Method, system, device and medium for automatic recognition of anatomical feature points in medical images

Abstract

The present invention discloses an automatic recognition method, system, device and medium for anatomical feature points of medical images. The automatic recognition method comprises: obtaining a target medical image, a preset image template and a preset atlas template; taking the target medical image as a reference image and the preset image template as a registration image, obtaining a registration deformation matrix based on pyramid registration of image mutual information measurement; deforming the preset atlas template based on the registration deformation matrix to obtain a deformed atlas template and matching it with the target medical image; overlapping the deformed atlas template with the target medical image, performing threshold segmentation on the deformed atlas template to obtain initial anatomical feature points, and matching the target anatomical feature points on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center. The present invention can realize the extraction of all anatomical feature points at one time, effectively improving the recognition speed, universality and recognition accuracy.

Description

Technical Field

[0001] The present invention relates to a method for automatically identifying anatomical feature points of medical images, and in particular to a method, system, equipment and medium for automatically identifying anatomical feature points of medical images. Background Art

[0002] Periacetabular osteotomy (PAO) is a commonly used procedure for acetabular reconstruction. Indications include adolescent acetabular dysplasia and femoral head subluxation. The procedure separates the acetabulum from the surrounding pelvis by performing a polygonal osteotomy around the acetabulum. The cut acetabulum can be moved significantly (changing position, angle, etc.), improving the coverage of the acetabulum on the femoral head, thereby reducing joint wear and relieving joint pain.

[0003] The key to PAO surgery is to determine the necessary amount of acetabular rotation to achieve the best match between the acetabulum and the femoral head. Before PAO surgery, doctors usually take CT (Computed Tomography) images of the patient's pelvis / femur, design a reasonable surgical plan based on the images, and determine the position and angle of the osteotomy and the amount of movement and rotation of the acetabulum. During PAO surgery, doctors can use auxiliary tools such as intraoperative images or surgical navigation systems to implement the surgical plan and complete the surgical procedure. After PAO surgery, doctors usually take X-rays or CT images of the patient's pelvis / femur again, and evaluate the completion of the surgical plan based on the comparison of preoperative and postoperative images.

[0004] In the above process, doctors usually need to identify and mark a large number of anatomical feature points in CT images to assist in the design of surgical plans, registration of surgical plans (surgical navigation registration), evaluation of surgical plans, etc. For example, when designing a surgical plan, doctors can define the anterior pelvic plane (APP) by marking the anterior superior iliac spine and pubic symphysis, and define the rotation center of the acetabulum by marking the center point of the acetabulum and the center point of the femoral head; when registering a surgical plan, doctors can define registration points by marking the anterior inferior iliac spine and the pubic trochanter; when evaluating a surgical plan, doctors can calculate the center edge angle (CE angle) by marking the boundary point of the acetabulum and the center point of the femur, calculate the AASA (Anterior Acetabular Sector Angle) by marking the anterior edge of the acetabulum and the center point of the femoral head, and calculate the PASA (Posterior Acetabular Sector Angle) by marking the posterior edge of the acetabulum and the center point of the femoral head, and so on.

[0005] The current solutions for the identification and marking of pelvic / femoral anatomical feature points mainly include the following:

[0006] 1. Based on the CT images, the doctor identifies and manually marks each anatomical feature point in turn according to his or her own experience and knowledge.

[0007] Manually marking anatomical feature points takes a lot of time and effort. Considering the large number of anatomical feature points (in actual cases, there may be dozens of anatomical feature points), the coordinates of each anatomical feature point need to be separately marked and recorded for later analysis and calculation. The accuracy and speed of identifying feature points are more dependent on the doctor's experience and proficiency. The process of manually marking anatomical feature points is a huge burden for clinicians.

[0008] 2. Based on the CT images, segment and reconstruct the 3D model of the pelvis / femur, find the local extreme value based on the surface curvature calculation of the model, extract the key area, screen the key area and detect the anatomical feature points. This method is suitable for the case where the anatomical feature point itself is located at the edge of the anatomical structure or the anatomical feature point is a local extreme point.

[0009] This method has certain defects in accuracy, speed and universality, and the accuracy of anatomical feature point recognition is greatly affected by the segmentation accuracy and the quality of 3D reconstruction (the number of triangular facets). Considering that the calculation of surface curvature and the solution of local extreme values ​​involve a large number of triangular facet traversal searches, the solution speed is generally not satisfactory. Moreover, the imaging representations of different anatomical feature points are not completely consistent, making it difficult to find a universal solution method.

[0010] 3. Based on a large number of pelvic / femur anatomical feature point annotation data sets, apply machine learning related technologies to train the data and generate a training model, and predict new data based on the training model to achieve automatic recognition and labeling of anatomical feature points.

[0011] This method requires a large amount of annotated data sets to ensure the accuracy and robustness of the training model. Considering that the annotation of anatomical feature points itself is a time-consuming task and highly dependent on the doctor's experience, it is not easy to obtain enough annotated samples. Given the large number of anatomical feature points and the differences in the imaging representations of each anatomical feature point, the complexity of the training model is high and it is easy to have the risk of overfitting. In addition, machine learning itself is a black box model, the logical relationship between input and output is relatively vague, the interpretability of the entire method is not strong, and there are still many limitations in practical applications. Summary of the invention

[0012] The technical problem to be solved by the present invention is to overcome the defects of the existing methods for identifying and marking anatomical feature points, which are limited, slow and of low accuracy, and to provide a method, system, device and medium for automatically identifying anatomical feature points in medical images.

