3D Image Detection Method, Device, Electronic Device and Storage Medium

Abstract

The present disclosure provides a three-dimensional image detection method, apparatus, electronic device, and storage medium. The method includes: obtaining a three-dimensional image of a target part and a three-dimensional segmentation image corresponding to the three-dimensional image, wherein the voxel information of the target voxels located on the target part in the three-dimensional segmentation image is different from the voxel information of other voxels located on other parts; generating a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image; wherein the target point cloud image includes each point cloud point located on the target part; and determining a detection result of the target part on the three-dimensional image based on the target point cloud image.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technologies, and in particular, to a three-dimensional image detection method, apparatus, electronic device, and storage medium. Background Art

[0002] Rib fractures are one of the common injuries after the chest is subjected to external force impacts. The diagnosis of rib fractures is very important for evaluating the severity of trauma and treatment. Since Computed Tomography (CT) scans have high sensitivity and specificity in detecting rib fractures, CT scans are of great significance for the medical identification of rib fractures. However, evaluating CT images is a time-consuming task that requires clinicians to have good patience and relatively rich experience, which in turn results in a low evaluation efficiency of CT images. Summary of the Invention

[0003] In view of this, the present disclosure provides at least a three-dimensional image detection method, apparatus, electronic device, and storage medium.

[0004] In a first aspect, the present disclosure provides a three-dimensional image detection method, including:

[0005] Obtain a three-dimensional image of a target part and a three-dimensional segmentation image corresponding to the three-dimensional image, where the voxel information of the target voxels located on the target part in the three-dimensional segmentation image is different from the voxel information of other voxels located on other parts;

[0006] Generate a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image; where each point cloud point located on the target part is included in the target point cloud image;

[0007] Determine a detection result of the target part on the three-dimensional image based on the target point cloud image.

[0008] In the above method, by obtaining a three-dimensional image and a three-dimensional segmentation image, and using the three-dimensional coordinate data of the target voxels on each target part in the three-dimensional segmentation image and the three-dimensional image to generate a target point cloud image corresponding to the three-dimensional image, since each point cloud point located on the target part is included in the target point cloud image, the screening of other feature information of other parts except the target part in the three-dimensional image is achieved, the interference of other feature information on the target feature information of the target part is alleviated, and then based on the target point cloud image, the detection result of the target part can be determined more accurately; and after the other feature information is screened out, when determining the detection result of the target part on the three-dimensional image based on the target point cloud image, the waste of computing resources caused by processing other feature information can be alleviated, and the detection efficiency of the three-dimensional image is improved.

[0009] In one possible implementation, after obtaining the three-dimensional image of the target part, the method further includes:

[0010] According to the set preprocessing voxel range, adjust the voxel information of at least one voxel within the preprocessing voxel range in the three-dimensional image to a preset range, and adjust the voxel information of at least one voxel outside the preprocessing voxel range in the three-dimensional image to a preset value, to generate a processed three-dimensional image;

[0011] Generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image includes:

[0012] Generate the target point cloud image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the processed three-dimensional image.

[0013] Here, preprocessing can be performed on each voxel in the three-dimensional image. For example, the voxel information of at least one voxel within the preprocessing voxel range in the three-dimensional image can be normalized, and the voxel information of at least one voxel outside the preprocessing voxel range in the three-dimensional image can be adjusted to a preset value, to generate a processed three-dimensional image, so that the detection result corresponding to the three-dimensional image can be determined more accurately using the processed three-dimensional image.

[0014] In one possible implementation, obtaining the three-dimensional segmentation image corresponding to the three-dimensional image includes:

[0015] Generate an initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image;

[0016] Perform dilation processing on the target voxels located on the target part in the initial three-dimensional segmentation image to generate the three-dimensional segmentation image.

[0017] In this implementation, after generating the initial three-dimensional segmentation image, dilation processing can be performed on the target voxels located on the target part in the initial three-dimensional segmentation image to generate a three-dimensional segmentation image, so that the segmentation area corresponding to the target part in the generated three-dimensional segmentation image increases, alleviating the omission of target feature information corresponding to the target part. Furthermore, using the generated three-dimensional segmentation image, the detection result corresponding to the three-dimensional image can be determined more accurately.

[0018] In one possible implementation, generating the initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image includes:

[0019] Based on the three-dimensional image and the trained first neural network for image segmentation, obtain the initial three-dimensional segmentation image; or,

[0020] Segment the three-dimensional image by using the shape data of the target part and the set voxel threshold to obtain the initial three-dimensional segmentation image.

[0021] Here, the first neural network can be used to accurately determine the initial three-dimensional segmentation image; or, the shape data of the target part and the set voxel threshold can also be used to simply and quickly determine the initial three-dimensional segmentation image.

[0022] In a possible implementation manner, generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image includes:

[0023] Determine a first voxel that matches the three-dimensional coordinate data of the target voxel from the multiple voxels included in the three-dimensional image;

[0024] Determine the three-dimensional coordinate data of the first voxel as the coordinate data of the point cloud point corresponding to the first voxel;

[0025] Determine the voxel information of the first voxel as the attribute data of the point cloud point;

[0026] Generate the target point cloud image corresponding to the three-dimensional image based on the coordinate data and the attribute data of the point cloud points respectively corresponding to each of the first voxels.

[0027] Here, by determining the target point cloud image corresponding to the three-dimensional image, the screening of other feature information of other parts except the target part is realized, and the detection efficiency of the three-dimensional image is improved when using the target point cloud image to determine the detection result corresponding to the three-dimensional image.

