CN115641324A - Cervical vertebra key point prediction method and system, electronic equipment and storage medium - Google Patents
Cervical vertebra key point prediction method and system, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115641324A CN115641324A CN202211377880.5A CN202211377880A CN115641324A CN 115641324 A CN115641324 A CN 115641324A CN 202211377880 A CN202211377880 A CN 202211377880A CN 115641324 A CN115641324 A CN 115641324A
- Authority
- CN
- China
- Prior art keywords
- image
- predicted
- cone
- heat map
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a method and a system for predicting cervical vertebra key points, electronic equipment and a storage medium, wherein the method for predicting the cervical vertebra key points comprises the following steps: performing image segmentation processing on the cervical vertebra medical image, and determining a plurality of segmented single cone images; selecting a single cone image to be predicted from the single cone image; and inputting the single cone image to be predicted into the key point prediction model to determine the key point of the single cone to be predicted. Because the cone structure of the cervical vertebra has great similarity and repeatability, in order to meet the prediction requirements of a plurality of key points on the surface of the cone, the invention carries out image segmentation on a cervical vertebra medical image, and then selects a single cone image to be predicted of a specific type to input into a key point prediction model to determine the key points of the single cone to be predicted, so as to improve the accuracy of predicting the cervical vertebra key points and reduce prediction ambiguity brought by a repeated structure.
Description
Technical Field
The invention relates to the technical field of electronic information, in particular to a method and a system for predicting cervical vertebra key points, electronic equipment and a storage medium.
Background
The existing cervical vertebra CBCT (Cone beam CT, i.e. Cone beam projection computed tomography) key point prediction technology is usually dedicated to solving the vertebral centre positioning problem, and the local precision is insufficient for the prediction requirements of a plurality of key points on the surface of the vertebral segment. The key point prediction technology of the cervical CBCT based on deep learning generally requires a large amount of training data, and the acquisition of medical image data has high requirements on the instrument and the labeling accuracy, so the prediction mode of deep learning is generally expensive and time-consuming. The keypoint prediction techniques of cervical CBCT using FCN (full convolution neural network) or reinforcement learning generally have a large number of parameters and a high demand for computational resources during training. Techniques for performing keypoint prediction for cervical CBCT using a heatmap regression framework typically require cumbersome post-processing steps to eliminate false-positive responses.
Disclosure of Invention
The invention aims to overcome the defects of insufficient local precision of key point prediction, complex processing steps and the like in the prior art, and provides a method and a system for predicting cervical vertebra key points, electronic equipment and a storage medium.
The invention solves the technical problems through the following technical scheme:
in a first aspect, the present invention provides a method for predicting a cervical spine key point, including:
performing image segmentation processing on the cervical vertebra medical image, and determining a plurality of segmented single cone images;
selecting a single cone image to be predicted from the single cone image;
and inputting the single conic node image to be predicted into a key point prediction model so as to determine the key point of the single conic node to be predicted.
Preferably, the step of performing image segmentation processing on the cervical vertebra medical image comprises:
carrying out image segmentation processing on the cervical vertebra medical image by adopting an image segmentation model; the image segmentation model is obtained by training nnU-Net or 3D U-Net of cervical vertebra image sample pairs marked with a single conical section segmentation result.
Preferably, the step of determining a plurality of segmented single cone images comprises:
expanding the single conic image to be predicted in the divided conic region, and determining the maximum communication region after expansion as a mask region;
and determining the cone image in the mask area as a single segmented cone image.
Preferably, the keypoint prediction model is obtained by training a CNN (convolutional neural network) -based heat map regression network by using a single cone image sample.
Preferably, the keypoint prediction model comprises a local appearance model, and the step of inputting the single conic node image to be predicted into the keypoint prediction model to determine the keypoint of the single conic node to be predicted comprises:
inputting the single cone image to be predicted into a local appearance model to output a three-dimensional heat map;
determining the peak coordinates of the three-dimensional heat map as key points of the single conic node to be predicted.