[0013] The present invention solves the above technical problems through the following technical solutions:

[0014] A first aspect provides a method for automatically identifying anatomical feature points in a medical image, the method comprising:

[0015] Obtain target medical images, preset image templates and preset atlas templates;

[0016] The preset image template matches the preset atlas template, the preset atlas template is marked with anatomical feature points, and the anatomical feature points are displayed by pixel values ​​different from the background of the preset atlas template;

[0017] Taking the target medical image as the reference image and the preset image template as the registration image, a registration deformation matrix is ​​obtained by pyramid registration based on image mutual information measurement;

[0018] Deforming the preset atlas template based on the registration deformation matrix to obtain the deformed atlas template and matching it with the target medical image;

[0019] The deformed atlas template overlaps with the target medical image, and the deformed atlas template is threshold segmented. The feature area is extracted according to the anatomical feature points, and the center of the feature area is calculated to obtain the initial anatomical feature points. The target anatomical feature points are matched on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center.

[0020] Preferably, the step of obtaining a registration deformation matrix by pyramid registration based on image mutual information measurement includes:

[0021] Taking the preset atlas template as the first registration image, performing the registration process through the pyramid model of affine transformation and calculating the mutual information between the first registration image and the reference image, and combining the deformation matrix in the registration to obtain the first registration deformation matrix;

[0022] Taking the deformed atlas generated based on the first registration deformation matrix as the second registration image, performing the registration process through the pyramid model of B-spline elastic transformation and calculating the mutual information between the second registration image and the reference image, and combining the deformation matrix in the registration process to obtain the second registration deformation matrix;

[0023] The step of deforming the atlas template based on the registration deformation matrix comprises:

[0024] The atlas template is sequentially applied with the first registration deformation matrix and the second registration deformation matrix to generate the deformed atlas template.

[0025] Preferably, the step of obtaining the first deformation matrix by performing a registration process on the pyramid model through affine transformation comprises:

[0026] Translating the preset image template so that the preset image template coincides with the geometric center of the target medical image;

[0027] Planning registration pyramid model;

[0028] The registration is performed by iteratively performing an affine transformation, the formula of which is:

[0029]

[0030] Among them, the spatial coordinates of any pixel point are marked as The spatial coordinates of the corresponding pixel points after affine transformation are marked as The spatial coordinates of the affine transformation center C are The translation matrix T of the affine transformation is denoted as The rotation of the affine transformation and the affine matrix M are recorded as

[0031] The mutual information between the registered image and the reference image is calculated for each affine transformation result. The mutual information calculation formula is:

[0032] I(A,B)=H(A)+H(B)-H(A,B)

[0033] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0034] H(A)=-∫p A (a)logp A (a)

[0035] where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A;

[0036] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0037] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0038] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B;

[0039] By recording and comparing the mutual information difference before and after each affine transformation, the parameters of the affine transformation are optimized and iterated to achieve local convergence and complete the registration.

[0040] And / or, the step of obtaining a second deformation matrix by performing a registration process on the pyramid model of the B-spline elastic transformation includes:

[0041] Planning registration pyramid model;

[0042] The registration is performed by iteratively executing the B-spline elastic transformation, and the formula of the B-spline elastic transformation is:

[0043]

[0044] in For each pixel The corresponding coordinate offset, P ijk Pixel The coordinates of adjacent spatial control points, B is the B-spline basis function, where the calculation formula of B(u) is:

[0045]

[0046]

[0047]

[0048]

[0049] Among them, the calculation formulas of B(v) and B(w) are the same as B(u);

[0050] The mutual information between the registration image and the reference image is calculated for each B-spline elastic transformation result. The mutual information calculation formula is:

[0051] I(A,B)=H(A)+H(B)-H(A,B)

[0052] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0053] H(A)=-∫p A (a)logp A (a)

[0054] where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A;

[0055] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0056] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0057] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B;

[0058] By recording and comparing the mutual information difference before and after each B-spline elastic transformation, the parameters of the B-spline elastic transformation are optimized and iterated to achieve local convergence and complete the registration.

[0059] Preferably, the background pixel value of the preset atlas template is 0, a plurality of the anatomical feature points are marked on the preset atlas template, and the pixel values ​​of the plurality of anatomical feature points are different.

[0060] A second aspect provides an automatic recognition system for anatomical feature points of medical images, the automatic recognition system comprising an acquisition module, a registration module, a deformation module and a matching module;

[0061] The acquisition module is used to acquire the target medical image, the preset image template and the preset atlas template;

[0062] The preset image template matches the preset atlas template, the preset atlas template is marked with anatomical feature points, and the anatomical feature points are displayed by pixel values ​​different from the background of the preset atlas template;

[0063] The registration module is used to use the target medical image as a reference image and the preset image template as a registration image, and obtain a registration deformation matrix based on pyramid registration measured by image mutual information;

[0064] The deformation module is used to deform the preset atlas template based on the registration deformation matrix to obtain the deformed atlas template and match it with the target medical image;

[0065] The matching module is used to overlap the deformed atlas template with the target medical image, perform threshold segmentation on the deformed atlas template, extract feature areas based on the anatomical feature points, calculate the center of the feature areas, obtain initial anatomical feature points, and match the target anatomical feature points on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center.

