WO2021078040A1 - 一种病灶的定位方法及装置 - Google Patents

一种病灶的定位方法及装置 Download PDF

Info

Publication number
WO2021078040A1
WO2021078040A1 PCT/CN2020/120627 CN2020120627W WO2021078040A1 WO 2021078040 A1 WO2021078040 A1 WO 2021078040A1 CN 2020120627 W CN2020120627 W CN 2020120627W WO 2021078040 A1 WO2021078040 A1 WO 2021078040A1
Authority
WO
WIPO (PCT)
Prior art keywords
point
spine
medical image
line
target
Prior art date
Application number
PCT/CN2020/120627
Other languages
English (en)
French (fr)
Inventor
李玉才
王少康
陈宽
Original Assignee
推想医疗科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 推想医疗科技股份有限公司 filed Critical 推想医疗科技股份有限公司
Publication of WO2021078040A1 publication Critical patent/WO2021078040A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30008—Bone
    • G06T2207/30012—Spine; Backbone

Definitions

  • This application relates to the technical field of medical diagnosis, and in particular to a method and device for locating a lesion.
  • the relative position of the lesion and the trunk that is, which area of the trunk the lesion is in (for example, the area includes the front, back, left, and right, etc.), is also of great significance for the diagnosis and treatment of the disease.
  • manual labeling or additional markers are usually used to determine the relative position of the lesion and the trunk.
  • the present application provides a method and device for locating a lesion, aiming to provide an efficient, accurate, and widely applicable lesion locating technology.
  • a method for locating lesions including:
  • the position parameter of the spine including the center point of the area occupied by the spine;
  • the extension line of the spine is a straight line extending forward of the spine from the center point, the spine
  • the forward direction is the direction opposite to the direction of the spinous process in the medical image.
  • the spinous process points from the first end to the second end, and the end of the spinous process closer to the vertebral body of the spine is The first end is the second end; the other end is the second end;
  • the area where the lesion is located is determined.
  • the location parameter further includes:
  • the angle of the spine, the angle of the spine is the angle between the extension line of the spine and the horizontal direction.
  • the determining the intersection of the extension line of the spine and the edge of the trunk includes:
  • the center point as the initial starting point, the reference point as the initial end point, and the midpoint between the starting point and the end point as the initial target point.
  • the target point located on the edge of the torso is taken as the intersection of the extension line of the spine and the edge of the torso.
  • the process of determining the positional relationship between the target point and the torso includes:
  • the target point is within the torso
  • the target point is outside the torso
  • the target point is on the edge of the torso.
  • the method before the update of the target point according to the following steps until the target point is a point on the edge of the torso, the method further includes:
  • the torso is segmented from the medical image to obtain a segmented image.
  • pixels of the torso are the target pixels, and other pixels are the background pixels.
  • the obtaining a reference point according to the equation of the center point, the angle, the straight line, and the size of the medical image includes:
  • the intersection of the extension line and the edge of the medical image is obtained as the reference point.
  • the detecting the position parameter of the spine from the medical image includes:
  • the position parameter of the spine in the medical image is determined, and the second parameter is the spine in each medical image before the medical image in the medical image sequence.
  • the positional parameters are determined, and the second parameter is the spine in each medical image before the medical image in the medical image sequence.
  • the determining the position parameter of the spine in the medical image according to the first parameter and the second parameter includes:
  • the weighted sum of the first parameter and the second parameter is used as the position of the spine in the medical image, wherein the weight of the first parameter is smaller than the weight of the second parameter.
  • a device for locating lesions including:
  • a spine detection unit configured to detect a position parameter of the spine from a medical image, the position parameter of the spine including the center point of the area occupied by the spine;
  • the intersection point determination unit is configured to determine the intersection point between the extension line of the spine and the edge of the trunk in the medical image, where the extension line of the spine starts from the center point and extends forward of the spine
  • the anterior direction of the spine is the direction opposite to the direction of the spinous process in the medical image.
  • the spinous process points from the first end to the second end, and the spinous process is a distance from the vertebral body of the spine. The closer end is the first end, and the other end is the second end;
  • the first boundary unit is configured to use the line connecting the center point and the intersection point as the boundary line between the left area and the right area of the medical image;
  • the second demarcation unit is used to determine the front-to-back demarcation line of the torso in the medical image according to the target perpendicular of the line connecting the center point and the intersection point, where the target perpendicular is the center point Among the perpendicular lines of the line with the intersection point, a perpendicular line passing through a predetermined point on the line between the center point and the intersection point;
  • the area determining unit is used to determine the area where the lesion is located according to the dividing line.
  • a processor for running a program wherein the above-mentioned method for locating a lesion is executed when the program is running.
  • a storage medium includes a stored program, wherein, when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned method for locating the lesion.
  • the lesion location method and device, processor, and storage medium provided in the present application detect the position parameters of the spine from a medical image, and in the medical image, determine the intersection of the extension line of the spine and the edge of the trunk.
  • the line connecting the center point and the intersection point is used as the dividing line between the left area and the right area of the medical image.
  • the front-to-back boundary line of the torso in the medical image is determined.
  • the vertical line is a vertical line that passes through a preset point on the line connecting the center point and the intersection point among the vertical lines connecting the center point and the intersection point. According to the dividing line, determine the area where the lesion is located.
  • Fig. 1a is a schematic diagram of a method for locating a lesion according to an embodiment of the application.
  • Fig. 1b is a schematic diagram of a CT image provided by an embodiment of the application.
  • Fig. 1c is a schematic diagram of another CT image provided by an embodiment of the application.
  • Fig. 1d is a schematic diagram of a medical image provided by an embodiment of the application.
  • Fig. 1e is a schematic diagram of another medical image provided by an embodiment of the application.
  • FIG. 1f is a schematic diagram of another medical image provided by an embodiment of the application.
  • FIG. 2 is a schematic diagram of a specific implementation manner of detecting the position parameter of the spine from a medical image provided by an embodiment of the application.
  • FIG. 3 is a schematic diagram of another method for locating a lesion according to an embodiment of the application.
  • FIG. 4 is a schematic structural diagram of a device for locating a lesion according to an embodiment of the application.
  • the embodiments of the present application provide an efficient, accurate, and widely applicable lesion location technology.
  • a schematic diagram of a method for locating a lesion provided in an embodiment of this application includes the following steps.
  • the position parameter of the spine includes the position coordinates of the center point of the area occupied by the spine.
  • the area occupied by the spine refers to the area of the human spine in a medical image (for example, a CT image).
  • a medical image for example, a CT image.
  • the circumscribed polygonal area of the area occupied by the imaging pixels of the spine in the cross-section of the human medical image for example, Rectangular area
  • the center point is the center of the circumscribed polygonal area (for example, rectangular area).
  • the polygonal area and the center point can be referred to the CT image shown in FIG. 1b.
  • the coordinate position of the center point of the area occupied by the spine in the medical image can be obtained preliminarily based on the deep learning algorithm.
  • S102 In the medical image, determine the intersection of the extension line of the spine and the edge of the trunk.
  • the extension line of the spine is a straight line extending forward of the spine from the center point of the area occupied by the spine.
  • the anterior direction of the spine is the direction opposite to the direction of the spinous process in the medical image.
  • the spinous process points from the first end to the second end, the end of the spinous process closer to the vertebral body of the spine is the first end, and the other end is the second end.
  • the extension line of the spine, the intersection of the extension line of the spine and the edge of the trunk, the anterior direction of the spine, the direction of the spinous process, and other physiological structures of the spine can be seen in the CT image shown in FIG. 1c.
  • the line connecting the center point and the intersection point divides the medical image into two regions.
  • the left and right are already marked in the existing CT images.
  • the dividing line between the left and right regions is given. It is also possible to mark the two areas as the left area and the right area according to the doctor's marking habits, which is not limited here.
  • the medical image shown in Figure 1d is not limited here.
  • S104 Determine the front-to-back dividing line of the torso in the medical image according to the target perpendicular line connecting the center point and the intersection point.
  • the target vertical line is a vertical line that passes through a preset point on the line connecting the center point and the intersection point among the vertical lines connecting the center point and the intersection point. Since the number of vertical lines connecting the center point and the intersection point is multiple, in this embodiment, the vertical line passing through the midpoint of the line connecting the center point and the intersection point can be used as the dividing line in the front-rear direction. In this case, the vertical line of the midpoint of the line between the center point and the intersection point further divides the medical image that has been divided into two areas into 4 areas, and the line is used as the left and right dividing line of the torso in the medical image.
