A method, device, and storage medium for segmenting a bone image

Abstract

The application discloses a skeleton image segmentation method, device and equipment and a storage medium. The method comprises the following steps: acquiring object bounding box data corresponding to at least one skeleton object in a skeleton image; determining distance proportion data sets corresponding to the skeleton objects based on the skeleton image and the object bounding box data; determining a skeleton segmentation result corresponding to the skeleton image based on the distance proportion data sets and a pre-trained target image segmentation model; wherein the object bounding box data represents the spatial information of the minimum circumscribed cuboid of the corresponding skeleton object in the skeleton image, the distance proportion data set comprises at least one distance proportion data, the distance proportion data represents the distance proportion of the image pixel point in the skeleton image relative to the minimum circumscribed cuboid, and the skeleton segmentation result comprises at least one segmentation image corresponding to the skeleton object. The embodiment of the application solves the problem of poor segmentation effect of the existing image segmentation model.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a bone image segmentation method, device, equipment and storage medium. Background Art

[0002] With the continuous development of information technology and high-end medical imaging technologies, represented by artificial intelligence, machine learning techniques are gaining increasing attention in the field of medical image processing. Machine learning algorithms can be used to extract implicit patterns from massive amounts of medical images and predict useful medical information, such as image segmentation, classification, and localization results.

[0003] For skeletal images, existing machine learning algorithms directly use skeletal images as input data for image segmentation models, training them so that the trained model sequentially outputs segmented images corresponding to one or more skeletal objects in the skeletal image. However, because the input data used by existing machine learning algorithms is overly simple, the segmentation results of the segmented images output by the trained model are poor, especially at the intersections of bones. Furthermore, the segmented images output by the model often exhibit significant errors in distinguishing the skeletal object category. Summary of the Invention

[0004] The embodiments of the present invention provide a bone image segmentation method, apparatus, device and storage medium to solve the problem of poor segmentation effect of existing image segmentation models, improve the accuracy of the segmented image output by the image segmentation model, and reduce the error rate of the image segmentation model in distinguishing bone objects.

[0005] According to one embodiment of the present invention, a method for segmenting a skeleton image is provided, the method comprising:

[0006] Obtaining object bounding box data corresponding to at least one skeletal object in the skeletal image;

[0007] Determining, based on the skeletal image and the bounding box data of each object, a distance ratio dataset corresponding to each skeletal object;

[0008] Determining a bone segmentation result corresponding to the bone image based on each of the distance ratio data sets and a pre-trained target image segmentation model;

[0009] In which, the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data set contains at least one distance ratio data, and the distance ratio data represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes a segmented image corresponding to at least one bone object.

[0010] According to another embodiment of the present invention, a bone image segmentation device is provided, the device comprising:

[0011] An object bounding box data acquisition module is used to acquire object bounding box data corresponding to at least one bone object in the bone image;

[0012] a distance ratio dataset determining module, configured to determine a distance ratio dataset corresponding to each of the skeletal objects based on the skeletal image and the bounding box data of each of the objects;

[0013] A skeleton segmentation result determination module is used to determine the skeleton segmentation result corresponding to the skeleton image based on each of the distance ratio data sets and a pre-trained target image segmentation model;

[0014] In which, the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data set contains at least one distance ratio data, and the distance ratio data represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes a segmented image corresponding to at least one bone object.

[0015] According to another embodiment of the present invention, an electronic device is provided, the electronic device including:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can perform the bone image segmentation method according to any embodiment of the present invention.

[0019] According to another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions. The computer instructions are used to enable a processor to implement the bone image segmentation method according to any embodiment of the present invention when executed.

[0020] The technical solution of the embodiment of the present invention obtains object bounding box data corresponding to at least one bone object in a bone image, determines a distance ratio data set corresponding to each bone object based on the bone image and each object bounding box data, and determines a bone segmentation result corresponding to the bone image based on each distance ratio data set and a pre-trained target image segmentation model, wherein the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data in the distance ratio data set represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes a segmented image corresponding to at least one bone object, thereby solving the problem of poor segmentation effect of the existing image segmentation model, improving the accuracy of the segmented image output by the image segmentation model, and reducing the image segmentation model's error rate in distinguishing bone objects.

[0021] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 This is a flowchart of a bone image segmentation method provided in Example 1 of the present invention;

[0024] Figure 2 This is a flowchart of a bone image segmentation method provided in the second embodiment of the present invention;

[0025] Figure 3 A schematic diagram of a method for determining a target skeleton average model provided by the second embodiment of the present invention;

[0026] Figure 4 is a schematic diagram of a model view of a femoral average model provided in the second embodiment of the present invention;

[0027] Figure 5 This is a flowchart of a bone image segmentation method provided in Example 3 of the present invention;

[0028] Figure 6 A schematic diagram of a bone projection image provided by the third embodiment of the present invention;

[0029] Figure 7A schematic diagram of a bone fuzzy image provided by the third embodiment of the present invention;

[0030] Figure 8 A schematic structural diagram of a bone image segmentation device provided in a fourth embodiment of the present invention;

[0031] Figure 9 This is a structural diagram of an electronic device provided in Example 5 of the present invention. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] Example 1

[0035] Figure 1 This is a flow chart of a bone image segmentation method provided in the first embodiment of the present invention. This embodiment is applicable to the case where an image segmentation model is used to segment bone objects in a bone image and to identify the bone objects to which the segmented image belongs. This method can be executed by a bone image segmentation device, which can be implemented in the form of hardware and / or software and can be configured in a terminal device. Figure 1 As shown, the method includes:

[0036] S110: Obtain object bounding box data corresponding to at least one skeletal object in the skeletal image.

[0037] Specifically, a bone image captured by a medical imaging device is obtained. Exemplary medical imaging devices include, but are not limited to, direct digital radiography (DR), computed tomography (CT), magnetic resonance imaging (MRI), or positron emission computed tomography (PET). The type of medical imaging device is not limited herein.

[0038] For example, when the skeletal image is of an upper limb, each skeletal object in the skeletal image includes at least one of the humerus, radius, and ulna. When the skeletal image is of a lower limb, each skeletal object in the skeletal image includes at least one of the femur, tibia, and fibula. Due to differences in the shooting position and shooting parameters, the skeletal image may contain one or more skeletal objects.

