CN112991452A - End-to-end centrum key point positioning measurement method and device based on centrum center point - Google Patents
End-to-end centrum key point positioning measurement method and device based on centrum center point Download PDFInfo
- Publication number
- CN112991452A CN112991452A CN202110348116.4A CN202110348116A CN112991452A CN 112991452 A CN112991452 A CN 112991452A CN 202110348116 A CN202110348116 A CN 202110348116A CN 112991452 A CN112991452 A CN 112991452A
- Authority
- CN
- China
- Prior art keywords
- centrum
- point
- vertebral body
- key point
- branch
- 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
- 238000000691 measurement method Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims description 39
- 238000000034 method Methods 0.000 claims description 22
- 238000005259 measurement Methods 0.000 claims description 18
- 238000005070 sampling Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 description 11
- 238000001514 detection method Methods 0.000 description 7
- 238000005457 optimization Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 210000004705 lumbosacral region Anatomy 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
- G06T2207/10124—Digitally reconstructed radiograph [DRR]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30008—Bone
- G06T2207/30012—Spine; Backbone
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Geometry (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention provides an end-to-end centrum key point positioning measurement method based on a centrum center point, which comprises the following steps of utilizing an end-to-end positioning model to obtain the centrum center point in a lumbar vertebra image and the relative position information of the centrum key point relative to the centrum center point end to end, and determining the centrum key point based on the relative position information.
Description
Technical Field
The invention relates to the field of image processing, in particular to an end-to-end centrum key point positioning measurement method and device based on centrum center points.
Background
At present, a large number of residents in China are troubled by lumbar related diseases, and with the continuous development of medical imaging technology, people can check the lumbar condition by using MRI, CT, DR, ultrasound and other technologies.
In traditional medical diagnosis field, medical staff turns over medical image picture and makes artificial subjective judgement to the lumbar vertebrae pathological change condition, and this kind of mode is consuming time and is hard and judge that the precision relies on medical staff's technical merit completely. The development of artificial intelligence has brought the possibility for the automatic positioning measurement of centrum key point, centrum key point positioning measurement model of prior art can be after the abundant training automatic positioning measurement centrum key point in the medical image, however for other positioning measurement tasks, the positioning measurement of lumbar vertebrae relates to lumbar vertebrae point key point quantity numerous, and some key point distribution is comparatively intensive, it is corresponding, there are training data's mark consuming time more, need more meticulous mark just can acquire accurate mark information scheduling problem, the mark cost of centrum key point positioning measurement model has just greatly been improved to the characteristics of these two big training data.
Moreover, the existing conventional centrum key point positioning model needs to position the centrum position firstly during training and measurement, and then the key point is positioned after the centrum position is cut. That is, the conventional vertebral body key point positioning model needs two step tasks for accomplishing the vertebral body key point positioning measurement, correspondingly, the training and predicting time is relatively long, and the model parameters are difficult to adjust.
Disclosure of Invention
The invention aims to provide an end-to-end centrum key point positioning measurement method and device based on a centrum center point, the method improves the traditional two-step measurement method, realizes the positioning measurement of the end-to-end centrum key point, and has the advantages of high efficiency in training and prediction processes, easiness in adjustment of model parameters, convenience in digging the mutual correlation among tasks and reduction of prediction time.
In order to achieve the above purposes, the technical scheme provides an end-to-end centrum key point positioning measurement method based on a centrum center point.
In some embodiments, the centrum center point coordinates of the centrum center point and the relative offset vector of the centrum key point relative to the centrum center point are obtained end-to-end, wherein the relative offset vector shows the offset vector relative to the centrum center point coordinates.
In some embodiments, the intersection of the line connecting the right point of the superior edge of the vertebral body and the left point of the inferior edge of the vertebral body and the line connecting the left point of the superior edge of the vertebral body and the right point of the inferior edge of the vertebral body is the center point of the vertebral body.
In some embodiments, the end-to-end positioning model adopts a multi-receptive-field model structure, and the feature decoding part of the multi-receptive-field model structure comprises a first branch positioned for the central point of the vertebral body and a second branch obtained for the relative position information of the key point of the vertebral body relative to the central point of the vertebral body, wherein a downsampling module and a convolution module are arranged in the first branch and the second branch, and the number of the downsampling module and the convolution module of the second branch is more than that of the downsampling module and the convolution module of the first branch.