Preferably, the keypoint prediction model includes a local appearance model and a spatial configuration model, and the step of inputting the single conic node image to be predicted into the keypoint prediction model to determine the keypoint of the single conic node to be predicted includes:
inputting the single cone image to be predicted into a local appearance model to output a three-dimensional heat map;
inputting the three-dimensional heat map into a spatial configuration model to output a spatial heat map;
multiplying the outputs of the three-dimensional heat map and the spatial heat map element by element to obtain a final heat map;
determining the peak coordinates of the final heat map as the key points of the single cone to be predicted.
Preferably, the local appearance model and the spatial configuration model are obtained by training the SCN with a single image sample of the cone pitch to be predicted.
In a second aspect, the present invention provides a cervical spine key point prediction system, including:
the determining module is used for carrying out image segmentation processing on the cervical vertebra medical image and determining a plurality of segmented single cone images;
the selection module is used for selecting a single conic node image to be predicted from the single conic node images;
the determining module is further used for inputting the single conic node image to be predicted into a key point prediction model so as to determine the key point of the single conic node to be predicted.
In a third aspect, the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for predicting a vital point of cervical vertebrae when executing the computer program.
In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, the computer program, when being executed by a processor, implementing the method for predicting a critical point of cervical spine as described above.
The positive progress effects of the invention are as follows:
because the cone structure of the cervical vertebra has great similarity and repeatability, in order to meet the prediction requirements of a plurality of key points on the surface of the cone, the invention carries out image segmentation on a cervical vertebra medical image, and then selects a single cone image to be predicted of a specific type to input into a key point prediction model to determine the key points of the single cone to be predicted, so as to improve the accuracy of predicting the cervical vertebra key points and reduce prediction ambiguity brought by a repeated structure.
Drawings
Fig. 1 is a first flowchart of a method for predicting a key point of a cervical vertebra according to embodiment 1 of the present invention;
fig. 2 is a second flowchart of a method for predicting key points of cervical vertebrae according to embodiment 1 of the present invention;
fig. 3 is a second flowchart of a method for predicting key points of cervical vertebrae according to embodiment 1 of the present invention;
fig. 4 is an effect diagram of a method for predicting a cervical spine key point according to embodiment 1 of the present invention;
fig. 5 is a first structural diagram of a cervical spine keypoint prediction system according to embodiment 2 of the present invention;
fig. 6 is a second structural diagram of a cervical spine key point prediction system according to embodiment 2 of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to embodiment 3 of the present invention.
Detailed Description
The invention is further illustrated by the following examples, which are not intended to limit the scope of the invention.
Example 1
The embodiment provides a method for predicting a cervical spine key point, where the position of the cervical spine key point is used to detect or monitor a health state of the cervical spine, and the method is applied to a cervical spine computerized reconstruction tomography image (e.g. a cervical spine CBCT image), and referring to fig. 1, the method for predicting the cervical spine key point includes:
s1, performing image segmentation processing on the cervical vertebra medical image, and determining a plurality of segmented single cone images.
The medical images of the cervical vertebrae can be obtained by scanning with CT (Computed Tomography) equipment, and can also be obtained by scanning with PET (Positron Emission Tomography) equipment.
And S2, selecting a single cone image to be predicted from the single cone image.
For example, the C2 cone image is selected from the C1-C7 cone images as the single cone image to be predicted of a specific type.
And S3, inputting the single conic node image to be predicted into the key point prediction model to determine the key point of the single conic node to be predicted.
In an alternative embodiment, the keypoint prediction model is obtained by training a CNN (convolutional neural network) -based heat map regression network with a single cone image sample.