[0039] In the present embodiment, the object bounding box data characterizes the spatial information of the minimum circumscribed cuboid of the corresponding skeleton object in the skeleton image. In an optional embodiment, each side of the minimum circumscribed cuboid is vertical to X-axis, Y-axis or Z-axis in the spatial coordinate system respectively. Wherein, exemplary, the object bounding box data can comprise 8 vertex position coordinates, center position coordinates, length data, width data and height data of the minimum circumscribed cuboid or the like.

[0040] In an optional embodiment, adopt object detection algorithm, at least one skeletal object in the skeletal image is located, and obtain the object bounding box data that each skeletal object corresponds to respectively.Wherein, exemplary, object detection algorithm can be YOLO algorithm.Here the object detection algorithm that adopts is not limited, and any object detection algorithm that can realize above-mentioned function is all within the protection of the application.

[0041] S120 : Determine a distance ratio data set corresponding to each skeletal object based on the skeletal image and the bounding box data of each object.

[0042] In this embodiment, the distance ratio data set includes at least one distance ratio data, and the distance ratio data represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid.

[0043] In an optional embodiment, based on the bone image and the bounding box data of each object, a distance ratio dataset corresponding to each bone object is determined, including: for each object bounding box data, based on the object bounding box data, determining the size data and at least one landmark coordinate of the minimum circumscribed cuboid; based on the landmark coordinates, the size data and the pixel coordinates corresponding to each image pixel in the bone image, determining the distance ratio dataset of the bone object corresponding to the object bounding box data.

[0044] Specifically, the size data includes length data, width data, and height data of the minimum circumscribed cuboid. In an optional embodiment, when the object bounding box data includes eight vertex position coordinates, the length data is equal to the difference between the maximum x-axis coordinate and the minimum x-axis coordinate of the eight vertex position coordinates, the width data is equal to the difference between the maximum y-axis coordinate and the minimum y-axis coordinate of the eight vertex position coordinates, and the height data is equal to the difference between the maximum z-axis coordinate and the minimum z-axis coordinate of the eight vertex position coordinates.

[0045] Specifically, based on the object bounding box data, the coordinates of the markers corresponding to at least one preset marker in the minimum circumscribed cuboid are determined. Exemplarily, the preset markers are any vertex position or center point position.

[0046] In an optional embodiment, based on each marker coordinate, dimension data and pixel coordinates corresponding to each image pixel in the skeleton image, a distance ratio dataset of the skeleton object corresponding to the object bounding box data is determined, including: determining at least one target pixel coordinate based on the object bounding box data and pixel coordinates corresponding to each image pixel in the skeleton image; for each marker coordinate, obtaining the difference coordinates corresponding to the marker coordinate and each target pixel coordinate; for each difference coordinate, using the ratio between the difference coordinate and the dimension data as distance ratio data, and adding the distance ratio data to the distance ratio dataset of the skeleton object corresponding to the object bounding box data.

[0047] In an optional embodiment, at least one target pixel coordinate is determined based on the object bounding box data and the pixel coordinates corresponding to each image pixel in the bone image, including: determining a pixel screening range based on a preset expansion ratio and the object bounding box data; taking each image pixel in the bone image whose pixel coordinates are within the pixel screening range as a target image pixel; and obtaining the target pixel coordinates corresponding to each target image pixel.

[0048] Specifically, when the preset expansion ratio is 100%, the pixel point screening range includes the three-dimensional spatial area represented by the entire skeletal image; when the preset expansion ratio is 0%, the pixel point screening range includes the internal spatial area of ​​the minimum circumscribed cuboid represented by the object bounding box data; when the preset expansion ratio is 0%-100%, the pixel point screening range includes the internal spatial area of ​​the minimum circumscribed cuboid represented by the object bounding box data and part of the external spatial area of ​​the minimum circumscribed cuboid. The preset expansion ratio is not limited here, and users can customize it according to actual needs.

[0049] Specifically, the difference coordinates include an x-difference coordinate, a y-difference coordinate, and a z-difference coordinate corresponding to the x-axis direction, the y-axis direction, and the z-axis direction, respectively.

[0050] Exemplarily, the distance ratio data corresponding to the target pixel coordinate A and the marker coordinate B satisfies the formula:

[0051]

[0052] Among them, p x 、p y and p z Respectively represent the distance ratio between the target pixel coordinate A and the marker coordinate B in the x-axis direction, y-axis direction and z-axis direction, x A 、y A and z A Respectively represent the axis coordinates of the target pixel point A in the x-axis direction, y-axis direction and z-axis direction, x B 、y B and z B They represent the axis coordinates of the marker coordinate B in the x-axis direction, y-axis direction, and z-axis direction respectively, maxx-minx represents the length data of the minimum circumscribed cuboid, maxy-miny represents the width data of the minimum circumscribed cuboid, and maxz-minz represents the height data of the minimum circumscribed cuboid.

[0053] For example, when the marker coordinate B is the coordinate of the front upper left corner, x B =minx,y B =miny.

[0054] Specifically, the distance ratio data set corresponding to the skeleton object includes distance ratio data corresponding to the coordinates of each target pixel point and the coordinates of at least one marker position.

[0055] S130 , determining a bone segmentation result corresponding to the bone image based on each distance ratio data set and a pre-trained target image segmentation model.

[0056] In an optional embodiment, based on each distance ratio data set and a pre-trained target image segmentation model, a bone segmentation result corresponding to the bone image is determined, including: obtaining image parameter data corresponding to the bone image; wherein the image parameter data includes image pixel value data of the bone image and / or image gradient data corresponding to at least one scale; inputting the image parameter data and each distance ratio data set into the pre-trained target image segmentation model to obtain the output bone segmentation result corresponding to the bone image.

[0057] Specifically, the image pixel value data can be used to represent the skeletal image, and the image gradient data can be used to represent the rate of change of the pixel value of a certain pixel point in the skeletal image in the X-axis direction, the Y-axis direction, and the Z-axis direction. In an optional embodiment, a Gaussian filter template is constructed based on at least one scale, and a Gaussian filter operation is performed on the skeletal image based on each filter template to obtain at least one image gradient data. For example, each scale can include 0, 1.2, 1.5, and 1.8.

[0058] In an optional embodiment, the image parameter data includes image gradient data corresponding to at least three scales. The advantage of this arrangement is that the segmentation quality of the target image segmentation model can be guaranteed.

[0059] Exemplarily, the model type of the target image segmentation model includes, but is not limited to, a two-dimensional random forest, a fully convolutional network model, a U-net, a SegNet, a PSPNet, or a dilated convolutional network model, etc. The model type of the target image segmentation model is not limited here.

