WO2021129325A1 - 超声图像病灶的分割方法、装置和计算机设备 - Google Patents

超声图像病灶的分割方法、装置和计算机设备 Download PDF

Info

Publication number
WO2021129325A1
WO2021129325A1 PCT/CN2020/133029 CN2020133029W WO2021129325A1 WO 2021129325 A1 WO2021129325 A1 WO 2021129325A1 CN 2020133029 W CN2020133029 W CN 2020133029W WO 2021129325 A1 WO2021129325 A1 WO 2021129325A1
Authority
WO
WIPO (PCT)
Prior art keywords
level set
region
lesion
interest
initial
Prior art date
Application number
PCT/CN2020/133029
Other languages
English (en)
French (fr)
Inventor
李焰驹
Original Assignee
飞依诺科技(苏州)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 飞依诺科技(苏州)有限公司 filed Critical 飞依诺科技(苏州)有限公司
Publication of WO2021129325A1 publication Critical patent/WO2021129325A1/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
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • 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
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • 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
    • G06T2207/10132—Ultrasound image

Definitions

  • This application relates to the technical field of medical image processing, and in particular to a method, device and computer equipment for segmenting ultrasound image lesions.
  • Breast cancer is a common malignant tumor in female diseases and has become one of the diseases that seriously threaten women's health.
  • Early detection, early diagnosis, and early treatment are the basic principles currently adopted in medicine for the prevention and treatment of breast cancer.
  • Ultrasound imaging has become one of the main methods of clinical diagnosis of breast tumors due to its advantages of non-invasiveness, non-radiation, and low cost.
  • ultrasound images often have large noise, low contrast, uneven grayscale, varying degrees of attenuation and infiltration effects, etc., making the surface of breast tumors more similar to the surrounding normal tissues, that is, ultrasound
  • the image has a weak ability to express the morphology of human organs, or the presentation of organs in the image is blurry and abstract; in addition, breast tumors vary greatly among individuals. Therefore, the judgment and reading of the lesion area in breast ultrasound images requires clinicians to have a higher professional level and rich experience. It is difficult for general doctors to accurately and quickly compare the breast tumor area with the normal surroundings in the ultrasound image. Distinguish the tissues and describe the results of the lesion.
  • an embodiment of the present application provides a method for segmenting an ultrasound image lesion, and the method includes:
  • a corresponding target zero level set binary image is obtained based on the target level set function, where the boundary pixels between the foreground and the background in the target zero level set binary image are the contour data of the lesion area.
  • estimating the centroid of the lesion area based on the initial binary image includes: using a set mask size to perform a morphological opening operation on the initial binary image to obtain a new binary image after the operation; according to the neighborhood
  • the connectivity criterion extracts each connected component from the new binary image, and calculates the area where each connected component is located; estimates the area where the largest connected component is located in the area where each connected component is located as the lesion area, and uses the image moment algorithm Calculate the center of mass of the lesion area.
  • constructing the initial zero level set function according to the centroid of the region of interest and the lesion area includes: creating an initial image of the same size according to the size of the region of interest; based on the centroid of the lesion area and the region of interest Determine the initial area of the level set in the initial image; obtain the binarized image of the initial image according to the initial area of the level set, and obtain the initial zero level set function based on the binarized image of the initial image.
  • the distance regularized level set evolution algorithm is used to iteratively calculate the initial zero level set function based on the region of interest, including: the evolution algorithm based on the distance regularized level set, and based on the region of interest and the initial zero level set
  • the function defines the energy functional function; the gradient descent method is used to solve the minimum value of the energy functional function to obtain the level set evolution equation; based on the level set evolution equation, the set step size and iteration number are used to calculate iteratively to obtain the target level set at the end of the evolution. function.
  • the method for setting the step size and the number of iterations includes: estimating the echo type of the lesion based on the histogram distribution of the region of interest; and determining the corresponding step size and the number of iterations based on the echo type.
  • the echo type includes anechoic and hypoechoic, and the corresponding step size and the number of iterations are determined based on the echo type, including: if the echo type is anechoic, the corresponding number of iterations is 120-260 , The time step is 1.0; if the echo type is low echo, the corresponding iteration number is 650-950, and the time step is 1.5.
  • the distance regularized level set evolution algorithm before using the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest, it also includes: comparing the Gaussian mask with a standard deviation of 2 and a size of 15*15.
  • the region of interest is convolved and smoothed to obtain a Gaussian smooth image after convolution; the distance regularized level set evolution algorithm is used to iteratively calculate the initial zero level set function based on the region of interest, including: using distance regularized level set evolution
  • the algorithm iteratively calculates the initial zero level set function based on the Gaussian smoothed image.
  • an embodiment of the present application provides a device for segmenting a lesion in an ultrasound image.
  • the device includes: a region of interest recognition module for recognizing a lesion in an ultrasound image to obtain a corresponding region of interest; a centroid estimation module , Used to threshold the region of interest to obtain the corresponding initial binary image, and estimate the centroid of the lesion area based on the initial binary image; the function construction module is used to construct the initial Zero level set function; regularized evolution module, used to use the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest to obtain the target level set function at the end of the evolution; the lesion segmentation result determination module, It is used to obtain the corresponding target zero level set binary image based on the target level set function, where the boundary pixels between the foreground and the background in the target zero level set binary image are the contour data of the lesion area.
  • an embodiment of the present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when the computer program is executed.
  • the embodiments of the present application also provide a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.
  • the above-mentioned ultrasound image lesion segmentation method, device and computer equipment identify the lesion in the ultrasound image to locate the corresponding region of interest, and then detect the region of interest by the lesion segmentation algorithm based on distance regularized level set evolution Drawing out the outline of the lesion to facilitate the subsequent diagnosis work can not only effectively reduce the workload of the doctor, but also improve the diagnosis efficiency.
  • FIG. 1 is an application environment diagram of a method for segmentation of an ultrasound image lesion in an embodiment
  • FIG. 2 is a schematic flowchart of a method for segmenting an ultrasound image lesion in an embodiment
  • Figure 3 is a schematic diagram of an ultrasound image of the original breast
  • Fig. 4 is a schematic diagram of a region of interest obtained by performing target detection on Fig. 3;
  • FIG. 5 is a schematic diagram of an initial binary image obtained after threshold segmentation is performed on FIG. 4;
  • Fig. 6 is a schematic diagram of a created binary image with the same size as Fig. 4;
  • Fig. 7 is a schematic diagram of the finally obtained binary image of the target zero level set
  • FIG. 8 is a schematic flow chart of the step of estimating the centroid of the lesion area in an embodiment
  • FIG. 9 is a schematic diagram of a new binary image obtained after performing a morphological opening operation on FIG. 5;
  • Figure 10 is a schematic diagram of roughly estimating the lesion area after analyzing Figure 9;
  • FIG. 11 is a schematic flowchart of the steps of constructing an initial zero level set function in an embodiment
  • FIG. 12 is a schematic flowchart of an iterative calculation step in an embodiment
  • Figure 13A is a schematic diagram of a low echo grayscale histogram
  • Figure 13B is a schematic diagram of an echoless grayscale histogram
  • Figure 14 is a schematic diagram showing the outline of the lesion on the original image
  • FIG. 15 is a schematic diagram of Gaussian smoothing after Gaussian smoothing is performed on FIG. 4;
  • Fig. 16 is a structural block diagram of an ultrasound image lesion segmentation device in an embodiment
  • Fig. 17 is a diagram of the internal structure of a computer device in an embodiment.
  • the method for segmenting an ultrasound image lesion provided in this application can be applied to the application environment as shown in FIG. 1.
  • the terminal 102 and the server 104 communicate through a network.
  • the terminal 102 may be a device with an ultrasound image collection function, or a device that stores the collected ultrasound images
  • the server 104 can be an independent server. Or it can be realized by a server cluster composed of multiple servers.
  • the terminal 102 is used to collect or store ultrasound images, and send the collected or stored ultrasound images to the server 104 via the network, and the server 104 recognizes the lesions in the ultrasound images to locate the corresponding region of interest, and then In the region of interest, the contour of the lesion is detected by the lesion segmentation algorithm based on Distance Regularized Level Set Evolution (DRLSE), so as to facilitate the subsequent diagnosis work, which not only can effectively reduce the workload of the doctor, but also Can improve the efficiency of diagnosis.
  • DRLSE Distance Regularized Level Set Evolution
  • a method for segmenting ultrasound image lesions is provided. Taking the method applied to the server in FIG. 1 as an example for description, the method includes the following steps:
  • Step 202 Identify the lesion in the ultrasound image to obtain a corresponding region of interest.
  • ultrasound images are clinically ultrasound images of breast nodules, breast tumors, thyroid nodules, thyroid tumors, liver cysts, kidney cysts, spleen cysts, etc.