In some embodiments, the end-to-end location model is trained by:
designing an end-to-end positioning model: designing a feature extraction part formed by stacking a plurality of convolution layers and a feature decoding part provided with different fields, wherein the feature decoding part comprises a first branch and a second branch of different numbers of down-sampling modules and convolution modules;
acquiring training data: marking the centrum central point and the centrum key point of the lumbar vertebra of the medical lumbar vertebra image, wherein the marking information of the centrum key point is as follows: the coordinate of the central point of the vertebral body + the relative offset vector, and the marking information of the central point of the vertebral body is as follows: coordinates of the center point of the vertebral body;
training an end-to-end positioning model: and training an iterative end-to-end positioning frame by using the training data to obtain an end-to-end positioning model.
In some embodiments, a mirror flip operation of the training data is randomly added to the training data when training the end-to-end positioning model.
In some embodiments, random mirror inversion operations are performed separately for prediction, the prediction results are fused, the predicted gaussian heatmaps are averaged, and then decoded into coordinate points.
An end-to-end centrum key point positioning measurement device based on a centrum center point adopts an end-to-end centrum key point positioning measurement method based on a centrum center point.
Compared with the prior art, the technical scheme has the following characteristics and beneficial effects:
1. the traditional multi-stage task of positioning the key points of the vertebral body is renovated and converted into positioning of the central point of the vertebral body +
The single-stage task of the relative position positioning of each centrum key point relative to the centrum center point utilizes a self-designed multi-sensing receiving field model structure, a plurality of branches adopted by the multi-sensing receiving field model structure provide different receiving fields, different targets are respectively predicted, the centrum center point and the centrum key point are simultaneously positioned based on the relative position of the central point of each centrum and the centrum key point, the position information of each centrum key point and the type of the centrum to which the centrum key point belongs are obtained, and the effect of performing the end-to-end positioning task on a plurality of centrums on the lumbar vertebra is achieved.
2. In the training stage, the training optimization mode of not supervising the category information of the key points can avoid the interference of the test enhancement technology on the semantic information of the image. Meanwhile, in the training stage, data enhancement strategies such as turning can be used on the premise of not damaging data semantic information, data can be predicted and positioned from multiple visual angles, and the robustness of the method can be greatly improved.
Drawings
FIG. 1 is a schematic illustration of a vertebral body center point based on an end-to-end vertebral body key point location measurement of the vertebral body center point according to an embodiment of the present invention.
Fig. 2 is a graph of the relationship between a vertebral body center point and vertebral body keypoints for an end-to-end vertebral body keypoint location measurement method based on a vertebral body center point, according to an embodiment of the invention.
Fig. 3 is a schematic structural diagram of an end-to-end vertebral body key point location measurement model based on a vertebral body center point according to an embodiment of the invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments that can be derived by one of ordinary skill in the art from the embodiments given herein are intended to be within the scope of the present invention.
The scheme provides an end-to-end centrum key point positioning measurement method and device based on a centrum center point, wherein the end-to-end centrum key point positioning measurement device based on the centrum center point refers to any electronic equipment capable of operating the end-to-end centrum key point positioning measurement method based on the centrum center point, and can be a computer terminal. The end-to-end vertebral body key point positioning measurement method based on the central point of the vertebral body can be applied to diagnosis and analysis of medical lumbar images, and information of the vertebral body key points can be efficiently acquired through the method, so that a doctor is assisted in further diagnosis and analysis.
The end-to-end vertebral body key point positioning measurement method based on the central point of the vertebral body is particularly used for processing DR/CR images, and positioning the central point of the vertebral body and the key point of the vertebral body by means of the relative position relation of the key point of the vertebral body and the central point of the vertebral body, so that the positioning measurement of the end-to-end vertebral body key point of the medical lumbar image is realized. The scheme is particularly suitable for DR/CR medical image processing, and the end-to-end single-task detection of the scheme can bring a lot of benefits compared with multi-stage task detection:
1. the training and prediction process is more efficient: the traditional multi-stage task detection needs to make training and optimization strategies aiming at independent tasks of each stage respectively, and carry out training iteration respectively; however, the single-phase task detection of the scheme only needs to make a training and optimizing strategy aiming at a single phase.
