于提取到的封闭区域，保留面积最大的n个封闭区域，其中n为大于0的整数；根据该n个封闭区域的大小关系，确定目标区 域的初始轮廓。需要说明的是，若仅有一个封闭区域，则直接将该封闭区域作为目标区域的初始轮廓。下面以n＝2为例对该过程进行说明：
首先，对于提取到的封闭区域，提取面积最大的2个封闭区域，若较大封闭区域的面积与较小封闭区域的面积之差小于预设阈值，则将该2个封闭区域同时确定为目标区域的初始轮廓；若较大封闭区域的面积与较小封闭区域的面积之差大于预设阈值，则仅将较大封闭区域确定为目标区域的初始轮廓。其中，预设阈值可以为较大封闭区域的面积的一半或者其他数值。
本实施例，通过以上处理，可以自动提取到目标区域的初始轮廓，而无需手动设置初始轮廓，因此能够提高CAD系统的自动化水平。
在图1中的步骤103中，主要采用基于区域的活动轮廓模型，来对目标区域的初始轮廓进行演化，从而获取目标区域的准确边界。其中，基于区域的活动轮廓模型也称为CV模型，是由Chan和Vese于2001年提出的一种基于图像全局区域信息的能量模型，其核心思想是利用分段常数去逼近图像的各个部分，通过变分法引入水平集建立方程，采用差分方法进行数值计算，最终将边界轮廓的演化问题转化为能量最小化问题。
而在本发明实施例中，并非直接采用CV模型，而是对CV模型进行了改进，以提高收敛速度。具体而言，在本发明实施例中，针对传统的CV模型，本发明实施例通过增加边缘指示函数(具体而言，是采用边缘指示函数来替换传统CV模中的Dirac函数)来检测远离演化曲线的目标边缘，以引导演化曲线停止在目标轮廓的边界处，从而提高收敛速度。
其中边缘指示函数可以表示为：
Figure PCTCN2017078782-appb-000003
或者，
Figure PCTCN2017078782-appb-000004
其中，α∈(0,1)，为图像边界强度场对演化速度的控制系数，β为比例常数，R为指数加权平均比率(The ratio of exponentially weighted averages，ROEWA)算子得到的边缘强度。此处，采用ROEWA算子而非一般图像分割算法采用的梯度值是考虑到超声图像中的斑点噪声为乘性噪声，而ROEWA算子能更好地适应噪声为乘性模型的图像。其中，所构建的边缘指示函数是一个单调递减函数，当R→∞时，g(R)→0，g的取值越趋向于0，表明该处越接近于病灶的真实边界轮廓。因此，边缘指示函数本身也可以加快CV模型的收敛速度。
另外，补充一点，经过以上改进，CV模型的能量函数为：
Figure PCTCN2017078782-appb-000005
其中，u0≥0，ν≥0，λ1，λ2＞0，且为给定的参数。
在图1的步骤103中，通过对传统的CV模型进行改进，可以提高CV模型的收敛速度，从而加快处理速度。例如，如下表所示，为在不同的迭代次数下，改进后的CV模型和传统CV模型的分割时间对比表。由该表可知，改进后的CV模型的分割时间显著地快于传统CV模型。
Figure PCTCN2017078782-appb-000006
再例如，对于同一幅超声肿瘤图像。在迭代次数均为500次的情况下，利用传统CV模型进行演化的结果如图4(a)所示，而利用改进后的CV模型进行演化后的结果如图4(b)所示。由图可见，利用改进后的CV模型进行演化后的结果与肿瘤的真实边界更接近。
以上对本发明的实施例的图像分析方法进行了详细说明，下面说明相应地该方法的装置。需要说明的是，由于相关细节已在前面详述，因此下面仅说明装置的主要架构，而忽略其细节描述。
如图5所示，是超声图像的分析装置的实施例的结构示意图。该分析装置500可以用于从超声图像中分割出目标区域。其中，该分析装置500包括：预处理模块501，用于对所述超声图像进行预处理；初始轮廓提取模块502，用于从预处理后的所述超声图像中提取出所述目标区域的初始轮廓；以及演化模块503，用于对所述目标区域的初始轮廓进行演化，从而获取所述目标区域的准确边界。
本发明实施例的超声图像的分析装置，能够在保证分割结果准确性的同时，提高分割的自动化水平和速度。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存储记忆体(Random Access Memory，RAM)等。
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. 一种超声图像的分析方法，用于从所述超声图像中分割出目标区域，其特征在于，所述分析方法包括：