[0028] In a possible implementation manner, determining the detection result of the target part on the three-dimensional image based on the target point cloud image includes:

[0029] Perform at least one convolution process on the target point cloud image to generate first intermediate feature data;

[0030] Perform at least one deconvolution process on the first intermediate feature data to generate second intermediate feature data;

[0031] Fuse the first intermediate feature data and the second intermediate feature data to generate target feature data;

[0032] Determine the detection result of detecting the target part on the three-dimensional image based on the target feature data.

[0033] Here, by generating first intermediate feature data and second intermediate feature data, the first intermediate feature data and the second intermediate feature data are fused to generate target feature data, which includes global features and local features in the three-dimensional image, making the feature information included in the target feature data relatively rich; furthermore, based on the target feature data, the detection result corresponding to the three-dimensional image can be determined more accurately.

[0034] In a possible implementation, the at least one convolution process performed on the target point cloud image to generate first intermediate feature data includes:

[0035] During the process of performing convolution on the target point cloud image, for each convolution calculation, it is judged whether there is any coordinate data of the point cloud points that matches the central position corresponding to the local feature data for which the convolution calculation is performed in the target point cloud image;

[0036] If so, the convolution kernel is used to perform this convolution calculation to obtain the calculation result corresponding to this convolution calculation, otherwise this convolution calculation is abandoned and the next convolution calculation is performed;

[0037] Based on the calculation results of at least one of the convolution calculations, first intermediate feature data is generated.

[0038] Here, for each convolution calculation, it can be judged whether there is any coordinate data of the point cloud points that matches the central position corresponding to the local feature data for which the convolution calculation is performed in the target point cloud image. If so, the convolution kernel is used to perform this convolution calculation, and if not, this convolution calculation is abandoned, which alleviates the resource consumption of performing convolution calculations on other feature information of other parts, reduces the computing resources consumed for determining the detection result, and improves the efficiency of determining the detection result.

[0039] For the effect descriptions of the following devices, electronic devices, etc., refer to the descriptions of the above methods and will not be elaborated here.

[0040] In a second aspect, the present disclosure provides a three-dimensional image detection device, including:

[0041] An acquisition module, configured to acquire a three-dimensional image of a target part and a three-dimensional segmentation image corresponding to the three-dimensional image, wherein the voxel information of the target voxels located on the target part in the three-dimensional segmentation image is different from the voxel information of other voxels located on other parts;

[0042] A generation module, configured to generate a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image; wherein, the target point cloud image includes each point cloud point located on the target part;

[0043] A determination module, configured to determine a detection result of the target part on the three-dimensional image based on the target point cloud image.

[0044] In a possible implementation manner, after obtaining the three-dimensional image of the target part, the apparatus further includes: a preprocessing module, configured to:

[0045] According to the set preprocessing voxel range, adjust the voxel information of at least one voxel within the preprocessing voxel range in the three-dimensional image to a preset range, and adjust the voxel information of at least one voxel outside the preprocessing voxel range in the three-dimensional image to a preset value, and generate a processed three-dimensional image;

[0046] When generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image, the generation module is configured to:

[0047] Generate the target point cloud image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the processed three-dimensional image.

[0048] In a possible implementation manner, when obtaining the three-dimensional segmentation image corresponding to the three-dimensional image, the obtaining module is configured to:

[0049] Generate an initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image;

[0050] Perform dilation processing on the target voxels located on the target part in the initial three-dimensional segmentation image to generate the three-dimensional segmentation image.

[0051] In a possible implementation manner, when generating the initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image, the obtaining module is configured to:

[0052] Obtain the initial three-dimensional segmentation image based on the three-dimensional image and a trained first neural network for image segmentation; or, 0]

[0053] Segment the three-dimensional image by using the shape data of the target part and the set voxel threshold to obtain the initial three-dimensional segmentation image.

[0054] In a possible implementation manner, when generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image, the generation module is configured to:

[0055] Determine a first voxel that matches the three-dimensional coordinate data of the target voxel from among the multiple voxels included in the three-dimensional image;

[0056] Determine the three-dimensional coordinate data of the first voxel as the coordinate data of the point cloud point corresponding to the first voxel;

[0057] Determine the voxel information of the first voxel as the attribute data of the point cloud point;

[0058] Generate a target point cloud image corresponding to the three-dimensional image based on the coordinate data and the attribute data of the point cloud points respectively corresponding to each of the first voxels.

[0059] In a possible implementation manner, when the determination module determines the detection result of the target part on the three-dimensional image based on the target point cloud image, it is used for:

[0060] Perform at least one convolution process on the target point cloud image to generate first intermediate feature data;

[0061] Perform at least one deconvolution process on the first intermediate feature data to generate second intermediate feature data;

[0062] Fuse the first intermediate feature data and the second intermediate feature data to generate target feature data;

[0063] Determine the detection result of detecting the target part on the three-dimensional image based on the target feature data.

[0064] In a possible implementation manner, when the determination module performs at least one convolution process on the target point cloud image to generate first intermediate feature data, it is used for:

[0065] During the process of performing convolution on the target point cloud image, for each convolution calculation, determine whether there is any coordinate data of the point cloud point that matches the central position corresponding to the local feature data in the target point cloud image where this convolution calculation is performed;

[0066] If there is a match, perform this convolution calculation using the convolution kernel to obtain the calculation result corresponding to this convolution calculation, otherwise abandon this convolution calculation and perform the next convolution calculation;

[0067] Generate first intermediate feature data based on the calculation results of at least one of the convolution calculations.

