Spine detecting, positioning and marking method and system and electronic equipment

Abstract

The invention relates to the technical field of image processing, in particular to a spine detection positioning marking method and system and electronic equipment, and the method comprises the steps: obtaining a spine image, carrying out the preprocessing of the spine image to obtain a standard spine image, determining a corresponding contrast image, stretching the standard spine image according to a contrast positioning point and a feature positioning point to obtain a positioning stretching image, determining a first type of center mark points of the positioning stretching image according to the contrast image, judging the movement distance of each feature positioning point according to the coordinates of each feature positioning point before and after movement so as to determine an overall matching state, and distinguishing a spine image and a background image of the positioning stretching image in combination with the contrast image, and inputting the image into a neural network model to determine a second type of central mark points of the spine image, determining a central point, judging the accuracy of the central point, and displaying the central point with an accurate judgment result at a corresponding position on the positioning stretching image. According to the invention, interference of peripheral tissues on spine positioning in the image is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a spine detection and positioning marking method and system and electronic equipment. BACKGROUND

[0002] In computer-aided diagnosis of spinal diseases, spine marker point positioning is a very critical step, which is usually applied to tasks such as Cobb angle calculation, biomechanical load analysis, and vertebral fracture detection. However, manually identifying 17 spines and positioning 4 corner points on each spine is very time-consuming, so there is a high demand for using computers to automatically position 68 points on the spine (17 segments of spine, 4 corner points on each spine). Although the task of automatically positioning spine marker points has been studied for decades, due to the problems of blurred imaging quality of X-ray images, overlapping of soft tissues, and high similarity of textures of adjacent vertebrae, this task still has great challenges.

[0003] Chinese Patent No. CN109919903B discloses a spine detection and positioning marking method, system and electronic equipment. The method comprises: inputting a to-be-detected image into a neural network model to obtain N+1 probability maps, wherein N is the number of to-be-detected spine center points; based on N probability maps, a plurality of candidate points corresponding to each to-be-detected spine center point are obtained; and a dynamic programming algorithm is used to obtain a spine detection and positioning result from the plurality of candidate points corresponding to each to-be-detected spine center point. The present application combines a deep neural network with a dynamic programming algorithm to realize detection and positioning marking of the spine. The detection process is very fast, and the accuracy and robustness of the detection result are relatively high. Therefore, the present application has the following problems:

[0004] Peripheral tissues are not removed, which interferes with the positioning of the spine in the image, and the neural network model is relied on to position and mark the spine center point, which cannot guarantee the accuracy of the positioned spine center point. SUMMARY

[0005] Therefore, the present application provides a spine detection and positioning marking method, system and electronic equipment to overcome the problem that peripheral tissues are not removed in the image, which interferes with the positioning of the spine, and the neural network model is relied on to position and mark the spine center point, which cannot guarantee the accuracy of the positioned spine center point.

[0006] To achieve the above-mentioned purpose, in one aspect, the present application provides a spine detection and positioning marking method, system and electronic equipment, comprising:

[0007] Obtaining a spine image and pre-processing it to obtain a standard spine image, and determining a plurality of feature positioning points of the standard spine image;

[0008] determining a corresponding control image according to the standard spine image;

[0009] stretching the standard spine image according to the control positioning points of the control image and the feature positioning points of the standard spine image to obtain a positioning stretched image;

[0010] determining first type of center marker points of the positioning stretched image according to the center point coordinates of each vertebra of the control image respectively;

[0011] judging the moving distance of each feature positioning point according to the coordinates before and after the moving of each feature positioning point to determine the overall matching state;

[0012] distinguishing the vertebra image and the background image of the positioning stretched image according to the overall matching state and the control image;

[0013] inputting the vertebra image into a neural network model to determine second type of center marker points of the vertebra image;

[0014] determining the center point of the marker name according to the coordinates of the first type of center marker points and the coordinates of the second type of center marker points with the same marker name and judging the accuracy of the center point;

[0015] displaying the corresponding position of the center point on the positioning stretched image with the judging result of accuracy.

[0016] Further, the preprocessing method comprises,

[0017] performing denoising processing on the spine image;

[0018] determining the atlas, the left humeral head, the right humeral head and the sacrum of the spine image after the denoising processing;

[0019] moving the atlas, the left humeral head, the right humeral head and the sacrum to the corresponding reference points to drive the spine image after the denoising processing to obtain the standard spine image after stretching.

[0020] Further, the shooting angle of the standard spine image determines the corresponding control image with the same angle.

[0021] Further, the method of stretching the standard spine image according to the control positioning points of the control image and the feature positioning points of the standard spine image to obtain the positioning stretched image comprises,

[0022] determining a plurality of feature positioning points on the standard spine image and determining the marker name and the feature positioning point coordinates thereof;

[0023] determining the coordinates of each control positioning point with the same marker name in the control image respectively;

[0024] moving each feature positioning point from the feature positioning point coordinate to the corresponding control positioning point coordinate to stretch the standard spine image and obtain a positioning stretched image.

[0025] Further, the method for determining the first type of center marker points of the positioning stretched image according to the center point coordinates of each spine of the control image comprises,

[0026] obtaining the marker name and coordinate of each center point of the spine of the control image;

[0027] determining the marker name and coordinate of each center point of the spine as the marker name and coordinate of each first type of center marker point of the positioning stretched image.