2. The model parameters are convenient to adjust: in the traditional multi-stage task detection, in the process of adjusting the performance of a parameter optimization model, the performance of a plurality of parameter schemes in a plurality of stages needs to be considered, and experimental adjustment needs to be carried out on a plurality of combinations, so that the difficulty of parameter adjustment is increased; however, the single-phase task of the present solution only needs to consider the parameter scheme of the individual task.
3. And (3) mining the correlation among tasks: according to the scheme, the task can be prevented from being separated, the correlation among various kinds of supervision information is fully utilized, optimization iteration is carried out while training, and the problem that supervision information is incomplete due to the fact that the task is split is avoided.
4. The scheme does not need to store and read the intermediate result of the multi-stage task when the prediction is used, and has obvious advantages in execution efficiency.
Specifically, the end-to-end vertebral body key point positioning measurement method based on the central point of the vertebral body comprises the following steps:
and obtaining the centrum central point in the lumbar image and the relative position information of the centrum key point relative to the centrum central point from end to end by utilizing the end-to-end positioning model, and determining the centrum key point based on the relative position information.
In the embodiment of the scheme, the centrum center point coordinate of the centrum center point and the relative offset vector of the centrum key point relative to the centrum center point are obtained end to end, wherein the relative offset vector displays the offset vector relative to the centrum center point coordinate.
Illustratively, the centrum center points are a (a, b), and the relative offset vector of a centrum key point is: a (a, b), Z ═ c + di, indicates that the vertebral body keypoints are shifted c units to the right and d units up from the vertebral body center point.
And the relative position information of the centrum center point and the centrum key point is displayed in the form of a Gaussian heat map.
In this scenario, a lumbar image of the patient is first required, and in this scenario the lumbar image is selected to be a DR/CR image. The scheme needs to perform positioning measurement operation on a plurality of indexes, the measurement of the indexes needs to be performed on a plain film (DR/CR image), and certain pre-processing operation is needed, wherein the pre-processing operation comprises reading original dcm data to obtain image pixel information in the dcm data. The pixel values are then normalized to the 0-1 range as input to the model. The lumbar image includes at least one vertebral body of the lumbar spine.
As shown in fig. 1, the center point of the vertebral body of the lumbar image is shown. Each lumbar vertebra comprises a plurality of vertebral bodies, and aiming at each vertebral body, the connection line of the right point of the upper edge and the left point of the lower edge of the vertebral body is selected in the scheme, and the intersection point of the connection line of the left point of the upper edge and the right point of the lower edge of the vertebral body is used as the central point of the vertebral body. The reason for choosing this point as the center point of the vertebral body is: the point is in a relatively middle position relative to the centrum key points, so that the offset of the centrum key points can be relatively balanced when the relative offset vector of the centrum key points relative to the centrum center point is calculated, the offset of each centrum key point in the end-to-end positioning model is relatively consistent in scale in the training process, the calculated loss is relatively close, and the situation that the offset of some centrum key points is preferentially optimized in the model optimization process due to overlarge loss provided by some centrum key points is avoided.
As shown in fig. 1 and 2, the central point of the vertebral body and the key points of the vertebral body are displayed by a gaussian heat map, and the offset vector of the key points of the vertebral body relative to the central point of the vertebral body is obtained, so as to obtain the relative position information of each key point of the vertebral body.
The vertebra comprises six vertebral bodies of T12, L1, L2, L3, L4 and L5, six key points corresponding to the right marginal point of the upper edge of the vertebral body, the left marginal point of the upper edge of the vertebral body, the right marginal point of the lower edge of the vertebral body, the left marginal point of the lower edge of the vertebral body, the inner side edge of the right side vertebral pedicle of the vertebral body and the inner side edge of the left side vertebral pedicle of the vertebral body are arranged on each vertebral body, and each key point is respectively distributed on the four vertexes of the vertebral body and.