[0068] In a third aspect, the present disclosure provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the three-dimensional image detection method described in the above first aspect or any one of the embodiments are executed.

[0069] In a fourth aspect, the present disclosure provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of the three-dimensional image detection method described in the above first aspect or any one of the embodiments are executed.

[0070] To make the above objects, features, and advantages of the present disclosure more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, the detailed description is as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings required to be used in the embodiments. The accompanying drawings are incorporated into the specification and constitute a part of this specification. These drawings show embodiments that conform to the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only show some embodiments of the present disclosure and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0072] Figure 1 A flowchart showing a three-dimensional image detection method provided by an embodiment of the present disclosure;

[0073] Figure 2 A flowchart showing a specific manner of determining a detection result of detecting a target part of the three-dimensional image in a three-dimensional image detection method provided by an embodiment of the present disclosure;

[0074] Figure 3 A schematic structural diagram showing a target neural network provided by an embodiment of the present disclosure;

[0075] Figure 4 A flowchart showing a convolution process provided by an embodiment of the present disclosure;

[0076] Figure 5 A schematic architecture diagram showing a three-dimensional image detection device provided by an embodiment of the present disclosure;

[0077] Figure 6 A schematic structural diagram showing an electronic device provided by an embodiment of the present disclosure. Specific Embodiments

[0078] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. Components of the embodiments of the present disclosure described and illustrated herein generally can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed present disclosure, but merely represents selected embodiments of the present disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of the present disclosure without creative efforts fall within the scope of protection of the present disclosure.

[0079] Rib fracture is one of the common injuries after the chest is subjected to external force impact. The diagnosis of rib fracture is very important for evaluating the severity of trauma and treatment. Since computed tomography (CT) scanning has high sensitivity and high specificity in detecting rib fractures, CT scanning is of great significance for the medical identification of rib fractures. However, evaluating CT images is a time-consuming task, which requires clinicians to have good patience and relatively rich experience, thus resulting in a low evaluation efficiency of CT images.

[0080] Generally, a trained neural network can be used to detect the scanned CT images. In one method, a two-dimensional convolutional neural network can be used to detect the axial slice images (two-dimensional images) in the CT images to obtain the detection results of the slice images; and the detection results of rib fractures in the CT images can be determined through the detection results of a series of consecutive slice images. However, the characteristic information of rib fractures in CT images is manifested as a three-dimensional information. When detecting two-dimensional slice images, some characteristic information will be lost, thereby reducing the accuracy of the detection results.

[0081] In another method, a three-dimensional convolutional neural network can be used to detect CT images. Since the computational complexity of the three-dimensional convolutional neural network is relatively large compared with the two-dimensional network, and there is redundant information in CT images, the calculation time of the detection results is relatively long, the video memory occupancy is relatively high, and the accuracy is relatively low.

[0082] To alleviate the above problems, the embodiments of the present disclosure provide a three-dimensional image detection method, device, electronic device, and storage medium.

[0083] Regarding the defects existing in the above solutions, they are all the results obtained by the inventors through practice and careful research. Therefore, the process of discovering the above problems and the solutions proposed by the present disclosure for the above problems in the following text should be the contributions made by the inventors to the present disclosure during the process of the present disclosure.

[0084] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0085] To facilitate the understanding of the embodiments of the present disclosure, first, a three-dimensional image detection method disclosed in the embodiments of the present disclosure will be introduced in detail. The execution subject of the three-dimensional image detection method provided by the embodiments of the present disclosure can be a terminal device, a server, etc. Among them, the terminal device can be, for example, a computer, a tablet, a mobile phone, a personal digital assistant (Personal Digital Assistant, PDA), etc.; the server can be, for example, a local server, a cloud server, etc. In some possible implementation manners, the three-dimensional image detection method can be implemented by a processor calling computer-readable instructions stored in a memory.

[0086] See Figure 1 As shown, it is a schematic flowchart of the three-dimensional image detection method provided by the embodiments of the present disclosure, including S101 - S103, where:

[0087] S101, obtain a three-dimensional image of a target part and a three-dimensional segmentation image corresponding to the three-dimensional image, where the voxel information of the target voxels located on the target part in the three-dimensional segmentation image is different from the voxel information of other voxels located on other parts;

[0088] S102, generate a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image; where the target point cloud image includes each point cloud point located on the target part;

[0089] S103, determine the detection result of the target part on the three-dimensional image based on the target point cloud image.

[0090] In the above method, by obtaining a three-dimensional image and a three-dimensional segmentation image, and using the three-dimensional coordinate data and the three-dimensional image of the target voxels on each target part in the three-dimensional segmentation image, a target point cloud image corresponding to the three-dimensional image is generated. Since the target point cloud image includes each point cloud point located on the target part, the screening of other feature information of other parts except the target part in the three-dimensional image is realized, the interference of other feature information on the target feature information of the target part is alleviated, and based on the target point cloud image, the detection result of the target part can be determined more accurately; and after the other feature information is screened out, when determining the detection result of the target part on the three-dimensional image based on the target point cloud image, the waste of computing resources caused by processing the other feature information can be alleviated, and the detection efficiency of the three-dimensional image is improved.

[0091] The following specifically describes S101 - S103.

[0092] Regarding S101:

[0093] In implementation, a three-dimensional image of the target part can be obtained. For example, the target part can be scanned by imaging to obtain the three-dimensional image corresponding to the target part. The three-dimensional image can be: medical images such as CT images and contrast-enhanced CT images; where the target part can be any part of the human body. For example, the target part can be: chest ribs, palms, etc.