[0028] Further, the method for determining the overall matching state according to the moving distance of each feature positioning point before and after the movement of the feature positioning point comprises,

[0029] obtaining the feature positioning point coordinate and the corresponding control positioning point coordinate of each feature positioning point;

[0030] determining the moving distance of the feature positioning point according to the feature positioning point coordinate and the corresponding control positioning point coordinate;

[0031] determining the average moving distance and the moving distance standard deviation according to the moving distance of each feature positioning point;

[0032] determining the overall matching state according to the distance difference between the average moving distance and the preset moving distance, wherein,

[0033] if the absolute value of the distance difference is greater than the moving distance standard deviation, determining that the overall matching state is poor;

[0034] if the absolute value of the distance difference is less than or equal to the moving distance standard deviation, determining that the overall matching state is good.

[0035] Further, the method for distinguishing the spine image and the background image of the positioning stretched image according to the overall matching state and the control image comprises,

[0036] if the overall matching state is good, the spine image of the positioning stretched image is the first control spine region of the control image;

[0037] if the overall matching state is poor, the spine image of the positioning stretched image is the second control spine region of the control image;

[0038] wherein, the area of the second control spine region is greater than the area of the first control spine region, and the second control spine region and the first control spine region are similar figures.

[0039] wherein the second control vertebral region has an area greater than the area of the first control vertebral region, and the second control vertebral region is a similar figure to the first control vertebral region.

[0040] Further, the method for determining the center point of the mark name and judging the accuracy of the center point according to the coordinates of the first type of center mark point and the coordinates of the second type of center mark point with the same mark name is,

[0041] determining the coordinates of the first type of center mark point and the coordinates of the second type of center mark point with the same mark name;

[0042] taking the midpoint of the line connecting the two coordinates as the center point of the mark name;

[0043] determining the accuracy of the center point according to the length of the line connecting the two coordinates, wherein,

[0044] if the length of the line connecting the two coordinates is greater than a preset length, judging that the center point is inaccurate;

[0045] if the length of the line connecting the two coordinates is less than or equal to the preset length, judging that the center point is accurate.

[0046] In a second aspect, the present application further provides a spine detection and positioning mark system for a spine detection and positioning mark method, comprising,

[0047] an image acquisition module for acquiring a spine image;

[0048] an image processing module connected to the image acquisition module, for pre-processing the spine image to obtain a standard spine image, and stretching the standard spine image according to the control positioning points of the control image and the feature positioning points of the standard spine image to obtain a positioning stretched image;

[0049] an image analysis module connected to the image processing module, for determining the atlas, left humeral head, right humeral head and sacrum of the denoised spine image, determining a plurality of feature positioning points of the standard spine image, and determining a corresponding control image according to the standard spine image;

[0050] an image marking module connected with the image processing module, configured to determine first type of center marking points of the positioning stretched image according to the coordinates of the center points of each vertebra of the contrast image, determine the moving distance of each feature positioning point according to the coordinates before and after the movement of each feature positioning point to determine the overall matching state, distinguish the vertebra image and the background image of the positioning stretched image according to the overall matching state and the contrast image, and determine the center point of the marking name according to the coordinates of the first type of center marking points and the coordinates of the second type of center marking points of the marking name and judge the accuracy of the center point;

[0051] an image output module connected with the image marking module, configured to output the positioning stretched image with the center point of the marking name.

[0052] In a third aspect, the present application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the vertebra detection and positioning marking method when executing the computer program.

[0053] Compared with the prior art, the vertebra detection and positioning marking method provided by the present application has the following advantages: the vertebra detection and positioning marking method provided by the present application determines the feature positioning points by acquiring and preprocessing the vertebra image, then acquires the contrast image according to the standard vertebra image, and stretches the positioning stretched image by means of the contrast positioning points and the feature positioning points, which can adjust the relationship between the vertebra and the surrounding tissues and reduce interference; in addition, the first type of center marking points and the second type of center marking points are determined based on the contrast image and the neural network model respectively, the positioning accuracy is improved by multi-source information fusion, and the neural network model is input after distinguishing the vertebra image and the background image according to the overall matching state, which can make the model focus on the vertebra features, further improve the accuracy of the vertebra positioning, reduce the interference of the surrounding tissues, finally determine the center point coordinates of the vertebra and display the accurate center point coordinates, and provide reliable basis for the vertebra related research, diagnosis and treatment.

[0054] Further, the noise interference in the spine image can be effectively removed through the denoising processing, so that the image is clearer, which helps to more accurately identify and locate various structural features of the spine in the subsequent process, and reduces the misjudgment or inaccurate positioning caused by noise; the machine learning model is used to quickly locate the atlas, left humeral head, right humeral head and sacrum, which are four points with obvious features in CT and easy to locate, and the center point thereof is determined as a to-be-moved reference point, which helps to establish a unified reference system in images taken in different directions, so that the subsequent image processing can be standardized and adjusted according to these key reference points, thereby improving the accuracy and consistency of the overall positioning, regardless of the changes in the shooting conditions of the original image, the stable reference points can be used as the basis for correction; moving the above four to-be-moved reference points to the respective reference points to drive the spine image to stretch to obtain a standard spine image, which can correct the deformation or positional deviation of the spine image caused by factors such as shooting angle and individual difference, so that the spine in the image is more consistent with the standard human body structure, which is convenient for accurate identification and analysis of the spine features in the subsequent process; the determined feature positioning points are all vertebral center points with unique anatomical structures and easy to identify in the CT image, and the determination of these feature positioning points can be used as an important reference for positioning other vertebrae, and the relative positional relationship between them can be used to more efficiently determine the position and sequence of the entire spine.