  • the vertical line is used as the front-rear dividing line of the torso in the front-rear direction in the medical image, and four areas of the front left area, front right area, rear left area, and rear right area of the torso are obtained, for example, the medical image shown in FIG. 1e.
  • the preset point is not limited to the midpoint of the line between the center point and the intersection, but can also be a point at 3/5 of the line between the center point and the intersection.
  • the specific location of the preset point can be determined by the technician according to the actual situation.
  • the settings are not limited in the embodiment of this application.
  • the region can be further divided.
  • the torso in the medical image is divided into 8 regions according to the angle line that forms a preset angle (for example, 45°) with the above-mentioned vertical line, and the front left area, front right area, middle front left area, middle front right area of the torso are obtained.
  • Area, middle and rear left area, middle and rear right area, rear left area and rear right area, the specific trunk area distribution map can be seen in Figure 1f.
  • S105 Determine the area where the lesion is located according to the dividing line.
  • the specific location of the lesion in the medical image can be determined through the existing lesion recognition technology. After determining the specific location of the lesion, based on the specific location and boundary of the lesion, the area where the lesion is located can be determined. For example, if the specific location of the lesion is on the left side of the left and right dividing line and on the upper side of the front and rear dividing line, the lesion is located in the front left area of the trunk.
  • the position parameter of the spine is detected from the medical image, and in the medical image, the intersection of the extension line of the spine and the edge of the trunk is determined.
  • the line connecting the center point and the intersection point is used as the dividing line between the left area and the right area of the medical image.
  • the front-to-back boundary line of the torso in the medical image is determined.
  • the vertical line is a vertical line that passes through a preset point on the line connecting the center point and the intersection point among the vertical lines connecting the center point and the intersection point. According to the dividing line, determine the area where the lesion is located.
  • this application based on the physiological structure characteristics of the spine, determines the intersection point between the extension line of the spine and the edge of the trunk, and further determines the line between the center point and the intersection point of the area occupied by the spine and the vertical line of the line, thereby determining the points of different regions.
  • the boundary line realizes the purpose of determining the area where the lesion is located based on the boundary line. Because the physiological structure of the spine is relatively stable, it has a higher accuracy compared with manual labeling, and because it realizes automatic realization, it has higher efficiency compared with human labeling. Further, because there is no need Additional annotations, so the scope of application is wider.
  • a schematic diagram of a specific implementation manner of detecting the position parameter of the spine from a medical image includes the following steps.
  • S201 Input the medical image into a preset model, and obtain the position parameter of the spine output by the model as the first parameter.
  • the preset model includes, but is not limited to, a deep learning model such as a Single Shot MultiBox Detector (SSD) model.
  • SSD Single Shot MultiBox Detector
  • the specific process of performing target detection on medical images and obtaining the position of the spine includes:
  • A1 Use a medical image (for example, a CT image) as the input of the feature extraction module in the target detection model, and perform feature extraction on the medical image to obtain the spine features in the medical image.
  • a medical image for example, a CT image
  • the feature extraction module can be constructed based on the ResNet50+FPN structure.
  • the spine feature is used as the input of the prediction module in the target detection model, and the direction and angle of the spine feature are predicted to obtain the direction and angle of the spine.
  • the angle of the spine is the angle between the extension line of the spine and the horizontal direction.
  • the specific process of predicting the direction of the spine features includes: global max pooling (GMP) processing of the spine features, and the processed spine features as the input of the fully connected network.
  • GMP global max pooling
  • the output result of the connection network is used as the direction of the spine feature.
  • the specific process of predicting the angle of the spine features includes: performing Global Max Pooling (GMP) processing on the spine feature, using the processed spine feature as the input of another fully connected network, and outputting the fully connected network The result is the angle of the spine feature.
  • GMP Global Max Pooling
  • the spine feature is used as the input of the target detection module in the target detection model, and the position of the spine feature is detected to obtain the position coordinates of the center point of the area occupied by the spine.
  • the training process of the above-mentioned target detection model is similar to the detection process of the above-mentioned target detection model, and only the sample medical image and the spine features marked with specific coordinate positions are used as the input of the initial target detection model.
  • S202 Determine the position parameter of the spine in the medical image according to the first parameter and the second parameter.
  • the second parameter is the position parameter of the spine in each medical image before the medical image in the medical image sequence, and the weight of the first parameter is smaller than the weight of the second parameter.
  • the medical images will be input into the model in the form of image frames.
  • the model performs spine detection on each image frame according to the arrangement order of each image frame in the medical image sequence, and obtains the position parameter of the spine in each image frame.
  • EWMA Exponential Weighted Moving Average
  • the weighted sum of the first parameter and the second parameter may be used as the position parameter of the spine in the medical image. Based on the characteristics of the EWMA algorithm, the calculation process of the weighted sum of the first parameter and the second parameter is shown in formula (1).
  • y t represents the weighted sum of the first parameter and the second parameter
  • ⁇ represents the weight corresponding to the second parameter
  • y t-1 represents the second parameter
  • (1- ⁇ ) represents the weight corresponding to the first parameter.
  • Weight, x t represents the first parameter.
  • the deviation can be reduced by the above formula.
  • the first parameter x t is (70, 90)
  • the second parameter y t-1 is (40, 32)
  • the weight ⁇ corresponding to the second parameter is 0.95.
  • the angle of the spine can also be obtained based on the model, and the EWMA algorithm can also be used to optimize the spine angle of each image frame, thereby correcting the single-level error of the spine angle output by the model, and smoothing the final The output result.
  • the above formula can reduce the amount of deviation.
  • the first angle parameter x t is 90°
  • the second angle parameter y t-1 is 32°
  • the weight ⁇ corresponding to the second angle parameter is 0.95.
  • weighted sum is only a specific implementation of S202.
  • other calculation methods can also be used to determine the final position parameter based on the first parameter and the second parameter.
  • a preset model is used to obtain the position parameter of the spine output by the model as the first parameter. More importantly, the position parameter of the spine in each medical image before the medical image in the medical image sequence is used as the second parameter. Using the first parameter obtained by the second parameter optimization model to obtain the final position parameter is beneficial to improve the accuracy of the position parameter.
  • a schematic diagram of another method for locating a lesion provided in an embodiment of this application includes the following steps.
  • the position parameter includes the position coordinates of the center point of the area occupied by the spine and the angle of the spine, and the angle of the spine is the angle between the extension line of the spine and the horizontal direction.
  • the position coordinates of the center point of the area occupied by the spine and the angle of the spine can be obtained based on the steps shown in FIG. 2 above.
  • other existing deep learning model algorithms can also be used to obtain the coordinate position of the center point of the area occupied by the spine and the angle of the spine.
  • the value of the parameter a in the linear equation is usually tan ⁇ , and ⁇ is the angle of the spine.
  • the intersection point of the extension line and the edge of the medical image is obtained, and the intersection point is used as the reference point.
  • a rectangular coordinate system is established based on the medical image.
  • the two edges of the medical image are respectively used as the x-axis and y-axis of the rectangular coordinate system.
  • intersection point (0, B) is used as the reference point. If the value of A*w+B is in the range of [0, h], the intersection point (w, A* w+B) as a reference point.
  • S303 Segment the torso from the medical image to obtain a segmented image.
  • the pixels of the torso are the target pixels, and the other pixels are the background pixels.
  • the target pixel is displayed as 1 in the segmented image of the CT image
  • the background pixel is displayed as 0 in the segmented image of the CT image.
  • a threshold segmentation algorithm is used to separate the body parts from the medical image, and the maximum connected domain method is used to exclude other non-trunk parts (such as shoulders, arms) in the body parts, so as to obtain a segmented image with only the torso remaining.
  • S303 is not only the execution sequence provided in the embodiment of the present application, but S303 can also be executed before S301 and/or S302.
  • S304 Use the center point as the initial starting point, the reference point as the initial end point, and the midpoint between the start point and the end point as the initial target point, and update the target point according to the preset steps until the target point is on the edge of the torso Point.
  • the preset steps include: if the target point is inside the torso, the target point is used as the new starting point, and the reference point is used as the end point to update the target point; if the target point is outside the torso, the target point is used as the new End point, using the center point as the starting point to update the target point.