[0094] After obtaining the three-dimensional image, the three-dimensional image can be subjected to target part segmentation processing to obtain a three-dimensional segmentation image, which includes the contour area of the target part. Among them, the three-dimensional segmentation image can be a binary image, that is, the voxel value of the target voxels located on the target part can be 1, and the voxel value of the other voxels except the target part can be 0.

[0095] Exemplarily, when the three-dimensional image is a CT image, the voxel information corresponding to each voxel in the CT image can be the Hounsfield unit (HU) value.

[0096] In an optional implementation manner, obtaining the three-dimensional segmentation image corresponding to the three-dimensional image may include step A1 and step A2, where:

[0097] Step A1, based on the three-dimensional image, generate an initial three-dimensional segmentation image corresponding to the three-dimensional image;

[0098] Step A2, perform dilation processing on the target voxels located on the target part in the initial three-dimensional segmentation image to generate the three-dimensional segmentation image.

[0099] In step A1, during implementation, voxels located on the target part in the three-dimensional image can be determined, and an initial three-dimensional segmentation image corresponding to the three-dimensional image can be generated by using the voxel positions of each voxel located on the target part.

[0100] In an optional implementation manner, generating the initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image may include the following two methods:

[0101] Method 1: Obtain the initial three-dimensional segmentation image based on the three-dimensional image and a trained first neural network for image segmentation.

[0102] Method 2: Segment the three-dimensional image by using the shape data of the target part and a set voxel threshold to obtain the initial three-dimensional segmentation image.

[0103] In Method 1, a trained first neural network for image segmentation can be used to process the three-dimensional image to obtain the initial three-dimensional segmentation image. Among them, the network structure of the first neural network can be set according to the actual situation.

[0104] In Method 2, it can be known that in the three-dimensional image, the tissue density of the target part is different from that of other parts, and the voxel information corresponding to parts with different tissue densities in the three-dimensional image varies greatly. Therefore, the three-dimensional image can be segmented by using the shape data of the target part and a set voxel threshold to obtain the initial three-dimensional segmentation image.

[0105] Exemplarily, for example, when the three-dimensional image is a CT image and the target part is the chest rib, since the voxel information in the rib area in the CT image is quite different from that of other parts, a voxel threshold can be set, and voxels smaller than the voxel threshold are considered as other voxels that do not belong to the rib. Then, other voxels included in the three-dimensional image can be deleted to obtain the candidate voxels after deletion. Since the chest rib has a specific shape structure, the candidate voxels can be further screened according to the determined shape data of the chest rib to obtain the three-dimensional coordinate data of the target voxels after screening. Then, the voxel values of voxels matching the three-dimensional coordinate data of the target voxels can be set to 1, and the voxel values of voxels not matching the three-dimensional coordinate data of the target voxels can be set to 0, thus generating the initial three-dimensional segmentation image.

[0106] Here, the first neural network can be used to accurately determine the initial three-dimensional segmentation image; or, the shape data of the target part and a set voxel threshold can also be used to simply and quickly determine the initial three-dimensional segmentation image.

[0107] In step A2, after obtaining the initial three-dimensional segmentation image, the target voxels located on the target part in the initial three-dimensional segmentation image can be dilated to obtain an adjusted three-dimensional segmentation image. Among them, the target contour of the target part included in the generated three-dimensional segmentation image is larger than the initial contour of the target part in the initial three-dimensional segmentation image, increasing the contour area range corresponding to the target part. Among them, the dilation process can be a graphics dilation process. For example, the graphics dilation process can be horizontal dilation, vertical dilation, omnidirectional dilation, etc.

[0108] In this embodiment, after generating the initial three-dimensional segmentation image, the target voxels located on the target part in the initial three-dimensional segmentation image can be dilated to generate a three-dimensional segmentation image, so that the segmentation area corresponding to the target part in the generated three-dimensional segmentation image increases, alleviating the omission of the target feature information corresponding to the target part. Furthermore, using the adjusted three-dimensional segmentation image, the detection result corresponding to the three-dimensional image can be determined more accurately.

[0109] In an optional embodiment, after obtaining the three-dimensional image of the target part, the method further includes: according to the set preprocessing voxel range, adjusting the voxel information of at least one voxel located within the preprocessing voxel range in the three-dimensional image to a preset range, and adjusting the voxel information of at least one voxel located outside the preprocessing voxel range in the three-dimensional image to a preset value to generate a processed three-dimensional image.

[0110] During implementation, according to the set preprocessing voxel range, the voxel information of at least one voxel located within the preprocessing voxel range in the three-dimensional image can be adjusted to a preset range, and the preset range can be between 0 and 1 (including 0 and 1), etc. And the voxel information of at least one voxel located outside the preprocessing voxel range in the three-dimensional image can be adjusted to a preset value, that is, the voxel information of at least one voxel located outside the preprocessing voxel range in the three-dimensional image is assigned a value to generate a processed three-dimensional image. Among them, the preset value can be set as needed. For example, the preset value can be 0 or 1.

[0111] Exemplarily, if the set preprocessing voxel range is [30 - 80], the voxel value (i.e., voxel information) of at least one voxel greater than 80 in the three-dimensional image can be assigned a value of 1, the voxel value of at least one voxel less than 30 can be assigned a value of 0, and the voxel values of the voxels between 30 and 80 are normalized, that is, the voxel values of each voxel in the processed three-dimensional image are all between 0 and 1. For example, if the voxel value of voxel one is 40, the voxel value of voxel one after normalization is 0.2.