[0066] Preferably, the registration module comprises a first registration unit and a second registration unit;

[0067] The first registration unit is used to use the preset atlas template as the first registration image, perform the registration process through the pyramid model of affine transformation and calculate the mutual information between the first registration image and the reference image, and merge the deformation matrix in the registration to obtain the first registration deformation matrix;

[0068] The second registration unit is used to use the deformed atlas generated based on the first registration deformation matrix as the second registration image, perform the registration process through the pyramid model of B-spline elastic transformation, calculate the mutual information between the second registration image and the reference image, and merge the deformation matrix in the registration process to obtain the second registration deformation matrix;

[0069] The registration module is also used to apply the first registration deformation matrix and the second registration deformation matrix to the atlas template in sequence to generate the deformed atlas template.

[0070] Preferably, the first registration unit includes a translation subunit, a first planning subunit, an affine transformation subunit, a first calculation subunit and a first iteration subunit;

[0071] The translation subunit is used to translate the preset image template so that the preset image template coincides with the geometric center of the target medical image;

[0072] The first planning subunit is used for planning and registering a first pyramid model;

[0073] The affine transformation subunit is used to perform registration by iteratively performing affine transformation, and the formula of the affine transformation is:

[0074]

[0075] Among them, the spatial coordinates of any pixel point are marked as The spatial coordinates of the corresponding pixel points after affine transformation are marked as The spatial coordinates of the affine transformation center C are The translation matrix T of the affine transformation is denoted as The rotation of the affine transformation and the affine matrix M are recorded as

[0076] The first calculation subunit is used to calculate the mutual information between the registration image and the reference image for each affine transformation result, and the mutual information calculation formula is:

[0077] I(A,B)=H(A)+H(B)-H(A,B)

[0078] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0079] H(A)=-∫p A (a)logp A (a)

[0080] where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A;

[0081] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0082] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0083] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B;

[0084] The first iterative subunit is used to optimize and iterate the parameters of the affine transformation by recording and comparing the mutual information difference before and after each affine transformation, so as to achieve local convergence and complete the registration.

[0085] And / or, the second registration unit includes a second planning subunit, a B-spline elastic transformation subunit, a second calculation subunit and a second iteration subunit;

[0086] The second planning subunit is used for planning and registering the second pyramid model;

[0087] The B-spline elastic transformation subunit is used to perform registration by iteratively executing the B-spline elastic transformation, and the formula of the B-spline elastic transformation is:

[0088]

[0089] in For each pixel The corresponding coordinate offset, P ijk Pixel The coordinates of adjacent spatial control points, B is the B-spline basis function, where the calculation formula of B(u) is:

[0090]

[0091]

[0092]

[0093]

[0094] Among them, the calculation formulas of B(v) and B(w) are the same as B(u);

[0095] The second calculation subunit is used to calculate the mutual information between the registration image and the reference image for each B-spline elastic transformation result, and the mutual information calculation formula is:

[0096] I(A,B)=H(A)+H(B)-H(A,B)

[0097] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0098] H(A)=-∫p A (a)logp A (a)

[0099] where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A;

[0100] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0101] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0102] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B;

[0103] The second iterative subunit is used to optimize and iterate the parameters of the B-spline elastic transformation by recording and comparing the mutual information difference before and after each B-spline elastic transformation, so as to achieve local convergence and complete the registration.

[0104] Preferably, the background pixel value of the preset atlas template acquired by the acquisition module is 0, a plurality of the anatomical feature points are marked on the preset atlas template, and the pixel values ​​of the plurality of anatomical feature points are all different.

[0105] A third aspect provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for automatically identifying anatomical feature points of medical images when executing the computer program.

[0106] A fourth aspect provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-mentioned method for automatically identifying anatomical feature points of medical images.

[0107] The positive and progressive effects of the present invention are: innovatively using the idea of ​​automatically identifying and extracting anatomical feature points by performing image registration, deformation, and nearest neighbor search based on a set of standard anatomical feature point recognition templates, converting the problem of automatic anatomical feature point recognition into a complex image registration problem, and using the results of image registration to extract all anatomical feature points at once, effectively improving the speed and universality of recognition. At the same time, image registration makes full use of image information (mutual information) and image features (anatomical feature areas) around anatomical feature points, effectively improving recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 This is a flow chart of a method for automatically identifying anatomical feature points in medical images according to Embodiment 1 of the present invention;

[0109] Figure 2 This is a schematic diagram of an image after the preset image template and the preset atlas template are overlapped in Example 1 of the present invention;

[0110] Figure 3 Schematic diagram of a registered image template according to Embodiment 1 of the present invention;

[0111] Figure 4 This is a schematic diagram of overlapping display of an image to be recognized and a deformed atlas template according to Embodiment 1 of the present invention;

[0112] Figure 5 This is a schematic diagram of the final anatomical feature point image in the medical image to be identified according to Embodiment 1 of the present invention;

[0113] Figure 6 This is a schematic diagram of the pyramid registration process of Embodiment 1 of the present invention;

[0114] Figure 7 This is a schematic diagram of the modules of the automatic recognition system of anatomical feature points in medical images according to Embodiment 2 of the present invention;

[0115] Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0116] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0117] Example 1

[0118] This embodiment provides a method for automatically identifying anatomical feature points in medical images. Figure 1 As shown, the automatic identification method includes:

[0119] S101, obtaining a target medical image, a preset image template, and a preset atlas template;

[0120] The preset image template matches the preset atlas template, the preset atlas template is marked with anatomical feature points, and the anatomical feature points are displayed by pixel values ​​different from the background of the preset atlas template;

[0121] S102, taking the target medical image as a reference image and the preset image template as a registration image, obtaining a registration deformation matrix through pyramid registration based on image mutual information measurement;

[0122] S103, deforming the preset atlas template based on the registration deformation matrix to obtain the deformed atlas template and matching it with the target medical image;

[0123] S104, the deformed atlas template overlaps with the target medical image, the deformed atlas template is threshold-segmented, the feature area is extracted according to the anatomical feature points, and the center of the feature area is calculated to obtain the initial anatomical feature points, and the target anatomical feature points are matched on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center.