  • the target point is within the torso
  • the target point is outside the torso
  • the target point is on the edge of the torso.
  • the center point is (x v , y v )
  • the reference point is (x i , y i )
  • (x v , y v ) is used as the starting point
  • (x i , y i ) is used as the end point
  • the midpoint (x m , y m ) between the two points is calculated.
  • Analyze the specific distribution of points in the (x m , y m ) window for example, a 3 ⁇ 3 pixel window
  • the points in the window of (x m , y m ) are all 1, and then (x m , y m ) is within the torso.
  • the selection of the target point is not only the midpoint between the start point and the end point, but also the preset division position point between the start point and the end point (for example, the 3/5 point on the line between the start point and the end point). point).
  • the coordinates of the start point and the end point are the same, or the coordinate position of the start point exceeds the coordinate position of the end point, reselect a new reference point and/or center point and execute the preset steps.
  • S305 Use the target point located on the edge of the trunk as the intersection of the extension line of the spine and the edge of the trunk.
  • the line connecting the center point and the intersection point divides the medical image into two regions, and according to the doctor's labeling habits, the two regions are respectively labeled as the left region and the right region.
  • S307 Determine the boundary line of the torso in the front and back direction of the torso in the medical image by connecting the target perpendicular line connecting the center point and the intersection point.
  • the target vertical line connecting the center point and the intersection point further divides the medical image that has been divided into two areas into 4 regions.
  • the line is the left and right dividing line of the torso in the medical image, and the target vertical line is used as
  • the front-to-back dividing line of the torso in the front-to-back direction in the medical image is obtained from the front left area, front right area, back left area, and back right area of the torso.
  • S308 Determine the area where the lesion is located according to the dividing line.
  • the area where the lesion is located can be determined according to the specific location and boundary of the lesion.
  • the position parameter of the spine is obtained from the medical image, and the position parameter includes the center point of the area occupied by the spine and the angle of the spine, and the torso is segmented from the medical image to obtain the segmented image.
  • the reference point is obtained, and the intersection point between the extension line of the spine and the edge of the torso is determined according to the position relationship of the reference point in the segmented image.
  • the dividing line of the medical image is obtained. In order to determine the area where the lesion is located according to the dividing line.
  • the reference point obtained based on the center point of the area occupied by the spine and the angle of the spine can accurately obtain the intersection point between the extension line of the spine and the edge of the torso.
  • the dividing line of the medical image is obtained, so as to realize the determination of the area where the lesion is located. Since the physiological structure of the spine is relatively stable, the dividing line is obtained based on the center point and the intersection point, and there will be no regional labeling errors or boundary deviations, so it has higher accuracy than human labeling. And because it realizes automatic realization, it has higher efficiency compared with human labeling. Furthermore, because no additional annotations are required, the scope of application is wide.
  • a schematic structural diagram of a lesion locating device provided in this embodiment of the present application includes:
  • the spine detection unit 100 is used to detect the position parameter of the spine from the medical image, and the position parameter of the spine includes the center point of the area occupied by the spine.
  • the position parameters mentioned in the spine detection unit 100 also include: the angle of the spine, which is the angle between the extension line of the spine and the horizontal direction.
  • the spine detection unit 100 is specifically configured to: input the medical image into a preset model, and obtain the position parameter of the spine output by the model as the first parameter. According to the first parameter and the second parameter, the position parameter of the spine in the medical image is determined.
  • the second parameter is the position parameter of the spine in each medical image before the medical image in the medical image sequence.
  • the spine detection unit 100 determines the position parameter of the spine in the medical image according to the first parameter and the second parameter.
  • the specific implementation method includes: taking the weighted sum of the first parameter and the second parameter as the position of the spine in the medical image, where , The weight of the first parameter is less than the weight of the second parameter.
  • the intersection point determination unit 200 is used to determine the intersection point between the extension line of the spine and the edge of the torso in the medical image.
  • the extension line of the spine is a straight line extending from the center point to the forward direction of the spine, and the forward direction of the spine is medical In the image, the direction opposite to the direction of the spinous process, the spinous process points from the first end to the second end, the end of the spinous process closer to the vertebral body of the spine is the first end, and the other end is the second end.
  • the specific implementation manner of the intersection point determination unit 200 determining the intersection point between the extension line of the spine and the edge of the trunk includes: obtaining the reference point according to the center point, the angle, the equation of the straight line, and the size of the medical image.
  • the torso is segmented from the medical image to obtain a segmented image.
  • the pixels of the torso are the target pixels, and the other pixels are the background pixels.
  • the target point will be used as the new starting point and the reference point will be used as the end point to update the target point. If the target point is outside the torso, the target point is used as the new end point, and the center point is used as the starting point to update the target point.
  • the target point located on the edge of the torso is regarded as the intersection of the extension line of the spine and the edge of the torso.
  • the intersection point determination unit 200 obtains the reference point according to the center point, the angle, the equation of the straight line, and the size of the medical image, and the specific implementation method includes: using the center point and the angle to solve the straight line equation to obtain the equation of the extension line. According to the equation of the extension line and the size of the medical image, the intersection point of the extension line and the edge of the medical image is obtained as a reference point.
  • the process of determining the positional relationship between the target point and the torso in the intersection point determining unit 200 includes: if the points in the window including the target point are all target pixel points, the target point is within the torso. If the points in the window including the target point are all background pixels, the target point is outside the torso. If the point in the window including the target point includes the background pixel point and the target pixel point, the target point is on the edge of the torso.
  • the first dividing unit 300 is used to connect the line between the center point and the intersection point as the dividing line between the left area and the right area of the medical image.
  • the second dividing unit 400 is used to determine the front-to-back dividing line of the torso in the medical image according to the target vertical line connecting the center point and the intersection point, where the target vertical line is the line connecting the center point and the intersection point In the vertical line, the vertical line passing through the preset point on the line connecting the center point and the intersection point.
  • the area determining unit 500 is used to determine the area where the lesion is located according to the dividing line.
  • the position parameter of the spine is detected from the medical image, and in the medical image, the intersection of the extension line of the spine and the edge of the trunk is determined.
  • the line connecting the center point and the intersection point is used as the dividing line between the left area and the right area of the medical image.
  • the front-to-back boundary line of the torso in the medical image is determined.
  • the vertical line is a vertical line that passes through a preset point on the line connecting the center point and the intersection point among the vertical lines connecting the center point and the intersection point. According to the dividing line, determine the area where the lesion is located.
  • this application determines the intersection point between the extension line of the spine and the edge of the trunk based on the physiological structure characteristics of the spine, and further determines the line between the center point and the intersection point of the area occupied by the spine and the vertical line of the line, thereby determining the points of different regions.
  • the boundary line achieves the purpose of determining the area where the lesion is located based on the boundary line. Because the physiological structure of the spine is relatively stable, it has a higher accuracy than artificial marking. Moreover, because of the realization of automatic realization, it has higher efficiency compared with manual labeling. Furthermore, because no additional annotations are required, the scope of application is wide.
  • an embodiment of the present application further provides a processor configured to run a program, wherein the above-mentioned method for locating the lesion disclosed in the embodiment of the present application is executed when the program is running.
  • an embodiment of the present application also provides a storage medium on which a program is stored, and when the program is executed by a processor, the above-mentioned method for locating the lesion disclosed in the embodiment of the present application is realized.
  • the functions described in the methods of the embodiments of the present application are implemented in the form of software functional units and sold or used as independent products, they can be stored in a storage medium readable by a computing device.
  • a computing device which may be a personal computer, a server, a mobile computing device, or a network device, etc.
  • a computing device which may be a personal computer, a server, a mobile computing device, or a network device, etc.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other media that can store program codes. .