[0112] Furthermore, the target point cloud image can be generated based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the processed three-dimensional image.

[0113] Here, each voxel in the three-dimensional image can be preprocessed. For example, the voxel information of at least one voxel within the preprocessing voxel range in the three-dimensional image can be normalized, and the voxel information of at least one voxel outside the preprocessing voxel range in the three-dimensional image can be adjusted to a preset value to generate a processed three-dimensional image, so that the detection result corresponding to the three-dimensional image can be determined more accurately using the processed three-dimensional image.

[0114] Regarding S102:

[0115] During implementation, the voxel positions and voxel information of the voxels in the three-dimensional image that match each target voxel can be determined according to the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image. Then, based on the voxel positions and voxel information of the voxels in the three-dimensional image that match each target voxel, the target point cloud image corresponding to the three-dimensional image can be generated.

[0116] In an optional implementation manner, generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image may include: Step B1 - Step B3, where:

[0117] Step B1: Determine a first voxel that matches the three-dimensional coordinate data of the target voxel from among the multiple voxels included in the three-dimensional image;

[0118] Step B2: Determine the three-dimensional coordinate data of the first voxel as the coordinate data of the point cloud point corresponding to the first voxel, and determine the voxel information of the first voxel as the attribute data of the point cloud point;

[0119] Step B3: Generate the target point cloud image corresponding to the three-dimensional image based on the coordinate data and the attribute data of the point cloud points respectively corresponding to each of the first voxels.

[0120] During implementation, the three-dimensional coordinate data of the target voxels included in the three-dimensional segmentation image can be determined first, where the three-dimensional coordinate data is the position information of the target voxels in the three-dimensional segmentation image; then, from the multiple voxels included in the three-dimensional image, the first voxel matching the three-dimensional coordinate data of the target voxels is determined; and the three-dimensional coordinate data of the first voxel in the three-dimensional image is determined as the coordinate data of the point cloud point corresponding to the first voxel, and the voxel information of the first voxel is determined as the attribute data of the point cloud point. For example, the voxel value of the first voxel is determined as the point value of the point cloud point. Furthermore, the coordinate data and attribute data of the point cloud points corresponding to each first voxel can be used to generate the target point cloud image corresponding to the three-dimensional image. Among them, the target point cloud image includes each point cloud point located on the target part.

[0121] Here, by determining the target point cloud image corresponding to the three-dimensional image, the screening of other feature information of other parts except the target part is realized, and the detection efficiency of the three-dimensional image is improved when using the target point cloud image to determine the detection result corresponding to the three-dimensional image.

[0122] For S103:

[0123] After obtaining the target point cloud image, the detection result of the target part on the three-dimensional image can be determined according to the target point cloud image. For example, the detection result of the target part on the three-dimensional image can be determined in response to the detection operation of the medical staff.

[0124] Alternatively, after obtaining the target point cloud image, the target point cloud image can be input into the trained target neural network to determine the detection result of the target part on the three-dimensional image. Among them, the network structure of the target neural network can be set according to actual needs.

[0125] The determined detection result can include the position information of the local area with damage on the target part. For example, the position information can be the position information of the detection box corresponding to the local area, the position information of the center point corresponding to the local area, etc.; and / or, the category information for characterizing the damage condition of the target part. For example, the category information can include: open fracture and closed fracture; or, it can include: mild fracture, severe fracture, severe fracture, etc.

[0126] In an optional implementation manner, as shown in Figure 2 The determination of the detection result of the target part on the three-dimensional image based on the target point cloud image may include:

[0127] S201, perform at least one convolution process on the target point cloud image to generate first intermediate feature data;

[0128] S202. Perform at least one deconvolution process on the first intermediate feature data to generate second intermediate feature data;

[0129] S203. Fuse the first intermediate feature data and the second intermediate feature data to generate target feature data;

[0130] S204. Based on the target feature data, determine the detection result of detecting the target part of the three-dimensional image.

[0131] Specifically, during implementation, the trained target neural network can be used to process the target point cloud image to determine the detection result of detecting the target part of the three-dimensional image. Refer to Figure 3 the structural schematic diagram of the target neural network shown in Figure 3 for a specific description of the process of S201 - S204. Among them, the network structure of the target neural network can be set as needed. This is only an exemplary description here. For example, the number of convolutional layers, the number of residual modules, the number of deconvolutional layers, etc. in the target neural network can be determined as needed, and the number of channels and feature sizes corresponding to each layer of the network structure can be set as needed.

[0132] The process of using the target neural network to process the target point cloud image to obtain the detection result may include:

[0133] First, the convolutional layer in the target neural network performs a convolution process on the target point cloud image (for example, the size can be 128×128×128 and the number of channels can be 1) to generate first intermediate feature data 1 with a size of 128×128×128 and a number of channels of 32.

[0134] Second, use the convolutional layer to perform a convolution process on the first intermediate feature data 1 to generate first intermediate feature data 2 with a size of 64×64×64 and a number of channels of 32.

[0135] Third, use the residual module to process the first intermediate feature data 2 to generate first intermediate feature data 3 with a size of 64×64×64 and a number of channels of 64.

[0136] Fourth, use the convolutional layer to perform a convolution process on the first intermediate feature data 3 to generate first intermediate feature data 4 with a size of 32×32×32 and a number of channels of 96.

[0137] Fifth, use the residual module to process the first intermediate feature data 4 to generate first intermediate feature data 5 with a size of 32×32×32 and a number of channels of 96.