[0060] In this embodiment, the bone segmentation result includes a segmented image corresponding to at least one bone object. Specifically, the target image segmentation model segments each bone object in the bone image in sequence, and determines the bone object to which each segmented image belongs according to the output order of the different segmented images. For example, assuming that the bone image is a lower limb image and the preset segmentation order is femur and tibia, the first segmented image output by the target image segmentation model is a segmented image of the femur, and the second segmented image is a segmented image of the tibia.

[0061] Based on the above embodiment, the method also includes: obtaining a training bone image set; wherein the training bone image set includes multiple training bone images; for each training bone image, obtaining object bounding box data corresponding to at least one bone object in the training bone image; determining a distance ratio data set corresponding to each bone object based on the training bone image and each object bounding box data; obtaining image parameter data corresponding to each training bone image, and inputting each image parameter data and each distance ratio data set into an initial image segmentation model to obtain output predicted segmentation results corresponding to each training bone image; adjusting the model parameters of the initial image segmentation model based on each predicted segmentation result and the standard segmentation results corresponding to each training bone image to obtain a trained target image segmentation model.

[0062] The technical solution of this embodiment obtains object bounding box data corresponding to at least one bone object in a bone image, determines a distance ratio data set corresponding to each bone object based on the bone image and each object bounding box data, and determines a bone segmentation result corresponding to the bone image based on each distance ratio data set and a pre-trained target image segmentation model, wherein the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data in the distance ratio data set represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes a segmented image corresponding to at least one bone object, which solves the problem of poor segmentation effect of the existing image segmentation model, improves the accuracy of the segmented image output by the image segmentation model, and reduces the error rate of the image segmentation model in distinguishing bone objects.

[0063] Example 2

[0064] Figure 2 This is a flow chart of a bone image segmentation method provided by the second embodiment of the present invention. This embodiment further refines the technical feature of "obtaining object bounding box data corresponding to at least one bone object in the bone image" in the above embodiment. Figure 2 As shown, the method includes:

[0065] S210: Obtain target skeleton average models corresponding to at least one skeleton object.

[0066] Specifically, the target skeleton average model can be used to represent the average atlas model of the corresponding skeleton object.

[0067] In an optional embodiment, a target bone average model corresponding to at least one bone object is obtained, including: for each bone object, obtaining a standard mask image and at least one reference mask image corresponding to the bone object, and registering each reference mask image with the standard mask image to obtain a registered mask image set; wherein the registered mask image set contains at least one registered mask image; based on the standard mask image and the registered mask image set, determining the target bone average model corresponding to the bone object.

[0068] Specifically, the standard mask image and the reference mask image can ensure the contour information of the skeletal object. For example, a mask image with higher image quality is manually selected from multiple mask images as the standard mask image. For example, the contour of the skeletal object in the standard mask image is the most standard, the size is the most appropriate, etc.

[0069] Specifically, in the process of registering the reference mask image with the standard mask image, the reference mask image serves as a floating image, the standard mask image serves as a reference image, and the registered mask image is a mask image obtained by registering the reference mask image with the standard mask image.

[0070] In an optional embodiment, based on the standard mask image and the registration mask image set, the target bone average model corresponding to the bone object is determined, including: obtaining the current registration mask image in the registration mask image set, and registering the standard mask image with the previous bone average model to obtain an intermediate mask image; determining the current bone summation model based on the current registration mask image and the previous bone summation model, and determining the current bone average model based on the intermediate mask image and the current bone summation model; using the current bone summation model as the previous bone summation model and the current bone average model as the previous bone average model, and repeating the step of obtaining the current registration mask image in the registration mask image set; until the number of iterations reaches the number of images corresponding to the registration mask image set, using the current bone average model as the target bone average model corresponding to the bone object; wherein, when the number of iterations is 1, the previous bone average model is the current registration mask image, and the previous bone summation model is 0.

[0071] Specifically, in the process of aligning the standard mask image with the previous average bone model, the standard mask image is used as a floating image, the previous average bone model is used as a reference image, and the intermediate mask image is the mask image obtained after the standard mask image is aligned with the previous average bone model.

[0072] Among them, exemplary, the current skeleton summation model satisfies the formula:

[0073] sumModel[i]=list[i]+sumModel[i-1]

[0074] Specifically, list[i+1] represents the i+1th registration mask image in the registration mask image set, and i is an integer greater than or equal to 1. When i=1, sumModel[0]=0.

[0075] Wherein, illustratively, the previous average skeleton model in the i-th iteration process satisfies the formula:

[0076]

[0077] Wherein, reg_ref[i-1] represents the intermediate mask image in the i-1th iteration process, list[1] represents the first registration mask image in the registration mask image set obtained in the first iteration process, and k represents the number of superpositions of the current bone summation model, and k increases with the increase of the number of iterations.

[0078] Figure 3 A schematic diagram of a method for determining a target skeleton average model provided in the second embodiment of the present invention. Specifically, during the i-th iteration, the i-th registration mask image (list[i]) in the registration mask image set is obtained, and the standard mask image (ref) is registered with the previous skeleton average model (meanModel[i-1]) to obtain the i-th intermediate mask image (reg_ref[i]). Based on the current registration mask image (list[i]) and the previous skeleton sum model (sumModel[i-1]), the current skeleton sum model (sumModel[i]) is determined, and based on the intermediate mask image (reg_ref[i]) and the current skeleton sum model (sumModel[i]), the current skeleton average model (meanModel[i]) is determined. Determine whether the current number of iterations is equal to the number of images (N) corresponding to the registration mask image set. If so, output the current bone mean model (meanModel[N]) as the target bone mean model corresponding to the bone object. If not, add 1 to i and 1 to k, and repeat the step of obtaining the i-th registration mask image (list[i]) in the registration mask image set.

[0079] Figure 4 Schematic diagram of a model view of a femoral average model provided by the second embodiment of the present invention. Specifically, Figure 4 The three figures in the figure represent, from left to right, the coronal model view, the sagittal model view, and the transverse model view of the average femoral model.

[0080] S220 . For each skeletal object, register the target skeletal average model corresponding to the skeletal object with the skeletal image to obtain object bounding box data corresponding to the skeletal object.

[0081] Wherein, exemplary, the registration algorithm used includes but is not limited to affine registration or rigid registration. The registration algorithm used is not limited here.