[0124] Specifically, the preset image template uses a set of computer tomography preset image templates (templateimage), and a corresponding three-dimensional atlas template (feature atlas) with complete annotation of all anatomical feature points is used, and the preset atlas template matches the preset image template. The scanning parameters of the preset image template are consistent with the scanning parameters of the conventional preoperative CT of PAO surgery, and on this basis, a high image resolution and image contrast are guaranteed as much as possible.

[0125] At the same time, the medical images of the anatomical feature points to be identified are preprocessed, taking the pelvis / femur CT image as an example, including:

[0126] Crop the area outside the pelvis / femur of the CT image to ensure that the image range roughly matches the preset image template;

[0127] Calibrate the grayscale histogram of CT images to ensure that the distribution range of the image grayscale histogram matches the preset image template;

[0128] Simple threshold segmentation is used to separate the patient from the scan background. The segmentation result is only used to determine the initial matching range of subsequent image registration. The segmentation accuracy does not directly affect the accuracy of anatomical feature point recognition.

[0129] Specifically, taking the identification of pelvic / femur CT images as an example, the atlas template pixel value size is 0.76*0.76mm, the scanning layer thickness is 0.5mm, and the spatial resolution is 512*512*301. A total of 70 anatomical feature points are included, and the index order is left / right anterior superior iliac spine, left / right anterior inferior iliac spine, left / right pubic tuberosity, left / right pubic symphysis, etc. Figure 2 As shown, the preset image template and the preset atlas template are displayed overlappingly.

[0130] The pelvis / femur CT image to be identified with anatomical feature points is cropped, and the main cropping range includes the entire pelvis and femur to below the lesser trochanter. The identification image pixel value used in the embodiment is 0.9*0.9mm, the scanning layer thickness is 0.625mm, and the spatial resolution is 512*512*331.

[0131] The pelvic / femoral CT images of the anatomical feature points to be identified were cropped, and the main cropping range included the entire pelvis and femur to below the lesser trochanter. The pixel value of the identified image used was 0.9*0.9mm, the scanning layer thickness was 0.625mm, and the spatial resolution was 512*512*331.

[0132] Simple threshold segmentation was used, where the calibrated patient region extraction threshold ranged from -700 to 3071 to separate the patient from the scan background area.

[0133] After registering the pelvis / femur CT images of the anatomical feature points to be identified, Figure 3 The registered image template shown is used to deform the preset atlas template based on the deformation matrix obtained by the registration to obtain the deformed atlas template.

[0134] like Figure 4 As shown, the medical image to be identified is overlapped with the deformed atlas template, and the deformed atlas template is threshold segmented and binarized using pixel thresholds in turn, corresponding to the anatomical feature areas of 70 anatomical feature points respectively, and the left side of the geometric center of each anatomical feature area is calculated in turn, the corresponding pixel point position is locked, and 70 initial anatomical feature points are obtained.

[0135] like Figure 5 As shown in the figure, the nearest neighbor pelvic / femoral surface matching method with gradually expanding search radius is used to optimize the position of anatomical feature points. Taking the initial anatomical feature point as the origin and a single pixel as the initial search radius, the pelvic / femoral surface pixel point closest to the origin within the search radius is found and used as the final anatomical feature point 1. 70 final anatomical feature points are obtained in sequence.

[0136] In this way, the recognition of anatomical feature points of medical images is performed based on a set of preset image templates and preset atlas templates that annotate anatomical feature recognition points. It does not require a large amount of annotated data sets and is simple and practical.

[0137] As a possible implementation method, the background pixel value of the preset atlas template is 0, and a plurality of anatomical feature points are marked on the preset atlas template, and the pixel values ​​of the plurality of anatomical feature points are all different.

[0138] Specifically, if 70 anatomical feature points are used, the pixels of the corresponding areas may be 1 to 70 in sequence.

[0139] As an implementable method, the step of obtaining the registration deformation matrix by pyramid registration based on image mutual information measurement in step S102 includes:

[0140] Taking the preset atlas template as the first registration image, performing the registration process through the pyramid model of affine transformation and calculating the mutual information between the first registration image and the reference image, and combining the deformation matrix in the registration to obtain the first registration deformation matrix;

[0141] Taking the deformed atlas generated based on the first registration deformation matrix as the second registration image, performing the registration process through the pyramid model of B-spline elastic transformation and calculating the mutual information between the second registration image and the reference image, and combining the deformation matrix in the registration process to obtain the second registration deformation matrix;

[0142] The step of deforming the atlas template based on the registration deformation matrix comprises:

[0143] The atlas template is sequentially applied with the first registration deformation matrix and the second registration deformation matrix to generate the deformed atlas template.