Abstract

一种病灶的定位方法，包括：从医学图像中检测脊椎的位置参数（S101），又在医学图像中，确定脊椎的延长线与躯干的边缘的交点（S102）；将中心点与交点的连线，作为医学图像的左区域和右区域的分界线（S103），依据中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线（S104）；依据分界线，确定病灶所在的区域（S105）。该方法具有较高的准确性和效率。

Description

一种病灶的定位方法及装置 技术领域
本申请涉及医疗诊断技术领域，尤其涉及一种病灶的定位方法及装置。
发明背景
目前，基于深度学习的一些算法，已经能够在医学影像例如CT图像中，确定出病灶的坐标位置。
但除了坐标位置，病灶与躯干的相对位置，即病灶处于躯干的哪个区域(如区域包括前后左右等)，也对于疾病的诊断和治疗方式的选择有重要意义。现有技术中，通常使用人工标注或者借助额外标记物的方式，确定病灶与躯干的相对位置。
但人工标注的方式的效率和准确性均不高。借助额外标记物的方式，需要在拍摄医学影像时将病区固定在标记物的附近，但有些病人因为外伤等原因，无法实现这种特殊位置的固定，所以，借助额外标记物的方式的适用性有限。
发明内容
本申请提供了一种病灶的定位方法及装置，目的在于提供一种高效准确且适用性广的病灶的定位技术。
为了实现上述目的，本申请提供了以下技术方案。
一种病灶的定位方法，包括：
从医学图像中检测脊椎的位置参数，所述脊椎的位置参数包括所述脊椎所占区域的中心点；
在所述医学图像中，确定所述脊椎的延长线与躯干的边缘的交点，所述脊椎的延长线为以所述中心点为起点，向所述脊椎的前向延伸的直线，所述脊椎的前向为所述医学图像中，与棘突的指向相反的方向，所述棘突从第一端指向第二端，所述棘突上距离所述脊椎的椎体较近的端为所述第一端，另一端为所述第二端；
将所述中心点与所述交点的连线，作为所述医学图像的左区域和右区域的分界线；
依据所述中心点与所述交点的连线的目标垂线，确定所述医学图像中所述躯干在前后方向的分界线，所述目标垂线为所述中心点与所述交点的连线的垂线中，经过所述中心点与所述交点的连线上的预设点的垂线；
依据所述分界线，确定病灶所在的区域。
可选的，所述位置参数还包括：
所述脊椎的角度，所述脊椎的角度为所述脊椎的延长线与水平方向的夹角。
可选的，所述确定所述脊椎的延长线与躯干的边缘的交点，包括：
依据所述中心点、所述角度、直线的方程和所述医学图像的尺寸，得到参考点；
将所述中心点作为初始的起点，将所述参考点作为初始的终点，将所述起点与所述终点的中点，作为初始的目标点，按照以下步骤更新目标点，直到所述目标点为所述躯干的边缘上的点：如果所述目标点在所述躯干之内，则将所述目标点作为新的起点，将所述参考点作为终点，更新所述目标点；如果所述目标点在所述躯干之外，则将所述目标点作为新的终点，将所述中心点作为起点，更新所述目标点；
将位于所述躯干的边缘上的目标点作为所述脊椎的延长线与所述躯干的边缘的交点。
可选的，确定所述目标点与躯干的位置关系的过程包括：
如果包括所述目标点的窗口中的点均为目标像素点，则所述目标点在所述躯干之内；
如果包括所述目标点的窗口中的点均为背景像素点，则所述目标点在所述躯干之外；
如果包括所述目标点的窗口中的点包括所述背景像素点和所述目标像素点，则所述目标点在所述躯干的边缘上。
可选的，在所述按照以下步骤更新目标点，直到所述目标点为所述躯干的边缘上的点之前，还包括：
从所述医学图像中分割所述躯干，得到分割图像，所述分割图像中，所述躯干的像素点为所述目标像素点，其它像素点为所述背景像素点。
可选的，所述依据所述中心点、所述角度、直线的方程和所述医学图像的尺寸，得到参考点，包括：
利用所述中心点与所述角度求解直线方程，得到所述延长线的方程；
依据所述延长线的方程与所述医学图像的尺寸，得到所述延长线与所述医学图像的边缘的交点，作为所述参考点。
可选的，所述从医学图像中检测脊椎的位置参数，包括：
将所述医学图像输入预设的模型，得到所述模型输出的所述脊椎的位置参数，作为第一参数；
依据所述第一参数和第二参数，确定所述医学图像中的所述脊椎的位置参数，所述第二参数为，医学图像序列中、所述医学图像之前的各个医学图像中所述脊椎的位置参数。
可选的，所述依据所述第一参数和第二参数，确定所述医学图像中的所述脊椎的位置参数，包括：
将所述第一参数与所述第二参数的加权和，作为所述医学图像中的所述脊椎的位置，其中，所述第一参数的权重小于所述第二参数的权重。
一种病灶的定位装置，包括：
脊椎检测单元，用于从医学图像中检测脊椎的位置参数，所述脊椎的位置参数包括所述脊椎所占区域的中心点；
交点确定单元，用于在所述医学图像中，确定所述脊椎的延长线与躯干的边缘的交点，所述脊椎的延长线为以所述中心点为起点，向所述脊椎的前向延伸的直线，所述脊椎的前向为所述医学图像中，与棘突的指向相反的方向，所述棘突从第一端指向第二端，所述棘突上距离所述脊椎的椎体较近的端为所述第一端，另一端为所述第二端；
第一分界单元，用于将所述中心点与所述交点的连线，作为所述医学图像的左区域和右区域的分界线；
第二分界单元，用于依据所述中心点与所述交点的连线的目标垂线，确定所述医学图像中所述躯干在前后方向的分界线，所述目标垂线为所述中心点与所述交点的连线的垂线中，经过所述中心点与所述交点的连线上的预设点的垂线；
区域确定单元，用于依据所述分界线，确定病灶所在的区域。
一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行上述的病灶的定位方法。
一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行上述的病灶的定位方法。
本申请提供的病灶的定位方法及装置、处理器、以及存储介质，从医学图像中检测脊椎的位置参数，又在医学图像中，确定脊椎的延长线与躯干的边缘的交点。将中心点与交点的连线，作为医学图像的左区域和右区域的分界线，依据中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线，其中，目标垂线为中心点与交点的连线的垂线中，经过中心点与交点的连线上的预设点的垂线。依据分界线，确定病灶所在的区域。可见，基于脊椎的生理结构特点，确定脊椎的延长线与躯干的边缘的交点，进一步确定脊椎所占区域的中心点与交点的连线以及连线的垂线，从而确定不同区域的分界线，实现依据分界线，确定病灶所在的区域的目的，因为脊椎的生理结构特点比较稳定，因此，与人为标注相比，具有较高的准确性。并且，因为实现了自动实现，所以与人为标注相比，具有较高的效率。进一步的，因为无需额外的标注物，所以适用范围较广。
附图简要说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1a为本申请实施例提供的一种病灶的定位方法的示意图。
图1b为本申请实施例提供的一种CT图像的示意图。
图1c为本申请实施例提供的另一种CT图像的示意图。
图1d为本申请实施例提供的一种医学图像的示意图。
图1e为本申请实施例提供的另一种医学图像的示意图。
图1f为本申请实施例提供的又一种医学图像的示意图。
图2为本申请实施例提供的一种从医学图像中检测脊椎的位置参数的具体实现方式的示意图。
图3为本申请实施例提供的另一种病灶的定位方法的示意图。
图4为本申请实施例提供的一种病灶的定位装置的结构示意图。
实施本发明的方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
由背景技术可知，医生为了提高疾病诊断结果的准确性，除了基于医学影像例如腹部或者胸部的CT图像，确定病灶在患者腹部或者胸部的坐标位置之外，还会标注病灶与躯体的相对位置，即病灶处于腹部或者胸部的哪个区域(如区域包括前后左右等)。其中，前后左右区域用于表示真实人体中的前后左右，映射在腹部或者胸部的CT图像中的区域。当然，也可以根据医生的标注习惯，将区域划分为多种(例如前中后左右)。
由于现有技术中病灶与躯干的相对位置的定位技术，定位的效率和准确性均不高。因此，本申请实施例提供了一种高效准确且适用性广的病灶的定位技术。
如图1a所示，为本申请实施例提供的一种病灶的定位方法的示意图，包括如下步骤。
S101：从医学图像中检测脊椎的位置参数。
其中，脊椎的位置参数包括脊椎所占区域的中心点的位置坐标。脊椎所占区域指的是人体脊椎在医学图像(例如CT图像)中的区域。通常情况下，依据医 学影像中的横断位(Axial)方向，(即人体的头顶至人体的脚底的观察方向)将人体医学图像的横截面中脊椎的成像像素所占区域的外接多边形区域(例如矩形区域)，作为脊椎所占区域。相应的，中心点则为该外接多边形区域(例如矩形区域)的中心。具体的，多边形区域和中心点可参见图1b示出的CT图像。
在本申请实施例中，可以基于深度学习算法，初步得到脊椎所占区域的中心点在医学图像中的坐标位置。
基于深度学习算法，得到脊椎所占区域的中心点在医学图像中的坐标位置这一过程的具体实现方式，可参见下述图2及图2示出的解释说明。当然，也可以采用现有的其他人体脊椎定位技术手段，得到该脊椎所占区域的中心点的具体位置。
S102：在医学图像中，确定脊椎的延长线与躯干的边缘的交点。
其中，脊椎的延长线为以脊椎所占区域的中心点为起点，向脊椎的前向延伸的直线。脊椎的前向为医学图像中，与棘突的指向相反的方向。棘突从第一端指向第二端，棘突上距离脊椎的椎体较近的端为第一端，另一端为第二端。具体的，脊椎的延长线、以及脊椎的延长线与躯干的边缘的交点，脊椎的前向、棘突的指向等脊椎生理结构，可参见图1c示出的CT图像。