It is worth mentioning that the scheme does not identify the centrum key points by identifying the category information of the centrum key points, but directly determines the category of the centrum key points according to the relative offset vector of the centrum key points relative to the centrum center point.
In order to realize the effect of end-to-end detection, the end-to-end positioning model of the scheme adopts a multi-receptive-field model structure, and multiple branches adopted by the multi-receptive-field model structure provide different receptive fields to respectively predict different targets. Specifically, the multi-sensing-field model structure comprises a feature extraction part and a feature decoding part, wherein the feature extraction part extracts medical image features in medical lumbar images by stacking a plurality of convolution layers, the feature decoding part performs decoding conversion on the medical image features to obtain a Gaussian heat map containing the relative position information of lumbar key points, and the Gaussian heat map is used for prediction in a training stage and decoding output in the prediction stage.
The feature decoding part is provided with a first branch positioned aiming at the central point of the cone and a second branch obtained aiming at the relative position information of the key point of the cone relative to the central point of the cone, wherein a downsampling module and a convolution module are arranged in the first branch and the second branch, and the quantity of the downsampling module and the convolution module of the second branch is more than that of the downsampling module and the convolution module of the first branch, so that the second branch can provide a larger receptive field for fitting global information; and the branch one can reserve more local image information and is suitable for distinguishing fine-grained characteristics.
Because the number of the cone labeling points is more and the distribution is denser, the cone labeling points are easy to be mixed, in order to better obtain the center point of the cone, the end-to-end positioning model is needed to process the local detail characteristics, therefore, the number of the down-sampling modules is relatively reduced by the first branch, and more image local information is reserved. However, the relative position information of the centrum key points relative to the centrum central points needs to control the global information, the centrum central points of a plurality of centrums are distinguished, two branches relatively reserve more down-sampling modules and convolution modules, and a larger receptive field is provided.
The end-to-end detection model designed by the scheme can give consideration to two different characteristics and reception field requirements, and simultaneously processes aiming at the first branch and the second branch. The difference of the cone central point positioning task and the cone key point positioning task in the receptive field and characteristic detail requirements is met. The two tasks are fitted in the end-to-end model, so that the training process of the model is simpler and is convenient to optimize. In the prediction stage, due to the adoption of an end-to-end structure, the utilization efficiency of hardware can be improved, and the time consumption is reduced.
The training mode of the end-to-end positioning model is as follows:
designing an end-to-end positioning model: designing a feature extraction part formed by stacking a plurality of convolution layers and a feature decoding part provided with different fields, wherein the feature decoding part comprises a first branch and a second branch of different numbers of down-sampling modules and convolution modules;
acquiring training data: marking the centrum central point and the centrum key point of the lumbar vertebra of the medical lumbar vertebra image, wherein the marking information of the centrum key point is as follows: the coordinate of the central point of the vertebral body + the relative offset vector, and the marking information of the central point of the vertebral body is as follows: the coordinates of the center point of the vertebral body.
Training an end-to-end positioning model: and training an iterative end-to-end positioning frame by using the training data to obtain an end-to-end positioning model.
Prediction of the end-to-end location model: and inputting the medical lumbar image into the end-to-end positioning model to obtain the coordinates of the central point of the vertebral body and the relative position information of the key point of the vertebral body, wherein the relative position information is a relative offset vector relative to the central point of the vertebral body.
In addition, when the end-to-end positioning model is trained, the class relationship of the centrum key points is not marked, and only the relative position information of the centrum key points is marked, so that the horizontal/vertical turning operation can be added randomly when the end-to-end positioning model is trained, and the class difference between the centrum key points is weakened manually. Conventionally, if the condition of labeling the category information of the cone key points is adopted, the original semantic information of the image is damaged by image turning operation, and the category information of different cone key points is influenced by predicting the turned image.
Specifically, in the scheme, a random probability is set, and the mirror image turning operation of the input training data is performed according to the probability in the training process. There are four states in total: and the probability is 1 and 4 without turning, horizontal turning, vertical turning and horizontal and vertical turning. And simultaneously carrying out mirror image overturning operation on the first branch and the second branch, wherein the first branch and the second branch share the same input layer, so that the operation taken on the input data acts on the first branch and the second branch simultaneously.