[0082] Specifically, the target skeletal average model contains model bounding box data of a skeletal object. The target skeletal average model is used to register the skeletal image, obtaining a registration transformation matrix corresponding to the target skeletal average model and the skeletal image. The model bounding box data is then transformed based on the registration transformation matrix to obtain skeletal bounding box data corresponding to the skeletal object in the skeletal image. During the registration of the skeletal image with the target skeletal average model, the floating image is the target skeletal average model, and the reference image is the skeletal image.

[0083] S230 : Determine a distance ratio data set corresponding to each skeletal object based on the skeletal image and the bounding box data of each object.

[0084] S240: Determine a bone segmentation result corresponding to the bone image based on each distance ratio data set and a pre-trained target image segmentation model.

[0085] Based on the above embodiment, optionally, the target skeleton average model includes model feature point coordinates corresponding to at least one skeleton feature point. Accordingly, the method further includes: for each skeleton object, obtaining a segmented image and a target skeleton average model corresponding to the skeleton object; aligning the target skeleton average model with the segmented image to obtain actual feature point coordinates corresponding to at least one skeleton feature point in the segmented image.

[0086] Specifically, skeletal feature points are used to characterize anatomically significant feature points on a skeletal object. Examples include, but are not limited to, the femoral head rotation center, the distal medial malleolus, the distal lateral malleolus, the medial ankle protuberance, the lateral ankle protuberance, and the tibial center. The setting of skeletal feature points is not limited here, and technicians can customize them based on actual needs.

[0087] Specifically, in the process of registering the target bone average model with the segmented image, the target bone average model is used as a floating image and the segmented image is used as a reference image to obtain the actual feature point coordinates corresponding to at least one bone feature point in the segmented image.

[0088] The benefit of being provided with like this is that the skeleton feature points in the existing segmented image need to be artificially marked one by one, and when the quantity of the segmented image obtained is more, can bring huge workload, consuming time.And because the gradient variation of the peripheral image where a certain feature point resides is very little, there is no obvious local feature, so that the labeling result of existing automatic labeling algorithm is not accurate enough.The present embodiment carries out an artificial labeling on the target skeleton average model and gets final product, while improving the labeling efficiency of the skeleton feature points, ensures the labeling accuracy of the skeleton feature points, for subsequent tasks such as the registration of the manipulator coordinate system provide reliable data support.Through experimental verification, the average coordinate error between the actual feature point coordinates of the skeleton feature points that the present embodiment obtains and the manual labeling result is 3-9mm.

[0089] The technical solution of the present embodiment, by obtaining the target skeleton average model corresponding to at least one skeletal object respectively, for each skeletal object, the target skeleton average model corresponding to the skeletal object is registered with the skeletal image, and the object bounding box data corresponding to the skeletal object is obtained, thereby solving the problem of obtaining the object bounding box data in the segmentation process of the skeletal image and ensuring the accuracy of the object bounding box data. Further, the present embodiment obtains the standard mask image and at least one reference mask image corresponding to the skeletal object for each skeletal object, and each reference mask image is registered with the standard mask image respectively to obtain a registration mask image set. Based on the standard mask image and the registration mask image set, the target skeleton average model corresponding to the skeletal object is determined, the determination method of the target skeleton average model is optimized, and the accuracy of the target skeleton average model is ensured. Therefore, the embodiment of the present invention further improves the segmentation effect of the image segmentation model.

[0090] Example 3

[0091] Figure 5 This is a flow chart of a bone image segmentation method provided by the third embodiment of the present invention. This embodiment further optimizes the registration process of the target bone average model and the bone image in the above embodiment. Figure 5 As shown, the method includes:

[0092] S310: Obtain target skeleton average models corresponding to at least one skeleton object.

[0093] S320: Acquire actual joint coordinates corresponding to the joint objects in the skeleton image and acquire model joint coordinates corresponding to the joint objects in the target skeleton average model.

[0094] Specifically, the joint object is a joint area between two bone objects. For example, when the bone image is a lower limb image, the joint object is a knee joint; when the bone image is an upper limb image, the joint object is an elbow joint.

[0095] Specifically, since the target average bone model is the average bone model corresponding to a single bone object, for example, when the bone image is a lower limb image and the joint object is a knee joint, if the target average bone model is a femoral average model, the model joint coordinates corresponding to the joint object in the target average bone model are the lower end position coordinates in the femoral average model; if the target average bone model is a tibia average model, the model joint coordinates corresponding to the joint object in the target average bone model are the upper end position coordinates in the femoral average model. The model joint coordinates corresponding to the joint object in the target average bone model can be obtained by manually marking in the target average bone model.

[0096] In an optional embodiment, the actual joint coordinates include first-axis coordinates, second-axis coordinates, and third-axis coordinates corresponding to the first projection direction, the second projection direction, and the third projection direction, respectively. Accordingly, obtaining the actual joint coordinates corresponding to the joint object in the bone image includes: projecting the bone image along the first projection direction and the second projection direction, respectively, to obtain a first bone projection image and a second bone projection image; based on the first bone projection image, determining the second-axis coordinates and third-axis coordinates corresponding to the joint object in the bone image and the second projection direction and the third projection direction, respectively; based on the second bone projection image, determining the first-axis coordinates corresponding to the joint object in the bone image and the first projection direction.

[0097] Specifically, the first projection direction and the second projection direction include the X-axis direction and the Y-axis direction, the X-axis direction and the Z-axis direction, or the Y-axis direction or the Z-axis direction. There is no limitation on the combination of any two projection directions.

[0098] In an optional embodiment, when the skeletal image is a lower limb image, and the lower limb image includes a left leg image and a right leg image, the first projection direction and the second projection direction include a Y-axis direction and a Z-axis direction. Figure 6 This is a schematic diagram of a bone projection image provided by the third embodiment of the present invention. Specifically, Figure 6 The left picture in the figure shows the first bone projection image obtained by projecting the lower limb image along the Y-axis direction, and the right picture shows the second bone projection image obtained by projecting the lower limb image along the Z-axis direction.

[0099] The advantage of this setting is that when the lower limb image contains a left leg image and a right leg image, if projected along the X-axis direction, the left leg and the right leg may overlap, making it impossible to distinguish the actual joint coordinates of the left knee joint and the right knee joint in the skeletal image.

[0100] Based on the above embodiment, optionally, before projecting the skeletal image along the second projection direction to obtain the second skeletal projection image, the method further includes: when the second projection direction is the Z-axis direction, cropping the lower limb image along the second projection direction based on a preset cropping ratio range to obtain a cropped lower limb image. Exemplarily, the preset cropping ratio range is 25%-75%, and the preset cropping ratio is not limited herein.