[0055] Further, the feature positioning points on the standard spine image and their coordinates and the coordinates of the corresponding positioning points on the control image are determined respectively, so that a corresponding relationship is established between the standard spine image and the control image, and the feature positioning points in the standard spine image are moved to the coordinates of the corresponding positioning points according to the above corresponding relationship, so as to drive the entire standard spine image to stretch to obtain a positioning stretched image, which can effectively correct the deformation, distortion or positional deviation that may exist in the standard spine image, so that the shape of the spine in the image is closer to the standard anatomical structure; by stretching based on the specific feature positioning points and the corresponding positioning points, the positions and names of each vertebra can be more accurately determined in the subsequent spine detection and positioning process.

[0056] Further, the movement distance, the average movement distance and the movement distance standard deviation of each feature positioning point are obtained, the matching degree between the standard spine image and the control image can be quantified, the matching state between the images is no longer a vague concept, but can be measured by specific numerical values, which provides an objective basis for subsequent judgment of the overall matching state and avoids errors caused by subjective judgment; the overall matching state is determined according to the distance difference between the average movement distance and the preset movement distance and the movement distance standard deviation, which can effectively distinguish between good matching and poor matching, and when the absolute value of the distance difference is greater than the movement distance standard deviation, the overall matching state is determined to be poor, otherwise the overall matching state is determined to be good, which enables the system to accurately determine the image with unsatisfactory matching, so as to take further correction measures (expand the area of the spine image) for subsequent analysis, thereby ensuring the quality of the image data used for spine detection and positioning markers;

[0057] Further, the spine image obtained by flexibly distinguishing the spine image and the background image according to the matching state and the control image is more in line with actual needs, whether it is a precise spine image when matching well or a relatively complete but targeted spine image when matching poorly, both of which provide a good foundation for further extracting detailed features of the spine, so that more accurate spine images can make feature extraction more accurate when diagnosing diseases of the spine, reducing feature misjudgment caused by background interference or incomplete spine images, thereby improving the accuracy of diagnosis; for the subsequent processing link of the neural network model, appropriate spine image input is also crucial, accurate and appropriate images can improve the learning and recognition efficiency of the model, reduce the model prediction deviation caused by poor input image quality or inaccurate range, and further improve the precision and reliability of the entire spine detection and positioning marker system;

[0058] Further, the first type of center marker point coordinates based on the control image and the second type of center marker point coordinates based on the neural network model are obtained respectively, combining two different positioning methods, the control image provides a positioning reference based on traditional anatomical knowledge and experience, and the neural network model uses the spine feature pattern learned from a large amount of data for positioning, which complement each other; then the midpoint of the line connecting the two coordinates is taken as the center point of the marker name, this method of determining the center point by combining the two positioning methods can integrate the advantages of both and reduce errors caused by a single method, thereby improving the accuracy of center point positioning;

[0059] Further, the present application judges the accuracy of the center point according to the comparison between the length of the two coordinate connecting lines and the preset length, and provides an objective and quantitative judgment standard: when the length of the connecting line is greater than the preset length, it is determined that the center point is inaccurate, and vice versa; this makes the present method can clearly identify the center point that may have positioning problems, find and exclude inaccurate positioning results in time, and avoid the influence of inaccurate center point information on subsequent analysis, diagnosis or treatment of the spine and other operations. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 Flowchart of the spine detection and positioning marking method of the embodiment of the present application;

[0061] Figure 2 Flowchart of determining the first column center marking point and the second column center marking point of the embodiment of the present application;

[0062] Figure 3 Step chart of determining the center point with the same marking name and judging the accuracy of the center point of the embodiment of the present application;

[0063] Figure 4 Connection diagram of the spine detection and positioning marking system of the embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose and advantages of the present application more clear and obvious, the present application will be further described below in combination with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0065] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application, and are not used to limit the protection scope of the present application.

[0066] It should be noted that in the description of the present application, the terms "upper", "lower", "left", "right", "inner", "outer" and the like indicate the direction or positional relationship of the terms based on the direction or positional relationship shown in the drawings, which is only for the convenience of description, and does not indicate or imply that the device or element must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation of the present application.