[0138] Sixth, perform convolution processing on the first intermediate feature data 5 using a convolutional layer to generate first intermediate feature data 6 with a size of 16×16×16 and 128 channels; and process the first intermediate feature data 5 using a residual module to generate first intermediate feature data 7 with a size of 32×32×32 and 96 channels.

[0139] Seventh, process the first intermediate feature data 6 using a residual module to generate first intermediate feature data 8 with a size of 16×16×16 and 128 channels.

[0140] Eighth, perform convolution processing on the first intermediate feature data 8 using a convolutional layer to generate first intermediate feature data 9 with a size of 8×8×8 and 160 channels; and process the first intermediate feature data 8 using a residual module to generate first intermediate feature data 10 with a size of 16×16×16 and 128 channels.

[0141] Ninth, process the first intermediate feature data 9 using a residual module to generate first intermediate feature data 11 with a size of 8×8×8 and 160 channels.

[0142] Tenth, perform deconvolution processing on the first intermediate feature data 11 using a deconvolutional layer to generate second intermediate feature data 1 with a size of 16×16×16 and 128 channels.

[0143] Eleventh, fuse the first intermediate feature data 10 and the second intermediate feature data 1 to generate candidate feature data 1 with a size of 16×16×16 and 128 channels.

[0144] Twelfth, process the candidate feature data 2 using a residual module to generate candidate feature data 2 with a size of 16×16×16 and 128 channels.

[0145] Thirteenth, perform deconvolution processing on the candidate feature data 2 using a deconvolutional layer to generate second intermediate feature data 2 with a size of 32×32×32 and 96 channels.

[0146] Fourteenth, fuse the first intermediate feature data 7 and the second intermediate feature data 2 to generate candidate feature data 3 with a size of 32×32×32 and 96 channels.

[0147] Fifteenth, process the candidate feature data 3 using a residual module to generate target feature data with a size of 32×32×32 and 96 channels.

[0148] Sixteenth, process the target feature data using the set classification branch and regression branch to generate a detection result.

[0149] For example, when training a target neural network, the set loss function may include a Focal classification loss function and an L1 location regression loss function.

[0150] Here, by generating first intermediate feature data and second intermediate feature data, fusing the first intermediate feature data and the second intermediate feature data, and generating target feature data, the target feature data includes global features and local features in the three-dimensional image, making the feature information included in the target feature data relatively rich; furthermore, based on the target feature data, the detection result corresponding to the three-dimensional image can be determined more accurately.

[0151] In an optional implementation manner, the at least one convolution process on the target point cloud image to generate first intermediate feature data may include steps C1 - C3, where:

[0152] Step C1, during the convolution process on the target point cloud image, for each convolution calculation, determine whether there is any coordinate data of the point cloud points that matches the central position corresponding to the local feature data in the target point cloud image where the convolution calculation is performed;

[0153] Step C2, if there is a match, perform this convolution calculation using the convolution kernel to obtain the calculation result corresponding to this convolution calculation, otherwise abandon this convolution calculation and perform the next convolution calculation;

[0154] Step C3, generate first intermediate feature data based on the calculation results of at least one convolution calculation.

[0155] See Figure 4 As shown, when performing a convolution process on the target point cloud image, it is necessary to use a convolution kernel and a set moving step size to perform multiple convolution operations on the target point cloud image. For example, when the step size is 2, it is necessary to perform a convolution operation on the feature values included in the first moving frame 41 in the target point cloud image with the convolution kernel, perform a convolution operation on the feature values included in the second moving frame 42 in the target point cloud image with the convolution kernel, and so on until performing a convolution operation on the feature values included in the nth moving frame 4n in the target point cloud image with the convolution kernel; then, the processing result of this convolution process can be obtained through the calculation results of at least one convolution operation.

[0156] During implementation, during the convolution process on the target point cloud image, for each convolution calculation, it can be determined whether there is any coordinate data of the point cloud points that matches the central position corresponding to the local feature data in the target point cloud image where the convolution calculation is performed.

[0157] For example, when performing a convolution operation on the local feature data included in the first moving frame 41 using a convolution kernel, it can be determined whether there is any coordinate data of a point cloud point that matches the central position of the local feature data included in the first moving frame 41, that is, to determine whether the central position of the local feature data included in the first moving frame 41 corresponds to the coordinate data of any point cloud point; if it exists, the convolution kernel is used to perform this convolution calculation to obtain the calculation result corresponding to this convolution calculation; if it does not exist, this convolution calculation is abandoned and the next convolution calculation is executed. Furthermore, based on the calculation results of at least one convolution calculation, the first intermediate feature data can be generated.

[0158] Here, for each convolution calculation, it can be determined whether there is any coordinate data of a point cloud point that matches the central position corresponding to the local feature data for which this convolution calculation is performed in the target point cloud image. If it exists, the convolution kernel is used to perform this convolution calculation. If it does not exist, this convolution calculation is abandoned, which alleviates the resource consumption for performing convolution calculations on other feature information of other parts, reduces the computational resources consumed for determining the detection result, and improves the efficiency of determining the detection result.

[0159] Those skilled in the art can understand that in the above method of the specific implementation manner, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined based on its function and possible internal logic.

[0160] Based on the same concept, the embodiments of the present disclosure also provide a three-dimensional image detection device. Refer to Figure 5 As shown, it is a schematic architecture diagram of the three-dimensional image detection device provided by the embodiments of the present disclosure, including an acquisition module 501, a generation module 502, and a determination module 503. Specifically:

[0161] The acquisition module 501 is configured to acquire a three-dimensional image of a target part and a three-dimensional segmentation image corresponding to the three-dimensional image. Among them, the voxel information of the target voxels located on the target part in the three-dimensional segmentation image is different from the voxel information of other voxels located on other parts;

[0162] The generation module 502 is configured to generate a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image; among them, the target point cloud image includes each point cloud point located on the target part;

[0163] The determination module 503 is configured to determine a detection result of the target part on the three-dimensional image based on the target point cloud image.