[0101] The advantage of such a setting is that the second skeleton projection image can include a partial skeleton projection image including the joint object, thereby reducing the amount of subsequent calculations and improving the accuracy of the subsequently determined first-axis coordinates.

[0102] In an optional embodiment, when the image type of the bone image is a CT image, before projecting the bone image along the first projection direction and the second projection direction to obtain the first bone projection image and the second bone projection image, the method further includes: obtaining the CT value corresponding to each image pixel in the bone image, and setting the pixel value of the image pixel corresponding to the CT value less than a first CT threshold to 0. Exemplarily, the first CT threshold can be 0 or 100. The first CT threshold is not limited herein.

[0103] The advantage of this setting is that while retaining the object area image corresponding to the bone object in the bone image, the non-object area image in the bone image is weakened, thereby achieving the purpose of reducing the noise information existing in the bone image.

[0104] In an optional embodiment, based on the first bone projection image, the second axis coordinates and third axis coordinates of the joint object in the bone image corresponding to the second projection direction and the third projection direction respectively are determined, including: performing Gaussian blur processing on the first bone projection image to obtain a first bone blurred image; based on the pixel coordinates of the pixel points corresponding to the pixel maximum values ​​in the first bone blurred image, determining the second axis coordinates and third axis coordinates of the joint object in the bone image corresponding to the second projection direction and the third projection direction respectively.

[0105] Specifically, the pixel coordinates of the pixel point corresponding to the pixel maximum in the first bone fuzzy image are obtained, and the axis coordinate corresponding to the second projection direction in the pixel coordinates is used as the second axis coordinate of the joint object, and the axis coordinate corresponding to the third projection direction is used as the third axis coordinate of the joint object.

[0106] In an optional embodiment, based on the second bone projection image, the first axis coordinates of the joint object in the bone image corresponding to the first projection direction are determined, including: performing Gaussian blur processing on the second bone projection image to obtain a second bone blurred image; based on the pixel coordinates of the pixel points corresponding to the pixel maximum value in the second bone blurred image, determining the first axis coordinates of the joint object in the bone image corresponding to the first projection direction.

[0107] Specifically, the pixel coordinates of the pixel corresponding to the pixel maximum in the second fuzzy skeleton image are obtained, and the axis coordinate corresponding to the first projection direction in the pixel coordinates is used as the first axis coordinate of the joint object.

[0108] Figure 7 This is a schematic diagram of a bone fuzzy image provided by the third embodiment of the present invention. Figure 7 by Figure 6 Taking the first bone projection image and the second bone projection image as examples, Figure 7 The left picture shows Figure 6 The first bone projection image shown corresponds to the first bone fuzzy image, and the right figure shows Figure 6 The second bone projection image shown corresponds to the second bone fuzzy image. Figure 7 The left and right figures can respectively obtain two local maxima, which correspond to the actual joint coordinates of the left knee joint and the actual joint coordinates of the right knee joint.

[0109] In an optional embodiment, when the bone image contains at least two joint objects and the first projection direction is the X-axis direction or the Y-axis direction, the pixel maximum in the first bone projection image is the pixel maximum within a preset ratio range. Specifically, based on the preset ratio range, the first bone fuzzy image is cropped in the X-axis direction or the Y-axis direction to obtain the cropped first bone fuzzy image, and the pixel maximum corresponding to the cropped first bone fuzzy image is determined. Exemplarily, the preset ratio range can be 40%-60%, and the preset ratio range is not limited here.

[0110] For example, if the skeleton image includes not only the knee joint but also the hip joint and / or ankle joint, the pixel maxima at the hip joint and / or ankle joint will interfere with the determination of the pixel maxima at the knee joint. This arrangement has the advantage of ensuring that the cropped first fuzzy skeleton image only contains the portion of the fuzzy skeleton image that includes the knee joint or elbow joint, reducing the amount of subsequent computation and improving the accuracy of the pixel coordinates corresponding to the subsequent pixel maxima.

[0111] On the basis of the above embodiment, optionally, when the skeletal image is a lower limb image, and the lower limb image contains a left leg image and a right leg image, the joint objects include a left knee joint and a right knee joint, the first skeletal projection image includes a first left leg projection image and a first right leg projection image, and the second skeletal projection image includes a second left leg projection image and a second right leg projection image. Accordingly, the skeletal image is projected along the first projection direction and the second projection direction to obtain a first skeletal projection image and a second skeletal projection image, including: performing a segmentation operation on the skeletal image to obtain a left leg image and a right leg image; projecting the left leg image along the first projection direction and the second projection direction to obtain a first left leg projection image and a second left leg projection image; projecting the right leg image along the first projection direction and the second projection direction to obtain a first right leg projection image and a second right leg projection image.

[0112] Exemplarily, the segmentation algorithms corresponding to the above segmentation operations include, but are not limited to, threshold-based segmentation algorithms, region growing algorithms, image edge segmentation algorithms, image threshold segmentation algorithms, region-based segmentation algorithms, or watershed algorithms, etc. The segmentation algorithms used are not limited here.

[0113] In an optional embodiment, the first projection direction is the Y-axis direction, and the second projection direction is the X-axis direction.

[0114] In this embodiment, the subsequent processing operations performed on the first left leg projection image and the first right leg projection image are similar to the above-mentioned processing operations corresponding to the first bone projection image, and the subsequent processing operations performed on the second left leg projection image and the second right leg projection image are similar to the above-mentioned processing operations corresponding to the second bone projection image, and will not be repeated here.

[0115] The advantage of this arrangement is that when the bending angles of the left and right legs of the subject are large and the first projection direction is the Z-axis direction, the Y-axis coordinates of the left and right leg projection images in the skeletal projection images projected along the Z-axis direction may have large errors. To prevent the left and right legs from overlapping due to the X-axis projection, this embodiment projects the left and right legs separately, and determines the actual joint coordinates of the left and right knee joints based on the first and second right leg projection images corresponding to the X-axis direction, respectively.

[0116] S330. Determine the model displacement difference based on the actual joint coordinates and the model joint coordinates, and adjust the spatial position of the target skeleton average model based on the model displacement difference to align the spatial position of the target skeleton average model with the spatial position of the skeleton image.