In the prediction stage, the four operations are respectively executed to respectively perform prediction. And then fusing the predicted results, directly averaging the Gaussian heat maps obtained by four times of prediction, and then decoding into coordinate points, wherein the operation is only carried out on the positioning task of each centrum key point.
In a second aspect, the present disclosure provides a device for operating the above end-to-end vertebral body key point positioning measurement method based on the vertebral body center point.
The computer system of the server for implementing the end-to-end vertebral body key point location measurement method based on the vertebral body center point according to the present embodiment includes a central processing unit CPU) which can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) or a program loaded from a storage part into a Random Access Memory (RAM). In the RAM, various programs and data necessary for system operation are also stored. The CPU, ROM, and RAM are connected to each other via a bus. An input/output (I/O) interface is also connected to the bus.
In particular, the process of the cone center point based end-to-end cone key point location measurement method described above with reference to the flow chart may be implemented as a computer software program, according to an embodiment of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication section, and/or installed from a removable medium. When executed by a Central Processing Unit (CPU), performs the functions corresponding to the above-described end-to-end vertebral body key point location measurement method based on the vertebral body center point as defined in the system of the present invention.
As another aspect, the present invention also provides a computer-readable medium that may be contained in the apparatus described in the above embodiments; or may be separate and not incorporated into the device. The computer readable medium carries one or more programs which, when executed by the apparatus, cause the apparatus to perform the process steps corresponding to the method for end-to-end vertebral body key location measurement based on a vertebral body center point.
The above-described embodiments should not be construed as limiting the scope of the invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions can occur, depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. An end-to-end centrum key point positioning measurement method based on centrum center points is characterized by comprising the following steps:
and obtaining the centrum central point in the lumbar image and the relative position information of the centrum key point relative to the centrum central point from end to end by utilizing the end-to-end positioning model, and determining the centrum key point based on the relative position information.
2. The method of claim 1, wherein the centrum center point coordinates of the centrum center point and the relative offset vector of the centrum key point with respect to the centrum center point are obtained end-to-end, wherein the relative offset vector shows the offset vector with respect to the centrum center point coordinates.
3. The method of claim 1, wherein the intersection of the line connecting the right point of the superior border and the left point of the inferior border of the vertebral body and the line connecting the left point of the superior border and the right point of the inferior border of the vertebral body is the center point of the vertebral body.
4. The method as claimed in claim 1, wherein the end-to-end positioning model adopts a multi-receptive-field model structure, the feature decoding portion of the multi-receptive-field model structure includes a first branch for positioning the central point of the vertebral body and a second branch for acquiring the relative position information of the key point of the vertebral body with respect to the central point of the vertebral body, wherein the first branch and the second branch are provided with down-sampling modules and convolution modules, and the number of the down-sampling modules and convolution modules of the second branch is greater than that of the down-sampling modules and convolution modules of the first branch.
5. The method of claim 1, wherein the training of the end-to-end positioning model comprises:
designing an end-to-end positioning model: designing a feature extraction part formed by stacking a plurality of convolution layers and a feature decoding part provided with different fields, wherein the feature decoding part comprises a first branch and a second branch of different numbers of down-sampling modules and convolution modules;
acquiring training data: marking the centrum central point and the centrum key point of the lumbar vertebra of the medical lumbar vertebra image, wherein the marking information of the centrum key point is as follows: the coordinate of the central point of the vertebral body + the relative offset vector, and the marking information of the central point of the vertebral body is as follows: coordinates of the center point of the vertebral body;
training an end-to-end positioning model: and training an iterative end-to-end positioning frame by using the training data to obtain an end-to-end positioning model.
6. The method of claim 5, wherein a mirror inversion operation of training data is added randomly to the training data when training the end-to-end positioning model.
7. An end-to-end vertebral body key point location measurement method based on a vertebral body center point as claimed in claim 5, characterized in that random mirror inversion operations are performed separately for prediction, the prediction results are fused, the predicted gaussian heat maps are averaged and then decoded into coordinate points.