[0144] Specifically, anatomical feature point recognition utilizes composite image registration, namely affine transformation and B-spline elastic transformation, which fully considers the overall matching and local deformation matching between the preset image template and the image to be identified, and can achieve the best matching effect with high recognition accuracy.

[0145] As an implementable method, the step of obtaining a first deformation matrix by performing a registration process on the pyramid model through affine transformation includes:

[0146] Translating the preset image template so that the preset image template coincides with the geometric center of the target medical image;

[0147] Planning registration pyramid model;

[0148] The registration is performed by iteratively performing an affine transformation, the formula of which is:

[0149]

[0150] Among them, the spatial coordinates of any pixel point are marked as The spatial coordinates of the corresponding pixel points after affine transformation are marked as The spatial coordinates of the affine transformation center C are The translation matrix T of the affine transformation is a 3×1 matrix, denoted as The rotation and affine matrix M of the affine transformation is a 3×3 matrix, denoted as

[0151] The mutual information between the registered image and the reference image is calculated for each affine transformation result. The mutual information calculation formula is:

[0152] I(A,B)=H(A)+H(B)-H(A,B)

[0153] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0154] H(A)=-∫p A (a)logp A (a)

[0155] where p A (a) is the grayscale distribution probability of the corresponding pixel value (or pixel value interval) a in image A (the statistical frequency of the pixel value or pixel value interval in the overall pixel grayscale histogram);

[0156] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0157] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0158] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair (or pixel value interval pair) ab in image A and image B (the statistical frequency of the pixel value pair or pixel value interval pair in the joint pixel grayscale histogram);

[0159] By recording and comparing the mutual information difference before and after each affine transformation, the parameters of the affine transformation are optimized and iterated to achieve local convergence and complete the registration.

[0160] The preset image template is translated so that the geometric center of the preset image template coincides with the geometric center of the image to be identified. This position is used as the initial position of the registration process, and the initial registration matrix is ​​recorded.

[0161] Planning registration pyramid model, such as Figure 6 As shown in the figure, the process of coarse registration using affine transformation uses a 4-layer isotropic pyramid model, that is, the resolution ratios of each layer are 8 8 8, 4 4 4, 2 2 2 and 1 1 1. Starting from the first layer, the registration image and the reference image are downsampled, and the resolution is reduced to 1 / 8 of the initial resolution (8 times the pixel size), the second layer is 1 / 4, the third layer is 1 / 2, and the fourth layer remains unchanged. When performing registration, starting from the first layer, the registration matrix is ​​calculated as the initial matrix for the second layer registration, and then it is performed in sequence until the four-layer registration is completed and the final registration matrix is ​​calculated.

[0162] As an implementable method, the step of obtaining a second deformation matrix by performing a registration process on the pyramid model of the B-spline elastic transformation includes:

[0163] Planning registration pyramid model;

[0164] The registration is performed by iteratively executing the B-spline elastic transformation, and the formula of the B-spline elastic transformation is:

[0165]

[0166] in For each pixel The corresponding coordinate offset, P ijk Pixel The coordinates of adjacent spatial control points, B is the B-spline basis function, where the calculation formula of B(u) is:

[0167]

[0168]

[0169]

[0170]

[0171] Among them, the calculation formula of B(v) and B(w) is the same as that of B(u), that is, v and w are substituted into the formula for calculation;

[0172] The mutual information between the registration image and the reference image is calculated for each B-spline elastic transformation result. The mutual information calculation formula is:

[0173] I(A,B)=H(A)+H(B)-H(A,B)

[0174] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0175] H(A)=-∫p A (a)logp A (a)

[0176] where p A (a) is the grayscale distribution probability of the corresponding pixel value (or pixel value interval) a in image A (the statistical frequency of the pixel value or pixel value interval in the overall pixel grayscale histogram);

[0177] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0178] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0179] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair (or pixel value interval pair) ab in image A and image B (the statistical frequency of the pixel value pair or pixel value interval pair in the joint pixel grayscale histogram);

[0180] By recording and comparing the mutual information difference before and after each B-spline elastic transformation, the parameters of the B-spline elastic transformation are optimized and iterated to achieve local convergence and complete the registration.

[0181] Among them, the B-spline elastic transformation is used for the process of precise registration. The B-spline elastic transformation adopts the same registration pyramid model as the affine transformation, and the first deformation matrix obtained by the affine transformation is used as the initial deformation matrix. The B-spline elastic transformation is a grid-type non-rigid deformation model, which distributes a series of grid points on the image at a certain interval, and uses the position of the grid points to calculate the coordinate offset of each pixel. Finally, the image is resampled according to the coordinate offset to achieve its non-rigid deformation.

[0182] The automatic recognition method of this embodiment innovatively uses the idea of ​​automatically identifying and extracting anatomical feature points by performing image registration, deformation, and nearest neighbor search based on a set of standard anatomical feature point recognition templates, converting the problem of automatic recognition of anatomical feature points into a complex image registration problem, and using the results of image registration to extract all anatomical feature points at one time, effectively improving the speed and universality of recognition. At the same time, image registration makes full use of image information (mutual information) and image features (anatomical feature areas) around anatomical feature points, effectively improving recognition accuracy.