[0067] Moreover, it needs to be explained that in the description of the present application, unless explicitly specified and limited, the terms "mount", "connect", "connection" should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0068] Please refer to Figure 1 As shown in the flow chart of the spinal detection and positioning marking method of the embodiment of the present application. The embodiment of the present application provides a spinal detection and positioning marking method, comprising:

[0069] Step S1, obtaining a standard spinal image by preprocessing the spinal image and determining a plurality of feature positioning points of the standard spinal image; it can be understood that obtaining a standard spinal image by preprocessing the spinal image and determining a plurality of feature positioning points lays a foundation for subsequent accurate processing; these feature positioning points can be some parts with obvious anatomical features on the spine, such as the apex of the spinous process of a specific vertebra, etc. By accurately identifying these points, the approximate range and key position of the spine can be preliminarily framed;

[0070] Step S2, determining a corresponding control image according to the standard spinal image;

[0071] Step S3, stretching the standard spinal image according to the control positioning points of the control image and the feature positioning points of the standard spinal image to obtain a positioning stretched image; it can be understood that the feature positioning points of the standard spinal image are stretched to obtain a positioning stretched image by using the control positioning points of the control image, and in this process, the standard image can be corrected and optimized according to the known control image information, so that the shape and position of the spine in the image are more consistent with the standard anatomical structure, thereby improving the accuracy of positioning;

[0072] Step S4, determining a first type of center marking point and a second type of center marking point of each spinal center point of the positioning stretched image;

[0073] Step S41, determining the first type of center marking point of the positioning stretched image according to the coordinates of each spinal center point of the control image respectively;

[0074] Step S421, determining the moving distance of each feature positioning point according to the coordinates before and after the movement of each feature positioning point to determine the overall matching state;

[0075] Step S422, distinguishing the spinal image and the background image of the positioning stretched image according to the overall matching state and the control image;

[0076] Step S423, input the vertebral image into the neural network model to determine the second type of center marker point of the vertebral image; it can be understood that the marker name and coordinates of each vertebral center point of the vertebral image are determined according to the neural network model respectively, such as determining the second type of center marker point coordinates of the second thoracic vertebra of the vertebral image as (x2, y2);

[0077] It can be understood that the first type of center marker point and the second type of center marker point of each vertebral center marker name are determined based on the contrast image and the neural network model respectively, the contrast image provides a positioning reference based on traditional anatomical knowledge and experience, and the neural network model uses the learned vertebral feature mode to position; the combination of the two and the mutual verification, in some complex spinal lesions or anatomical variations, the neural network model can make accurate judgment according to its learned many cases experience, and the positioning point of the contrast image can play a role in auxiliary calibration, greatly improving the accuracy of the positioning of the vertebral center point, and then improving the accuracy of the whole vertebral positioning;

[0078] Step S5, determining the center point of the marker name according to the coordinates of the first type of center marker point and the coordinates of the second type of center marker point with the same marker name and judging the accuracy of the center point; it can be understood that the final center point is determined according to the coordinates of the two types of center marker points with the same marker name, and its accuracy is judged, and this kind of multi-source information fusion and verification method can effectively reduce the error caused by single method, and further ensure the high accuracy of vertebral positioning;

[0079] Step S6, display the corresponding position of the center point with accurate judgment result on the positioning stretched image.

[0080] It can be understood that step S3 can adjust the relative position relationship of the spine and the surrounding tissue to a certain extent by stretching the standard vertebral image to obtain the positioning stretched image, so that the structure of the spine is more prominent and clear; for example, in the original image, there are more muscle, blood vessel and other soft tissue images around the spine, which overlap or interfere with each other, the stretching process can change the spatial distribution between them, reduce the overlapping and interference phenomenon, which is beneficial to the subsequent accurate positioning and identification of the spine; when step S42 separates the vertebral image from the complex background and determines the second type of center marker point by using the neural network model, the model only needs to focus on the vertebral image part, avoiding the interference of the surrounding tissue on the model judgment; if image separation is not performed in the neural network training process, the image features of the surrounding tissue may interfere with the learning and identification of the vertebral features by the model, and after image separation, the model can more accurately capture the key features of the spine, thereby improving the positioning accuracy and reducing the interference of the surrounding tissue.

[0081] Specifically, in step S1, the preprocessing method comprises,

[0082] Step S11, denoising the spine image;

[0083] Step S12, determining the atlas, left humeral head, right humeral head and sacrum of the denoised spine image; it can be understood that the atlas (i.e. the first cervical vertebra), the left humeral head, the right humeral head (the humerus is the longest and strongest long bone in the upper limb, and its upper end forms the shoulder joint with the scapula, and its upper end includes the humeral head) and the sacrum (a triangular bone fused by five sacral vertebrae) are all points that have characteristics and can be quickly located in CT;

[0084] In implementation, the atlas, left humeral head, right humeral head and sacrum of the denoised spine image can be quickly located by a machine learning model, and the center points of the four positions are determined as the to-be-moved reference points;

[0085] Step S13, moving the four to-be-moved reference points corresponding to the atlas, the left humeral head, the right humeral head and the sacrum to the respective reference points to drive the denoised spine image to stretch and obtain a standard spine image; it can be understood that the spine images obtained in different shooting directions each correspond to a control image, and the control image is a standard image generated according to the standard human body structure. The center points of the atlas, left humeral head, right humeral head and sacrum on the control image are the reference points;

[0086] In implementation, the to-be-moved reference point of the atlas is moved to the position of the reference point (there are coordinates on the image, which can be directly moved according to the coordinates), the to-be-moved reference point of the left humeral head is moved to the position of the reference point, the to-be-moved reference point of the right humeral head is moved to the position of the reference point, and the to-be-moved reference point of the sacrum is moved to the position of the reference point. These points will drive the entire spine image to move when moving, thus forming a standard spine image.