The following introduces a training method of a key point prediction model for predicting the key point positions of a single cone joint to be predicted end to end based on a heat map regression framework of CNN:
a d-dimensional target heat map (i.e., a single image sample of the cone node to be detected) is defined by using a gaussian formula, and the d-dimensional target heat map refers to a d-dimensional matrix which is formed by radiating (decreasing in value) to the periphery by taking a target key point as a center (at a peak position). The value at each coordinate of the matrix can be calculated by a gaussian formula. For x i Representing the coordinates of the target key points on a target heat map, which is defined as follows:
wherein x is i And σ i For a parameter defined in a Gaussian distribution, x i Means, σ, representing the Gaussian distribution i The standard deviation of the gaussian distribution is indicated. x represents an argument, which in this embodiment can be considered as any point coordinate in the target heat map. Gamma is used to avoid instability of the network values during training. Unlike other heat map regression methods that fix the value of the gaussian function σ, this embodiment sets σ as one of the parameters that need to be trained. Because σ affects the width of the peak in the gaussian function, in the training of the network, a different σ value depending on the confidence of the prediction is set for each keypoint. The network synchronously predicts N heat maps (N represents the number of key points of a single cone image to be detected), and the aim is to minimize the sum of the distances between the predicted heat maps and the target heat maps corresponding to all target key points. The target heat map is obtained according to the Gaussian formula, is calculated according to the real key point position and is used for representing the real information of the key point; the prediction heat map is an output result of the final key point prediction model; the target heat map is centered on the target keypoints (i.e., the real keypoints), and the center of the prediction heat map is the prediction keypoints. The penalty function for calculating the difference between the predicted and true case is set as follows:
where w, b represent the weight and deviation of the network, respectively, h i (x; w, b) represents the prediction heatmap (i.e., the output of the keypoint prediction model), and α, λ are equilibrium parameters.
In an alternative embodiment, the keypoint prediction model comprises a local appearance model, see fig. 2, step S3 comprising:
and S31, inputting the single cone image to be predicted into the local appearance model so as to output a three-dimensional heat map.
The three-dimensional heat map is the three-dimensional heat map of a single cone to be predicted, wherein the local key points are accurate, but ambiguous key points exist.
Specifically, the method comprises the following steps: the local appearance model takes a multi-level structure of a contraction path and an expansion path, each level being composed of a plurality of successive convolution layers. In the shrink path, averaging pooling is used before the last convolutional layer of each stage, down-sampled to half the current resolution as input to the next stage. In the extended path, the output of each stage is linearly up-sampled to double the resolution, then connected to the convolutional layer output of the previous stage, until the original resolution is reached again, and the three-dimensional heat map is output through the last convolutional layer.
And S32, determining the peak coordinates of the three-dimensional heat map as key points of a single cone to be predicted.
In the embodiment, the single cone image to be predicted is input into the local appearance model, and the key point of the single cone to be predicted is determined, so that the number of parameters is simplified, the calculation complexity is reduced, and the efficiency of predicting the key point of the single cone to be predicted is improved.
In an alternative embodiment, the keypoint prediction model comprises a local appearance model and a spatial configuration model, see fig. 3, and step S3 comprises:
and S34, inputting the single cone image to be predicted into the local appearance model so as to output a three-dimensional heat map.
The three-dimensional heat map is the three-dimensional heat map of a single cone to be predicted, wherein the local key points are accurate, but ambiguous key points exist.
And S35, inputting the three-dimensional heat map into the space configuration model to output the space heat map.
The spatial heat map is a heat map containing spatial information obtained after disambiguating the three-dimensional heat map.
And S36, multiplying the outputs of the three-dimensional heat map and the spatial heat map element by element to obtain a final heat map.
The three-dimensional heat map and the space heat map are three-dimensional matrixes, each coordinate position in the three-dimensional matrixes is multiplied element by element, the value of the coordinate position on the three-dimensional heat map is multiplied by the value of the coordinate position on the space heat map to be used as the value of the coordinate position corresponding to the final heat map, and the final heat map can be obtained after the values of all the coordinate positions are determined. The value ranges of the three-dimensional heat map and the space heat map at all coordinate positions are [0,1], and after multiplication, only the coordinate position with a larger value on the three-dimensional heat map and the space heat map can have a larger value on the final heat map.
And S37, determining the peak coordinates of the final heat map as key points of a single cone to be predicted.
Fig. 4 is a comparison diagram of the manually labeled key point positions of the single vertebral segment to be predicted and the key point positions of the single vertebral segment to be predicted obtained by applying the cervical spine key point prediction method of the embodiment. The dots represent the key points of the single vertebral segment to be predicted through manual marking, and the triangles represent the key points of the single vertebral segment to be predicted, which are obtained by applying the cervical vertebra key point prediction method. The positions of the key points marked by the cervical vertebra key point prediction method are deviated from the positions of the key points marked manually, and the positions of the key points marked by the cervical vertebra key point prediction method are more accurate from experimental data.