S103：将中心点与交点的连线，作为医学图像的左区域和右区域的分界线。
其中，中心点与交点的连线，将医学图像划分为两个区域。需要说明的是，通常，现有的CT图像中已经标注出左和右，本实施例中，以给出左右区域之间的分界线。也可以，依据医生的标注习惯，将两个区域分别标注为左区域和右区域，这里不做限定。例如，图1d所示出的医学图像。
S104：依据中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线。
其中，目标垂线为中心点与交点的连线的垂线中，经过中心点与交点的连线上的预设点的垂线。由于中心点与交点的连线的垂线的数量为多个，本实施例中，可以将过中心点与交点的连线的中点的垂线，作为在前后方向的分界线。在此情况下，中心点与交点的连线的中点的垂线，将已经划分为两个区域的医学图像进一步划分为4个区域，连线作为医学图像中躯干在左右方向的左右分界线，垂线作为医学图像中躯干在前后方向的前后分界线，得到躯干的前左区域、前右区域、后左区域和后右区域这4个区域，例如，图1e所示出的医学图像。当然，预设点不仅仅局限于中心点与交点的连线的中点，也可以是中心点与交点的连线的3/5处的点，预设点的具体位置可由技术人员根据实际情况进行设置，本申请实施例不做限定。
需要说明的是，在将医学图像中躯干划分为4个区域的基础上，还可以进一步的进行区域划分。例如，依据与上述垂线成预设角度(例如45°)的夹角线将 医学图像中躯干划分为8个区域，得到躯干的前左区域、前右区域、中前左区域、中前右区域、中后左区域、中后右区域、后左区域和后右区域，具体躯干区域分布图可参见图1f。
S105：依据分界线，确定病灶所在的区域。
其中，可通过现有的病灶识别技术手段，确定医学图像中病灶的具体位置。在确定病灶的具体位置后，依据病灶的具体位置和分界线，便能够确定病灶所在的区域。例如，病灶的具体位置位于左右分界线的左边，并且位于前后分界线的上边，则病灶处于躯干的前左区域。
在本申请实施例中，从医学图像中检测脊椎的位置参数，又在医学图像中，确定脊椎的延长线与躯干的边缘的交点。将中心点与交点的连线，作为医学图像的左区域和右区域的分界线，依据中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线，其中，目标垂线为中心点与交点的连线的垂线中，经过中心点与交点的连线上的预设点的垂线。依据分界线，确定病灶所在的区域。可见本申请，基于脊椎的生理结构特点，确定脊椎的延长线与躯干的边缘的交点，进一步确定脊椎所占区域的中心点与交点的连线以及连线的垂线，从而确定不同区域的分界线，实现依据分界线，确定病灶所在的区域的目的。因为脊椎的生理结构特点比较稳定，因此，与人为标注相比，具有较高的准确性，并且，因为实现了自动实现，所以与人为标注相比，具有较高的效率，进一步的，因为无需额外的标注物，所以适用范围较广。
可选的，如图2所示，为本申请实施例提供的一种从医学图像中检测脊椎的位置参数的具体实现方式的示意图，包括如下步骤。
S201：将医学图像输入预设的模型，得到模型输出的脊椎的位置参数，作为第一参数。
其中，预设的模型包括但不限于是目标检测(Single ShotMultiBox Detector，SSD)模型等深度学习模型。通过目标检测模型，对医学图像进行目标检测，得到脊椎位置的具体过程包括：
A1、将医学图像(例如CT图像)作为目标检测模型中特征提取模块的输入，对医学图像进行特征提取，得到医学图像中的脊椎特征。
需要说明的是，可以基于ResNet50+FPN结构构建该特征提取模块。
A2、将脊椎特征作为目标检测模型中预测模块的输入，对脊椎特征进行方向和角度的预测，得到脊椎的方向和角度。其中，脊椎的角度为脊椎的延长线与水平方向的夹角。
需要说明的是，对脊椎特征进行方向的预测的具体过程包括：将脊椎特征进行全局最大池化(Global Max Pooling，GMP)处理，将处理后的脊椎特征作为全连接网络的输入，将该全连接网络输出的结果作为该脊椎特征的方向。
对脊椎特征进行角度的预测的具体过程包括：将脊椎特征进行全局最大池化(Global Max Pooling，GMP)处理，将处理后的脊椎特征作为另一全连接网络的输入，将该全连接网络输出的结果作为该脊椎特征的角度。
A3、将脊椎特征作为目标检测模型中目标检测模块的输入，对脊椎特征进行位置检测，得到脊椎所占区域的中心点的位置坐标。
需要强调的是，上述目标检测模型的训练过程，与上述目标检测模型的检测过程类似，仅使用了样本医学图像和标注具体坐标位置的脊椎特征作为初始目标检测模型的输入。
S202：依据第一参数与第二参数，确定医学图像中的脊椎的位置参数。
其中，第二参数为，医学图像序列中、医学图像之前的各个医学图像中脊椎的位置参数，并且，第一参数的权重小于第二参数的权重。
需要说明的是，在将医学图像输入预设的模型的过程中，医学图像会以图像帧的方式输入到模型中。模型依据医学图像序列中各个图像帧的排列顺序，对各个图像帧进行脊椎检测，得到每个图像帧中的脊椎的位置参数。但深度学习的模型难免会出现错误，并且因为一次拍摄得到的医学图像帧序列中，脊椎的位置参数具有连续性，因此，本实施例中，利用指数加权移动平均(Exponential Weighted Moving Average，EWMA)算法，为每个图像帧中的脊椎位置参数进行优化。
具体的，可以将第一参数和第二参数的加权和，作为医学图像中的脊椎的位置参数。基于EWMA算法的特性，第一参数和第二参数的加权和的计算过程如公式(1)所示。
y t＝γy t-1+(1-γ)x t    (1)
在公式(1)中，y t表示第一参数和第二参数的加权和，γ表示第二参数对应的权重，y t-1表示第二参数，(1-γ)表示第一参数对应的权重，x t表示第一参数。
假设模型预测的脊椎所占区域的中心点的位置坐标发生了较大偏差，则通过上述公式，能够减少偏差量。例如，第一参数x t为(70，90)，第二参数y t-1为(40，32)，第二参数对应的权重γ为0.95。则加权和为(40×0.95+(1-0.95)×70，32×0.95+(1-0.95)×90)＝(41.5，34.9)。可见，优化后的位置坐标与上一帧中的脊椎所占区域的位置坐标相比，偏差不大。
此外，不仅仅是脊椎的位置，脊椎的角度也可基于模型得到，并且也可以利用EWMA算法，为每个图像帧的脊椎角度进行优化，从而纠正模型输出脊椎角度的单一层面误差，并平滑最终的输出结果。
假设模型预测的脊椎的角度发生了较大偏差，则通过上述公式，能够减少偏差量。例如，第一角度参数x t为90°，第二角度参数y t-1为32°，第二角度参数对应的权重γ为0.95。相应的，参照公式(1)的计算过程，第一角度参数和第二角度参数的加权和为0.95×32+(1-0.95)×90＝34.9。可见，优化后的角度与上一 帧中的脊椎的角度相比，偏差不大。
需要说明的是，加权和仅仅为S202的一种具体实现方式，除了加权和之外，也可以使用其他运算方式，基于第一参数和第二参数确定最终的位置参数。
在本申请实施例中，使用预设的模型，得到模型输出的脊椎的位置参数，作为第一参数。更重要的是，将医学图像序列中、医学图像之前的各个医学图像中脊椎的位置参数，作为第二参数。使用第二参数优化模型得到的第一参数，得到最终的位置参数，有利于提高位置参数的准确性。
可选的，如图3所示，为本申请实施例提供的另一种病灶的定位方法的示意图，包括如下步骤。
S301：从医学图像中检测脊椎的位置参数。
其中，位置参数包括脊椎所占区域的中心点的位置坐标、以及脊椎的角度，脊椎的角度为脊椎的延长线与水平方向的夹角。在本申请实施例中，可以基于上述图2示出的步骤，得到脊椎所占区域的中心点的位置坐标和脊椎的角度。当然，也可以采用现有的其它深度学习模型算法，得到脊椎所占区域的中心点的坐标位置和脊椎的角度。
S302：依据中心点、角度、直线的方程和医学图像的尺寸，得到参考点。
其中，直线的方程为ax+b＝y，利用中心点与角度求解直线方程中的参数a的值和参数b的值，得到脊椎的延长线的方程。需要说明的是，直线方程中的参数a的值通常为tanθ，θ为脊椎的角度。
进一步的，依据脊椎的延长线的方程与医学图像的尺寸，得到延长线与医学图像的边缘的交点，并将交点作为参考点。
具体的，依据医学图像建立直角坐标系，医学图像的两条边缘分别作为直角坐标系的x轴和y轴，医学图像的尺寸的宽度为w、高度为h，脊椎的延长线方程为Ax+B＝y。分别将x＝0,x＝w代入延长线方程中，计算得到延长线与医学图像的边缘(该边缘与y轴平行)的交点(0，B)和交点(w，A*w+B)，并分别判断B的值和A*w+B的值是否处于[0，h]的范围内。若B的值处于[0，h]的范围内，则交点(0，B)作为参考点，若A*w+B的值处于[0，h]的范围内，则交点(w，A*w+B)作为参考点。
此外，分别将y＝0,y＝h代入延长线方程中，计算得到延长线与医学图像的边缘(该边缘与x轴平行)的交点(-B/A，0)和交点((h-B)/A，h)，并分别判断-B/A的值和(h-B)/A的值是否处于[0，w]的范围内。若-B/A的值处于[0，w]的范围内，则交点(-B/A，0)作为参考点，若(h-B)/A的值处于[0，w]的范围内，则交点((h-B)/A，h)作为参考点。
S303：从医学图像中分割躯干，得到分割图像。
其中，在分割图像中，躯干的像素点为目标像素点，其它像素点为背景像素 点。以CT图像的分割图像(通常为二值图像)为例，目标像素点在CT图像的分割图像中显示为1，背景像素点在CT图像的分割图像中显示为0。