[0164] In a possible implementation, after obtaining the three-dimensional image of the target part, the device further includes: a preprocessing module 504, configured to:

[0165] According to the set preprocessing voxel range, adjust the voxel information of at least one voxel within the preprocessing voxel range in the three-dimensional image to a preset range, and adjust the voxel information of at least one voxel outside the preprocessing voxel range in the three-dimensional image to a preset value, and generate a processed three-dimensional image;

[0166] When generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image, the generating module 502 is configured to:

[0167] Generate the target point cloud image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the processed three-dimensional image.

[0168] In a possible implementation, when obtaining the three-dimensional segmentation image corresponding to the three-dimensional image, the obtaining module 501 is configured to:

[0169] Generate an initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image;

[0170] Perform dilation processing on the target voxels located on the target part in the initial three-dimensional segmentation image to generate the three-dimensional segmentation image.

[0171] In a possible implementation, when generating the initial three-dimensional segmentation image corresponding to the three-dimensional image based on the three-dimensional image, the obtaining module 501 is configured to:

[0172] Obtain the initial three-dimensional segmentation image based on the three-dimensional image and a trained first neural network for image segmentation; or,

[0173] Segment the three-dimensional image using the shape data of the target part and the set voxel threshold to obtain the initial three-dimensional segmentation image.

[0174] In a possible implementation, when generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image, the generating module 502 is configured to:

[0175] Determine a first voxel in the three-dimensional image that matches the three-dimensional coordinate data of the target voxel from among the multiple voxels included in the three-dimensional image;

[0176] Determine the three-dimensional coordinate data of the first voxel as the coordinate data of the point cloud point corresponding to the first voxel, and determine the voxel information of the first voxel as the attribute data of the point cloud point;

[0177] Generate a target point cloud image corresponding to the three-dimensional image based on the coordinate data and the attribute data of the point cloud points respectively corresponding to each of the first voxels.

[0178] In a possible implementation manner, when determining the detection result of the target part on the three-dimensional image based on the target point cloud image, the determination module 503 is configured to:

[0179] Perform at least one convolution process on the target point cloud image to generate first intermediate feature data;

[0180] Perform at least one deconvolution process on the first intermediate feature data to generate second intermediate feature data;

[0181] Fuse the first intermediate feature data and the second intermediate feature data to generate target feature data;

[0182] Determine the detection result of detecting the target part on the three-dimensional image based on the target feature data.

[0183] In a possible implementation manner, when performing at least one convolution process on the target point cloud image to generate first intermediate feature data, the determination module 503 is configured to:

[0184] During the process of performing convolution on the target point cloud image, for each convolution calculation, determine whether there is a match between the coordinate data of any one of the point cloud points and the central position corresponding to the local feature data in the target point cloud image where the convolution calculation is performed;

[0185] If there is a match, perform this convolution calculation using the convolution kernel to obtain the calculation result corresponding to this convolution calculation; otherwise, abandon this convolution calculation and perform the next convolution calculation;

[0186] Generate first intermediate feature data based on the calculation results of at least one convolution calculation.

[0187] In some embodiments, the functions or templates included in the device provided by the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0188] Based on the same inventive concept, the embodiments of the present disclosure also provide an electronic device. Refer to Figure 6As shown in the figure, it is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure, including a processor 601, a memory 602, and a bus 603. Among them, the memory 602 is used to store execution instructions, including an internal memory 6021 and an external memory 6022; here, the internal memory 6021 is also called the main memory, which is used to temporarily store the operation data in the processor 601 and the data exchanged with the external memory 6022 such as a hard disk. The processor 601 exchanges data with the external memory 6022 through the internal memory 6021. When the electronic device 600 runs, the processor 601 communicates with the memory 602 through the bus 603, so that the processor 601 executes the following instructions:

[0189] Obtain a three-dimensional image of a target part and a three-dimensional segmentation image corresponding to the three-dimensional image. Among them, the voxel information of the target voxels located on the target part in the three-dimensional segmentation image is different from the voxel information of other voxels located on other parts;

[0190] Generate a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmentation image and the three-dimensional image; among them, each point cloud point located on the target part is included in the target point cloud image;

[0191] Determine the detection result of the target part on the three-dimensional image based on the target point cloud image.

[0192] Among them, the specific processing flow of the processor 601 can refer to the description in the above method embodiment and will not be elaborated here.

[0193] In addition, an embodiment of the present disclosure also provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is run by a processor, it executes the steps of the three-dimensional image detection method described in the above method embodiment. Among them, the storage medium can be a volatile or non-volatile computer-readable storage medium.

[0194] An embodiment of the present disclosure also provides a computer program product. The computer program product carries program codes. The instructions included in the program codes can be used to execute the steps of the three-dimensional image detection method described in the above method embodiment. Specifically, refer to the above method embodiment and will not be elaborated here.

[0195] Among them, the above computer program product can be specifically implemented by means of hardware, software, or a combination thereof. In an optional embodiment, the computer program product is specifically embodied as a computer storage medium. In another optional embodiment, the computer program product is specifically embodied as a software product, such as a Software Development Kit (SDK), etc.