[0087] In implementation, step S1 further comprises step S14 of determining a plurality of characteristic positioning points of the standard spine image;

[0088] In the implementation, the feature positioning points refer to the center points of the most easily identified and most characteristic several vertebrae, including: the atlas (second cervical vertebra), the seventh cervical vertebra (vertebra prominens), the first thoracic vertebra, the twelfth thoracic vertebra, the third lumbar vertebra and the fifth lumbar vertebra; wherein (1) the atlas has an upward tooth process, which is inserted into the posterior arch of the anterior arch of the atlas to form the atlantoaxial joint, and this special structure of the atlas can be easily identified on the CT image. From the transverse CT image, the ring structure of the atlas is clearly visible, the position and shape of the lateral mass are clear, and the tooth process of the atlas is a clear bony protrusion, and its joint relationship with the atlas is obvious; (2) the spinous process of the seventh cervical vertebra is the longest and most prominent, which can be easily touched at the back of the neck when bending down. On the CT image, its spinous process is particularly eye-catching in the cervical spine sequence. Whether it is a sagittal or transverse CT image, the spinous process of the seventh cervical vertebra can be used as an important anatomical landmark to help locate other cervical vertebrae or distinguish cervical vertebrae from thoracic vertebrae. From the sagittal CT image, the shape of the spinous process extending downward is very typical; (3) the first thoracic vertebra is adjacent to the cervical vertebrae, and its vertebral body shape starts to change to the typical thoracic vertebra shape, and has a rib notch related to the first rib. On the CT image, its junction relationship with the seventh cervical vertebra and other thoracic vertebrae is easy to identify. In the sagittal CT image, the vertebral body of the first thoracic vertebra is slightly larger than that of the cervical vertebra, and the transition area between its upper edge and the lower edge of the seventh cervical vertebra is an important identification point when observing the sequence of the spine. In the transverse CT image, its rib notch structure is also an important basis for identification; (4) the twelfth thoracic vertebra is the last piece of thoracic vertebra, and its shape is between thoracic vertebra and lumbar vertebra. Its inferior articular process is related to the superior articular process of the lumbar vertebra. In the CT image, it can be easily identified. In the sagittal CT image, the vertebral body shape, intervertebral disc height and transition area with the lower lumbar vertebra of the twelfth thoracic vertebra are important identification features. In the transverse CT image, the shape and direction of its articular process help to distinguish it from other thoracic vertebrae; (5) the third lumbar vertebra is usually located in the middle position of the lumbar vertebrae, and its vertebral body is thick and the transverse process is long. In the CT image, its vertebral body is easy to locate in the lumbar spine sequence. From the sagittal CT image, the vertebral body height, intervertebral disc thickness and spinous process shape of the third lumbar vertebra are helpful for identification. In the transverse CT image, the shape and length of its transverse process can be used as one of the bases for distinguishing different lumbar vertebrae; (6) the fifth lumbar vertebra is the lowermost piece of the lumbar vertebra, which is connected with the sacrum, and its vertebral body is large, and its inferior articular process forms a joint with the superior articular process of the sacrum. In the CT image, whether it is sagittal or transverse, the connection relationship between the fifth lumbar vertebra and the sacrum is a very obvious identification feature. In the sagittal CT image, the intervertebral disc between the fifth lumbar vertebra and the sacrum and their bony connection can be seen. In the transverse CT image, the joint shape of its inferior articular process and the superior articular process of the sacrum can be observed to identify it;

[0089] It can be understood that the above six vertebrae are all characteristic and relatively easy to identify in the CT image, so the trained machine learning model can be used to automatically identify and determine the above characteristic positioning points; in implementation, at least four characteristic positioning points are selected, preferably the seventh cervical vertebra (sacrum), the twelfth thoracic vertebra, the third lumbar vertebra and the fifth lumbar vertebra.

[0090] Specifically, in step S2, the shooting angle of the standard spine image (i.e. the spine image) determines the corresponding control image with the same angle.

[0091] It can be understood that the shooting angle of the standard spine image includes sagittal images, coronal images and cross-sectional images (axial images); wherein the sagittal position is a term in human anatomy, the sagittal image is a tomographic image obtained by cutting the human body in the direction from front to back (or from back to front), just like cutting the human body from the side to observe the internal structure, in medical imaging, such images can clearly show the arrangement and morphology of the body structure in the sagittal direction; the coronal position is a tomographic image obtained by cutting the human body in the left-right direction, similar to cutting the human body from the front to the back to observe, such images can show the morphology and relationship of the human body structure in the coronal direction; the cross section is a tomographic image obtained by cutting the human body in the direction from top to bottom (or from bottom to top), just like cutting the human body from the top of the head to the bottom of the feet to observe, which is the most common type.

[0092] In implementation, the control image of the spine image taken by sagittal position should also be the standard image generated according to the standard human structure in sagittal position, the control image of the spine image taken by coronal position should also be the standard image generated according to the standard human structure in coronal position, and the control image of the spine image taken by cross section should also be the standard image generated according to the standard human structure in cross section.