[0117] Specifically, after the target skeleton average model is established, it will have default spatial starting point coordinates, such as the default spatial starting point coordinates of (0,0,0). Exemplarily, the model displacement difference can be added to the spatial starting point coordinates of the target skeleton average model to obtain the target skeleton average model aligned with the spatial position of the skeleton image.

[0118] It should be noted that this step achieves a rough alignment of the spatial positions of the bone image and the target bone average model. In order to improve the spatial alignment effect of the bone image and the target bone average model, subsequent registration operations need to be performed.

[0119] S340: For each skeletal object, register the target skeletal average model corresponding to the skeletal object with the skeletal image to obtain object bounding box data corresponding to the skeletal object.

[0120] S350: Determine the distance ratio data set corresponding to each skeletal object based on the skeletal image and the bounding box data of each object.

[0121] S360: Determine a bone segmentation result corresponding to the bone image based on each distance ratio data set and a pre-trained target image segmentation model.

[0122] After being established, the target skeletal average model has its default spatial position, and the skeletal image collected by medical imaging equipment also has a corresponding spatial position. When the spatial position difference between the two is large, the accuracy of the object bounding box data obtained by direct registration is very poor. The technical solution of this embodiment obtains a target skeletal average model corresponding to at least one skeletal object, obtains the actual joint coordinates corresponding to the joint object in the skeletal image, and obtains the model joint coordinates corresponding to the joint object in the target skeletal average model. Based on the actual joint coordinates and the model joint coordinates, the model displacement difference is determined, and the spatial position of the target skeletal average model is adjusted based on the model displacement difference to align the spatial position of the target skeletal average model with the spatial position of the skeletal image. For each skeletal object, the target skeletal average model corresponding to the skeletal object is registered with the skeletal image to obtain the object bounding box data corresponding to the skeletal object. This solves the problem of large errors in the object bounding box data obtained by registering the target skeletal average model with the skeletal image, and further improves the segmentation effect of the subsequent image segmentation model.

[0123] Example 4

[0124] Figure 8 This is a schematic diagram of the structure of a bone image segmentation device provided by the fourth embodiment of the present invention. Figure 8 As shown, the apparatus includes: an object bounding box data acquisition module 410 , a distance ratio data set determination module 420 and a skeleton segmentation result determination module 430 .

[0125] The object bounding box data acquisition module 410 is used to acquire object bounding box data corresponding to at least one skeletal object in the skeletal image;

[0126] a distance ratio dataset determination module 420 for determining a distance ratio dataset corresponding to each skeletal object based on the skeletal image and the bounding box data of each object;

[0127] A skeleton segmentation result determination module 430 is configured to determine a skeleton segmentation result corresponding to the skeleton image based on each distance ratio dataset and a pre-trained target image segmentation model;

[0128] Among them, the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data set contains at least one distance ratio data, the distance ratio data represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes at least one segmented image corresponding to each bone object.

[0129] The technical solution of this embodiment obtains object bounding box data corresponding to at least one bone object in a bone image, determines a distance ratio data set corresponding to each bone object based on the bone image and each object bounding box data, and determines a bone segmentation result corresponding to the bone image based on each distance ratio data set and a pre-trained target image segmentation model, wherein the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data in the distance ratio data set represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes a segmented image corresponding to at least one bone object, which solves the problem of poor segmentation effect of the existing image segmentation model, improves the accuracy of the segmented image output by the image segmentation model, and reduces the error rate of the image segmentation model in distinguishing bone objects.

[0130] Based on the above embodiment, optionally, the distance ratio dataset determining module 420 includes:

[0131] a marker coordinate determining unit, configured to determine, for each object bounding box data, size data of a minimum circumscribed cuboid and at least one marker coordinate based on the object bounding box data;

[0132] The distance ratio data set determining unit is used to determine the distance ratio data set of the skeletal object corresponding to the object bounding box data based on the coordinates of each landmark, the size data and the pixel coordinates corresponding to each image pixel in the skeletal image.

[0133] Based on the above embodiment, optionally, the distance ratio dataset determining unit is specifically configured to:

[0134] Determining at least one target pixel coordinate based on the object bounding box data and the pixel coordinates corresponding to each image pixel in the bone image;

[0135] For each marker coordinate, obtain the difference coordinates corresponding to the marker coordinate and each target pixel coordinate;

[0136] For each difference coordinate, the ratio between the difference coordinate and the size data is used as distance ratio data, and the distance ratio data is added to the distance ratio data set of the bone object corresponding to the object bounding box data.

[0137] Based on the above embodiment, optionally, the skeleton segmentation result determination module 430 is specifically configured to:

[0138] Acquiring image parameter data corresponding to the bone image; wherein the image parameter data includes image pixel value data of the bone image and / or image gradient data corresponding to at least one scale;

[0139] The image parameter data and each distance ratio data set are input into the pre-trained target image segmentation model to obtain the bone segmentation result corresponding to the output bone image.

[0140] Based on the above embodiment, optionally, the object bounding box data acquisition module 410 includes:

[0141] a target skeleton average model acquisition unit, configured to acquire target skeleton average models corresponding to at least one skeleton object;

[0142] The object bounding box data determining unit is used to align the target bone average model corresponding to each bone object with the bone image to obtain the object bounding box data corresponding to the bone object.

[0143] Based on the above embodiment, optionally, the object bounding box data acquisition module 410 further includes:

[0144] an actual joint coordinate acquisition unit, configured to acquire the actual joint coordinates corresponding to the joint object in the skeletal image and the model joint coordinates corresponding to the joint object in the target skeletal average model before registering the target skeletal average model corresponding to the skeletal object with the skeletal image to obtain object bounding box data corresponding to the skeletal object;

[0145] The target skeleton average model alignment unit is used to determine the model displacement difference based on the actual joint coordinates and the model joint coordinates, and adjust the spatial position of the target skeleton average model based on the model displacement difference so that the spatial position of the target skeleton average model is aligned with the spatial position of the skeleton image.

[0146] Based on the above embodiment, optionally, the actual joint coordinates include first axis coordinates, second axis coordinates, and third axis coordinates corresponding to the first projection direction, the second projection direction, and the third projection direction, respectively, and the actual joint coordinate acquisition unit includes:

[0147] The bone image projection subunit is used to project the bone image along a first projection direction and a second projection direction respectively to obtain a first bone projection image and a second bone projection image;

[0148] A second axis coordinate determination subunit is configured to determine, based on the first skeleton projection image, second axis coordinates and third axis coordinates of the joint object in the skeleton image corresponding to the second projection direction and the third projection direction, respectively;

[0149] The first axis coordinate determining subunit is used to determine the first axis coordinate corresponding to the joint object in the skeleton image and the first projection direction based on the second skeleton projection image.