[0196] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems and devices described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein. In several embodiments provided by the present disclosure, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical functional division, and there may be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some communication interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0197] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0198] In addition, in each embodiment of the present disclosure, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0199] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such an understanding, the technical solution of the present disclosure, in essence, or the part that contributes to the prior art or part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present disclosure. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs that can store program codes.

[0200] If the technical solution of this application involves personal information, before the product applying the technical solution of this application processes personal information, it has clearly informed the personal information processing rules and obtained the individual's independent consent. If the technical solution of this application involves sensitive personal information, before the product applying the technical solution of this application processes sensitive personal information, it has obtained the individual's separate consent and at the same time meets the requirements of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If an individual voluntarily enters the collection scope, it is regarded as consenting to the collection of their personal information; or on the device for personal information processing, when the personal information processing rules are informed by obvious signs / information, personal authorization is obtained through pop-up messages or by asking the individual to upload their personal information themselves, etc.; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0201] The above are only specific implementation manners of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present disclosure can easily think of changes or substitutions, which should all be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the protection scope of the claims.

Claims

1. A three-dimensional image detection method, characterized in that: include: Acquire a three-dimensional image of the target part and a three-dimensional segmented image corresponding to the three-dimensional image, wherein voxel information of a target voxel located at the target part in the three-dimensional segmented image is different from voxel information of other voxels located at other parts; generating a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmented image and the three-dimensional image; wherein the target point cloud image includes each point cloud point located on the target part; Determining a detection result of the target part on the three-dimensional image based on the target point cloud image; The step of generating a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmented image and the three-dimensional image includes: Determining a first voxel that matches the three-dimensional coordinate data of the target voxel from a plurality of voxels included in the three-dimensional image; Determining the three-dimensional coordinate data of the first voxel as coordinate data of a point cloud point corresponding to the first voxel; and determining the voxel information of the first voxel as attribute data of the point cloud point; A target point cloud image corresponding to the three-dimensional image is generated based on the coordinate data and the attribute data of the point cloud points corresponding to each of the first voxels.

2. The method according to claim 1, characterized in that After acquiring the three-dimensional image of the target part, the method further includes: adjusting, according to the set preprocessing voxel range, voxel information of at least one voxel in the three-dimensional image that is within the preprocessing voxel range to be within a preset range, and adjusting voxel information of at least one voxel in the three-dimensional image that is outside the preprocessing voxel range to a preset value, thereby generating a processed three-dimensional image; Generating a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmented image and the three-dimensional image includes: The target point cloud image is generated based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmented image and the processed three-dimensional image.

3. The method according to claim 1 or 2, characterized in that Acquiring a three-dimensional segmented image corresponding to the three-dimensional image includes: Based on the three-dimensional image, generating an initial three-dimensional segmented image corresponding to the three-dimensional image; The target voxels located on the target part in the initial three-dimensional segmentation image are expanded to generate the three-dimensional segmentation image.

4. The method according to claim 3, characterized in that Generating an initial three-dimensional segmented image corresponding to the three-dimensional image based on the three-dimensional image includes: obtaining the initial three-dimensional segmented image based on the three-dimensional image and a trained first neural network for image segmentation; or The three-dimensional image is segmented using the shape data of the target part and a set voxel threshold to obtain the initial three-dimensional segmented image.

5. The method according to claim 1, wherein Determining a detection result of the target part on the three-dimensional image based on the target point cloud image includes: Performing at least one convolution process on the target point cloud image to generate first intermediate feature data; performing at least one deconvolution process on the first intermediate feature data to generate second intermediate feature data; fusing the first intermediate feature data and the second intermediate feature data to generate target feature data; Based on the target feature data, a detection result of target part detection on the three-dimensional image is determined.

6. The method according to claim 5, characterized in that The performing at least one convolution process on the target point cloud image to generate first intermediate feature data includes: During the convolution processing of the target point cloud image, for each convolution calculation, determining whether the coordinate data of any point cloud point matches the center position corresponding to the local feature data for performing the convolution calculation in the target point cloud image; If it exists, the convolution kernel is used to perform the convolution calculation and obtain the calculation result corresponding to the convolution calculation. Otherwise, the convolution calculation is abandoned and the next convolution calculation is performed. Based on the calculation result of at least one convolution calculation, first intermediate feature data is generated.

7. A three-dimensional image detection device, characterized in that: include: an acquisition module, configured to acquire a three-dimensional image of a target part and a three-dimensional segmented image corresponding to the three-dimensional image, wherein voxel information of a target voxel located at the target part in the three-dimensional segmented image is different from voxel information of other voxels located at other parts; a generating module, configured to generate a target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmented image and the three-dimensional image; wherein the target point cloud image includes each point cloud point located on the target part; generating the target point cloud image corresponding to the three-dimensional image based on the three-dimensional coordinate data of each target voxel in the three-dimensional segmented image and the three-dimensional image comprises: determining a first voxel that matches the three-dimensional coordinate data of the target voxel from a plurality of voxels included in the three-dimensional image; determining the three-dimensional coordinate data of the first voxel as coordinate data of a point cloud point corresponding to the first voxel; determining the voxel information of the first voxel as attribute data of the point cloud point; and generating the target point cloud image corresponding to the three-dimensional image based on the coordinate data and the attribute data of the point cloud points corresponding to each of the first voxels; A determination module is used to determine a detection result of the target part on the three-dimensional image based on the target point cloud image.

8. An electronic device, characterized in that: include: A processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the three-dimensional image detection method according to any one of claims 1 to 6 are performed.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the three-dimensional image detection method according to any one of claims 1 to 6.

Images