[0183] Example 2

[0184] This embodiment provides a system 2 for automatically identifying anatomical feature points in medical images. Figure 7As shown, the automatic recognition system 2 comprises an acquisition module 21, a registration module 22, a deformation module 23 and a matching module 24;

[0185] The acquisition module 21 is used to acquire the target medical image, the preset image template and the preset atlas template;

[0186] The preset image template matches the preset atlas template, the preset atlas template is marked with anatomical feature points, and the anatomical feature points are displayed by pixel values ​​different from the background of the preset atlas template;

[0187] The registration module 22 is used to use the target medical image as a reference image and the preset image template as a registration image, and obtain a registration deformation matrix based on pyramid registration measured by image mutual information;

[0188] The deformation module 23 is used to deform the preset atlas template based on the registration deformation matrix to obtain the deformed atlas template and match it with the target medical image;

[0189] The matching module 24 is used to overlap the deformed atlas template with the target medical image, perform threshold segmentation on the deformed atlas template, extract feature areas based on the anatomical feature points, calculate the center of the feature areas, obtain initial anatomical feature points, and match the target anatomical feature points on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center.

[0190] As an implementable embodiment, the registration module 22 includes a first registration unit and a second registration unit;

[0191] The first registration unit is used to use the preset atlas template as the first registration image, perform the registration process through the pyramid model of affine transformation and calculate the mutual information between the first registration image and the reference image, and merge the deformation matrix in the registration to obtain the first registration deformation matrix;

[0192] The second registration unit is used to use the deformed atlas generated based on the first registration deformation matrix as the second registration image, perform the registration process through the pyramid model of B-spline elastic transformation, calculate the mutual information between the second registration image and the reference image, and merge the deformation matrix in the registration process to obtain the second registration deformation matrix;

[0193] The registration module is also used to apply the first registration deformation matrix and the second registration deformation matrix to the atlas template in sequence to generate the deformed atlas template.

[0194] As an implementable embodiment, the first registration unit includes a translation subunit, a first planning subunit, an affine transformation subunit, a first calculation subunit and a first iteration subunit;

[0195] The translation subunit is used to translate the preset image template so that the preset image template coincides with the geometric center of the target medical image;

[0196] The first planning subunit is used for planning and registering a first pyramid model;

[0197] The affine transformation subunit is used to perform registration by iteratively performing affine transformation, and the formula of the affine transformation is:

[0198]

[0199] Among them, the spatial coordinates of any pixel point are marked as The spatial coordinates of the corresponding pixel points after affine transformation are marked as The spatial coordinates of the affine transformation center C are The translation matrix T of the affine transformation is a 3×1 matrix, denoted as The rotation and affine matrix M of the affine transformation is a 3×3 matrix, denoted as

[0200] The first calculation subunit is used to calculate the mutual information between the registration image and the reference image for each affine transformation result, and the mutual information calculation formula is:

[0201] I(A,B)=H(A)+H(B)-H(A,B)

[0202] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0203] H(A)=-∫p A (a)logp A (a)

[0204] where p A (a) is the grayscale distribution probability of the corresponding pixel value (or pixel value interval) a in image A (the statistical frequency of the pixel value or pixel value interval in the overall pixel grayscale histogram);

[0205] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0206] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0207] where p AB(a, b) is the joint grayscale distribution probability of the corresponding pixel value pair (or pixel value interval pair) ab in image A and image B (the statistical frequency of the pixel value pair or pixel value interval pair in the joint pixel grayscale histogram);

[0208] The first iterative subunit is used to optimize and iterate the parameters of the affine transformation by recording and comparing the mutual information difference before and after each affine transformation, so as to achieve local convergence and complete the registration.

[0209] As an implementable embodiment, the second registration unit includes a second planning subunit, a B-spline elastic transformation subunit, a second calculation subunit and a second iteration subunit;

[0210] The second planning subunit is used for planning and registering the second pyramid model;

[0211] The B-spline elastic transformation subunit is used to perform registration by iteratively executing the B-spline elastic transformation, and the formula of the B-spline elastic transformation is:

[0212]

[0213] in For each pixel The corresponding coordinate offset, P ijk Pixel The coordinates of adjacent spatial control points, B is the B-spline basis function, where the calculation formula of B(u) is:

[0214]

[0215]

[0216]

[0217]

[0218] Among them, the calculation formula of B(v) and B(w) is the same as that of B(u), that is, v and w are substituted into the formula for calculation;

[0219] The second calculation subunit is used to calculate the mutual information between the registration image and the reference image for each B-spline elastic transformation result, and the mutual information calculation formula is:

[0220] I(A,B)=H(A)+H(B)-H(A,B)

[0221] Where I(A,B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is:

[0222] H(A)=-∫p A (a)logp A (a)

[0223] where p A (a) is the grayscale distribution probability of the corresponding pixel value (or pixel value interval) a in image A (the statistical frequency of the pixel value or pixel value interval in the overall pixel grayscale histogram);

[0224] H(A,B) is the joint entropy value of image A and image B, and the calculation formula is:

[0225] H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb

[0226] where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair (or pixel value interval pair) ab in image A and image B (the statistical frequency of the pixel value pair or pixel value interval pair in the joint pixel grayscale histogram);

[0227] The second iterative subunit is used to optimize and iterate the parameters of the B-spline elastic transformation by recording and comparing the mutual information difference before and after each B-spline elastic transformation, so as to achieve local convergence and complete the registration.

[0228] As an implementable embodiment, the background pixel value of the preset atlas template acquired by the acquisition module 21 is 0, a plurality of anatomical feature points are marked on the preset atlas template, and the pixel values ​​of the plurality of anatomical feature points are different.