[0093] Specifically, in step S3, the method of stretching the standard spine image according to the control positioning points of the control image and the characteristic positioning points of the standard spine image to obtain the positioning stretched image comprises,

[0094] Step S31, determining a plurality of characteristic positioning points on the standard spine image and determining their mark names and characteristic positioning point coordinates; in one implementation, the coordinates of four characteristic positioning points are determined by machine learning, and the four characteristic positioning points are sacrum, twelfth thoracic vertebra, third lumbar vertebra and fifth lumbar vertebra, i.e. the mark names of the four characteristic positioning points are sacrum, twelfth thoracic vertebra, third lumbar vertebra and fifth lumbar vertebra;

[0095] Step S32, respectively determine the coordinates of each control positioning point with the same marker name in the control image; it can be understood that the control image has been determined in step S2, and the position coordinates and marker name of the vertebral region and all vertebrae (the center point of each vertebra) have been marked on the control image, so in an embodiment, the control positioning points are the sacrum, the twelfth thoracic vertebra, the third lumbar vertebra and the fifth lumbar vertebra of the control image, and the coordinates (control positioning point coordinates) of the control positioning points can be obtained according to the marker name of the control positioning points respectively;

[0096] Step S33, respectively determine the feature positioning point coordinates and the control positioning point coordinates of the same marker name on the standard vertebral image, and move each feature positioning point from the feature positioning point coordinates to the corresponding (with the same marker name) control positioning point coordinates, so as to drive the standard vertebral image to stretch and obtain the positioning stretched image.

[0097] Please refer to Figure 2 The flow chart for determining the first column center marker point and the second column center marker point is shown in the figure. In an embodiment, step S4 includes step S41 and step S42, and step S41 and step S42 are performed simultaneously; wherein step S41 includes step S411 and step S412, and step S42 includes step S421 to step S423;

[0098] Specifically, in step S41, the method for determining the first column center marker point of the positioning stretched image according to the center point coordinates of each vertebra of the control image includes,

[0099] Step S411, obtaining the marker name and coordinates of each center point of the vertebra of the control image;

[0100] Step S412, respectively determining the marker name and coordinates of each center point of the vertebra as the marker name and coordinates of each first column center marker point of the positioning stretched image.

[0101] In an embodiment, after the positioning stretched image is determined, the center point position of the vertebra of any control image is determined as the corresponding first column center marker point in the positioning stretched image, for example, the coordinates of the second thoracic vertebra of the control image are (x1, y1), and then the same coordinate position (x1, y1) in the positioning stretched image is marked as the first column center marker point of the second thoracic vertebra.

[0102] Specifically, in step S421, the method for determining the overall matching state according to the moving distance of each feature positioning point according to the coordinates before and after the movement of each feature positioning point includes,

[0103] Step S4211, obtaining the feature positioning point coordinates and the corresponding control positioning point coordinates of each feature positioning point;

[0104] Step S4212, determining the moving distance of each feature point according to the feature point coordinate and the corresponding control point coordinate; it can be understood that the distance between two points is determined according to the coordinates of the two points, which is a prior art and will not be described here;

[0105] Step S4213, determining the average moving distance and the moving distance standard deviation according to the moving distance of each feature point;

[0106] It can be understood that the average moving distance is the average of the moving distance of each feature point, and the moving distance standard deviation is the standard deviation of the moving distance of each feature point; the method of calculating the average and the standard deviation is also a prior art and will not be described here;

[0107] Step S4214, determining the overall matching state according to the distance difference between the average moving distance and the preset moving distance, wherein,

[0108] If the absolute value of the distance difference is greater than the moving distance standard deviation, it is determined that the overall matching state is poor;

[0109] If the absolute value of the distance difference is less than or equal to the moving distance standard deviation, it is determined that the overall matching state is good;

[0110] In implementation, the preset moving distance is determined according to specific spine image data features and application scenarios; generally, its value should be less than the maximum reasonable range of relative position change between each feature point of the spine in normal physiological state, and greater than the position fluctuation range of the feature point due to image noise, slight shooting angle difference and other factors; in high-quality spine CT image and normal physiological state, the relative position of each feature point is relatively stable, and the preset moving distance can be set within a few millimeters (such as 3mm-5mm); if the image quality is slightly poor or some special situations are considered, this value can be appropriately relaxed, but generally it should not exceed 10mm-15mm, otherwise it may lead to misjudgment, because the too large moving distance may mean that the image matching has a serious problem, such as incorrect feature recognition or image data anomaly.

[0111] Therefore, generally, the preset moving distance ∈ [5mm, 10mm], and preferably set to 7mm.

[0112] Specifically, in step S422, the spine image and the background image of the positioning stretched image are distinguished according to the overall matching state and the control image, comprising,

[0113] If the overall matching state is good, the spine image of the positioning stretched image is the first control spine region of the control image, and the remaining region is the background image;

[0114] If the overall matching state is poor, the spine image of the positioning stretched image corresponds to a second control spine region of the control image, and the remaining region is a background image;

[0115] The area of the second control spine region is greater than the area of the first control spine region, which is greater than the spine region of the control image, and the second control spine region and the first control spine region are similar figures with the spine region of the control image.

[0116] In implementation, the spine region of the control image < the first control spine region ≤ 1.3 times the spine region of the control image < the second control spine region ≤ 2 times the spine region of the control image.

[0117] In implementation, the first control spine region is preferably set to 1.3 times the spine region of the control image; the second control spine region is determined according to the ratio of the distance difference to the moving distance standard deviation, and the second control spine region = distance difference ÷ moving distance standard deviation × 1.3 times the spine region of the control image (after rounding to two decimal places according to the "rounding" principle), and when the calculated second control spine region is greater than 2 times the spine region of the control image, the second control spine region is 2 times the spine region of the control image.