    对所述超声图像进行预处理；
    从预处理后的所述超声图像中提取出所述目标区域的初始轮廓；以及
    对所述目标区域的初始轮廓进行演化，从而获取所述目标区域的准确边界。
  2. 如权利要求1所述的超声图像的分析方法，其特征在于，对所述超声图像进行预处理的步骤包括：选取所述超声图像的有效区域；以及对选取的所述有效区域进行去噪处理。
  3. 如权利要求1所述的超声图像的分析方法，其特征在于，所述从预处理后的所述超声图像中提取出所述目标区域的初始轮廓的步骤包括：
    对所述预处理后的所述超声图像进行预分割，得到预分割后的图像；
    对所述预分割后的图像依次进行一系列处理，其中该一系列处理包括如下至少一项：形态学处理、空洞填充、以及去除与边界相连的区域；
    提取经过所述一系列处理后的图像中的封闭区域；以及
    根据所述提取到的封闭区域的大小，确定所述目标区域的初始轮廓。
  4. 如权利要求3所述的超声图像的分析方法，其特征在于，对所述预处理后的所述超声图像进行预分割的步骤包括：
    采用最大类间方差法，来处理所述预处理后的所述超声图像，从而得到第一阈值；
    根据所述第一阈值，从所述预处理后的所述超声图像中划分出前景图像；
    采用所述最大类间方差法，来处理所述前景图像，以得到第二阈值；以及
    利用所述第二阈值来对所述预处理后的所述超声图像进行预分割，以得到所述预分割后的图像。
  5. 如权利要求3所述的超声图像的分析方法，其特征在于，对所述预处理后的所述超声图像进行预分割的步骤包括：
    采用最大类间方差法，来处理所述预处理后的所述超声图像，从而得到第一阈值；以及
    根据所述第一阈值，来对所述预处理后的所述超声图像进行预分割，以得到所述预分割后的图像；
    所述对所述预分割后的图像依次进行一系列处理的步骤包括：对所述预分割后的图像多次执行所述形状学处理。
  6. 如权利要求3所述的超声图像的分析方法，其特征在于，根据所述提取到的封闭区域的大小，确定所述目标区域的初始轮廓的步骤包括：
    对于所述提取到的封闭区域，保留面积最大的n个封闭区域，其中n为大于0的整数；以及
    根据该n个封闭区域的大小关系，确定所述目标区域的初始轮廓。
  7. 如权利要求6所述的超声图像的分析方法，其特征在于，所述根据该n个封闭区域的大小关系，确定所述目标区域的初始轮廓的步骤包括：
    当n＝2时，若该2个封闭区域中，较大封闭区域的面积与较小封闭区域的面积之差小于预设阈值，则将该2个封闭区域同时确定为所述目标区域的初始轮廓；若所述较大封闭区域的面积与所述较小封闭区域的面积之差大于预设阈值，则将所述较大封闭区域确定为所述目标区域的初始轮廓。
  8. 如权利要求1所述的超声图像的分析方法，其特征在于，对所述目标区域的初始轮廓进行演化的步骤包括：
    采用基于区域的活动轮廓模型，来对所述目标区域的初始轮廓进行演化。
  9. 如权利要求8所述的超声图像的分析方法，其特征在于，在所述基于区域的活动轮廓模型中，通过边缘指示函数检测远离演化曲线的目标边缘，以引导所述演化曲线停止在目标轮廓的边界处，从而提高收敛速度；
    其中边缘指示函数为：
    Figure PCTCN2017078782-appb-100001
    或者，
    Figure PCTCN2017078782-appb-100002
    其中，α∈(0,1)，为图像边界强度场对演化速度的控制系数，β为比例常数，R为指数加权平均比率算子得到的边缘强度。
  10. 一种超声图像的分析装置，用于从所述超声图像中分割出目标区域，其特征在于，所述分析装置包括：
    预处理模块，用于对所述超声图像进行预处理；
    初始轮廓提取模块，用于从预处理后的所述超声图像中提取出所述目标区域的初始轮廓；以及
    演化模块，用于对所述目标区域的初始轮廓进行演化，从而获取所述目标区域的准确边界。
PCT/CN2017/078782 2017-03-27 2017-03-30 超声图像的分析方法及装置 Ceased WO2018176319A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710186624.0 2017-03-27
CN201710186624.0A CN107169975B (zh) 2017-03-27 2017-03-27 超声图像的分析方法及装置