[0229] The automatic recognition system provided in this embodiment innovatively uses a set of standard anatomical feature point recognition templates to perform image registration, deformation and nearest neighbor search to achieve automatic recognition and extraction of anatomical feature points.

[0230] The problem of automatic recognition of anatomical feature points is converted into a complex image registration problem, and the results of image registration are used to extract all anatomical feature points at one time, which effectively improves the speed and universality of recognition. At the same time, image registration makes full use of image information (mutual information) and image features around anatomical feature points (anatomical feature areas), effectively improving recognition accuracy.

[0231] Example 3

[0232] like Figure 8 FIG. 3 is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for automatically identifying anatomical feature points of medical images in Embodiment 1 above is implemented. Figure 8The electronic device 30 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0233] The electronic device 30 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0234] The bus 33 includes a data bus, an address bus, and a control bus.

[0235] The memory 32 may include a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322 , and may further include a read-only memory (ROM) 323 .

[0236] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, such program modules 324 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0237] The processor 31 executes various functional applications and data processing by running the computer programs stored in the memory 32, such as the automatic recognition method of anatomical feature points of medical images in embodiment 1 of the present invention.

[0238] The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 35. Furthermore, the model-generated device 30 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. As shown, the network adapter 36 communicates with other modules of the model-generated device 30 via a bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (RAID) systems, tape drives, and data backup storage systems.

[0239] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0240] Example 4

[0241] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the method for automatically identifying anatomical feature points of medical images described in Example 1 is implemented.

[0242] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0243] In a possible implementation, the present invention can also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the method for automatically identifying anatomical feature points in medical images described in Example 1.

[0244] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on a user device, partially on a user device, as an independent software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0245] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A method for automatically identifying anatomical feature points in medical images, characterized in that: The automatic identification method comprises: Obtain target medical images, preset image templates and preset atlas templates; The preset image template matches the preset atlas template, the preset atlas template is marked with anatomical feature points, and the anatomical feature points are displayed by pixel values ​​different from the background of the preset atlas template; Taking the target medical image as the reference image and the preset image template as the registration image, a registration deformation matrix is ​​obtained by pyramid registration based on image mutual information measurement; Deforming the preset atlas template based on the registration deformation matrix to obtain the deformed atlas template and matching it with the target medical image; The deformed atlas template overlaps with the target medical image, and the deformed atlas template is threshold segmented. The feature area is extracted according to the anatomical feature points, and the center of the feature area is calculated to obtain the initial anatomical feature points. The target anatomical feature points are matched on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center.

2. The method for automatically identifying anatomical feature points in medical images according to claim 1, characterized in that: The step of obtaining a registration deformation matrix by pyramid registration based on image mutual information measurement includes: Taking the preset atlas template as the first registration image, performing the registration process through the pyramid model of affine transformation and calculating the mutual information between the first registration image and the reference image, and combining the deformation matrix in the registration process to obtain the first registration deformation matrix; Taking the deformed atlas generated based on the first registration deformation matrix as the second registration image, performing the registration process through the pyramid model of B-spline elastic transformation and calculating the mutual information between the second registration image and the reference image, and combining the deformation matrix in the registration process to obtain the second registration deformation matrix; The step of deforming the atlas template based on the registration deformation matrix comprises: The atlas template is sequentially applied with the first registration deformation matrix and the second registration deformation matrix to generate the deformed atlas template.

3. The method for automatically identifying anatomical feature points in medical images according to claim 2, characterized in that: The step of obtaining a first registration deformation matrix by performing a registration process through an affine transformed pyramid model includes: Translating the preset image template so that the preset image template coincides with the geometric center of the target medical image; Planning registration pyramid model; The registration is performed by iteratively performing an affine transformation, the formula of which is: Among them, the spatial coordinates of any pixel point are marked as The spatial coordinates of the corresponding pixel points after affine transformation are marked as The spatial coordinates of the affine transformation center C are The translation matrix T of the affine transformation is denoted as The rotation of the affine transformation and the affine matrix M are recorded as The mutual information between the registered image and the reference image is calculated for each affine transformation result. The mutual information calculation formula is: I(A, B) = H(A) + H(B) - H(A, B) Where I(A, B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is: H(A)=-∫p A (a)logp A (a)from where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A; H(A, B) is the joint entropy value of image A and image B, and the calculation formula is: H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B; By recording and comparing the mutual information difference before and after each affine transformation, the parameters of the affine transformation are optimized and iterated to achieve local convergence and complete the registration; And / or, the step of performing a registration process through a pyramid model of B-spline elastic transformation to obtain a second registration deformation matrix includes: Planning registration pyramid model; The registration is performed by iteratively executing the B-spline elastic transformation, and the formula of the B-spline elastic transformation is: in For each pixel The corresponding coordinate offset, P ijk Pixel The coordinates of adjacent spatial control points, B is the B-spline basis function, where the calculation formula of B(u) is: Among them, the calculation formula of B(v) and B(w) is the same as that of B(u); The mutual information between the registration image and the reference image is calculated for each B-spline elastic transformation result. The mutual information calculation formula is: I(A, B) = H(A) + H(B) - H(A, B) Where I(A, B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is: H(A)=-∫p A (a)logp A (a)from where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A; H(A, B) is the joint entropy value of image A and image B, and the calculation formula is: H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B; By recording and comparing the mutual information difference before and after each B-spline elastic transformation, the parameters of the B-spline elastic transformation are optimized and iterated to achieve local convergence and complete the registration.