In an optional embodiment, the local appearance model and the spatial configuration model are both trained by using a single cone image sample pair to be predicted, SCN (spatial configuration-Net, a spatial configuration network, which is a CNN).
Specifically, the method comprises the following steps:
the local appearance model takes a multi-level structure of a contraction path and an expansion path, each level being composed of a plurality of successive convolution layers. In the systolic path, averaging pooling is used before the last convolutional layer of each stage, down-sampled to half the current resolution as input to the next stage. In the extended path, the output of each stage is linearly up-sampled to double the resolution, then connected to the convolutional layer output of the previous stage, until the original resolution is reached again, and the three-dimensional heat map is output through the last convolutional layer.
The spatial configuration model takes a three-dimensional heat map as input, implicitly incorporates a geometric model representing the spatial relationship between keypoints, provides a response in the vicinity of target keypoint locations, and suppresses false positive reactions in the three-dimensional heat map by learning how to robustly predict the location of individual keypoints from local position prediction. Because the spatial configuration module has low requirement on the accuracy of local information, in order to reduce the calculation amount, down sampling is firstly carried out on the three-dimensional heat map, then continuous convolutional layers are used for modeling the complex relation between key points, and finally the up sampling is carried out to the original resolution.
In this embodiment, the key point prediction model includes a local appearance model and a spatial configuration model, where the local appearance model predicts a three-dimensional heat map of a single image of a cone to be predicted, where a local key point is accurate but the local key point is ambiguous, from an input three-dimensional image of the single image of the cone to be predicted. And the spatial configuration model learns the spatial relationship among the key points, eliminates ambiguous key points from the three-dimensional heat map obtained by the local appearance module and obtains a spatial heat map containing spatial information. And multiplying the outputs of the two models element by element to obtain a final heat map, and taking the peak value coordinate as the predicted key point coordinate. The number of parameters is simplified, the calculation complexity is reduced, and the accuracy and efficiency of predicting the key point of the single cone node to be predicted are improved.
Because the cone structure of the cervical vertebra has great similarity and repeatability, in order to meet the prediction requirements of multiple key points on the surface of the cone, the image segmentation processing is carried out on the medical image of the cervical vertebra, and then a single cone image to be predicted of a specific type is selected and input into the key point prediction model to determine the key points of the single cone to be predicted, so that the accuracy of predicting the key points of the cervical vertebra is improved, and the prediction ambiguity caused by the repeated structure is reduced.
In an alternative embodiment, step S1 comprises:
and carrying out image segmentation processing on the cervical vertebra medical image by adopting an image segmentation model. The image segmentation model is obtained by training a pair of cervical vertebra image samples nnU-Net (a deep learning network) or 3D U-Net (a deep learning network) marked with a single cone segmentation result.
In the embodiment, the image segmentation model is adopted to perform image segmentation processing on the cervical vertebra medical image and is used as a pre-processing step of the prediction key point, so that the prediction accuracy is improved, and the prediction ambiguity caused by similar and repeated cone structure is reduced.
In an alternative embodiment, referring to fig. 2, step S1 comprises:
and S11, carrying out image segmentation processing on the cervical vertebra medical image.
And S12, performing expansion processing on the divided cone joint area, and determining the maximum communication area after the expansion processing as a mask area.
Wherein a cone region refers to the actual single cone region. The problem that the segmented single vertebral segment region is hollow, discontinuous, slightly smaller than the actual vertebral segment region and the like possibly exists, and the position of a key point of the finally predicted single vertebral segment image to be predicted can be influenced. Therefore, the expansion processing is carried out on the segmented single vertebral segment region, and the maximum communication region after the expansion processing is taken as a mask region which is not smaller than (preferably slightly larger than) the actual single vertebral segment region. The maximum communication zone refers to the largest continuous zone (i.e., the location of the actual single vertebral segment) after the expansion process has been performed on the single vertebral segment region.
And S13, determining the cone image in the mask area as a single divided cone image.