[0150] Based on the above embodiment, optionally, the second axis coordinate determination subunit is specifically configured to:

[0151] Performing Gaussian blur processing on the first bone projection image to obtain a first bone blurred image;

[0152] Based on the pixel coordinates of the pixel points corresponding to the pixel maximum values ​​in the first fuzzy skeleton image, the second axis coordinates and the third axis coordinates of the joint object in the skeleton image corresponding to the second projection direction and the third projection direction are determined respectively.

[0153] Based on the above embodiment, optionally, when the skeleton image is a lower limb image, and the lower limb image includes a left leg image and a right leg image, the joint object includes a left knee joint and a right knee joint, the first skeleton projection image includes a first left leg projection image and a first right leg projection image, and the second skeleton projection image includes a second left leg projection image and a second right leg projection image. Accordingly,

[0154] Based on the above embodiment, optionally, the skeleton image projection subunit is specifically configured to:

[0155] Perform segmentation on the skeleton image to obtain left leg images and right leg images;

[0156] Projecting the left leg image along a first projection direction and a second projection direction respectively to obtain a first left leg projection image and a second left leg projection image;

[0157] The right leg image is projected along a first projection direction and a second projection direction respectively to obtain a first right leg projection image and a second right leg projection image.

[0158] Based on the above embodiment, optionally, the first projection direction is the Y-axis direction, and the second projection direction is the X-axis direction.

[0159] Based on the above embodiment, optionally, the target skeleton average model acquisition unit includes:

[0160] a registration mask image set determination subunit, configured to obtain, for each skeletal object, a standard mask image and at least one reference mask image corresponding to the skeletal object, and register each reference mask image with the standard mask image to obtain a registration mask image set; wherein the registration mask image set includes at least one registration mask image;

[0161] The target skeleton average model determination subunit is used to determine the target skeleton average model corresponding to the skeleton object based on the standard mask image and the registration mask image set.

[0162] Based on the above embodiment, optionally, the target skeleton average model determination subunit is specifically used to:

[0163] Obtain the current registration mask image in the registration mask image set, and register the standard mask image with the previous skeleton average model to obtain an intermediate mask image;

[0164] Determining a current bone sum model based on the current registration mask image and the previous bone sum model, and determining a current bone average model based on the intermediate mask image and the current bone sum model;

[0165] The current bone summation model is used as the previous bone summation model and the current bone average model is used as the previous bone average model, and the step of obtaining the current registration mask image in the registration mask image set is repeated;

[0166] Until the number of iterations reaches the number of images corresponding to the registration mask image set, the current average skeleton model is used as the target average skeleton model corresponding to the skeleton object;

[0167] Among them, when the number of iterations is 1, the previous skeleton average model is the current registration mask image, and the previous skeleton summation model is 0.

[0168] Based on the above embodiment, optionally, the target skeleton average model includes model feature point coordinates corresponding to at least one skeleton feature point, and accordingly, the device further includes:

[0169] The actual feature point coordinate determination module is used to obtain the segmented image corresponding to each skeletal object and the target skeletal average model;

[0170] The target skeleton average model is aligned with the segmented image to obtain the actual feature point coordinates corresponding to at least one skeleton feature point in the segmented image.

[0171] The bone image segmentation device provided in the embodiment of the present invention can execute the bone image segmentation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0172] Example 5

[0173] Figure 9 Schematic diagram of the structure of an electronic device provided in the fifth embodiment of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown in the embodiments of the present invention, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0174] like Figure 9 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 and a random access memory (RAM) 13, that is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor 11. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0175] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0176] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the bone image segmentation method.

[0177] In some embodiments, the method for segmenting a skeletal image can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for segmenting a skeletal image described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the method for segmenting a skeletal image by any other appropriate means (e.g., by means of firmware).

[0178] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0179] Computer programs for implementing the skeletal image segmentation method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs implement the functions / operations specified in the flowcharts and / or block diagrams. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0180] Example 6

[0181] Embodiment 6 of the present invention further provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a processor to execute a bone image segmentation method, the method comprising:

[0182] Obtaining object bounding box data corresponding to at least one skeletal object in the skeletal image;

[0183] Determine a distance ratio dataset corresponding to each skeletal object based on the skeletal image and the bounding box data of each object;

[0184] Determine the bone segmentation result corresponding to the bone image based on each distance ratio dataset and the pre-trained target image segmentation model;

[0185] Among them, the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data set contains at least one distance ratio data, the distance ratio data represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes at least one segmented image corresponding to each bone object.

[0186] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0187] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0188] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0189] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0190] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0191] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for segmenting a bone image, characterized in that: include: Obtaining object bounding box data corresponding to at least one skeletal object in the skeletal image; For each object bounding box data, determining size data of a minimum circumscribed cuboid and at least one marker coordinate based on the object bounding box data; Determining at least one target pixel coordinate based on the object bounding box data and pixel coordinates corresponding to each image pixel in the bone image; For each marker coordinate, obtaining the difference coordinates corresponding to the marker coordinate and each target pixel coordinate; For each difference coordinate, taking a ratio between the difference coordinate and the size data as distance ratio data, and adding the distance ratio data to a distance ratio data set of the skeletal object corresponding to the object bounding box data; Acquiring image parameter data corresponding to the bone image; wherein the image parameter data includes image pixel value data of the bone image and / or image gradient data corresponding to at least one scale; Inputting the image parameter data and each of the distance ratio data sets into a pre-trained target image segmentation model to obtain a bone segmentation result corresponding to the output bone image; In which, the object bounding box data represents the spatial information of the minimum circumscribed cuboid corresponding to the bone object in the bone image, the distance ratio data set contains at least one distance ratio data, and the distance ratio data represents the distance ratio of the image pixel point in the bone image relative to the minimum circumscribed cuboid, and the bone segmentation result includes a segmented image corresponding to at least one bone object.

2. The method according to claim 1, characterized in that The obtaining of object bounding box data corresponding to at least one skeletal object in the skeletal image includes: Obtaining target skeleton average models corresponding to at least one skeleton object; For each skeletal object, the target skeletal average model corresponding to the skeletal object is registered with the skeletal image to obtain object bounding box data corresponding to the skeletal object.