  • the lesions refer to breast nodules, breast tumors, thyroid nodules, Thyroid tumors, liver cysts, kidney cysts, spleen cysts and other diseased parts.
  • a region of interest region of interest, ROI for short
  • the lesion in the ultrasound image can be automatically identified and located through target detection, so that the corresponding region of interest can be obtained in the ultrasound image.
  • the corresponding ultrasound image may be a breast ultrasound image (as shown in FIG. 3), so that the corresponding region of interest is obtained in the breast ultrasound image (as shown in FIG. 4).
  • Step 204 Perform threshold segmentation on the region of interest to obtain a corresponding initial binary image, and estimate the centroid of the lesion area based on the initial binary image.
  • the Otsu method can be based on the Otsu method (OTSU) to threshold the region of interest to obtain the corresponding initial binary image (as shown in Figure 5), and perform image processing based on the initial binary image, so that the initial binary image
  • Otsu method Otsu method
  • Step 206 Construct an initial zero level set function according to the centroid of the region of interest and the lesion region.
  • a binary image with the same size as the ROI (as shown in FIG. 6) is created, the foreground is a circular area with a pixel gray value of -2, and the background is an area with a pixel gray value of 2.
  • the center coordinate of the circular area is the center of mass of the lesion area
  • the binary image is the initial zero level set image and the initial object of the evolution algorithm.
  • the data type of the corresponding pixel gray value is floating point, then The corresponding function is the initial zero level set function.
  • Step 208 Use the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest to obtain the target level set function at the end of the evolution.
  • the active contour of the distance regularized level set evolution (hereinafter referred to as DRLSE) algorithm is used to iteratively calculate the initial zero level set function, so that the initial zero level set evolves toward the target boundary, thereby obtaining the evolution
  • the target level set function at the end time is to express a closed two-dimensional curve in a three-dimensional space.
  • Regularization is essentially an active contour model based on level set evolution.
  • the "active contour” refers to the two-dimensional curve represented by the zero level set.
  • Step 210 Obtain a corresponding binary image of the target zero level set based on the target level set function.
  • a digital image can be understood as a binary function
  • a binary function can be expressed as a continuous surface in the Cartesian three-dimensional coordinate system. Therefore, in this embodiment, the corresponding target zero level set binary image (as shown in Figure 7) can be obtained based on the target level set function, where the boundary pixels between the foreground and the background in the target zero level set binary image are Contour data of the lesion.
  • the above-mentioned ultrasound image lesion segmentation method recognizes the lesion in the ultrasound image to locate the corresponding region of interest, and then detects the outline of the tumor lesion in the region of interest through the lesion segmentation algorithm based on distance regularized level set evolution. In order to facilitate the subsequent diagnosis work, not only can effectively reduce the workload of the doctor, but also can improve the diagnosis efficiency.
  • the above-mentioned estimation of the center of mass of the lesion area based on the initial binary image may specifically include the following steps:
  • Step 802 Perform a morphological opening operation on the initial binary image by using the set mask size to obtain a new binary image after the operation.
  • a morphological opening operation is performed on the initial binary image as shown in FIG. 5 based on the mask size set above, so as to obtain a new binary image after the operation (as shown in FIG. 9).
  • Step 804 Extract each connected component from the new binary image according to the neighborhood connectivity criterion, and calculate the area of the area where each connected component is located.
  • each neighborhood is used to extract the connected components of each neighborhood in the new binary image.
  • Step 806 Estimate the area of the connected component with the largest area among the areas where the connected components are located as the lesion area, and calculate the centroid of the lesion area by using the image moment algorithm.
  • the area where the connected component with the largest area is located among the areas where each connected component is located is roughly estimated as the lesion area (as shown in FIG. 10).
  • the coarsely segmented area is ideal.
  • the coarsely segmented area can only roughly determine the location of the lesion and cannot accurately reflect the shape of the lesion. Therefore, it is necessary to determine the precise contour data of the lesion area through the subsequent steps shown in FIG. 2.
  • the image moment algorithm can be used to calculate the centroid of the lesion area.
  • the image moment algorithm can specifically use the Hu moment algorithm, that is, the centroid coordinates of the lesion area are obtained by the Hu moment algorithm O(X0,Y0 ).
  • constructing an initial zero level set function according to the centroids of the region of interest and the lesion region specifically includes the following steps:
  • Step 1102 Create an initial image with the same size according to the size of the region of interest.
  • Step 1104 Determine the initial area of the level set in the initial image based on the centroid of the lesion area and the minimum side length of the area of interest.
  • an initial image with the same size as the ROI is created according to the size of the ROI, and then based on the centroid of the lesion area and the minimum side length of the region of interest, the initial area of the level set is determined in the initial image. That is, a circular area with a pixel gray value of -2 (displayed as black) is constructed in the initial image, the gray value of pixels outside the circular area is 2 (displayed as white), and the center coordinates of the circular area are The center of mass of the lesion area, the radius is the side length of the smaller side of the ROI divided by 5.
  • the circular boundary of the circular area is the initial contour of the level set evolution, which is the initial area of the level set.
  • Step 1106 Obtain a binarized image of the initial image according to the initial region of the level set, and obtain an initial zero level set function based on the binarized image of the initial image.
  • the binary image of the initial image is obtained, that is, the image shown in Figure 6.
  • the valued image is the initial zero-level set image, and it is also the initial object of the evolutionary algorithm.
  • the binary image shown in Figure 6 can be represented by the following function:
  • x, y are the horizontal and vertical coordinates of the image
  • R 0 represents the ROI image area
  • using the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest includes the following steps:
  • Step 1202 Define an energy functional function based on the distance regularized level set evolution algorithm, and based on the region of interest and the initial zero level set function.
  • the energy functional function of the image information is defined as:
  • E( ⁇ ) ⁇ R p ( ⁇ )+E ext ( ⁇ )(2), where ⁇ >0 is a constant, E ext ( ⁇ ) is the external energy functional, which makes the zero level set evolve toward the target boundary, R p ( ⁇ ) is the regularization term of the target level set function.
  • ⁇ is an arbitrary real number, that is, a constant, and ⁇ is a positive real number, which are the weights of the length term and the area term on the right side of the formula (3).
  • G ⁇ represents a function with a standard deviation of ⁇ , that is, the function corresponding to the region of interest as shown in FIG. 4, I represents the binary image as shown in FIG. 6, and * represents a convolution operator.
  • the edge stop function in the DRLSE algorithm can be rewritten as:
  • p is the potential energy, defined as follows:
  • Step 1204 Solve the minimum value of the energy functional function by using the gradient descent method to obtain the level set evolution equation.
  • the formula (9) is solved by the gradient descent method to achieve the purpose of solving the minimum value of the energy functional. Approximately transforming the partial differential equation in formula (9) into a discrete finite difference form, the level set evolution equation is obtained, that is, the DRLSE model:
  • Step 1206 Iteratively calculates based on the level set evolution equation using the set step length and the number of iterations to obtain the target level set function at the end of the evolution.
  • the setting of the step size and the number of iterations includes: estimating the echo type of the lesion according to the histogram distribution of the region of interest, that is, in FIG. 4, and then determining the corresponding step size and the number of iterations according to the echo type.
  • the echo types include low echo and anechoic, as shown in FIG. 13A and FIG. 13B, which respectively represent the gray histogram of low echo and anechoic.
  • the horizontal axis represents a total of 256 intervals from 0 to 255 (that is, the possible gray values), and the vertical axis represents the frequency of each gray value.
  • the frequency of the most frequent gray values in the echo-free histogram is very different from the average frequency of other gray values. Therefore, the distribution feature of the histogram can be used to distinguish the echo type of the lesion.
  • the following formula can be used to quantitatively express:
  • ratio maxFrequency/mean_num, where maxFrequency is the frequency of the gray value that appears most frequently in the histogram, and mean_num is the average frequency of other gray values.
  • maxFrequency is the frequency of the gray value that appears most frequently in the histogram
  • mean_num is the average frequency of other gray values. The larger the ratio, the more likely it is an anechoic lesion. Specifically, when the ratio is greater than 7, it can be determined as an anechoic lesion, and when the ratio is less than 7, it can be determined as a hypoechoic lesion.
  • the corresponding iteration number is 120 to 260, and the time step is 1.0; if the echo type is low echo, the corresponding iteration number is 650 to 950, the time step is 1.5.
  • the corresponding target zero level set binary image (as shown in Figure 7) is obtained.
  • the boundary pixels between the foreground and the background are the contour data of the lesion.
  • the outline of the lesion (as shown in Figure 14) is displayed on the region of interest (ie Figure 4).