Publications (1)

Publication Number Publication Date
WO2018176319A1 true WO2018176319A1 (zh) 2018-10-04

Family

ID=59848781

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/078782 Ceased WO2018176319A1 (zh) 2017-03-27 2017-03-30 超声图像的分析方法及装置

Country Status (2)

Country Link
CN (1) CN107169975B (zh)
WO (1) WO2018176319A1 (zh)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108013904B (zh) * 2017-12-15 2020-12-25 无锡祥生医疗科技股份有限公司 心脏超声成像方法
CN111093525A (zh) * 2018-08-07 2020-05-01 温州医科大学 光学相干断层图像处理方法
CN109602304A (zh) * 2018-11-30 2019-04-12 余姚市腾翔电子科技有限公司 人体参数解析系统
CN112233122B (zh) * 2019-06-28 2025-07-15 深圳市理邦精密仪器股份有限公司 超声图像中对象提取、测量方法及装置
CN112419222B (zh) * 2019-08-22 2024-12-31 深圳市理邦精密仪器股份有限公司 超声胎儿颈部透明层图像分割、检测方法及装置
CN119360048A (zh) * 2024-12-26 2025-01-24 深圳市睿达科技有限公司 一种轮廓的识别方法、系统、终端设备及介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1793350A1 (en) * 2005-12-01 2007-06-06 Medison Co., Ltd. Ultrasound imaging system and method for forming a 3D ultrasound image of a target object
CN101599174A (zh) * 2009-08-13 2009-12-09 哈尔滨工业大学 基于边缘和统计特征的水平集医学超声图像区域轮廓提取方法
CN101702236A (zh) * 2009-10-30 2010-05-05 无锡景象数字技术有限公司 一种多目标前景分割方法
CN106340022A (zh) * 2015-07-08 2017-01-18 中国科学院沈阳自动化研究所 一种基于区域相关性的图像分割方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104637047B (zh) * 2013-11-13 2018-07-06 北京慧眼智行科技有限公司 一种图像处理方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1793350A1 (en) * 2005-12-01 2007-06-06 Medison Co., Ltd. Ultrasound imaging system and method for forming a 3D ultrasound image of a target object
CN101599174A (zh) * 2009-08-13 2009-12-09 哈尔滨工业大学 基于边缘和统计特征的水平集医学超声图像区域轮廓提取方法
CN101702236A (zh) * 2009-10-30 2010-05-05 无锡景象数字技术有限公司 一种多目标前景分割方法
CN106340022A (zh) * 2015-07-08 2017-01-18 中国科学院沈阳自动化研究所 一种基于区域相关性的图像分割方法

Also Published As

Publication number Publication date
CN107169975B (zh) 2019-07-30
CN107169975A (zh) 2017-09-15

Similar Documents

Publication Publication Date Title
WO2018176319A1 (zh) 超声图像的分析方法及装置
US7756316B2 (en) Method and system for automatic lung segmentation
US10229488B2 (en) Method and system for determining a stage of fibrosis in a liver
US9940706B2 (en) Image processing apparatus, image processing method and storage medium for extracting a lesion having a ground glass opactiy
JP4767957B2 (ja) 結合空間強度尤度比検定を使用する体積腫瘍分断
US8831328B2 (en) Method and system for segmenting a brain image
US20080040083A1 (en) System and Method for Solid Component Evaluation in Mixed Ground Glass Nodules
CN112907581A (zh) 一种基于深度学习的mri多类脊髓肿瘤分割方法
US8073217B2 (en) Structure segmentation via MAR-cut
CN102855483B (zh) 处理超声图像的方法和装置以及乳腺癌诊断设备
CN113269764A (zh) 颅内动脉瘤自动分割方法、系统、样本处理及模型训练方法
Sahakyan et al. Segmentation of the breast region in digital mammograms and detection of masses
CN116912823A (zh) 一种基于深度学习的肾脏病理切片细胞识别方法及系统
KR102332032B1 (ko) 복부장기 자동분할을 이용한 복부장기 볼륨 변화 분석에 기반한 복부장기 상태 판단 방법
CN112258449A (zh) 一种基于结节特征的快速结节匹配方法
KR101474162B1 (ko) 흉부 컴퓨터 단층 영상의 간유리음영 결절 자동 분할 시스템 및 그 분할 방법
CN106778793B (zh) 一种影像特征的可重复性测量方法及装置
CN118071621A (zh) 基于导向滤波和梯度增强的多层荧光图像融合与信号点精确提取及计数方法及其应用
CN110060246A (zh) 一种图像处理方法、设备及存储介质
WO2018098697A1 (zh) 一种影像特征的可重复性测量方法及装置
CN120047418B (zh) 一种甲状腺结节分级识别方法及系统
CN120563535A (zh) 一种超声图像肝肾比的计算方法和装置
RU2656761C1 (ru) Способ и система сегментации изображений очагов легких
Macenko et al. Lesion detection using morphological watershed segmentation and modelbased inverse filtering
CN112634280B (zh) 基于能量泛函的mri图像脑肿瘤分割方法

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: 17904275

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: 17904275

Country of ref document: EP

Kind code of ref document: A1