[0118] It can be understood that when the overall matching state is good, the spine image of the positioning stretched image corresponds to a smaller first control spine region of the control image, because in the case of a good matching state, it indicates that the positioning stretched image and the control image have smaller differences, and the position and shape of the spine are more accurate, so a relatively smaller and accurate region can be used to define the spine image, reducing the false inclusion of background images and improving the accuracy of spine image extraction. For example, in image processing with normal shooting and no obvious abnormal deformation of the spine structure, such accurate division can effectively focus on the spine itself and exclude the interference of surrounding irrelevant tissues, making the subsequent analysis of the spine more targeted and accurate. When the overall matching state is poor, the spine image of the positioning stretched image corresponds to a larger second control spine region, because a poor matching state may mean that the image has a large deviation or deformation, and a larger region can ensure the complete inclusion of the possible deviated spine part, avoiding the omission of some parts of the spine due to the small region, and ensuring the integrity of the spine image extraction. Although some background images may be included, in such a poor matching complex situation, integrity is more critical for subsequent further correction and analysis.

[0119] Please refer to Figure 3As shown, it is the step diagram of determining the center point of the same mark name and judging the accuracy of the center point by the embodiment of the present application. Specifically, in step S5, the method of determining the center point of the mark name and judging the accuracy of the center point according to the coordinates of the first type of center mark point and the coordinates of the second type of center mark point with the same mark name is,

[0120] In step S51, the coordinates of the first type of center mark point and the coordinates of the second type of center mark point with the same mark name are determined; it can be understood that the center point of the same mark name should correspond to a first type of center mark point coordinate and a second type of center mark point coordinate, the first type of center mark point coordinate is determined directly according to the contrast image, and the second type of center mark point coordinate is obtained by inputting the determined spine region into the neural network model; for example, the first type of center mark point coordinate of the second thoracic vertebra is (x1, y1), and the second type of center mark point coordinate is (x2, y2);

[0121] In step S52, the midpoint of the line connecting the two coordinates is recorded as the center point (x, y) of the mark name, x=(x1+x2)÷2, y=(y1+y2)÷2;

[0122] In step S53, the accuracy of the center point is determined according to the length of the line connecting the two coordinates, wherein,

[0123] If the length of the line connecting the two coordinates is greater than the preset length, it indicates that the distance between the two center mark point coordinates of the same mark name is large, and cannot be considered as being in the same position, so it is judged that the center point is not accurate;

[0124] If the length of the line connecting the two coordinates is less than or equal to the preset length, it indicates that the distance between the two center mark point coordinates of the same mark name is small, and can be considered as being in the same position, so it is judged that the center point is accurate.

[0125] It can be understood that the preset length needs to be comprehensively considered various factors, including the resolution of the spine image, the requirement of the actual application scene on the accuracy, and the fluctuation range of the position of each center point under the normal physiological structure of the human spine, etc. In the case of high-resolution spine image (such as the image obtained by modern advanced CT device) and high requirement on positioning accuracy (such as used for fine spine surgery planning, etc.), the preset length can be set to a smaller value, which is generally about 1mm-3mm. Because the high-resolution image can clearly present the spine structure, the smaller preset length can more strictly ensure the accuracy of the center point positioning, and meet the high-precision application requirement. For some conventional spine examination or preliminary diagnosis scene, the image resolution is relatively not so high, the accuracy requirement is relatively not so extreme, but the center point positioning should be reasonable, and the preset length can be appropriately relaxed to avoid obvious positioning error affecting the subsequent analysis, so the value is about 3mm-5mm. Preferably, the preset length is set to 3mm.

[0126] Referring to Figure 4 The embodiment of the present application further provides a spine detection and positioning marking system for a spine detection and positioning marking method, which comprises,

[0127] An image acquisition module is configured to acquire a spine image.

[0128] An image processing module is connected with the image acquisition module, and is configured to pre-process the spine image to obtain a standard spine image, and stretch the standard spine image according to a contrast positioning point of a contrast image and the feature positioning point of the standard spine image to obtain a positioning stretched image.

[0129] An image analysis module is connected with the image processing module, and is configured to determine the atlas, left humeral head, right humeral head and sacrum of the spine image after denoising processing, determine a plurality of feature positioning points of the standard spine image, and determine a corresponding contrast image according to the standard spine image.

[0130] An image marking module is connected with the image processing module, and is configured to determine a first type of center marking point of the positioning stretched image according to the coordinates of each spine center point of the contrast image, determine the moving distance of each feature positioning point according to the coordinates of each feature positioning point before and after movement to determine the overall matching state, distinguish the spine image and the background image of the positioning stretched image according to the overall matching state and the contrast image, and determine the center point of the marking name according to the coordinates of the first type of center marking point and the coordinates of the second type of center marking point of the marking name, and judge the accuracy of the center point.

[0131] An image output module, which is connected with the image marking module, is used to output the positioning stretched image with the marked name and the accurate center point.

[0132] The embodiment of the present application further provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the spine detection and positioning marking method when executing the computer program.

[0133] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will fall within the protection scope of the present application.