  • Gaussian denoising can be performed on the region of interest, that is, the region of interest shown in Figure 4 can be smoothed by convolution and smoothing with a Gaussian mask with a standard deviation of 2 and a size of 15*15 to obtain a smoothed convolution Gaussian smooth image (as shown in Figure 15).
  • the distance-regularized level set evolution algorithm is specifically used to iteratively calculate the initial zero level set function based on the Gaussian smooth image. That is, the above G ⁇ represents a function whose standard deviation is ⁇ , that is, the function corresponding to the Gaussian smooth image as shown in FIG. 15.
  • a device for segmenting an ultrasound image lesion including: a region of interest recognition module 1601, a centroid estimation module 1602, a function construction module 1603, a regularization evolution module 1604, and lesion segmentation
  • the result determination module 1605 where:
  • the region of interest identification module 1601 is used to identify the lesion in the ultrasound image to obtain the corresponding region of interest;
  • the centroid estimation module 1602 is used to threshold the region of interest to obtain the corresponding initial binary image, and estimate the centroid of the lesion area according to the initial binary image;
  • the function construction module 1603 is used to construct an initial zero-level set function according to the centroid of the region of interest and the lesion area;
  • the regularized evolution module 1604 is used to use the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest to obtain the target level set function at the end of the evolution;
  • the lesion segmentation result determination module 1605 is configured to obtain a corresponding target zero level set binary image based on the target level set function, where the boundary pixels between the foreground and the background in the target zero level set binary image are the contour data of the lesion.
  • the centroid estimation module is specifically used to: use the set mask size to perform morphological opening operation on the initial binary image to obtain a new binary image after the operation; Each connected component is extracted from the value image, and the area of each connected component is calculated; the area of the connected component with the largest area in the area of each connected component is estimated as the lesion area, and the image moment algorithm is used to calculate the centroid of the lesion area.
  • the function construction module is specifically used to: create an initial image of the same size according to the size of the region of interest; determine the initial level set in the initial image based on the centroid of the lesion area and the minimum side length of the region of interest Area: The binarized image of the initial image is obtained according to the initial area of the level set, and the initial zero level set function is obtained based on the binarized image of the initial image.
  • the regularized evolution module is specifically used to: define an energy functional function based on the distance regularized level set evolution algorithm, and based on the region of interest and the initial zero level set function; use the gradient descent method to find the minimum value of the energy functional function , Get the level set evolution equation; based on the level set evolution equation, iteratively calculate with the set step size and the number of iterations to obtain the target level set function at the end of the evolution.
  • it also includes a Gaussian smoothing module, which is used to calculate the standard deviation of 2 and the size of 15*15 before using the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest.
  • the Gaussian mask performs convolution smoothing on the region of interest to obtain a Gaussian smooth image after convolution smoothing.
  • Each module in the above-mentioned ultrasound image lesion segmentation device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 17.
  • the computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, a computer program, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium.
  • the database of the computer equipment is used to store ultrasound image data.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • FIG. 17 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a computer device including a memory and a processor, a computer program is stored in the memory, and the processor implements the following steps when the processor executes the computer program:
  • a corresponding target zero level set binary image is obtained based on the target level set function, where the boundary pixels between the foreground and the background in the target zero level set binary image are the contour data of the lesion area.
  • the processor further implements the following steps when executing the computer program: using a set mask size to perform a morphological opening operation on the initial binary image to obtain a new binary image after the operation; Extract each connected component from the new binary image, and calculate the area where each connected component is located; estimate the area where the connected component with the largest area is located in the area where each connected component is located as the focus area, and use the image moment algorithm to calculate the focus area The center of mass.
  • the processor further implements the following steps when executing the computer program: creating an initial image of the same size according to the size of the region of interest; based on the centroid of the lesion area and the minimum side length of the region of interest, in the initial image Determine the initial area of the level set; obtain the binarized image of the initial image according to the initial area of the level set, and obtain the initial zero level set function based on the binarized image of the initial image.
  • the processor further implements the following steps when executing the computer program: according to the distance regularization level set evolution algorithm, and based on the region of interest and the initial zero level set function to define the energy functional function; using the gradient descent method to solve the energy functional function The minimum value of, obtains the level set evolution equation; based on the level set evolution equation, iteratively calculates with the set step size and number of iterations to obtain the target level set function at the end of the evolution.
  • the processor further implements the following steps when executing the computer program: according to the histogram distribution of the region of interest, the echo type of the lesion is estimated; and the corresponding step size and the number of iterations are determined based on the echo type.
  • the echo types include anechoic and hypoechoic
  • the processor also implements the following steps when executing the computer program: if the echo type is anechoic, the corresponding iteration number is 120-260, and the time step is 1.0 ; If the echo type is low echo, the corresponding iteration number is 650-950, and the time step is 1.5.
  • the processor further implements the following steps when executing the computer program: before using the distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on the region of interest, set the standard deviation to 2 and the size to 15 *15 Gaussian mask performs convolution smoothing on the region of interest to obtain a Gaussian smoothed image after convolution; then the distance regularized level set evolution algorithm is used to iteratively calculate the initial zero level set function based on the region of interest, including : Use distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on Gaussian smoothed image.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
  • a corresponding target zero level set binary image is obtained based on the target level set function, where the boundary pixels between the foreground and the background in the target zero level set binary image are the contour data of the lesion area.
  • the following steps are also implemented: use the set mask size to perform morphological opening operation on the initial binary image to obtain a new binary image after the operation; according to the neighborhood connectivity criterion Extract each connected component from the new binary image, and calculate the area where each connected component is located; estimate the area where the connected component with the largest area is located in the area where each connected component is located as the lesion area, and use the image moment algorithm to calculate the lesion The centroid of the region.
  • the following steps are also implemented: create an initial image of the same size according to the size of the region of interest; based on the centroid of the lesion area and the minimum side length of the region of interest, the initial image Determine the initial area of the level set; obtain the binarized image of the initial image according to the initial area of the level set, and obtain the initial zero level set function based on the binarized image of the initial image.
  • the following steps are also implemented: according to the distance regularized level set evolution algorithm, the energy functional function is defined based on the region of interest and the initial zero level set function; and the gradient descent method is used to solve the energy pan The minimum value of the function is used to obtain the level set evolution equation; based on the level set evolution equation, the set step size and number of iterations are iteratively calculated to obtain the target level set function at the end of the evolution.
  • the following steps are further implemented: according to the histogram distribution of the region of interest, the echo type of the lesion is estimated; and the corresponding step size and the number of iterations are determined based on the echo type.
  • the echo types include anechoic and hypoechoic, and when the computer program is executed by the processor, the following steps are also implemented: if the echo type is anechoic, the corresponding number of iterations is 120-260, and the time step is 1.0; If the echo type is low echo, the corresponding iteration number is 650-950, and the time step is 1.5.
  • the following steps are also implemented: using the distance regularized level set evolution algorithm to calculate the initial zero level set function based on the region of interest iteratively, set the standard deviation to 2 and the size to 15
  • the Gaussian mask of *15 performs convolution smoothing on the region of interest to obtain a Gaussian smooth image after convolution;
  • the distance regularized level set evolution algorithm is used to iteratively calculate the initial zero level set function based on the region of interest, including :
  • Use distance regularized level set evolution algorithm to iteratively calculate the initial zero level set function based on Gaussian smoothed image.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Abstract