4. The method for automatically identifying anatomical feature points in medical images according to claim 1, characterized in that: The background pixel value of the preset atlas template is 0, and a plurality of anatomical feature points are marked on the preset atlas template, and the pixel values ​​of the plurality of anatomical feature points are all different.

5. An automatic recognition system for anatomical feature points in medical images, characterized in that: The automatic recognition system comprises an acquisition module, a registration module, a deformation module and a matching module; The acquisition module is used to acquire the target medical image, the preset image template and the preset atlas template; The preset image template matches the preset atlas template, the preset atlas template is marked with anatomical feature points, and the anatomical feature points are displayed by pixel values ​​different from the background of the preset atlas template; The registration module is used to use the target medical image as a reference image and the preset image template as a registration image, and obtain a registration deformation matrix based on pyramid registration measured by image mutual information; The deformation module is used to deform the preset atlas template based on the registration deformation matrix to obtain the deformed atlas template and match it with the target medical image; The matching module is used to overlap the deformed atlas template with the target medical image, perform threshold segmentation on the deformed atlas template, extract feature areas based on the anatomical feature points, calculate the center of the feature areas, obtain initial anatomical feature points, and match the target anatomical feature points on the anatomical plane in the target medical image within a preset radius with the initial anatomical feature points as the center.

6. The automatic recognition system for anatomical feature points of medical images according to claim 5, characterized in that: The registration module includes a first registration unit and a second registration unit; The first registration unit is used to use the preset atlas template as the first registration image, perform the registration process through the pyramid model of affine transformation and calculate the mutual information between the first registration image and the reference image, and merge the deformation matrix in the registration to obtain the first registration deformation matrix; The second registration unit is used to use the deformed atlas generated based on the first registration deformation matrix as the second registration image, perform the registration process through the pyramid model of B-spline elastic transformation, calculate the mutual information between the second registration image and the reference image, and merge the deformation matrix in the registration process to obtain the second registration deformation matrix; The registration module is also used to apply the first registration deformation matrix and the second registration deformation matrix to the atlas template in sequence to generate the deformed atlas template.

7. The automatic recognition system for anatomical feature points of medical images according to claim 6, characterized in that: The first registration unit includes a translation subunit, a first planning subunit, an affine transformation subunit, a first calculation subunit and a first iteration subunit; The translation subunit is used to translate the preset image template so that the preset image template coincides with the geometric center of the target medical image; The first planning subunit is used for planning and registering a first pyramid model; The affine transformation subunit is used to perform registration by iteratively performing affine transformation, and the formula of the affine transformation is: Among them, the spatial coordinates of any pixel point are marked as The spatial coordinates of the corresponding pixel points after affine transformation are marked as The spatial coordinates of the affine transformation center C are The translation matrix T of the affine transformation is denoted as The rotation of the affine transformation and the affine matrix M are recorded as The first calculation subunit is used to calculate the mutual information between the registration image and the reference image for each affine transformation result, and the mutual information calculation formula is: I(A, B) = H(A) + H(B) - H(A, B) Where I(A, B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is: H(A)=-∫p A (a)logp A (a)from where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A; H(A, B) is the joint entropy value of image A and image B, and the calculation formula is: H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B; The first iterative subunit is used to optimize and iterate the parameters of the affine transformation by recording and comparing the mutual information difference before and after each affine transformation, so as to achieve local convergence and complete the registration; And / or, the second registration unit includes a second planning subunit, a B-spline elastic transformation subunit, a second calculation subunit and a second iteration subunit; The second planning subunit is used for planning and registering the second pyramid model; The B-spline elastic transformation subunit is used to perform registration by iteratively executing the B-spline elastic transformation, and the formula of the B-spline elastic transformation is: in For each pixel The corresponding coordinate offset, P ijk Pixel The coordinates of adjacent spatial control points, B is the B-spline basis function, where the calculation formula of B(u) is: Among them, the calculation formula of B(v) and B(w) is the same as that of B(u); The second calculation subunit is used to calculate the mutual information between the registration image and the reference image for each B-spline elastic transformation result, and the mutual information calculation formula is: I(A, B) = H(A) + H(B) - H(A, B) Where I(A, B) is the mutual information between image A and image B, image A and image B are the registration image and the reference image respectively, H(A) and H(B) are the entropy values ​​of image A and image B respectively, and the calculation formula is: H(A)=-∫p A (a)logp A (a)from where p A (a) is the grayscale distribution probability corresponding to pixel value a in image A; H(A, B) is the joint entropy value of image A and image B, and the calculation formula is: H(A,B)=-∫p AB (a,b)logp AB (a,b)dadb where p AB (a, b) is the joint grayscale distribution probability of the corresponding pixel value pair ab in image A and image B; The second iterative subunit is used to optimize and iterate the parameters of the B-spline elastic transformation by recording and comparing the mutual information difference before and after each B-spline elastic transformation, so as to achieve local convergence and complete the registration.

8. The automatic recognition system for anatomical feature points of medical images according to claim 5, characterized in that: The background pixel value of the preset atlas template acquired by the acquisition module is 0, and a plurality of anatomical feature points are marked on the preset atlas template, and the pixel values ​​of the plurality of anatomical feature points are all different.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for automatically identifying anatomical feature points in medical images according to any one of claims 1 to 4 is implemented.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for automatically identifying anatomical feature points in medical images according to any one of claims 1 to 4 is implemented.

Images