In this embodiment, since the segmented single vertebral segment region may have the problems of cavities, discontinuity, slightly smaller than the actual region of the vertebral segment, and the like, and may affect the position of the key point of the finally predicted single vertebral segment image to be predicted, the segmented vertebral segment region is expanded, and the maximum communication region is determined as the mask region, thereby further improving the accuracy of predicting the key point of the single vertebral segment to be predicted.
In an optional embodiment, before step S3, the single image of the cone to be predicted is input into the normalization model to obtain a single image of the cone to be predicted with a standard size, so as to improve the efficiency of predicting the key point of the single cone to be predicted.
The normalized model is obtained by training according to the following steps:
firstly, sampling a plurality of cervical vertebra medical image samples to the same sampling interval so as to obtain a plurality of vertebral segment regions of the same type. The cone region images in the mask region are determined to be single cone images of the same type (for example, a plurality of cone images from C1 to C7 are obtained respectively). Then, a single cone image to be predicted is selected from the single cone images (for example, the C2 cone image is selected from the cone images from C1 to C7 as the single cone image to be predicted).
Secondly, determining the image size of the single vertebral image of the same type according to the actual vertebral size of the single vertebral image of the same type and the initial key point position marked on the single vertebral image of the same type. The processed single vertebral level images of the same type are aligned to a coordinate system with the origin (0,0,0).
And finally, performing image data enhancement processing on the masked and aligned single vertebral image of the same type. The method is characterized in that the random cutting, rotation and scaling processing are carried out on the single vertebral image of the same type, and the coordinate value of the cut image is normalized to the range of [ -1,1] to obtain the training data input of the normalized model. Thereby achieving higher robustness on the basis of a small amount of training data.
Example 2
The present embodiment provides a cervical spine key point prediction system, referring to fig. 5, the cervical spine key point prediction system includes:
the determining module 1 is used for performing image segmentation processing on the cervical vertebra medical image and determining a plurality of segmented single cone images.
And the selection module 2 is used for selecting a single conic node image to be predicted from the single conic node images.
The determining module 1 is further configured to input the single conic node image to be predicted into the key point prediction model to determine a key point of the single conic node to be predicted.
In an alternative embodiment, referring to fig. 6, the cervical spine keypoint prediction system further comprises:
the segmentation module 3 is used for carrying out image segmentation processing on the cervical vertebra medical image by adopting an image segmentation model; the image segmentation model is obtained by training nnU-Net or 3D U-Net by using cervical vertebra image sample pairs marked with a single conical section segmentation result.
In an optional embodiment, the determining module 1 is further configured to perform expansion processing on the segmented cone region, and determine a maximum communication region after the expansion processing as a mask region; and determining the cone image in the mask area as a single segmented cone image.
In an alternative embodiment, the keypoint prediction model is obtained by training a CNN-based heat map regression network with a single cone image sample.
In an alternative embodiment, the keypoint prediction model comprises a local appearance model, see fig. 6, the cervical spine keypoint prediction system further comprising: and the acquisition module 4 is used for inputting the single cone image to be predicted into the local appearance model so as to output the three-dimensional heat map.
The determining module 1 is further configured to determine the peak coordinates of the three-dimensional heat map as key points of a single cone to be predicted.
In an optional embodiment, the keypoint prediction model includes a local appearance model and a spatial configuration model, and the obtaining module 4 is further configured to input a single image of a cone joint to be predicted into the local appearance model to output a three-dimensional heat map; the three-dimensional heat map is also input into the spatial configuration model to output a spatial heat map; and also for multiplying the outputs of the three-dimensional heat map and the spatial heat map element-by-element to obtain a final heat map.
The determining module 1 is further configured to determine the peak coordinates of the final heat map as key points of a single cone to be predicted.
In an alternative embodiment, the local appearance model and the spatial configuration model are obtained by training the SCN by using a single image sample of the cone node to be predicted.
It should be noted that, the implementation principle and the technical effect of each module of the cervical spine key point prediction system of this embodiment may refer to the corresponding parts of embodiment 1, and are not described herein again.
Example 3
This embodiment provides an electronic device, and fig. 7 is a schematic block diagram of the electronic device. The electronic device comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the cervical vertebra key point prediction method of the embodiment 1. The electronic device 30 shown in fig. 7 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiment of the present invention.