一种超声图像病灶的分割方法、装置和计算机设备，通过识别超声图像中的病灶，以得到对应的感兴趣区域；对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心；根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数；利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；基于目标水平集函数得到对应的目标零水平集二值图像，即检测出乳腺肿瘤病灶的轮廓数据，以便后续诊断工作的进行，不仅能够有效地减少医生的工作量，且能提高诊断效率。

Description

超声图像病灶的分割方法、装置和计算机设备 技术领域
本申请涉及医学图像处理技术领域，特别是涉及一种超声图像病灶的分割方法、装置和计算机设备。
背景技术
乳腺癌是女性疾病中常见的恶性肿瘤，已成为严重威胁女性健康的病症之一。早发现、早诊断、早治疗是目前医学上对防治乳腺癌采取的基本原则。超声成像凭借其无创伤、无辐射、费用低廉等优势，已成为乳腺肿瘤临床诊断的主要手段之一。
然而，由于受成像设备的影响，超声图像常常具有较大的噪声、低对比度、灰度不均匀、不同程度的衰减以及浸润效应等，使得乳腺肿瘤在表面上与周围正常组织较为相似，即超声图像对人体器官形态的表达能力较弱，或者说器官在图像中的呈现形式较为模糊与抽象；此外，不同个体间乳腺肿瘤差异也较大。因此，乳腺超声图像中对于病灶区域的判断和阅读需要临床医生具有较高的专业水平和较丰富的经验，一般医生较难在超声图像中用肉眼准确、快速地将乳腺肿瘤区域与其周围的正常组织区分开来并作出病灶描述结果。
发明内容
基于此，有必要针对上述一般医生较难在超声图像中快速准确的定位病灶区域的问题，提供一种超声图像病灶的分割方法、装置和计算机设备。
为了实现上述目的，一方面，本申请实施例提供了一种超声图像病灶的分 割方法，所述方法包括：
识别超声图像中的病灶，以得到对应的感兴趣区域；
对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心；
根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数；
利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；
基于目标水平集函数得到对应的目标零水平集二值图像，其中，目标零水平集二值图像中前景与背景的边界像素为病灶区域的轮廓数据。
在其中一个实施例中，根据初始二值图像估计病灶区域的质心，包括：采用设定掩模尺寸对初始二值图像进行形态学开运算，得到运算后的新的二值图像；根据邻域连通准则在新的二值图像中提取各连通分量，并计算各连通分量所在区域的面积；将各连通分量所在区域的面积中面积最大的连通分量所在区域估计为病灶区域，并采用图像矩算法计算病灶区域的质心。
在其中一个实施例中，根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数，包括：根据感兴趣区域的尺寸创建一幅尺寸相同的初始图像；基于病灶区域的质心以及感兴趣区域的最小边长，在初始图像中确定水平集初始区域；根据水平集初始区域得到初始图像的二值化图像，基于初始图像的二值化图像得到初始零水平集函数。
在其中一个实施例中，利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，包括：根据距离正则化水平集演化算法，并基于感兴趣区域以及初始零水平集函数定义能量泛函数；采用梯度下降法求解能量泛函数的最小值，得到水平集演化方程；基于水平集演化方程采用设定的 步长和迭代次数迭代计算，以得到演化终止时刻的目标水平集函数。
在其中一个实施例中，步长和迭代次数的设定方法包括：根据感兴趣区域的直方图分布，估计病灶的回声类型；基于回声类型确定对应的步长和迭代次数。
在其中一个实施例中，回声类型包括无回声和低回声，则基于所述回声类型确定对应的步长和迭代次数，包括：若回声类型为无回声时，则对应的迭代次数为120～260，时间步长为1.0；若回声类型为低回声时，则对应的迭代次数为650～950，时间步长为1.5。
在其中一个实施例中，利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算之前，还包括：将标准差为2，尺寸为15*15的高斯掩模对感兴趣区域进行卷积平滑，以得到卷积平滑后的高斯平滑图像；则利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，包括：利用距离正则化水平集演化算法基于高斯平滑图像对初始零水平集函数进行迭代计算。
另一方面，本申请实施例提供了一种超声图像病灶的分割装置，所述装置包括：感兴趣区域识别模块，用于识别超声图像中的病灶，以得到对应的感兴趣区域；质心估计模块，用于对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心；函数构造模块，用于根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数；正则化演化模块，用于利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；病灶分割结果确定模块，用于基于目标水平集函数得到对应的目标零水平集二值图像，其中，目标零水平集二值图像中前景与背景的边界像素为病灶区域的轮廓数据。
又一方面，本申请实施例还提供了一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现如上所述方法的步骤。
再一方面，本申请实施例还提供了一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现如上所述方法的步骤。
上述超声图像病灶的分割方法、装置和计算机设备，通过对超声图像中的病灶进行识别，以定位对应的感兴趣区域，进而在感兴趣区域中通过基于距离正则化水平集演化的病灶分割算法检测出病灶轮廓，以便于后续诊断工作的进行，不仅能够有效地减少医生的工作量，且能提高诊断效率。
附图说明
图1为一个实施例中超声图像病灶的分割方法的应用环境图；
图2为一个实施例中超声图像病灶的分割方法的流程示意图；
图3为原乳腺超声图像示意图；
图4为对图3进行目标检测得到的感兴趣区域的示意图；
图5为对图4进行阈值分割后得到的初始二值图像的示意图；
图6为创建的与图4尺寸相同的二值图像的示意图；
图7为最终得到的目标零水平集二值图像的示意图；
图8为一个实施例中估计病灶区域的质心步骤的流程示意图；
图9为对图5进行形态学开运算后得到的新的二值图像的示意图；
图10为对图9进行分析后粗略估计出病灶区域的示意图；
图11为一个实施例中构造初始零水平集函数步骤的流程示意图；
图12为一个实施例中进行迭代计算步骤的流程示意图；
图13A为低回声灰度直方图的示意图；
图13B为无回声灰度直方图的示意图；
图14为在原图上显示病灶的轮廓示意图；
图15是对图4进行高斯平滑处理后的高斯平滑示意图；
图16为一个实施例中超声图像病灶的分割装置的结构框图；
图17为一个实施例中计算机设备的内部结构图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的超声图像病灶的分割方法，可以应用于如图1所示的应用环境中。其中，终端102与服务器104通过网络进行通信，在本实施例中，终端102可以是具有超声图像采集功能的设备，也可以是对采集的超声图像进行存储的设备，服务器104可以用独立的服务器或者是多个服务器组成的服务器集群来实现。具体的，终端102用于采集或存储超声图像，并将采集或存储的超声图像通过网络发送至服务器104，服务器104则对超声图像中的病灶进行识别，以定位对应的感兴趣区域，进而在感兴趣区域中通过基于距离正则化水平集演化(Distance Regularized Level Set Evolution，简称DRLSE)的病灶分割算法检测出病灶轮廓，以便于后续诊断工作的进行，不仅能够有效地减少医生的工作量，且能提高诊断效率。
在一个实施例中，如图2所示，提供了一种超声图像病灶的分割方法，以该方法应用于图1中的服务器为例进行说明，包括以下步骤：
步骤202，识别超声图像中的病灶，以得到对应的感兴趣区域。