具体的，通过阈值分割算法将从医学图像中分离出身体部位，并利用最大连通域方法排除身体部位中的其他非躯干部分(例如肩部、手臂)，从而得到仅保留躯干的分割图像。
需要说明的是，上述具体实现过程仅仅用于举例说明，从医学图像中分割躯干的具体实现过程当然还可以采用现有的其他图像分割手段来实现，这里不再赘述。
此外，需要强调的是，S303的执行顺序不仅仅是本申请实施例所提供的执行顺序，S303也可以在S301和/或S302之前执行。
S304：将中心点作为初始的起点，将参考点作为初始的终点，将起点与终点的中点，作为初始的目标点，并按照预设的步骤更新目标点，直到目标点为躯干的边缘上的点。
其中，预设的步骤包括：如果目标点在躯干之内，则将目标点作为新的起点，将参考点作为终点，更新目标点；如果目标点在躯干之外，则将目标点作为新的终点，将中心点作为起点，更新目标点。
可选的，
如果包括目标点的窗口中的点均为目标像素点，则目标点在躯干之内；
如果包括目标点的窗口中的点均为背景像素点，则目标点在躯干之外；
如果包括目标点的窗口中的点包括背景像素点和目标像素点，则目标点在躯干的边缘上。
具体的，以CT图像的分割图像为例，中心点为(x v，y v)，参考点为(x i，y i)，将(x v，y v)作为起点，将(x i，y i)作为终点，计算得到两点之间的中点(x m，y m)。分析(x m，y m)的窗口(例如3×3的像素窗口)中的点的具体分布情况(窗口中的点只由0或1表示)。(x m，y m)的窗口中的点均为1，则(x m，y m)在躯干之内。
接着，将(x m，y m)作为新的起点，将(x i，y i)作为新的终点，计算得到两点之间的中点(x n，y n)。分析(x n，y n)的窗口(例如3×3的像素窗口)中的点的具体分布情况。(x n，y n)的窗口中的点均为0，则(x n，y n)在躯干之外。
其次，将(x m，y m)作为新的起点，将(x n，y n)作为新的终点，计算得到两点之间的中点(x k，y k)。分析(x k，y k)的窗口(例如3×3的像素窗口)中的点的具体分布情况。(x k，y k)的窗口中的点包括0和1，则(x k，y k)在躯干的边缘上。
需要说明的是，上述具体实现过程仅仅用于举例说明。
需要强调的是，目标点的选取不仅仅是起点和终点之间的中点，也可以是起点和终点之间的预设划分位置点(例如起点和终点之间连线上3/5处的点)。此外，若是起点的坐标和终点的坐标相同，或者是，起点的坐标位置超过终点的坐标位 置，则重新选择新的参考点和/或中心点，并执行预设的步骤。
S305：将位于躯干的边缘上的目标点作为脊椎的延长线与躯干的边缘的交点。
S306：将中心点与交点的连线，作为医学图像的左区域和右区域的分界线。
其中，中心点与交点的连线，将医学图像划分为两个区域，并依据医生的标注习惯，将两个区域分别标注为左区域和右区域。
S307：将中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线。
其中，中心点与交点的连线的目标垂线，将已经划分为两个区域的医学图像进一步划分为4个区域，连线作为医学图像中躯干在左右方向的左右分界线，目标垂线作为医学图像中躯干在前后方向的前后分界线，得到躯干的前左区域、前右区域、后左区域和后右区域这4个区域。
S308：依据分界线，确定病灶所在区域。
其中，在确定病灶的具体位置后，依据病灶的具体位置和分界线，便能够确定病灶所在的区域。
在本申请实施例中，从医学图像中获取脊椎的位置参数，该位置参数包括脊椎所占区域的中心点和脊椎的角度，并从医学图像中分割出躯干，得到分割图像。依据中心点和角度，得到参考点，依据参考点在分割图像中的位置关系，确定脊椎的延长线与躯干边缘的交点。基于中心点和交点，得到医学图像的分界线。从而依据分界线确定病灶所在区域。可见，无论医学图像中躯干如何摆布，基于脊椎所占区域的中心点和脊椎的角度所得到的参考点，能够准确得到脊椎的延长线与躯干边缘的交点。依据中心点和交点，得到医学图像的分界线，从而实现对病灶所在区域的确定。由于脊椎的生理结构特点比较稳定，依据中心点和交点得到分界线，不会出现区域标注错误或者分界偏差，因此与人为标注相比，具有较高的准确性。并且因为实现了自动实现，所以与人为标注相比，具有较高的效率。进一步的，因为无需额外的标注物，所以适用范围较广。
与上述本申请实施例提供的病灶的定位方法相对应，如图4所示，为本申请实施例提供的一种病灶的定位装置的结构示意图，包括：
脊椎检测单元100，用于从医学图像中检测脊椎的位置参数，脊椎的位置参数包括脊椎所占区域的中心点。
其中，脊椎检测单元100中所提及的位置参数还包括：脊椎的角度，脊椎的角度为脊椎的延长线与水平方向的夹角。
脊椎检测单元100具体用于：将医学图像输入预设的模型，得到模型输出的脊椎的位置参数，作为第一参数。依据第一参数和第二参数，确定医学图像中的脊椎的位置参数，第二参数为，医学图像序列中、医学图像之前的各个医学图像中脊椎的位置参数。
脊椎检测单元100依据第一参数和第二参数，确定医学图像中的脊椎的位置参数的具体实现方式包括：将第一参数与第二参数的加权和，作为医学图像中的脊椎的位置，其中，第一参数的权重小于第二参数的权重。
交点确定单元200，用于在医学图像中，确定脊椎的延长线与躯干的边缘的交点，脊椎的延长线为以中心点为起点，向脊椎的前向延伸的直线，脊椎的前向为医学图像中，与棘突的指向相反的方向，棘突从第一端指向第二端，棘突上距离脊椎的椎体较近的端为第一端，另一端为第二端。
其中，交点确定单元200确定脊椎的延长线与躯干的边缘的交点的具体实现方式包括：依据中心点、角度、直线的方程和医学图像的尺寸，得到参考点。从医学图像中分割躯干，得到分割图像，分割图像中，躯干的像素点为目标像素点，其它像素点为背景像素点。将中心点作为初始的起点，将参考点作为初始的终点，将起点与终点的中点，作为初始的目标点，按照以下步骤更新目标点，直到目标点为躯干的边缘上的点：如果目标点在躯干之内，则将目标点作为新的起点，将参考点作为终点，更新目标点。如果目标点在躯干之外，则将目标点作为新的终点，将中心点作为起点，更新目标点。将位于躯干的边缘上的目标点作为脊椎的延长线与躯干的边缘的交点。
交点确定单元200依据中心点、角度、直线的方程和医学图像的尺寸，得到参考点的具体实现方式包括：利用中心点与角度求解直线方程，得到延长线的方程。依据延长线的方程与医学图像的尺寸，得到延长线与医学图像的边缘的交点，作为参考点。
此外，交点确定单元200中确定目标点与躯干的位置关系的过程包括：如果包括目标点的窗口中的点均为目标像素点，则目标点在躯干之内。如果包括目标点的窗口中的点均为背景像素点，则目标点在躯干之外。如果包括目标点的窗口中的点包括背景像素点和目标像素点，则目标点在躯干的边缘上。
第一分界单元300，用于将中心点与交点的连线，作为医学图像的左区域和右区域的分界线。
第二分界单元400，用于依据中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线，所述目标垂线为所述中心点与所述交点的连线的垂线中，经过所述中心点与所述交点的连线上的预设点的垂线。
区域确定单元500，用于依据分界线，确定病灶所在的区域。
在本申请实施例中，从医学图像中检测脊椎的位置参数，又在医学图像中，确定脊椎的延长线与躯干的边缘的交点。将中心点与交点的连线，作为医学图像的左区域和右区域的分界线，依据中心点与交点的连线的目标垂线，确定医学图像中躯干在前后方向的分界线，其中，目标垂线为中心点与交点的连线的垂线中，经过中心点与交点的连线上的预设点的垂线。依据分界线，确定病灶所在的区域。 可见本申请，基于脊椎的生理结构特点，确定脊椎的延长线与躯干的边缘的交点，进一步确定脊椎所占区域的中心点与交点的连线以及连线的垂线，从而确定不同区域的分界线，实现依据分界线，确定病灶所在的区域的目的，因为脊椎的生理结构特点比较稳定，因此，与人为标注相比，具有较高的准确性。并且，因为实现了自动实现，所以与人为标注相比，具有较高的效率。进一步的，因为无需额外的标注物，所以适用范围较广。
进一步的，本申请实施例还提供了一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行上述本申请实施例公开的病灶的定位方法。
进一步的，本申请实施例还提供了一种存储介质，其上存储有程序，该程序被处理器执行时实现上述本申请实施例公开的病灶的定位方法。
本申请实施例方法所述的功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算设备可读取存储介质中。基于这样的理解，本申请实施例对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该软件产品存储在一个存储介质中，包括若干指令用以使得一台计算设备(可以是个人计算机，服务器，移动计算设备或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似部分互相参见即可。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本申请。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本申请的精神或范围的情况下，在其它实施例中实现。因此，本申请将不会被限制于本文所示的这些实施例，而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。