As shown in fig. 7, the electronic device 30 may be embodied in the form of a general purpose computing device, which may be, for example, a server device. The components of the electronic device 30 may include, but are not limited to: the at least one processor 31, the at least one memory 32, and a bus 33 connecting the various system components (including the memory 32 and the processor 31).
The bus 33 includes a data bus, an address bus, and a control bus.
The memory 32 may include volatile memory, such as Random Access Memory (RAM) 321 and/or cache memory 322, and may further include Read Only Memory (ROM) 323.
The processor 31 executes various functional applications and data processing, such as the cervical spine keypoint prediction method of embodiment 1 of the present invention, by running a computer program stored in the memory 32.
The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). Such communication may be through input/output (I/O) interfaces 35. Also, model-generating device 30 may also communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) via network adapter 36. As shown in FIG. 7, network adapter 36 communicates with the other modules of model-generating device 30 via bus 33. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the model-generating device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.
It should be noted that although in the above detailed description several units/modules or sub-units/modules of the electronic device are mentioned, such a division is merely exemplary and not mandatory. Indeed, the features and functions of two or more of the units/modules described above may be embodied in one unit/module according to embodiments of the invention. Conversely, the features and functions of one unit/module described above may be further divided into embodiments by a plurality of units/modules.
Example 4
The present embodiment provides a computer-readable storage medium on which a computer program is stored, which when executed by a processor, implements the cervical spine keypoint prediction method of embodiment 1.
More specific examples, among others, that the readable storage medium may employ may include, but are not limited to: a portable disk, a hard disk, random access memory, read only memory, erasable programmable read only memory, optical storage device, magnetic storage device, or any suitable combination of the foregoing.
In a possible implementation, the present invention can also be implemented in the form of a program product including program code for causing a terminal device to execute the method for predicting a cervical spine keypoint of embodiment 1 when the program product is run on the terminal device.
Where program code for carrying out the invention is written in any combination of one or more programming languages, the program code may execute entirely on the user device, partly on the user device, as a stand-alone software package, partly on the user device and partly on a remote device or entirely on the remote device.
While specific embodiments of the invention have been described above, it will be appreciated by those skilled in the art that this is by way of example only, and that the scope of the invention is defined by the appended claims. Various changes and modifications to these embodiments may be made by those skilled in the art without departing from the spirit and scope of the invention, and these changes and modifications are within the scope of the invention.
Claims (10)
1. A cervical spine key point prediction method is characterized by comprising the following steps:
performing image segmentation processing on the cervical vertebra medical image, and determining a plurality of segmented single cone images;
selecting a single cone image to be predicted from the single cone image;
and inputting the single conic node image to be predicted into a key point prediction model so as to determine the key point of the single conic node to be predicted.
2. The method for predicting the key points of the cervical vertebrae as set forth in claim 1, wherein the step of performing image segmentation processing on the medical image of the cervical vertebrae includes:
carrying out image segmentation processing on the cervical vertebra medical image by adopting an image segmentation model; the image segmentation model is obtained by training nnU-Net or 3D U-Net by using cervical vertebra image sample pairs marked with a single conical section segmentation result.
3. The method of claim 1, wherein the step of determining the plurality of segmented single cone images comprises:
expanding the segmented conical section area, and determining the maximum communication area after expansion as a mask area;
and determining the cone image in the mask area as a single segmented cone image.
4. The method of predicting cervical spine keypoints according to claim 1, wherein the keypoint prediction model is obtained by training a CNN-based heat map regression network using a single cone image sample.
5. The method as claimed in claim 1, wherein the keypoint prediction model comprises a local appearance model, and the step of inputting the image of the single cone to be predicted into the keypoint prediction model to determine the keypoint of the single cone to be predicted comprises:
inputting the single cone image to be predicted into a local appearance model to output a three-dimensional heat map;
determining the peak coordinates of the three-dimensional heat map as key points of the single conic node to be predicted.