其中，超声图像为临床上乳腺结节、乳腺肿瘤、甲状腺结节、甲状腺肿瘤、肝脏囊肿、肾脏囊肿、脾囊肿等病灶的超声图像，病灶则是指乳腺结节、乳腺肿瘤、甲状腺结节、甲状腺肿瘤、肝脏囊肿、肾脏囊肿、脾囊肿等病变部分。感兴趣区域(region of interest，简称ROI)是从超声图像中选择的一个需要处理的图像区域，这个区域是进行图像分析所关注的重点。具体的，可以通过目标检测对超声图像中的病灶进行自动识别、定位，从而在超声图像中得到对应的感兴趣区域。以下以病灶为乳腺肿瘤为例进行说明，则对应的超声图像可以为乳腺超声图像(如图3所示)，从而在乳腺超声图像中得到对应的感兴趣区域(如图4所示)。
步骤204，对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心。
具体的，可以是基于大津法(OTSU)对感兴趣区域进行阈值分割，从而得到对应的初始二值图像(如图5所示)，并基于初始二值图像进行图像处理，从而在初始二值图像中粗略估计出病灶区域的质心，其中，质心是指病灶区域的中心。
步骤206，根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数。
具体的，创建一幅与ROI尺寸相同的二值图像(如图6所示)，其前景是像素灰度值为-2的圆形区域，背景是像素灰度值为2的区域。其中，圆形区域的圆心坐标即为病灶区域的质心，该二值图像即为初始零水平集图像，也是演化算法的初始对象，其对应的像素灰度值的数据类型是浮点型，则对应的函数即为初始零水平集函数。
步骤208，利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集 函数进行迭代计算，得到演化终止时刻的目标水平集函数。
在本实施例中，基于感兴趣区域，利用距离正则化水平集演化(以下简称DRLSE)算法的活动轮廓对初始零水平集函数进行迭代计算，使得初始零水平集朝目标边界演化，从而得到演化终止时刻的目标水平集函数。其中，零水平集就是将一个封闭的二维曲线表达于三维的空间里。“正则化”在本质上是一种基于水平集演化的活动轮廓模型，该“活动轮廓(active contour)”就是指零水平集表示的那个二维曲线。
步骤210，基于目标水平集函数得到对应的目标零水平集二值图像。
由于一幅数字图像可以理解为一个二元函数，则一个二元函数在笛卡尔三维坐标系中可以表示为一个连续的曲面。因此，在本实施例中，基于目标水平集函数可以得到对应的目标零水平集二值图像(如图7所示)，其中，目标零水平集二值图像中前景与背景的边界像素则为病灶的轮廓数据。
上述超声图像病灶的分割方法，通过对超声图像中的病灶进行识别，以定位对应的感兴趣区域，进而在感兴趣区域中通过基于距离正则化水平集演化的病灶分割算法检测出肿瘤病灶轮廓，以便于后续诊断工作的进行，不仅能够有效地减少医生的工作量，且能提高诊断效率。
在一个实施例中，如图8所示，上述根据初始二值图像估计病灶区域的质心，具体可以包括如下步骤：
步骤802，采用设定掩模尺寸对初始二值图像进行形态学开运算，得到运算后的新的二值图像。
其中，设定掩模尺寸为eleSize＝2*(tempSize/14)+1，tempSize则是ROI区域中较小一边的边长。具体的，基于上述设定的掩模尺寸对如图5所示的初始二值图像进行形态学开运算，从而得到一个运算后的新的二值图像 (如图9所示)。
步骤804，根据邻域连通准则在新的二值图像中提取各连通分量，并计算各连通分量所在区域的面积。
具体的，根据邻域连通准则采用8邻域在新的二值图像中提取各邻域的连通分量，此时图9中有8个连通分量，即8个彼此独立的轮廓(每个白色区域为一个轮廓)。进而计算各连通分量所在区域的面积(即该连通分量区域所占的像素数)。
步骤806，将各连通分量所在区域的面积中面积最大的连通分量所在区域估计为病灶区域，并采用图像矩算法计算病灶区域的质心。
在本实施例中，将各连通分量所在区域的面积中面积最大的连通分量所在区域粗略估计为病灶区域(如图10所示)。需要说明的是，由于本实施例中的示例图像病灶较为清晰，故粗分割的区域较为理想，但很多情况下的粗分割区域只能大概确定病灶的位置，并不能精确地反映病灶的形状，因此，需要通过图2所示的后续步骤确定病灶区域的精确轮廓数据。具体的，在粗略估计出病灶区域后，可以采用图像矩算法计算病灶区域的质心，其中，图像矩算法具体可以采用Hu矩算法，即通过Hu矩算法得到病灶区域的质心坐标O(X0,Y0)。
在一个实施例中，如图11所示，根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数，具体包括如下步骤：
步骤1102，根据感兴趣区域的尺寸创建一幅尺寸相同的初始图像。
步骤1104，基于病灶区域的质心以及感兴趣区域的最小边长，在初始图像中确定水平集初始区域。
具体的，根据ROI的尺寸创建一幅与ROI尺寸相同的初始图像，进而基于病灶区域的质心以及感兴趣区域的最小边长，在初始图像中确定水平集初始区 域。即在初始图像中构造一个像素灰度值为-2(显示为黑色)的圆形区域，该圆形区域以外的像素灰度值为2(显示为白色)，且圆形区域的圆心坐标为病灶区域的质心，半径为ROI较小一边的边长除以5，该圆形区域的圆形边界即水平集演化的初始轮廓，也即为水平集初始区域。
步骤1106，根据水平集初始区域得到初始图像的二值化图像，基于初始图像的二值化图像得到初始零水平集函数。
由于上述初始图像中水平集初始区域的像素灰度值为-2，而其他区域的像素灰度值为2，因此，得到初始图像的二值化图像，即图6所示的图像，该二值化图像即为初始零水平集图像，也是演化算法的初始对象。又由于数字图像可以通过二元函数表示，因此，对于图6所示的二值化图像可通过如下函数表示：
Figure PCTCN2020133029-appb-000001
其中，x,y为图像的横纵坐标，R 0表示ROI图像域。
在一个实施例中，如图12所示，利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，包括如下步骤：
步骤1202，根据距离正则化水平集演化算法，并基于感兴趣区域以及初始零水平集函数定义能量泛函数。
具体的，在DRLSE算法中，定义图像信息的能量泛函数为：
E(φ)＝μR p(φ)+E ext(φ)(2)，式中μ＞0，为常数，E ext(φ)为外部能量泛函，使得零水平集朝目标边界演化，R p(φ)为目标水平集函数的正则化项。
具体的，
Figure PCTCN2020133029-appb-000002
式中，α是任意实数，即常数，λ为正实数，分别是式(3)中右侧长度项和面积项的权重，本实施例中可以取λ＝4、α＝3。
Figure PCTCN2020133029-appb-000003
是对目标水平集函数φ(x,y)求梯度，x,y为图像的横纵坐标；δ ε(·)和H ε(·)是一维正则化的Dirac函数和 Heaviside函数。则有：
Figure PCTCN2020133029-appb-000004
g是边缘停止函数，定义
Figure PCTCN2020133029-appb-000005
其中，G σ表示标准差为σ的函数，即如图4所示的感兴趣区域对应的函数，I表示如图6所示的二值图像，*表示卷积运算符。在本实施例中，则可以将DRLSE算法中的边缘停止函数改写为：
Figure PCTCN2020133029-appb-000006
公式(2)中，定义
Figure PCTCN2020133029-appb-000007
其中，p为势能，定义如下：
Figure PCTCN2020133029-appb-000008
将公式(3)、(6)代入公式(2)中，即可得到下式：
Figure PCTCN2020133029-appb-000009
该式是要求解的能量泛函，很显然，能量泛函E(φ)的自变量是φ，而φ是个函数，因此，对式(8)等号两边同时做微分，则得到如下式(9)：
Figure PCTCN2020133029-appb-000010
步骤1204，采用梯度下降法求解能量泛函数的最小值，得到水平集演化方程。
通过梯度下降法求解公式(9)，以达到求解能量泛函的最小值的目的。将公式(9)中的偏微分方程近似转化为离散的有限差分形式，则得到水平集演化方程， 即DRLSE模型：
Figure PCTCN2020133029-appb-000011
其中，式(10)中的
Figure PCTCN2020133029-appb-000012