3. The method according to claim 2, characterized in that Before registering the target skeleton average model corresponding to the skeleton object with the skeleton image to obtain object bounding box data corresponding to the skeleton object, the method further includes: Acquire actual joint coordinates corresponding to the joint object in the skeleton image and acquire model joint coordinates corresponding to the joint object in the target skeleton average model; Based on the actual joint coordinates and the model joint coordinates, a model displacement difference is determined, and the spatial position of the target skeleton average model is adjusted based on the model displacement difference so that the spatial position of the target skeleton average model is aligned with the spatial position of the skeleton image.

4. The method according to claim 3, characterized in that The actual joint coordinates include first axis coordinates, second axis coordinates, and third axis coordinates corresponding to the first projection direction, the second projection direction, and the third projection direction, respectively. Accordingly, obtaining the actual joint coordinates corresponding to the joint object in the skeletal image includes: Projecting the bone image along a first projection direction and a second projection direction respectively to obtain a first bone projection image and a second bone projection image; Based on the first skeleton projection image, determining the second axis coordinate and the third axis coordinate corresponding to the second projection direction and the third projection direction of the joint object in the skeleton image respectively; Based on the second skeleton projection image, a first axis coordinate corresponding to the joint object in the skeleton image and the first projection direction is determined.

5. The method according to claim 4, characterized in that The determining, based on the first skeleton projection image, second-axis coordinates and third-axis coordinates corresponding to the second projection direction and the third projection direction of the joint object in the skeleton image, respectively, includes: Performing Gaussian blur processing on the first bone projection image to obtain a first bone blurred image; Based on the pixel coordinates of the pixel points corresponding to the pixel maximum values ​​in the first fuzzy skeleton image, the second axis coordinates and the third axis coordinates of the joint object in the skeleton image corresponding to the second projection direction and the third projection direction respectively are determined.

6. The method according to claim 4, characterized in that When the skeletal image is a lower limb image, and the lower limb image includes a left leg image and a right leg image, the joint objects include a left knee joint and a right knee joint, the first skeletal projection image includes a first left leg projection image and a first right leg projection image, and the second skeletal projection image includes a second left leg projection image and a second right leg projection image. Accordingly, projecting the skeletal image along the first projection direction and the second projection direction, respectively, to obtain the first skeletal projection image and the second skeletal projection image, includes: Performing a segmentation operation on the skeleton image to obtain a left leg image and a right leg image; Projecting the left leg image along a first projection direction and a second projection direction respectively to obtain a first left leg projection image and a second left leg projection image; The right leg image is projected along a first projection direction and a second projection direction respectively to obtain a first right leg projection image and a second right leg projection image.

7. The method according to claim 6, characterized in that The first projection direction is the Y-axis direction, and the second projection direction is the X-axis direction.

8. The method according to claim 2, characterized in that The obtaining of target skeleton average models corresponding to at least one skeleton object includes: For each skeletal object, a standard mask image and at least one reference mask image corresponding to the skeletal object are obtained, and each reference mask image is registered with the standard mask image to obtain a registered mask image set; wherein the registered mask image set includes at least one registered mask image; Based on the standard mask image and the registration mask image set, a target skeleton average model corresponding to the skeleton object is determined.

9. The method according to claim 8, characterized in that Determining a target skeleton average model corresponding to the skeleton object based on the standard mask image and the registration mask image set includes: Acquire a current registration mask image in the registration mask image set, and register the standard mask image with a previous bone average model to obtain an intermediate mask image; Determining a current bone sum model based on the current registration mask image and the previous bone sum model, and determining a current bone average model based on the intermediate mask image and the current bone sum model; Using the current bone sum model as the previous bone sum model and the current bone average model as the previous bone average model, and repeatedly performing the step of obtaining the current registration mask image in the registration mask image set; When the number of iterations reaches the number of images corresponding to the registration mask image set, the current average skeleton model is used as the target average skeleton model corresponding to the skeleton object; Among them, when the number of iterations is 1, the previous skeleton average model is the current registration mask image, and the previous skeleton summation model is 0.

10. The method according to claim 2, characterized in that The target skeleton average model includes model feature point coordinates corresponding to at least one skeleton feature point. Accordingly, the method further includes: For each skeletal object, obtaining a segmented image and a target skeletal average model corresponding to the skeletal object; The target skeleton average model is registered with the segmented image to obtain actual feature point coordinates corresponding to at least one skeleton feature point in the segmented image.

11. A bone image segmentation device, characterized in that: include: An object bounding box data acquisition module is used to acquire object bounding box data corresponding to at least one bone object in the bone image; a distance ratio dataset determining module, configured to determine a distance ratio dataset corresponding to each of the skeletal objects based on the skeletal image and the bounding box data of each of the objects; A skeleton segmentation result determination module is used to determine the skeleton segmentation result corresponding to the skeleton image based on each of the distance ratio data sets and a pre-trained target image segmentation model; The object bounding box data represents spatial information of a minimum circumscribed cuboid corresponding to a skeletal object in the skeletal image, the distance ratio data set includes at least one distance ratio data, the distance ratio data represents a distance ratio of an image pixel point in the skeletal image relative to the minimum circumscribed cuboid, and the skeletal segmentation result includes a segmented image corresponding to at least one skeletal object. The distance ratio dataset determination module includes: a marker coordinate determining unit, configured to determine, for each object bounding box data, size data of the minimum circumscribed cuboid and at least one marker coordinate based on the object bounding box data; a distance ratio data set determining unit, configured to determine a distance ratio data set of the skeletal object corresponding to the object bounding box data based on the landmark coordinates, the size data, and the pixel coordinates corresponding to the image pixels in the skeletal image; The distance ratio dataset determining unit is specifically configured to: Determining at least one target pixel coordinate based on the object bounding box data and the pixel coordinates corresponding to each image pixel in the bone image; obtaining, for each marker coordinate, a difference coordinate corresponding to the marker coordinate and each target pixel coordinate; and using, for each difference coordinate, a ratio between the difference coordinate and the size data as distance ratio data, and adding the distance ratio data to a distance ratio data set of the bone object corresponding to the object bounding box data; The skeleton segmentation result determination module is specifically used to: Obtain image parameter data corresponding to the bone image; wherein the image parameter data includes image pixel value data of the bone image and / or image gradient data corresponding to at least one scale; input the image parameter data and each of the distance ratio data sets into a pre-trained target image segmentation model to obtain a bone segmentation result corresponding to the output bone image.

12. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the bone image segmentation method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the bone image segmentation method according to any one of claims 1 to 10 when executed.

Images