8. An end-to-end centrum key point positioning measurement device based on centrum center point, characterized in that, the end-to-end centrum key point positioning measurement method based on centrum center point of any claim 1 to 7 is adopted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110348116.4A CN112991452A (en) | 2021-03-31 | 2021-03-31 | End-to-end centrum key point positioning measurement method and device based on centrum center point |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110348116.4A CN112991452A (en) | 2021-03-31 | 2021-03-31 | End-to-end centrum key point positioning measurement method and device based on centrum center point |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112991452A true CN112991452A (en) | 2021-06-18 |
Family
ID=76338619
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110348116.4A Pending CN112991452A (en) | 2021-03-31 | 2021-03-31 | End-to-end centrum key point positioning measurement method and device based on centrum center point |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112991452A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114092963A (en) * | 2021-10-14 | 2022-02-25 | 北京百度网讯科技有限公司 | Key point detection and model training method, device, equipment and storage medium |
| CN116721159A (en) * | 2023-08-04 | 2023-09-08 | 北京智源人工智能研究院 | Ultrasonic carotid artery central point coordinate prediction method and carotid artery cross section tracking method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111524188A (en) * | 2020-04-24 | 2020-08-11 | 杭州健培科技有限公司 | Lumbar positioning point acquisition method, equipment and medium |
| CN112132013A (en) * | 2020-09-22 | 2020-12-25 | 中国科学技术大学 | Vehicle key point detection method |
| CN112365523A (en) * | 2020-11-05 | 2021-02-12 | 常州工学院 | Target tracking method and device based on anchor-free twin network key point detection |
-
2021
- 2021-03-31 CN CN202110348116.4A patent/CN112991452A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111524188A (en) * | 2020-04-24 | 2020-08-11 | 杭州健培科技有限公司 | Lumbar positioning point acquisition method, equipment and medium |
| CN112132013A (en) * | 2020-09-22 | 2020-12-25 | 中国科学技术大学 | Vehicle key point detection method |
| CN112365523A (en) * | 2020-11-05 | 2021-02-12 | 常州工学院 | Target tracking method and device based on anchor-free twin network key point detection |
Non-Patent Citations (14)
| Title |
|---|
| HUSEN SHI等: "Improved Stacked Hourglass Network with Offset Learning for Robust Facial Landmark Detection", 《2019 9TH INTERNATIONAL CONFERENCE ON INFORMATION SCIENCE AND TECHNOLOGY (ICIST)》, 16 September 2019 (2019-09-16), pages 58 - 64 * |
| JINGRU YI等: "VERTEBRA-FOCUSED LANDMARK DETECTION FOR SCOLIOSIS ASSESSMENT", 《2020 IEEE 17TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING (ISBI)》 * |
| JINGRU YI等: "VERTEBRA-FOCUSED LANDMARK DETECTION FOR SCOLIOSIS ASSESSMENT", 《2020 IEEE 17TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING (ISBI)》, 22 May 2020 (2020-05-22), pages 737 * |
| JUNFENG ZHANG等: "Robust Facial Landmark Detection via Heatmap-Offset Regression", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| JUNFENG ZHANG等: "Robust Facial Landmark Detection via Heatmap-Offset Regression", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》, vol. 29, 11 March 2020 (2020-03-11), pages 5050 - 5064, XP011779361, DOI: 10.1109/TIP.2020.2976765 * |
| LUCAS STOFFL等: "End-to-End Trainable Multi-Instance Pose", 《ARXIV.ORG》 * |
| LUCAS STOFFL等: "End-to-End Trainable Multi-Instance Pose", 《ARXIV.ORG》, 22 March 2021 (2021-03-22), pages 1 - 8 * |
| RHYDIAN WINDSOR等: "A Convolutional Approach to Vertebrae Detection and Labelling in Whole Spine MRI", 《MICCAI 2020:MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION》 * |
| RHYDIAN WINDSOR等: "A Convolutional Approach to Vertebrae Detection and Labelling in Whole Spine MRI", 《MICCAI 2020:MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION》, 29 September 2020 (2020-09-29), pages 712 - 722 * |
| SHENMING FENG等: "Learning Joint Structure for Human Pose Estimation", 《ACM TRANSACTIONS ON MULTIMEDIA COMPUTING, COMMUNICATIONS, AND APPLICATIONS》 * |
| SHENMING FENG等: "Learning Joint Structure for Human Pose Estimation", 《ACM TRANSACTIONS ON MULTIMEDIA COMPUTING, COMMUNICATIONS, AND APPLICATIONS》, vol. 16, no. 3, 31 August 2020 (2020-08-31), pages 1 - 17, XP058687765, DOI: 10.1145/3392302 * |