Claims (11)

  1. 一种病灶的定位方法，包括：
    从医学图像中检测脊椎的位置参数，所述脊椎的位置参数包括所述脊椎所占区域的中心点；
    在所述医学图像中，确定所述脊椎的延长线与躯干的边缘的交点，所述脊椎的延长线为以所述中心点为起点，向所述脊椎的前向延伸的直线，所述脊椎的前向为所述医学图像中，与棘突的指向相反的方向，所述棘突从第一端指向第二端，所述棘突上距离所述脊椎的椎体较近的端为所述第一端，另一端为所述第二端；
    将所述中心点与所述交点的连线，作为所述医学图像的左区域和右区域的分界线；
    依据所述中心点与所述交点的连线的目标垂线，确定所述医学图像中所述躯干在前后方向的分界线，所述目标垂线为所述中心点与所述交点的连线的垂线中，经过所述中心点与所述交点的连线上的预设点的垂线；
    依据所述分界线，确定病灶所在的区域。
  2. 根据权利要求1所述的方法，其中，所述位置参数还包括：
    所述脊椎的角度，所述脊椎的角度为所述脊椎的延长线与水平方向的夹角。
  3. 根据权利要求2所述的方法，其中，所述确定所述脊椎的延长线与躯干的边缘的交点，包括：
    依据所述中心点、所述角度、直线的方程和所述医学图像的尺寸，得到参考点；
    将所述中心点作为初始的起点，将所述参考点作为初始的终点，将所述起点与所述终点的中点，作为初始的目标点，按照以下步骤更新目标点，直到所述目标点为所述躯干的边缘上的点：如果所述目标点在所述躯干之内，则将所述目标点作为新的起点，将所述参考点作为终点，更新所述目标点；
    如果所述目标点在所述躯干之外，则将所述目标点作为新的终点，将所述中心点作为起点，更新所述目标点；
    将位于所述躯干的边缘上的目标点作为所述脊椎的延长线与所述躯干的边缘的交点。
  4. 根据权利要求3所述的方法，其中，确定所述目标点与躯干的位置关系的过程包括：
    如果包括所述目标点的窗口中的点均为目标像素点，则所述目标点在所述躯干之内；
    如果包括所述目标点的窗口中的点均为背景像素点，则所述目标点在所述躯 干之外；
    如果包括所述目标点的窗口中的点包括所述背景像素点和所述目标像素点，则所述目标点在所述躯干的边缘上。
  5. 根据权利要求4所述的方法，其中，在所述按照以下步骤更新目标点，直到所述目标点为所述躯干的边缘上的点之前，还包括：
    从所述医学图像中分割所述躯干，得到分割图像，所述分割图像中，所述躯干的像素点为所述目标像素点，其它像素点为所述背景像素点。
  6. 根据权利要求3至5任一项所述的方法，其中，所述依据所述中心点、所述角度、直线的方程和所述医学图像的尺寸，得到参考点，包括：
    利用所述中心点与所述角度求解直线方程，得到所述延长线的方程；
    依据所述延长线的方程与所述医学图像的尺寸，得到所述延长线与所述医学图像的边缘的交点，作为所述参考点。
  7. 根据权利要求1-6任一项所述的方法，其中，所述从医学图像中检测脊椎的位置参数，包括：
    将所述医学图像输入预设的模型，得到所述模型输出的所述脊椎的位置参数，作为第一参数；
    依据所述第一参数和第二参数，确定所述医学图像中的所述脊椎的位置参数，所述第二参数为，医学图像序列中、所述医学图像之前的各个医学图像中所述脊椎的位置参数。
  8. 根据权利要求7所述的方法，其中，所述依据所述第一参数和第二参数，确定所述医学图像中的所述脊椎的位置参数，包括：
    将所述第一参数与所述第二参数的加权和，作为所述医学图像中的所述脊椎的位置，其中，所述第一参数的权重小于所述第二参数的权重。
  9. 一种病灶的定位装置，包括：
    脊椎检测单元，用于从医学图像中检测脊椎的位置参数，所述脊椎的位置参数包括所述脊椎所占区域的中心点；
    交点确定单元，用于在所述医学图像中，确定所述脊椎的延长线与躯干的边缘的交点，所述脊椎的延长线为以所述中心点为起点，向所述脊椎的前向延伸的直线，所述脊椎的前向为所述医学图像中，与棘突的指向相反的方向，所述棘突从第一端指向第二端，所述棘突上距离所述脊椎的椎体较近的端为所述第一端，另一端为所述第二端；
    第一分界单元，用于将所述中心点与所述交点的连线，作为所述医学图像的左区域和右区域的分界线；
    第二分界单元，用于依据所述中心点与所述交点的连线的目标垂线，确定所述医学图像中所述躯干在前后方向的分界线，所述目标垂线为所述中心点与所述 交点的连线的垂线中，经过所述中心点与所述交点的连线上的预设点的垂线；
    区域确定单元，用于依据所述分界线，确定病灶所在的区域。
  10. 一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行权利要求1-8中任一项所述的病灶的定位方法。
  11. 一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行如权利要求1-8中任一项所述的病灶的定位方法。
PCT/CN2020/120627 2019-10-21 2020-10-13 一种病灶的定位方法及装置 WO2021078040A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911002514.X 2019-10-21
CN201911002514.XA CN110752029B (zh) 2019-10-21 2019-10-21 一种病灶的定位方法及装置

Publications (1)

Publication Number Publication Date
WO2021078040A1 true WO2021078040A1 (zh) 2021-04-29

Family

ID=69279201

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/120627 WO2021078040A1 (zh) 2019-10-21 2020-10-13 一种病灶的定位方法及装置

Country Status (2)

Country Link
CN (1) CN110752029B (zh)
WO (1) WO2021078040A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113283361A (zh) * 2021-06-02 2021-08-20 广东电网有限责任公司广州供电局 一种绝缘层破损识别模型训练方法、识别方法和装置

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110752029B (zh) * 2019-10-21 2020-08-28 北京推想科技有限公司 一种病灶的定位方法及装置
CN113112467B (zh) * 2021-04-06 2023-04-07 上海深至信息科技有限公司 一种平面图标注系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110164798A1 (en) * 2008-04-03 2011-07-07 Fujifilm Corporation Apparatus, method, and program for detecting three dimenmsional abdominal cavity regions
US20120207268A1 (en) * 2011-01-04 2012-08-16 Edda Technology (Suzhou) Ltd. System and methods for functional analysis of soft organ segments in spect-ct images
CN104582579A (zh) * 2012-10-23 2015-04-29 株式会社日立医疗器械 图像处理装置及椎管评价方法
CN105496563A (zh) * 2015-12-04 2016-04-20 上海联影医疗科技有限公司 标定医学图像定位线的方法
CN107292928A (zh) * 2017-06-16 2017-10-24 沈阳东软医疗系统有限公司 一种血管定位的方法及装置
CN109509186A (zh) * 2018-11-09 2019-03-22 北京邮电大学 基于大脑ct图像的缺血性脑卒中病灶检测方法及装置
CN110752029A (zh) * 2019-10-21 2020-02-04 北京推想科技有限公司 一种病灶的定位方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8374892B2 (en) * 2010-01-25 2013-02-12 Amcad Biomed Corporation Method for retrieving a tumor contour of an image processing system
WO2015158372A1 (en) * 2014-04-15 2015-10-22 Elekta Ab (Publ) Method and system for calibration
KR102233966B1 (ko) * 2014-05-12 2021-03-31 삼성전자주식회사 의료 영상 정합 방법 및 그 장치
CN106600591B (zh) * 2016-12-13 2019-12-03 上海联影医疗科技有限公司 一种医学图像方位显示方法及装置
CN107808377B (zh) * 2017-10-31 2019-02-12 北京青燕祥云科技有限公司 一种肺叶中病灶的定位装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110164798A1 (en) * 2008-04-03 2011-07-07 Fujifilm Corporation Apparatus, method, and program for detecting three dimenmsional abdominal cavity regions
US20120207268A1 (en) * 2011-01-04 2012-08-16 Edda Technology (Suzhou) Ltd. System and methods for functional analysis of soft organ segments in spect-ct images
CN104582579A (zh) * 2012-10-23 2015-04-29 株式会社日立医疗器械 图像处理装置及椎管评价方法
CN105496563A (zh) * 2015-12-04 2016-04-20 上海联影医疗科技有限公司 标定医学图像定位线的方法
CN107292928A (zh) * 2017-06-16 2017-10-24 沈阳东软医疗系统有限公司 一种血管定位的方法及装置
CN109509186A (zh) * 2018-11-09 2019-03-22 北京邮电大学 基于大脑ct图像的缺血性脑卒中病灶检测方法及装置
CN110752029A (zh) * 2019-10-21 2020-02-04 北京推想科技有限公司 一种病灶的定位方法及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113283361A (zh) * 2021-06-02 2021-08-20 广东电网有限责任公司广州供电局 一种绝缘层破损识别模型训练方法、识别方法和装置
CN113283361B (zh) * 2021-06-02 2022-08-12 广东电网有限责任公司广州供电局 一种绝缘层破损识别模型训练方法、识别方法和装置

Also Published As

Publication number Publication date
CN110752029B (zh) 2020-08-28
CN110752029A (zh) 2020-02-04

Similar Documents

Publication Publication Date Title
WO2021078040A1 (zh) 一种病灶的定位方法及装置
CN111046717B (zh) 眼底图像黄斑中心定位方法、装置、电子设备及存储介质
US8150132B2 (en) Image analysis apparatus, image analysis method, and computer-readable recording medium storing image analysis program
US10332267B2 (en) Registration of fluoroscopic images of the chest and corresponding 3D image data based on the ribs and spine
CN110021025B (zh) 感兴趣区域的匹配和显示方法、装置、设备及存储介质
JP6564018B2 (ja) 放射線画像の肺野セグメンテーション技術及び骨減弱技術
WO2021184600A1 (zh) 一种图像分割方法及装置、设备及计算机可读存储介质
US9299148B2 (en) Method and system for automatically determining a localizer in a scout image
WO2019196099A1 (zh) 医学图像内目标对象的边界定位方法、存储介质及终端
US20120271162A1 (en) Constrained Registration for Motion Compensation in Atrial Fibrillation Ablation Procedures
US9514384B2 (en) Image display apparatus, image display method and storage medium storing image display program
JP6001783B2 (ja) 神経繊維構造の定位
CN110969698A (zh) 颞骨空间坐标系的构建方法、空间定位方法及电子设备
US20210271914A1 (en) Image processing apparatus, image processing method, and program
CN108846830A (zh) 对ct中腰椎自动定位的方法、装置以及存储介质
JP2015136566A (ja) 画像処理装置、およびプログラム
CN114092475B (zh) 病灶长径确定方法、图像标注方法、装置及计算机设备
CN113613562A (zh) 对x射线成像系统进行定位
CN111311655A (zh) 多模态图像配准方法、装置、电子设备、存储介质
CN113643176A (zh) 一种肋骨显示方法和装置
WO2011163414A2 (en) Mechanism for advanced structure generation and editing
CN105825519A (zh) 用于处理医学影像的方法和装置
Jeon et al. Maximum a posteriori estimation method for aorta localization and coronary seed identification
Fatima et al. Vertebrae localization and spine segmentation on radiographic images for feature‐based curvature classification for scoliosis
JP2001118058A (ja) 画像処理装置及び放射線治療計画システム

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: 20878422

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: 20878422

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 01/09/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20878422

Country of ref document: EP

Kind code of ref document: A1