就是式(9)中的等式的右侧表达式，k为迭代次数iter，Δt为步长step。
步骤1206，基于水平集演化方程采用设定的步长和迭代次数迭代计算，以得到演化终止时刻的目标水平集函数。
其中，步长和迭代次数的设定包括：根据感兴趣区域即图4的直方图分布，估计病灶的回声类型，进而根据回声类型确定对应的步长和迭代次数。通常，回声类型包括低回声和无回声，如图13A和图13B所示，分别表示低回声和无回声的灰度直方图。其横轴表示0到255共256个区间(即灰度可能的取值)，纵轴为每个灰度值出现的频数。很显然，无回声的直方图中出现频数最多的灰度值的频数与其它灰度值出现的平均频数相差很大。因此可用直方图的这种分布特征来区分病灶的回声类型。具体可以采用如下公式来定量表示：
ratio＝maxFrequency/mean_num，其中，maxFrequency是直方图中出现频数最多的灰度值的频数，mean_num是其它灰度值出现的平均频数。比值ratio越大，则越可能是无回声病灶。具体的，当比值大于7时，则可以确定为无回声病灶，当比值小于7时，则可以确定为低回声病灶。
在本实施例中，若通过上述方法确定回声类型为无回声时，则对应的迭代次数为120～260，时间步长为1.0；若回声类型为低回声时，则对应的迭代次数为650～950，时间步长为1.5。将设定的步长Δt和迭代次数k代入式(10)进行计算，从而得到演化终止时刻的目标水平集函数φ(x,y)。
进而基于目标水平集函数φ(x,y)得到对应的目标零水平集二值图像(如图7所示)，其前景与背景的边界像素则为病灶的轮廓数据，基于该轮廓数据在原感 兴趣区域(即图4)上显示病灶的轮廓(如图14所示)。
在一个实施例中，由于原始的超声图像中会存在一定的噪声，则在本实施例中，在利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算之前，还可以对感兴趣区域进行高斯去噪处理，即通过将标准差为2，尺寸为15*15的高斯掩模对如图4所示的感兴趣区域进行卷积平滑，以得到卷积平滑后的高斯平滑图像(如图15所示)。则在在迭代计算时，具体利用距离正则化水平集演化算法基于高斯平滑图像对初始零水平集函数进行迭代计算。即上述G σ表示标准差为σ的函数，即如图15所示的高斯平滑图像对应的函数。
应该理解的是，虽然图1-14的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图1-14中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在一个实施例中，如图16所示，提供了一种超声图像病灶的分割装置，包括：感兴趣区域识别模块1601、质心估计模块1602、函数构造模块1603、正则化演化模块1604和病灶分割结果确定模块1605，其中：
感兴趣区域识别模块1601，用于识别超声图像中的病灶，以得到对应的感兴趣区域；
质心估计模块1602，用于对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心；
函数构造模块1603，用于根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数；
正则化演化模块1604，用于利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；
病灶分割结果确定模块1605，用于基于目标水平集函数得到对应的目标零水平集二值图像，其中，目标零水平集二值图像中前景与背景的边界像素为病灶的轮廓数据。
在一个实施例中，质心估计模块具本用于：采用设定掩模尺寸对初始二值图像进行形态学开运算，得到运算后的新的二值图像；根据邻域连通准则在新的二值图像中提取各连通分量，并计算各连通分量所在区域的面积；将各连通分量所在区域的面积中面积最大的连通分量所在区域估计为病灶区域，并采用图像矩算法计算病灶区域的质心。
在一个实施例中，函数构造模块具体用于：根据感兴趣区域的尺寸创建一幅尺寸相同的初始图像；基于病灶区域的质心以及感兴趣区域的最小边长，在初始图像中确定水平集初始区域；根据水平集初始区域得到初始图像的二值化图像，基于初始图像的二值化图像得到初始零水平集函数。
在一个实施例中，正则化演化模块具体用于：根据距离正则化水平集演化算法，并基于感兴趣区域以及初始零水平集函数定义能量泛函数；采用梯度下降法求解能量泛函数的最小值，得到水平集演化方程；基于水平集演化方程采用设定的步长和迭代次数迭代计算，以得到演化终止时刻的目标水平集函数。
在一个实施例中，还包括高斯平滑模块，用于在利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算之前，将标准差为2，尺寸为15*15的高斯掩模对感兴趣区域进行卷积平滑，以得到卷积平滑后的高 斯平滑图像。
关于超声图像病灶的分割装置的具体限定可以参见上文中对于超声图像病灶的分割方法的限定，在此不再赘述。上述超声图像病灶的分割装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图17所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的数据库用于存储超声图像数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种超声图像病灶的分割方法。
本领域技术人员可以理解，图17中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种计算机设备，包括存储器和处理器，存储器中存储有计算机程序，该处理器执行计算机程序时实现以下步骤：
识别超声图像中的病灶，以得到对应的感兴趣区域；
对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心；
根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数；
利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；
基于目标水平集函数得到对应的目标零水平集二值图像，其中，目标零水平集二值图像中前景与背景的边界像素为病灶区域的轮廓数据。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：采用设定掩模尺寸对初始二值图像进行形态学开运算，得到运算后的新的二值图像；根据邻域连通准则在新的二值图像中提取各连通分量，并计算各连通分量所在区域的面积；将各连通分量所在区域的面积中面积最大的连通分量所在区域估计为病灶区域，并采用图像矩算法计算病灶区域的质心。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：根据感兴趣区域的尺寸创建一幅尺寸相同的初始图像；基于病灶区域的质心以及感兴趣区域的最小边长，在初始图像中确定水平集初始区域；根据水平集初始区域得到初始图像的二值化图像，基于初始图像的二值化图像得到初始零水平集函数。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：根据距离正则化水平集演化算法，并基于感兴趣区域以及初始零水平集函数定义能量泛函数；采用梯度下降法求解能量泛函数的最小值，得到水平集演化方程；基于水平集演化方程采用设定的步长和迭代次数迭代计算，以得到演化终止时刻的目标水平集函数。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：根据感兴趣区域的直方图分布，估计病灶的回声类型；基于回声类型确定对应的步长和迭代次数。
在一个实施例中，回声类型包括无回声和低回声，则处理器执行计算机程 序时还实现以下步骤：若回声类型为无回声时，则对应的迭代次数为120～260，时间步长为1.0；若回声类型为低回声时，则对应的迭代次数为650～950，时间步长为1.5。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：在利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算之前，将标准差为2，尺寸为15*15的高斯掩模对感兴趣区域进行卷积平滑，以得到卷积平滑后的高斯平滑图像；则利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，包括：利用距离正则化水平集演化算法基于高斯平滑图像对初始零水平集函数进行迭代计算。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：