| XINGYI ZHOU等: "Bottom-up Object Detection by Grouping Extreme and Center Points", 《2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| XINGYI ZHOU等: "Bottom-up Object Detection by Grouping Extreme and Center Points", 《2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》, 9 January 2020 (2020-01-09), pages 2 - 3 * |
| 朱秀昌等: "《数字图像处理和图像通信》", 北京邮电大学出版社, pages: 271 - 275 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114092963A (en) * | 2021-10-14 | 2022-02-25 | 北京百度网讯科技有限公司 | Key point detection and model training method, device, equipment and storage medium |
| CN114092963B (en) * | 2021-10-14 | 2023-09-22 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for key point detection and model training |
| CN116721159A (en) * | 2023-08-04 | 2023-09-08 | 北京智源人工智能研究院 | Ultrasonic carotid artery central point coordinate prediction method and carotid artery cross section tracking method |
| CN116721159B (en) * | 2023-08-04 | 2023-11-03 | 北京智源人工智能研究院 | Ultrasonic carotid artery central point coordinate prediction method and carotid artery cross section tracking method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109344833A (en) | Medical image cutting method, segmenting system and computer readable storage medium | |
| CN109712127B (en) | Power transmission line fault detection method for machine inspection video stream | |
| CN112991452A (en) | End-to-end centrum key point positioning measurement method and device based on centrum center point | |
| CN111291825A (en) | Focus classification model training method and device, computer equipment and storage medium | |
| CN111476776B (en) | Chest lesion position determination method, system, readable storage medium and apparatus | |
| CN111784762A (en) | Method and device for extracting blood vessel center line of X-ray contrast image | |
| CN111080639A (en) | Multi-scene digestive tract endoscope image identification method and system based on artificial intelligence | |
| CN112037146A (en) | Medical image artifact automatic correction method and device and computer equipment | |
| CN109285147B (en) | Image processing method and device for breast molybdenum target calcification detection and server | |
| CN116503668A (en) | Medical image classification method based on small sample element learning | |
| CN114998591A (en) | Semi-supervised learning method of human skeleton semantic segmentation model | |
| CN113450363B (en) | Meta-learning cell nucleus segmentation system and method based on label correction | |
| CN105102607A (en) | Image processing device, program, storage medium, and image processing method | |
| CN117975101A (en) | Traditional Chinese medicine disease classification method and system based on tongue picture and text information fusion | |
| CN111209946B (en) | Three-dimensional image processing method, image processing model training method and medium | |
| CN117437423A (en) | Weak supervision medical image segmentation method and device based on SAM collaborative learning and cross-layer feature aggregation enhancement | |
| Qin et al. | Residual block-based multi-label classification and localization network with integral regression for vertebrae labeling | |
| CN113076806A (en) | Structure-enhanced semi-supervised online map generation method | |
| Wang et al. | Semantic Segmentation of Foggy Scenes Based on Progressive Domain Gap Decoupling | |
| CN116563305A (en) | Segmentation method and device for abnormal region of blood vessel and electronic equipment | |
| CN112750124B (en) | Model generation method, image segmentation method, model generation device, image segmentation device, electronic equipment and storage medium | |
| CN116912872A (en) | Drawing identification method, device, equipment and readable storage medium | |
| CN115910385A (en) | Pathological degree prediction method, system, medium, equipment and terminal | |
| CN115937492A (en) | Transformer equipment infrared image identification method based on feature identification | |
| CN113792740B (en) | Artery and vein segmentation method, system, equipment and medium for fundus color illumination |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210618 |