[0134] The above description is only the preferred embodiments of the present application and is not used to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for locating a marker for a spinal procedure, comprising: The method comprises: acquiring a spine image and pre-processing the spine image to obtain a standard spine image, and determining a plurality of feature positioning points of the standard spine image; determining a corresponding control image according to the standard spine image; stretching the standard spine image according to control positioning points of the control image and the feature positioning points of the standard spine image to obtain a positioning stretched image; determining a first type of center marker point of the positioning stretched image according to a center point coordinate of each vertebra of the control image; judging a moving distance of each feature positioning point according to coordinates before and after the movement of each feature positioning point to determine an overall matching state; distinguishing a spine image and a background image of the positioning stretched image according to the overall matching state and the control image; inputting the spine image into a neural network model to determine a second type of center marker point of the spine image; determining a center point of a marker name according to coordinates of the first type of center marker point and coordinates of the second type of center marker point of the marker name, and judging the accuracy of the center point; displaying the corresponding position of the center point on the positioning stretched image.

2. The method of claim 1, wherein, The pre-processing method comprises: performing denoising processing on the spine image; determining the atlas, the left humeral head, the right humeral head and the sacrum of the spine image after denoising processing; moving the atlas, the left humeral head, the right humeral head and the sacrum to corresponding reference points to drive the spine image after denoising processing to stretch to obtain the standard spine image.

3. The method of claim 1, wherein, determining a corresponding control image with the same angle according to the shooting angle of the standard spine image.

4. The method of claim 1, wherein, The method for stretching the standard spine image according to control positioning points of the control image and the feature positioning points of the standard spine image to obtain a positioning stretched image comprises: determining a plurality of feature positioning points on the standard spine image and determining the marker name and feature positioning point coordinates thereof; determining the coordinates of each control positioning point with the same marker name in the control image; moving each feature positioning point from the feature positioning point coordinates to the corresponding control positioning point coordinates to drive the standard spine image to stretch and obtain the positioning stretched image.

5. The method of claim 1, wherein, The method for determining a first type of center marker point of the positioning stretched image according to a center point coordinate of each vertebra of the control image comprises: obtaining the marker name and coordinates of each center point of the vertebra of the control image; determining the marker name and coordinates of each center point of the vertebra as the marker name and coordinates of each first type of center marker point of the positioning stretched image.

6. The method of claim 4, wherein, The method for judging the moving distance of each feature positioning point according to the coordinates before and after the movement of each feature positioning point to determine an overall matching state comprises: obtaining the feature positioning point coordinates and the corresponding control positioning point coordinates of each feature positioning point; determining the moving distance of the feature positioning point according to the feature positioning point coordinates and the corresponding control positioning point coordinates; determining the average moving distance and the moving distance standard deviation according to the moving distance of each feature positioning point; determining the overall matching state according to the distance difference between the average moving distance and the preset moving distance, wherein, if the absolute value of the distance difference is greater than the standard deviation of the movement distance, determining that the overall matching state is poor; if the absolute value of the distance difference is less than or equal to the standard deviation of the movement distance, determining that the overall matching state is good.

7. The method of claim 1, wherein, distinguishing the vertebral image and the background image of the positioning stretched image according to the overall matching state and the contrast image, including, if the overall matching state is good, the vertebral image of the positioning stretched image is the first contrast vertebral region of the contrast image; if the overall matching state is poor, the vertebral image of the positioning stretched image is the second contrast vertebral region of the contrast image; wherein the area of the second contrast vertebral region is greater than the area of the first contrast vertebral region, and the second contrast vertebral region and the first contrast vertebral region are similar figures.

8. The method of claim 1, wherein, a method for determining the center point of the same mark name according to the coordinates of the first type of center mark point and the coordinates of the second type of center mark point, and judging the accuracy of the center point, including, determining the coordinates of the first type of center mark point and the coordinates of the second type of center mark point with the same mark name; taking the midpoint of the line connecting the two coordinates as the center point of the mark name; determining the accuracy of the center point according to the length of the line connecting the two coordinates, wherein, if the length of the line connecting the two coordinates is greater than the preset length, determining that the center point is inaccurate; if the length of the line connecting the two coordinates is less than or equal to the preset length, determining that the center point is accurate.

9. A spinal fixation locator system for use with the spinal fixation locator method of any one of claims 1-8, wherein, including, an image acquisition module for acquiring a vertebral image; an image processing module connected with the image acquisition module, for pre-processing the vertebral image to obtain a standard vertebral image, and stretching the standard vertebral image according to the contrast positioning points of the contrast image and the feature positioning points of the standard vertebral image to obtain a positioning stretched image; an image analysis module connected with the image processing module, for determining the atlas, left humeral head, right humeral head and sacrum of the vertebral image after denoising processing, determining a plurality of feature positioning points of the standard vertebral image, and determining a corresponding contrast image according to the standard vertebral image; an image marking module connected with the image processing module, for determining a first type of center mark point of the positioning stretched image according to the coordinates of each vertebral center point of the contrast image, judging the movement distance of each feature positioning point according to the coordinates before and after the movement of each feature positioning point to determine an overall matching state, distinguishing the vertebral image and the background image of the positioning stretched image according to the overall matching state and the contrast image, and determining the center point of the mark name according to the coordinates of the first type of center mark point and the coordinates of the second type of center mark point with the same mark name, and judging the accuracy of the center point; an image output module connected with the image marking module, for outputting the positioning stretched image with the accurate center point of the mark name.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the method for detecting and positioning the mark of the vertebral column according to any one of claims 1-8.

Images