识别超声图像中的病灶，以得到对应的感兴趣区域；
对感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据初始二值图像估计病灶区域的质心；
根据感兴趣区域以及病灶区域的质心，构造初始零水平集函数；
利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；
基于目标水平集函数得到对应的目标零水平集二值图像，其中，目标零水平集二值图像中前景与背景的边界像素为病灶区域的轮廓数据。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：采用设定掩模尺寸对初始二值图像进行形态学开运算，得到运算后的新的二值图像；根据邻域连通准则在新的二值图像中提取各连通分量，并计算各连通分量所在区域的面积；将各连通分量所在区域的面积中面积最大的连通分量所在区域估计 为病灶区域，并采用图像矩算法计算病灶区域的质心。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：根据感兴趣区域的尺寸创建一幅尺寸相同的初始图像；基于病灶区域的质心以及感兴趣区域的最小边长，在初始图像中确定水平集初始区域；根据水平集初始区域得到初始图像的二值化图像，基于初始图像的二值化图像得到初始零水平集函数。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：根据距离正则化水平集演化算法，并基于感兴趣区域以及初始零水平集函数定义能量泛函数；采用梯度下降法求解能量泛函数的最小值，得到水平集演化方程；基于水平集演化方程采用设定的步长和迭代次数迭代计算，以得到演化终止时刻的目标水平集函数。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：根据感兴趣区域的直方图分布，估计病灶的回声类型；基于回声类型确定对应的步长和迭代次数。
在一个实施例中，回声类型包括无回声和低回声，则计算机程序被处理器执行时还实现以下步骤：若回声类型为无回声时，则对应的迭代次数为120～260，时间步长为1.0；若回声类型为低回声时，则对应的迭代次数为650～950，时间步长为1.5。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算之前，将标准差为2，尺寸为15*15的高斯掩模对感兴趣区域进行卷积平滑，以得到卷积平滑后的高斯平滑图像；则利用距离正则化水平集演化算法基于感兴趣区域对初始零水平集函数进行迭代计算，包括：利用距离正则化水平集演化算法基于高斯平滑图像对初始零水平集函数进行迭代计算。
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，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种超声图像病灶的分割方法，其特征在于，所述方法包括：
    识别超声图像中的病灶，以得到对应的感兴趣区域；
    对所述感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据所述初始二值图像估计病灶区域的质心；
    根据所述感兴趣区域以及所述病灶区域的质心，构造初始零水平集函数；
    利用距离正则化水平集演化算法基于所述感兴趣区域对所述初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；
    基于所述目标水平集函数得到对应的目标零水平集二值图像，所述目标零水平集二值图像中前景与背景的边界像素为所述病灶区域的轮廓数据。
  2. 根据权利要求1所述的超声图像病灶的分割方法，其特征在于，所述根据所述初始二值图像估计病灶区域的质心，包括：
    采用设定掩模尺寸对所述初始二值图像进行形态学开运算，得到运算后的新的二值图像；
    根据邻域连通准则在所述新的二值图像中提取各连通分量，并计算所述各连通分量所在区域的面积；
    将所述各连通分量所在区域的面积中面积最大的连通分量所在区域估计为病灶区域，并采用图像矩算法计算所述病灶区域的质心。
  3. 根据权利要求1所述的超声图像病灶的分割方法，其特征在于，所述根据所述感兴趣区域以及所述病灶区域的质心，构造初始零水平集函数，包括：
    根据所述感兴趣区域的尺寸创建一幅尺寸相同的初始图像；
    基于所述病灶区域的质心以及所述感兴趣区域的最小边长，在所述初始图像中确定水平集初始区域；
    根据所述水平集初始区域得到所述初始图像的二值化图像，基于所述初始 图像的二值化图像得到初始零水平集函数。
  4. 根据权利要求1所述的超声图像病灶的分割方法，其特征在于，所述利用距离正则化水平集演化算法基于所述感兴趣区域对所述初始零水平集函数进行迭代计算，包括：
    根据距离正则化水平集演化算法，并基于所述感兴趣区域以及所述初始零水平集函数定义能量泛函数；
    采用梯度下降法求解所述能量泛函数的最小值，得到水平集演化方程；
    基于所述水平集演化方程采用设定的步长和迭代次数迭代计算，以得到演化终止时刻的目标水平集函数。
  5. 根据权利要求4所述的超声图像病灶的分割方法，其特征在于，所述步长和迭代次数的设定方法包括：
    根据所述感兴趣区域的直方图分布，估计所述病灶的回声类型；
    基于所述回声类型确定对应的步长和迭代次数。
  6. 根据权利要求4所述的超声图像病灶的分割方法，其特征在于，所述回声类型包括无回声和低回声，所述基于所述回声类型确定对应的步长和迭代次数，包括：
    若所述回声类型为无回声时，则对应的迭代次数为120～260，时间步长为1.0；
    若所述回声类型为低回声时，则对应的迭代次数为650～950，时间步长为1.5。
  7. 根据权利要求1至6任一项所述的超声图像病灶的分割方法，其特征在于，所述利用距离正则化水平集演化算法基于所述感兴趣区域对所述初始零水平集函数进行迭代计算之前，还包括：
    将标准差为2，尺寸为15*15的高斯掩模对所述感兴趣区域进行卷积平滑，以得到卷积平滑后的高斯平滑图像；
    所述利用距离正则化水平集演化算法基于所述感兴趣区域对所述初始零水平集函数进行迭代计算，包括：
    利用距离正则化水平集演化算法基于所述高斯平滑图像对所述初始零水平集函数进行迭代计算。
  8. 一种超声图像病灶的分割装置，其特征在于，所述装置包括：
    感兴趣区域识别模块，用于识别超声图像中的病灶，以得到对应的感兴趣区域；
    质心估计模块，用于对所述感兴趣区域进行阈值分割，以得到对应的初始二值图像，并根据所述初始二值图像估计病灶区域的质心；
    函数构造模块，用于根据所述感兴趣区域以及所述病灶区域的质心，构造初始零水平集函数；
    正则化演化模块，用于利用距离正则化水平集演化算法基于所述感兴趣区域对所述初始零水平集函数进行迭代计算，得到演化终止时刻的目标水平集函数；
    病灶分割结果确定模块，用于基于所述目标水平集函数得到对应的目标零水平集二值图像，所述目标零水平集二值图像中前景与背景的边界像素为所述病灶区域的轮廓数据。
  9. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至7中任一项所述方法的步骤。
  10. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所 述计算机程序被处理器执行时实现权利要求1至7中任一项所述方法的步骤。
PCT/CN2020/133029 2019-12-25 2020-12-01 超声图像病灶的分割方法、装置和计算机设备 WO2021129325A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911358130.1A CN113112443B (zh) 2019-12-25 2019-12-25 超声图像病灶的分割方法、装置和计算机设备
CN201911358130.1 2019-12-25

Publications (1)

Publication Number Publication Date
WO2021129325A1 true WO2021129325A1 (zh) 2021-07-01

Family

ID=76575168

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/133029 WO2021129325A1 (zh) 2019-12-25 2020-12-01 超声图像病灶的分割方法、装置和计算机设备

Country Status (2)

Country Link
CN (1) CN113112443B (zh)
WO (1) WO2021129325A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113379739A (zh) * 2021-07-23 2021-09-10 平安科技（深圳）有限公司 超声图像的识别方法、装置、设备及存储介质
CN113724827A (zh) * 2021-09-03 2021-11-30 上海深至信息科技有限公司 一种超声报告中病灶区域自动标注的方法和系统
CN113780282A (zh) * 2021-09-15 2021-12-10 上海联影医疗科技股份有限公司 等中心位置确定方法、装置、计算机设备和可读存储介质
CN114757952A (zh) * 2022-06-15 2022-07-15 深圳瀚维智能医疗科技有限公司 超声图像处理方法、装置、设备及存储介质
CN115205246A (zh) * 2022-07-14 2022-10-18 中国南方电网有限责任公司超高压输电公司广州局 换流阀电晕放电紫外图像特征提取方法和装置
CN115797356A (zh) * 2023-02-09 2023-03-14 山东第一医科大学附属省立医院(山东省立医院) 一种核磁共振肿瘤区域提取方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113781477A (zh) * 2021-10-29 2021-12-10 平安科技（深圳）有限公司 基于人工智能的结石图像识别方法、装置及设备
CN114092475B (zh) * 2022-01-19 2022-04-22 浙江太美医疗科技股份有限公司 病灶长径确定方法、图像标注方法、装置及计算机设备
CN115797373A (zh) * 2023-01-09 2023-03-14 苏州浪潮智能科技有限公司 一种图像分割方法、装置、电子设备及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106570860A (zh) * 2016-10-31 2017-04-19 浙江工业大学 一种改进的基于距离正则化水平集的核磁共振图像咬肌分割方法
CN108875829A (zh) * 2018-06-20 2018-11-23 鲁东大学 一种乳腺肿瘤图像的分类方法及系统
CN110009645A (zh) * 2019-04-11 2019-07-12 东北大学 一种肝癌病灶影像的双层轮廓分割方法
CN110276356A (zh) * 2019-06-18 2019-09-24 南京邮电大学 基于r-cnn的眼底图像微动脉瘤识别方法
CN110288581A (zh) * 2019-06-26 2019-09-27 电子科技大学 一种基于保持形状凸性水平集模型的分割方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104732213B (zh) * 2015-03-23 2018-04-20 中山大学 一种基于乳腺磁共振图像的计算机辅助肿块检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106570860A (zh) * 2016-10-31 2017-04-19 浙江工业大学 一种改进的基于距离正则化水平集的核磁共振图像咬肌分割方法
CN108875829A (zh) * 2018-06-20 2018-11-23 鲁东大学 一种乳腺肿瘤图像的分类方法及系统
CN110009645A (zh) * 2019-04-11 2019-07-12 东北大学 一种肝癌病灶影像的双层轮廓分割方法
CN110276356A (zh) * 2019-06-18 2019-09-24 南京邮电大学 基于r-cnn的眼底图像微动脉瘤识别方法
CN110288581A (zh) * 2019-06-26 2019-09-27 电子科技大学 一种基于保持形状凸性水平集模型的分割方法

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
LI, CHUNMING ET AL.: "Distance Regularized Level Set Evolution and Its Application to Image Segmentation", IEEE TRANSACTIONS ON IMAGE PROCESSING, vol. 19, no. 12, 31 December 2010 (2010-12-31), XP011374153, DOI: 10.1109/TIP.2010.2069690 *
SELVATHI D.; BAMA S.: "Phase based distance regularized level set for the segmentation of ultrasound kidney images", PATTERN RECOGNITION LETTERS., vol. 86, 9 December 2016 (2016-12-09), NL, pages 9 - 17, XP029902545, ISSN: 0167-8655, DOI: 10.1016/j.patrec.2016.12.002 *
YANG, YI ET AL.: "Ultrasound Galactophore Lesion Extraction Based on Revised Level Set Image Segmentation", COMPUTER APPLICATIONS AND SOFTWARE, vol. 31, no. 11, 30 November 2014 (2014-11-30), pages 217 - 221,240, XP055824075 *
YU CHUN-YAN; ZHANG WEI-SHI; YU YING-YING; LI YING: "A novel active contour model for image segmentation using distance regularization term", COMPUTERS & MATHEMATICS WITH APPLICATIONS, vol. 65, no. 11, 17 April 2013 (2013-04-17), AMSTERDAM, NL, pages 1746 - 1759, XP028546275, ISSN: 0898-1221, DOI: 10.1016/j.camwa.2013.03.021 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113379739A (zh) * 2021-07-23 2021-09-10 平安科技（深圳）有限公司 超声图像的识别方法、装置、设备及存储介质
CN113379739B (zh) * 2021-07-23 2022-03-25 平安科技（深圳）有限公司 超声图像的识别方法、装置、设备及存储介质
CN113724827A (zh) * 2021-09-03 2021-11-30 上海深至信息科技有限公司 一种超声报告中病灶区域自动标注的方法和系统
CN113780282A (zh) * 2021-09-15 2021-12-10 上海联影医疗科技股份有限公司 等中心位置确定方法、装置、计算机设备和可读存储介质
CN113780282B (zh) * 2021-09-15 2024-03-26 上海联影医疗科技股份有限公司 等中心位置确定方法、装置、计算机设备和可读存储介质
CN114757952A (zh) * 2022-06-15 2022-07-15 深圳瀚维智能医疗科技有限公司 超声图像处理方法、装置、设备及存储介质
CN115205246A (zh) * 2022-07-14 2022-10-18 中国南方电网有限责任公司超高压输电公司广州局 换流阀电晕放电紫外图像特征提取方法和装置
CN115205246B (zh) * 2022-07-14 2024-04-09 中国南方电网有限责任公司超高压输电公司广州局 换流阀电晕放电紫外图像特征提取方法和装置
CN115797356A (zh) * 2023-02-09 2023-03-14 山东第一医科大学附属省立医院(山东省立医院) 一种核磁共振肿瘤区域提取方法

Also Published As

Publication number Publication date
CN113112443B (zh) 2024-03-22
CN113112443A (zh) 2021-07-13

Similar Documents

Publication Publication Date Title
WO2021129325A1 (zh) 超声图像病灶的分割方法、装置和计算机设备
WO2021129323A1 (zh) 超声图像病灶描述方法、装置、计算机设备和存储介质
Huang et al. Breast ultrasound image segmentation: a survey
Beeravolu et al. Preprocessing of breast cancer images to create datasets for deep-CNN
CN108010021B (zh) 一种医学图像处理系统和方法
Foruzan et al. Improved segmentation of low-contrast lesions using sigmoid edge model
George et al. Automated cell nuclei segmentation for breast fine needle aspiration cytology
Luo et al. A novel segmentation approach combining region-and edge-based information for ultrasound images
JP2010504129A (ja) 肺結節の高度コンピュータ支援診断
Militello et al. A fully automatic 2D segmentation method for uterine fibroid in MRgFUS treatment evaluation
Abbas et al. Combined spline and B-spline for an improved automatic skin lesion segmentation in dermoscopic images using optimal color channel
JP2006325937A (ja) 画像判定装置、画像判定方法およびそのプログラム
JP2006006359A (ja) 画像生成装置、画像生成装置方法、およびそのプログラム
Sethi et al. Segmentation of cancerous regions in liver using an edge-based and phase congruent region enhancement method
WO2021129324A1 (zh) 超声图像病灶的分割方法、装置和计算机设备
Cavalcanti et al. A coarse-to-fine approach for segmenting melanocytic skin lesions in standard camera images
Sagar et al. Color channel based segmentation of skin lesion from clinical images for the detection of melanoma
Iqbal et al. Potential lung nodules identification for characterization by variable multistep threshold and shape indices from CT images
Kaur et al. Computer-aided diagnosis of renal lesions in CT images: a comprehensive survey and future prospects
Lee et al. Unsupervised segmentation of lung fields in chest radiographs using multiresolution fractal feature vector and deformable models
Masood et al. Automated segmentation of skin lesions: Modified Fuzzy C mean thresholding based level set method
Faisal et al. Computer assisted diagnostic system in tumor radiography
KR101892949B1 (ko) 복부 컴퓨터단층촬영 영상에서 다중 아틀라스 기반 형상 및 밝기값 정보를 이용한 계층적 장기 분할 방법 및 장치
Jalab et al. Fractional Renyi entropy image enhancement for deep segmentation of kidney MRI
Joda et al. Digital mammogram enhancement based on automatic histogram clipping

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: 20907048

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: 20907048

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20907048

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 26.01.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20907048

Country of ref document: EP

Kind code of ref document: A1