6. The method as claimed in claim 1, wherein the keypoint prediction model includes a local appearance model and a spatial configuration model, and the step of inputting the image of the single cone node to be predicted into the keypoint prediction model to determine the keypoint of the single cone node to be predicted includes:
inputting the single cone image to be predicted into a local appearance model to output a three-dimensional heat map;
inputting the three-dimensional heat map into a spatial configuration model to output a spatial heat map;
multiplying the outputs of the three-dimensional heat map and the spatial heat map element by element to obtain a final heat map;
determining the peak coordinates of the final heat map as the key points of the single cone to be predicted.
7. The method of predicting cervical spine keypoints according to claim 6, wherein the local appearance model and the spatial configuration model are obtained by training an SCN with a single image sample of the cone joint to be predicted.
8. A cervical spine keypoint prediction system, comprising:
the determining module is used for carrying out image segmentation processing on the cervical vertebra medical image and determining a plurality of segmented single cone images;
the selection module is used for selecting a single conic node image to be predicted from the single conic node images;
the determining module is further used for inputting the single conic node image to be predicted into a key point prediction model so as to determine the key point of the single conic node to be predicted.
9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the cervical spine keypoint prediction method of any of claims 1-7 when executing the computer program.
10. A computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, implements the cervical spine keypoint prediction method of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211377880.5A CN115641324A (en) | 2022-11-04 | 2022-11-04 | Cervical vertebra key point prediction method and system, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211377880.5A CN115641324A (en) | 2022-11-04 | 2022-11-04 | Cervical vertebra key point prediction method and system, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115641324A true CN115641324A (en) | 2023-01-24 |
Family
ID=84948625
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211377880.5A Pending CN115641324A (en) | 2022-11-04 | 2022-11-04 | Cervical vertebra key point prediction method and system, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115641324A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116862869A (en) * | 2023-07-07 | 2023-10-10 | 东北大学 | Automatic detection method for mandible fracture based on mark point detection |
-
2022
- 2022-11-04 CN CN202211377880.5A patent/CN115641324A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116862869A (en) * | 2023-07-07 | 2023-10-10 | 东北大学 | Automatic detection method for mandible fracture based on mark point detection |
| CN116862869B (en) * | 2023-07-07 | 2024-04-19 | 东北大学 | Automatic detection method for mandible fracture based on mark point detection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dewi et al. | Synthetic Data generation using DCGAN for improved traffic sign recognition | |
| CN108764048B (en) | Face key point detection method and device | |
| CN111369681B (en) | Three-dimensional model reconstruction method, device, equipment and storage medium | |
| CN110689038A (en) | Training method and device of neural network model and medical image processing system | |
| CN110599526B (en) | Image registration method, computer device, and storage medium | |
| CN109272442B (en) | Method, device and equipment for processing panoramic spherical image and storage medium | |
| CN113326851B (en) | Image feature extraction method and device, electronic equipment and storage medium | |
| CN112102294A (en) | Training method and device for generating countermeasure network, and image registration method and device | |
| CN111192320B (en) | Position information determining method, device, equipment and storage medium | |
| CN115457492A (en) | Target detection method and device, computer equipment and storage medium | |
| CN111860484B (en) | Region labeling method, device, equipment and storage medium | |
| CN115641324A (en) | Cervical vertebra key point prediction method and system, electronic equipment and storage medium | |
| CN110517221B (en) | Gap positioning method and device based on real coordinates and storage medium | |
| CN108597589B (en) | Model generation method, target detection method and medical imaging system | |
| CN113516697A (en) | Image registration method and device, electronic equipment and computer-readable storage medium | |
| CN111915676B (en) | Image generation method, device, computer equipment and storage medium | |
| CN113610856B (en) | Method and device for training image segmentation model and image segmentation | |
| CN110956131A (en) | Single-target tracking method, device and system | |
| KR102519114B1 (en) | Apparatus and Method for Providing a Surgical Environment based on a Virtual Reality | |
| CN113096104A (en) | Training method and device of target segmentation model and target segmentation method and device | |
| CN112435284A (en) | Image registration method, electronic device and storage medium | |
| Zhao et al. | Place recognition with deep superpixel features for brain-inspired navigation | |
| Wang et al. | Liver medical image registration based on biomechanical model | |
| CN112883844B (en) | Unsupervised learning-based power transmission tower point cloud key point extraction method | |
| CN115841596B (en) | Multi-label image classification method and training method and device for model thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |