Segmentation recognition method, device and equipment for spine image

Abstract

The application provides a spine image segmentation and recognition method, device and equipment, which can reduce the computing resources and improve the accuracy of the vertebral segmentation. The application takes part of the vertebrae of the spine as the key vertebrae, performs key vertebra segmentation and recognition on the spine image, obtains the key vertebra segmentation result and the recognition result, determines the vertebra type corresponding to each vertebra positioning point according to the position of the vertebra positioning point in the spine image, the key vertebra segmentation result and the recognition result, takes one of the vertebra positioning points as a cutting reference point, determines a cutting boundary based on the vertebra positioning points adjacent to the upper and lower sides of the cutting reference point, cuts an image sub-block including a single complete vertebra from the spine image, inputs the image sub-block and a Gaussian distribution probability map generated with the cutting reference point as the center into a vertebra segmentation model to obtain the segmentation result of the single vertebra, and obtains the recognition result of the single vertebra according to the vertebra type corresponding to the vertebra positioning point serving as the cutting reference point.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method, apparatus, computer device, storage medium, and computer program product for segmenting and recognizing spinal images. Background Technology

[0002] Segmentation and recognition of spinal images include vertebral segmentation and vertebral recognition. Vertebral segmentation mainly determines which regions of the spinal image correspond to vertebrae, thus obtaining the vertebral regions in the spinal image. Vertebral recognition mainly determines the vertebral category to which the vertebral region belongs.

[0003] Vertebral segmentation and vertebral recognition are two crucial stages in understanding spinal images and are fundamental to the diagnosis of spinal diseases (such as spinal deformities and fractures) and robotic surgical planning (such as pedicle screw fixation). Because of the complex morphology of vertebrae, manual segmentation and recognition are extremely time-consuming. Therefore, fully automated spinal segmentation and recognition is essential for saving doctors' time and effort and improving the efficiency of disease diagnosis and surgery.

[0004] The mainstream methods for automated spine segmentation and recognition are mostly based on deep neural networks, with vertebral segmentation and vertebral recognition as the tasks of the deep neural network. Since there are many types of vertebrae, the network output channels for vertebral recognition are often multiple, which consumes a lot of computing resources. Furthermore, it is difficult to make the deep neural network focus on vertebral segmentation, resulting in a reduction in segmentation accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer device, storage medium, and computer program product for segmenting and recognizing spinal images to address the aforementioned technical problems.

[0006] A method for segmenting and recognizing spine images, the method comprising:

[0007] Using some vertebrae of the spine as key vertebrae, the key vertebrae are segmented and identified in the spinal image to obtain the key vertebrae segmentation results and key vertebrae identification results.

[0008] The vertebrae included in the spinal image are located to obtain a sequence of vertebral location points;

[0009] Based on the location of the vertebral positioning points in the spinal image, the segmentation results of the key vertebrae, and the identification results of the key vertebrae, the vertebral category corresponding to each vertebral positioning point is determined.

[0010] Using one of the vertebral positioning points in the vertebral positioning point sequence as the cropping reference point, and based on the vertebral positioning points adjacent to the cropping reference point, the cropping boundary is determined, and an image sub-block including a single complete vertebra is cropped from the spinal image.

[0011] The image sub-blocks and the Gaussian probability distribution map generated with the cropping reference point as the center are input into the vertebral segmentation model to obtain the segmentation result of a single vertebra. Based on the vertebral category corresponding to the vertebral positioning point that serves as the cropping reference point, the recognition result of a single vertebra is obtained.

[0012] In one embodiment, the vertebral category corresponding to each vertebral location point is determined based on the position of the vertebral location point in the spinal image, the key vertebral segmentation result, and the key vertebral identification result, including:

[0013] Based on the location of the vertebral positioning points in the spinal image and the segmentation results of the key vertebrae, the vertebral positioning points in the key vertebrae region are determined.

[0014] Based on the key vertebral identification results, the vertebral categories corresponding to the key vertebral regions and the constraints of the spinal physiological structure are determined, thereby obtaining the vertebral categories corresponding to the vertebral positioning points in the key vertebral regions and the vertebral categories corresponding to the vertebral positioning points in the non-key vertebral regions.

[0015] In one embodiment, the vertebrae that are key vertebrae include the upper cervical vertebrae;

[0016] The vertebral category corresponding to the key vertebral region determined based on the key vertebral identification results, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the key vertebral region and the vertebral category corresponding to the vertebral location point in the non-key vertebral region, including:

[0017] If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the upper cervical spine region, then, using the vertebral positioning point corresponding to the upper cervical spine region as a reference, and according to the constraints of the spinal physiological structure, the vertebral category corresponding to the vertebral positioning point after the vertebral positioning point corresponding to the upper cervical spine region is determined.

[0018] In one embodiment, the vertebrae that are key vertebrae include the thoracic vertebrae;

[0019] The vertebral category corresponding to the key vertebral region determined based on the key vertebral identification results, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the key vertebral region and the vertebral category corresponding to the vertebral location point in the non-key vertebral region, including:

[0020] If, based on the key vertebral segmentation results and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the thoracic vertebral region, then the vertebral positioning point whose upper adjacent vertebral positioning point is not in the thoracic vertebral region and whose lower adjacent vertebral positioning point is in the thoracic vertebral region is taken as the vertebral positioning point corresponding to the first thoracic vertebra.

[0021] Using the vertebral positioning point corresponding to the first thoracic vertebra as a reference, and based on the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the first thoracic vertebra is determined.

[0022] In one embodiment, the vertebrae that are key vertebrae include the thoracic vertebrae;

[0023] The vertebral category corresponding to the key vertebral region determined based on the key vertebral identification results, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the key vertebral region and the vertebral category corresponding to the vertebral location point in the non-key vertebral region, including:

[0024] If, based on the key vertebrae segmentation results and key vertebrae identification results, it is determined that the key vertebrae region of the spinal image includes the thoracic vertebrae region, then the lumbar vertebrae region is obtained based on the region below the thoracic vertebrae region.

[0025] The vertebral positioning point that is adjacent to both the vertebral positioning point in the thoracic region and the vertebral positioning point in the lumbar region is taken as the vertebral positioning point corresponding to the first lumbar vertebra.

[0026] Based on the vertebral positioning point corresponding to the first lumbar vertebra, and according to the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the first thoracic vertebra is determined.

[0027] In one embodiment, the vertebrae that are key vertebrae include the sacrum;

[0028] The vertebral category corresponding to the key vertebral region determined based on the key vertebral identification results, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the key vertebral region and the vertebral category corresponding to the vertebral location point in the non-key vertebral region, including:

[0029] If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the sacral region, then, using the vertebral positioning point corresponding to the sacral region as a reference, and according to the constraints of the spinal physiological structure, the vertebral category corresponding to the vertebral positioning point above the vertebral positioning point in the sacral region is determined.

[0030] In one embodiment, before determining the vertebral category corresponding to each vertebral location point based on the position of the vertebral location point in the spinal image, the key vertebral segmentation result, and the key vertebral identification result, the method further includes:

[0031] Based on the segmentation results of the key vertebrae, the non-spinal regions in the spinal image are determined;

[0032] Based on the position of the vertebral positioning points in the spinal image, vertebral positioning points located in the non-spinal region are removed from the vertebral positioning point sequence.

[0033] In one embodiment, where the sacrum is included among the key vertebrae, the method further includes, before determining the vertebrae category corresponding to each vertebrae location point based on the position of the vertebrae location points in the spinal image, the key vertebrae segmentation result, and the key vertebrae identification result:

[0034] If, based on the key vertebrae segmentation results and the key vertebrae identification results, it is determined that the spinal image includes the sacral region;

[0035] If, based on the position of the vertebral positioning point in the spinal image, it is determined that there is no vertebral positioning point located in the sacral region in the vertebral positioning point sequence, then the centroid of the sacral region is taken as the vertebral positioning point corresponding to the sacrum, and the vertebral positioning point corresponding to the sacrum is added to the vertebral positioning point sequence.

[0036] A segmentation and recognition device for spine images, the device comprising:

[0037] The key vertebra segmentation and recognition module is used to segment and recognize key vertebrae in spinal images, taking some vertebrae of the spine as key vertebrae, and to obtain key vertebra segmentation results and key vertebra recognition results.

[0038] The vertebral localization module is used to locate the vertebrae included in the spinal image and obtain a sequence of vertebral localization points.

[0039] The positioning point marking module is used to determine the vertebral category corresponding to each vertebral positioning point based on the position of the vertebral positioning point in the spinal image, the segmentation result of the key vertebra, and the identification result of the key vertebra;

[0040] The image cropping module is used to take one of the vertebral positioning points in the vertebral positioning point sequence as the cropping reference point, determine the cropping boundary based on the vertebral positioning points adjacent to the cropping reference point, and crop out an image sub-block including a single complete vertebra from the spinal image.

[0041] The single vertebra segmentation and recognition module is used to input the image sub-block and the Gaussian distribution probability map generated with the cropping reference point as the center into the vertebra segmentation model to obtain the segmentation result of a single vertebra, and to obtain the recognition result of a single vertebra according to the vertebra category corresponding to the vertebra positioning point used as the cropping reference point.

[0042] A computer device includes a memory and a processor, the memory storing a computer program and the processor performing the above-described method.

[0043] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor using the methods described above.

[0044] A computer program product having a computer program stored thereon, the computer program being executed by a processor using the above-described method.

[0045] The aforementioned methods, devices, computer equipment, storage media, and computer programs for segmenting and recognizing spinal images transform the complex problem of spinal segmentation and recognition into several simpler tasks. First, vertebral recognition is transformed into the segmentation of key vertebrae. Based on the segmentation results of key vertebrae, the recognition results of key vertebrae, and the location of vertebral positioning points in the spinal image, vertebral recognition is performed. This eliminates the need for deep neural networks to perform full-category vertebral recognition, greatly reducing the difficulty of vertebral recognition and computational resources. Second, spinal segmentation is transformed into the segmentation of individual vertebrae. Based on the vertebral positioning points and the positioning points of adjacent vertebrae, the spinal image is cropped to obtain image sub-blocks including individual complete vertebrae. This allows the vertebral segmentation model to focus on segmenting individual vertebrae within the image sub-blocks under the guidance of the corresponding Gaussian probability distribution map, improving the accuracy of vertebral segmentation. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating a method for segmenting and recognizing a spine image in one embodiment;

[0047] Figure 2 This is a flowchart illustrating a method for segmenting and recognizing spine images in another embodiment;

[0048] Figure 3 This is a schematic diagram of a spine image obtained based on other scanning fields in one embodiment;

[0049] Figure 4 This is a schematic diagram of a spine image obtained based on other scanning fields in another embodiment;

[0050] Figure 5 This is a flowchart illustrating the segmentation and recognition method for a spine image in yet another embodiment;

[0051] Figure 6 This is a structural block diagram of a spine image segmentation and recognition device in one embodiment;

[0052] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0055] Segmentation and recognition of spinal images are crucial steps in the automated analysis of spinal morphology and pathology, and are also important technologies in the automated planning of spinal surgical robots. This application provides a method for segmenting and recognizing spinal images, which can be automated by computer equipment to segment and recognize spinal images, providing an effective technical foundation for the diagnosis of spinal diseases and spinal surgical robots, thereby greatly saving doctors' time and improving surgical efficiency.

[0056] Combination Figure 1 and Figure 2 This application describes the steps involved in the segmentation and recognition method for spine images.

[0057] Step S201: Using some vertebrae of the spine as key vertebrae, segment and identify the key vertebrae in the spinal image to obtain the key vertebrae segmentation results and key vertebrae identification results.

[0058] Segmentation and recognition of key vertebrae is a spinal segmentation and recognition task. The segmentation and recognition results of key vertebrae obtained by this task can be used to: (1) remove non-spine regions from spinal images; (2) identify whether a spinal image includes key vertebrae, and if the spinal image includes key vertebrae, identify the vertebrae category and the region corresponding to the key vertebrae; (3) determine the vertebrae category corresponding to the vertebrae location point.

[0059] The segmentation and identification of key vertebrae can be accomplished using a multi-class segmentation model. The multi-class segmentation algorithm can be, but is not limited to, a deep convolutional neural network model based on 3DUNet, or other artificial intelligence algorithms such as RNN, Transformer, or graph cut algorithms.

[0060] When training a multi-class segmentation model, the training samples input to the model are spine images. When the key vertebral regions are specifically represented as masks, the output of the multi-class segmentation model is a segmentation mask with key vertebral category labels. To improve the running speed of the multi-class segmentation model, some embodiments can resample the original spine image at intervals of [x, y, z] (x, y, z can be, but are not limited to, 4mm, 4mm, 4mm, such as 5mm, 5mm, 5mm, etc.), reducing the size of the spine image. Larger sampling intervals result in smaller image sizes, making the multi-class segmentation model run faster, but also reducing accuracy.

[0061] The key vertebrae can be arbitrarily selected from all the vertebrae included in the spine. To reduce the difficulty of vertebral identification, vertebrae with high morphological distinctiveness can be selected from the vertebrae included in the spine, such as at least one of the first cervical vertebra, second cervical vertebra, thoracic vertebrae, or sacrum. Considering that the scanning field of view may vary in different scanning scenarios, in order to adapt the method provided in this application to various scanning fields of view and reduce the limitation of the scanning field of view on the method provided in this application, the first cervical vertebra, second cervical vertebra, thoracic vertebrae, and sacrum can be used as key vertebrae. Thus, by segmenting and identifying the key vertebrae in the spinal image obtained from any scanning field of view, it can be determined whether the spinal image includes the cervical region, thoracic region, lumbar region, or sacral region.

[0062] The segmentation results of key vertebrae are mainly used to describe which regions of the spinal image correspond to key vertebrae. The regions corresponding to key vertebrae can be called key vertebra regions. Based on the key vertebra regions, non-key vertebra regions in the spinal image can be identified.

[0063] The identification results of key vertebrae are used to describe the key vertebrae category to which the key vertebrae region belongs. For example, regarding key vertebrae categories: if the first cervical vertebra, the second cervical vertebra, the thoracic vertebrae, and the sacrum are selected as key vertebrae, then the key vertebrae categories include the first cervical vertebra, the second cervical vertebra, the thoracic vertebrae, and the sacrum.

[0064] Using the first cervical vertebra as a key vertebra: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] Figure 1 After performing key vertebrae segmentation and identification on the spinal image shown, key vertebrae segmentation results and key vertebrae identification results are obtained. The key vertebrae segmentation results represent region 10 of the spinal image as a key vertebrae region, and the key vertebrae identification results represent the key vertebrae category to which the key vertebrae region belongs.

[0065] Step S202: Locate the vertebrae included in the spinal image to obtain a sequence of vertebral location points.

[0066] Vertebral localization points are used to characterize vertebrae in spinal images. The specific form of a vertebral localization point can be at least one or more of the following: vertebral center point, vertebral centroid, etc. Vertebral localization points can be obtained through a vertebral localization model, which only needs to output the vertebral localization points; therefore, the output channel number of the vertebral localization model is 1. This vertebral localization model can employ, but is not limited to, a deep convolutional neural network model based on an improved 3D UNet, such as SpatialConfiguration-Net.

[0067] When the specific form of the vertebral positioning point is the vertebral center point, if the vertebral center point is detected through the vertebral positioning model, the training process of the vertebral positioning model includes: acquiring the spine image as input, positioning and marking the vertebrae in the spine image, taking the points obtained from the positioning and marking as the vertebral center points of the corresponding vertebrae, and generating a Gaussian distribution probability map with the vertebral center points as the standard for subsequent loss calculation (the Gaussian distribution probability map generated with the vertebral center points is called the standard map); then, inputting the spine image into the vertebral positioning model so that the vertebral positioning model outputs the corresponding Gaussian distribution probability map (called the prediction map); and adjusting the parameters of the vertebral positioning model based on the loss values ​​determined by the standard map and the prediction map until training is completed.

[0068] When the input spinal image is a CT spinal image, the spinal image can be corrected to the LPS position (i.e., in a three-dimensional image coordinate system, the left hand direction of the human body is the positive X-axis, the back direction of the human body is the positive Y-axis, and the head direction of the human body is the positive Z-axis). The vertebral positioning points obtained from the Gaussian distribution probability map output by the vertebral positioning model are arranged in descending order of Z-axis coordinate values, i.e., from head to toe.

[0069] In addition, before vertebral localization, the non-vertebral regions in the spinal image can be removed by expanding the outer rectangle of the mask of the key vertebra obtained in step S201 by a preset distance (the preset distance can be, but is not limited to, 30 mm; the purpose of such expansion is to improve the robustness of the algorithm), thereby obtaining a spinal region image, and vertebral localization can be performed on the spinal region image.

[0070] In some embodiments, after obtaining the vertebral positioning point sequence, before identifying the vertebral category, vertebral positioning points located in non-spinal regions can be removed. Specific steps include: determining non-spinal regions in the spinal image based on the key vertebral segmentation results; and removing vertebral positioning points located in the non-spinal regions from the vertebral positioning point sequence based on their positions in the spinal image.

[0071] Based on the segmentation results of key vertebrae, the spinal region in the spinal image can be determined. The specific form of the spinal region can be a mask. Then, the area outside the spinal region in the spinal image is regarded as the non-spine region. Based on the position of the vertebral positioning point in the spinal image, the vertebral positioning points located in the non-spine region in the vertebral positioning point sequence are removed.

[0072] By using the above method, vertebral category identification can be avoided for vertebral location points in non-spinal regions, thus improving identification efficiency.

[0073] In some embodiments, after obtaining the vertebral positioning point sequence, before identifying the vertebral category of the vertebral positioning points, vertebral positioning points that have not been detected can be added. For example, if it is determined that there is a sacral region in the spinal image based on the key vertebral segmentation results and key vertebral recognition results, but there are no vertebral positioning points in the sacral region in the vertebral positioning point sequence, then vertebral positioning points corresponding to the sacrum can be added.

[0074] The specific steps include: if the key vertebral segmentation result and the key vertebral recognition result determine that the spinal image includes the sacral region; if the position of the vertebral positioning point in the spinal image determines that there is no vertebral positioning point located in the sacral region in the vertebral positioning point sequence, then the centroid of the sacral region is taken as the vertebral positioning point corresponding to the sacrum, and the vertebral positioning point corresponding to the sacrum is added to the vertebral positioning point sequence.

[0075] Step S203: Based on the position of the vertebral positioning point in the spinal image, the key vertebral segmentation result, and the key vertebral identification result, determine the vertebral category corresponding to each vertebral positioning point.

[0076] In one embodiment, step S203 may specifically include the following steps: determining the vertebral positioning points in the critical vertebral region based on the position of the vertebral positioning points in the spinal image and the critical vertebral segmentation results; obtaining the vertebral category corresponding to the vertebral positioning points in the critical vertebral region and the vertebral category corresponding to the vertebral positioning points in the non-critical vertebral region based on the vertebral category category corresponding to the critical vertebral region and the spinal physiological structure constraints determined by the critical vertebral identification results.

[0077] For example, in the case of Figure 1After vertebral localization of the spinal image shown, multiple vertebral localization points are obtained. These points are then sorted from head to toe, resulting in a sequence of vertebral localization points 101 to 110. Matching the positions of these vertebral localization points with the positions of key vertebral regions in the spinal image determines that vertebral localization point 101 is located in key vertebral region 10, while other vertebral localization points are located in non-key vertebral regions. Therefore, the vertebral category corresponding to vertebral localization point 101 is the key vertebral category to which key vertebral region 10 belongs.

[0078] If the critical vertebral region 10 belongs to the first cervical vertebra, then the vertebral category corresponding to vertebral positioning point 101 is the first cervical vertebra. Then, according to the constraints of the physiological structure of the spine, the vertebral positioning points 102 to 110 after vertebral positioning point 101 can be determined to correspond to the second, third, fourth, fifth, sixth, seventh, first, second, and third cervical vertebrae, respectively.

[0079] By using the above method, the task of recognizing spinal images is transformed into the task of segmenting and recognizing a few key vertebrae. The deep neural network only needs to segment and recognize a few key vertebrae, without having to recognize all types of vertebrae, which greatly reduces the difficulty of vertebrae recognition and reduces computing resources.

[0080] When some vertebrae considered as critical vertebrae include the upper cervical vertebrae, based on the critical vertebrae identification results, the vertebrae category corresponding to the critical vertebrae region and the constraints of the spinal physiological structure are used to obtain the vertebrae category corresponding to the vertebrae location points in the critical vertebrae region and the vertebrae category corresponding to the vertebrae location points in non-critical vertebrae regions. Specifically, this may include:

[0081] If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the upper cervical spine region, then, using the vertebral positioning point corresponding to the upper cervical spine region as a reference, and according to the constraints of the spinal physiological structure, the vertebral category corresponding to the vertebral positioning point after the vertebral positioning point corresponding to the upper cervical spine region is determined.

[0082] The upper cervical vertebrae can be the first cervical vertebra and / or the second cervical vertebrae. If there are vertebral positioning points corresponding to the upper cervical vertebrae in the vertebral positioning point sequence, then the vertebral category corresponding to the vertebral positioning point after the upper cervical vertebrae can be determined sequentially as the third cervical vertebra, the fourth cervical vertebra, the fifth cervical vertebra, etc., based on the constraints of the physiological structure of the spine.

[0083] When the vertebrae considered as critical vertebrae include the thoracic vertebrae, based on the critical vertebrae identification results, the vertebrae category corresponding to the critical vertebrae region, and the constraints of the spinal physiological structure, are used to obtain the vertebrae category corresponding to the vertebrae location points in the critical vertebrae region and the vertebrae category corresponding to the vertebrae location points in non-critical vertebrae regions. Specifically, this may include:

[0084] If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the thoracic vertebral region, then the vertebral positioning point whose upper adjacent vertebral positioning point is not located in the thoracic vertebral region and whose lower adjacent vertebral positioning point is located in the thoracic vertebral region is taken as the vertebral positioning point corresponding to the first thoracic vertebra; based on the vertebral positioning point corresponding to the first thoracic vertebra, and according to the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the vertebral positioning point corresponding to the first thoracic vertebra is determined.

[0085] When the vertebrae considered as critical vertebrae include the thoracic vertebrae, based on the critical vertebrae identification results, the vertebrae category corresponding to the critical vertebrae region, and the constraints of the spinal physiological structure, are used to obtain the vertebrae category corresponding to the vertebrae location points in the critical vertebrae region and the vertebrae category corresponding to the vertebrae location points in non-critical vertebrae regions. Specifically, this may include:

[0086] If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the thoracic vertebral region, then the lumbar vertebral region is obtained based on the region below the thoracic vertebral region. The vertebral positioning points that are adjacent to both the vertebral positioning points in the thoracic vertebral region and the vertebral positioning points in the lumbar vertebral region are taken as the vertebral positioning points corresponding to the first lumbar vertebra. Based on the vertebral positioning points corresponding to the first lumbar vertebra, and according to the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning points after the vertebral positioning points corresponding to the first thoracic vertebra is determined.

[0087] In cases where the vertebrae considered as critical vertebrae include the sacrum, based on the critical vertebrae identification results, the vertebrae category corresponding to the critical vertebrae region, and the constraints of the spinal physiological structure, are used to obtain the vertebrae category corresponding to the vertebrae location points in the critical vertebrae region, and the vertebrae category corresponding to the vertebrae location points in non-critical vertebrae regions, including:

[0088] If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the sacral region, then, using the vertebral positioning point corresponding to the sacral region as a reference, and according to the constraints of the spinal physiological structure, the vertebral category corresponding to the vertebral positioning point above the vertebral positioning point in the sacral region is determined.

[0089] In the following description, C1 to C7 represent the first to seventh cervical vertebrae, respectively; T1 to T12 represent the first to twelfth thoracic vertebrae, respectively. If there are variations in the thoracic vertebrae, such as the thirteenth or more thoracic vertebrae, they will be designated as T13, T14, etc.; L1 to L5 represent the first to fifth lumbar vertebrae. If there are variations in the lumbar vertebrae, such as the sixth or more lumbar vertebrae, they will be designated as L6, L7, etc.; S represents the sacrum.

[0090] When the key vertebrae include the first cervical vertebra, the second cervical vertebra, the thoracic vertebrae, and the sacrum, if the vertebral positioning point sequence in the spinal image is determined to include C1 or C2 based on the location of the vertebral positioning points in the spinal image, the key vertebra segmentation results, and the key vertebra identification results, then it can be further divided into the following four cases:

[0091] (1) If the vertebral positioning point sequence only includes vertebral positioning points corresponding to C1 or C2, and does not include vertebral positioning points corresponding to T1, L1 or S, then the vertebral positioning points located after C1 or C2 are determined according to the constraints of the physiological structure of the spine, in order from head to toe, to determine the vertebral category corresponding to each vertebral positioning point.

[0092] (2) If the vertebral positioning point sequence includes not only the vertebral positioning points corresponding to C1 or C2, but also the vertebral positioning points corresponding to T1, then the vertebral positioning points between C1 or C2 and T1 (excluding T1) shall be determined according to the vertebral positioning points corresponding to C1 or C2, in order from head to toe, based on the constraints of the physiological structure of the spine; the vertebral positioning points located after T1 shall be determined according to the vertebral positioning points corresponding to T1, in order from head to toe, based on the constraints of the physiological structure of the spine.

[0093] (3) If the vertebral positioning point sequence includes not only vertebral positioning points corresponding to C1 or C2, but also vertebral positioning points corresponding to T1 and L1, then the vertebral positioning points between C1 or C2 and T1 (excluding T1) are determined according to the vertebral positioning points corresponding to C1 or C2, based on the constraints of the spinal physiological structure, in order from head to toe, to determine the corresponding vertebral category; the vertebral positioning points between T1 and L1 (excluding L1) are determined according to the vertebral positioning points corresponding to T1, based on the constraints of the spinal physiological structure, in order from head to toe, to determine the corresponding vertebral category; the vertebral positioning points located after L1 are determined according to the vertebral positioning points corresponding to L1, based on the constraints of the spinal physiological structure, in order from head to toe, to determine the corresponding vertebral category.

[0094] (4) If the vertebral positioning point sequence includes not only vertebral positioning points corresponding to C1 or C2, but also vertebral positioning points corresponding to T1, L1, and S, then the vertebral positioning points between C1 or C2 and T1 (excluding T1) are determined according to the vertebral positioning points corresponding to C1 or C2, based on the constraints of the spinal physiological structure, in a head-to-toe order; the vertebral positioning points between T1 and L1 (excluding L1) are determined according to the vertebral positioning points corresponding to T1, based on the constraints of the spinal physiological structure, in a head-to-toe order; and the vertebral positioning points between L1 and S (excluding S) are determined according to the vertebral positioning points corresponding to L1, based on the constraints of the spinal physiological structure, in a head-to-toe order.

[0095] When the key vertebrae include the first cervical vertebra, the second cervical vertebra, the thoracic vertebrae, and the sacrum, if the vertebral positioning point sequence in the spinal image does not include C1 and C2, based on the location of the vertebral positioning points in the spinal image, the key vertebra segmentation results, and the key vertebra identification results, then it can be further divided into the following three cases:

[0096] (1) If the vertebral positioning point sequence only includes the vertebral positioning points corresponding to T1, then the vertebral positioning points before T1 are determined according to the vertebral positioning points corresponding to T1 in the order from foot to head; the vertebral positioning points after T1 are determined according to the vertebral positioning points corresponding to T1 in the order from head to foot based on the physiological structure constraints of the spine.

[0097] (2) If the vertebral positioning point sequence includes vertebral positioning points corresponding to T1 and L1, then the vertebral positioning points before T1 are determined according to the vertebral positioning points corresponding to T1, in order from foot to head; the vertebral positioning points between T1 and L1 (excluding L1) are determined according to the vertebral positioning points corresponding to T1, in order from head to foot according to the constraints of the spinal physiological structure; the vertebral positioning points after L1 are determined according to the vertebral positioning points corresponding to L1, in order from head to foot according to the constraints of the spinal physiological structure.

[0098] (3) If the vertebral positioning point sequence includes vertebral positioning points corresponding to T1, L1 and S, then the vertebral positioning points before T1 are determined according to the vertebral positioning points corresponding to T1, in order from foot to head; the vertebral positioning points between T1 and L1 (excluding L1) are determined according to the vertebral positioning points corresponding to T1, in order from head to foot according to the constraints of the spinal physiological structure; the vertebral positioning points between L1 and S (excluding S) are determined according to the vertebral positioning points corresponding to L1, in order from head to foot according to the constraints of the spinal physiological structure.

[0099] When the key vertebrae include the first cervical vertebra, the second cervical vertebra, the thoracic vertebrae, and the sacrum, if the vertebral positioning point sequence in the spinal image does not include C1, C2, and T1, based on the location of the vertebral positioning points in the spinal image, the segmentation results of the key vertebrae, and the identification results of the key vertebrae, it can be further divided into the following two cases:

[0100] (1) If the vertebral positioning point sequence only includes the vertebral positioning points corresponding to L1, then the vertebral positioning points before L1 are determined according to the vertebral positioning points corresponding to L1 in the order from foot to head; the vertebral positioning points after L1 are determined according to the vertebral positioning points corresponding to L1 in the order from head to foot based on the physiological structure constraints of the spine.

[0101] (2) If the vertebral positioning point sequence includes vertebral positioning points corresponding to L1 and S, then the vertebral positioning points before L1 are determined according to the vertebral positioning points corresponding to L1, in order from foot to head, to determine the corresponding vertebral category; the vertebral positioning points between L1 and S (excluding S) are determined according to the vertebral positioning points corresponding to L1, in order from head to foot, based on the constraints of the physiological structure of the spine.

[0102] In cases where key vertebrae include the first cervical vertebra, second cervical vertebra, thoracic vertebrae, and sacrum, if, based on the location of vertebral positioning points in the spinal image, the key vertebra segmentation results, and the key vertebra identification results, it is determined that the sequence of vertebral positioning points in the spinal image does not include vertebral positioning points corresponding to C1, C2, T1, and L1, but includes vertebral positioning points corresponding to S, then the vertebral positioning points preceding S are used as a reference, and the corresponding vertebral category is determined in order from foot to head.

[0103] When key vertebrae include the first cervical vertebra, second cervical vertebra, thoracic vertebrae, and sacrum, if the vertebral positioning point sequence in the spinal image does not include vertebral positioning points corresponding to C1, C2, T1, L1, and S, based on the location of the vertebral positioning points in the spinal image, the key vertebral segmentation results, and the key vertebral identification results, then the vertebral category corresponding to each positioning point is determined sequentially from head to toe, according to the three major categories of cervical, thoracic, and lumbar vertebrae. In this case, the vertebral category corresponding to each positioning point is not determined based on at least one of the vertebral positioning points corresponding to C1, C2, T1, L1, and S. Therefore, a corresponding hint can be given, indicating that the vertebral category corresponding to the positioning point has high uncertainty and may be incorrect.

[0104] Step S204: Using one of the vertebral positioning points in the vertebral positioning point sequence as the cropping reference point, and based on the vertebral positioning points adjacent to the cropping reference point, the cropping boundary is determined, and an image sub-block including a single complete vertebra is cropped from the spinal image.

[0105] Multiple image sub-blocks can be cropped from the spinal image, each image sub-block including a single complete vertebra, and the cropping method of each image sub-block is as described in step S204.

[0106] like Figure 1 As shown, if vertebral positioning point 104 is taken as the cropping reference point, the vertebral positioning point adjacent to and above vertebral positioning point 104 is vertebral positioning point 103, and the vertebral positioning point adjacent to and below vertebral positioning point 104 is vertebral positioning point 105. Therefore, the cropping boundary is determined based on vertebral positioning points 103 and 105, ensuring that the cropped image sub-block includes a single complete vertebra; furthermore, as... Figure 1 As shown, the image sub-block may include an incomplete vertebra, with a complete vertebra located in the middle of the incomplete vertebra.

[0107] Step S205: Input the image sub-block and the Gaussian distribution probability map generated with the cropping reference point as the center into the vertebral segmentation model to obtain the segmentation result of a single vertebra, and obtain the recognition result of a single vertebra according to the vertebral category corresponding to the vertebral positioning point used as the cropping reference point.

[0108] Since each image sub-block is cropped based on a corresponding cropping reference point, each sub-block has a corresponding Gaussian probability map. This Gaussian probability map is generated centered on the cropping reference point, and its size is the same as the size of the image sub-block. The pixel value corresponding to the cropping reference point is the largest, which can be set to 1, indicating the highest probability. The pixel value gradually decreases as the distance from the cropping reference point increases, and the decay follows a Gaussian distribution. The purpose of the Gaussian probability map is to guide the vertebral segmentation model to segment only the complete single vertebra corresponding to the cropping reference point.

[0109] After inputting the image sub-blocks and their corresponding Gaussian probability distribution maps into the vertebral segmentation model, the segmentation result of a single complete vertebra can be obtained. Furthermore, the vertebral category corresponding to the vertebral positioning point within the image sub-block, which serves as the cropping reference point, can be used as the identification result of a single complete vertebra.

[0110] The aforementioned spinal image segmentation and recognition method transforms the complex problem of spinal segmentation and recognition into several simpler tasks. First, vertebral recognition is transformed into the segmentation of key vertebrae. Based on the segmentation results of key vertebrae, the recognition results of key vertebrae, and the position of vertebral positioning points in the spinal image, vertebral recognition is performed. This eliminates the need for deep neural networks to perform full-category vertebral recognition, greatly reducing the difficulty of vertebral recognition and computational resources. Second, spinal segmentation is transformed into the segmentation of individual vertebrae. Based on the vertebral positioning points and the positioning points of adjacent vertebrae, the spinal image is cropped to obtain image sub-blocks including individual complete vertebrae. This allows the vertebral segmentation model to focus on segmenting individual vertebrae within the image sub-blocks under the guidance of the corresponding Gaussian probability distribution map, improving the accuracy of vertebral segmentation.

[0111] It should be noted that, Figure 1 The spinal images in the image can be based on images obtained from other scanning fields, such as... Figure 3 and Figure 4 .

[0112] Regarding steps S204 and S205, the following describes a more detailed implementation:

[0113] Segmenting a single vertebra using a vertebral segmentation model is considered fine vertebral segmentation. The vertebral segmentation model only needs to distinguish between non-vertebral regions (i.e., background) and vertebral regions, and only segments the complete vertebra in the middle.

[0114] After obtaining the vertebral positioning point sequence of the spinal image, the spinal image is cropped based on the vertebral positioning point sequence to obtain multiple image sub-blocks, each of which contains only one complete vertebra. To obtain an image sub-block containing a single complete vertebra, the specific cropping method is as follows: Using the center point of any vertebra as the cropping reference point, crop along the Z-axis both vertically and horizontally. The upper boundary of the Z-axis cropping is the Z-axis coordinate value of the adjacent vertebral positioning points at the reference point, and the lower boundary is the Z-axis coordinate value of the adjacent vertebral positioning points below the reference point. The cropping range for the X and Y axes is defined by expanding the bounding rectangle of the key vertebral region outwards by a preset distance (the preset distance can be, but is not limited to, 20 mm; this expansion aims to improve the robustness of the vertebral segmentation model, as the accuracy of the key vertebral region may not be very high. Expansion maximizes the chance of the cropped image sub-block containing the complete vertebra; a larger preset distance increases the likelihood of including the complete vertebra, but also increases the size of the image sub-block. If the expansion exceeds the image boundary, the image boundary is removed). The corresponding boundary values ​​for the X and Y axes are then defined. For the upper end vertebral location point in the vertebral location point sequence, the upper boundary of the Z-axis clipping is set to the upper boundary value of the Z-axis of the spinal image; for the lower end vertebral location point in the vertebral location point sequence, the lower boundary of the Z-axis clipping is set to the lower boundary value of the Z-axis of the spinal image.

[0115] The vertebral segmentation model can be a binary classification segmentation model (the binary classification segmentation algorithm can be, but is not limited to, a deep convolutional neural network model based on 3DUNet, or other artificial intelligence algorithms such as RNN and Transformer). The vertebral segmentation model has two inputs: one is an image sub-block, and the other is the Gaussian probability map corresponding to the image sub-block (since vertebrae are connected by intervertebral discs, an image sub-block may contain multiple vertebrae, but each image sub-block contains only one complete vertebra. The purpose of the Gaussian probability map is to guide the vertebral segmentation model to segment only a single complete vertebra). The output of the vertebral segmentation model is the segmentation mask of a single complete vertebra, i.e., the segmentation result of a single vertebra, completing the segmentation of a single vertebra. The vertebral category corresponding to the segmentation mask of this single complete vertebra is the vertebral category corresponding to the cropping reference point used to obtain the image sub-block, thus completing the labeling of the segmentation mask of the single complete vertebra and completing the identification of the single vertebra. To improve segmentation accuracy, image sub-blocks can be resampled at intervals of [x, y, z] (x, y, z can be, but are not limited to, 1mm, 1mm, 1mm, such as 0.6mm, 0.6mm, 0.6mm, etc.). Smaller intervals will result in better segmentation accuracy, but will also lead to a greater computational load.

[0116] After completing step S205, the segmentation mask of a single complete vertebra of each image sub-block can be obtained. Then, according to the position of the image sub-block in the spinal image, the segmentation mask of a single complete vertebra of each image sub-block is reconstructed onto the spinal image.

[0117] This application also provides an embodiment in which the spinal image is a CT image, with the first cervical vertebra, the second cervical vertebra, the thoracic vertebra and the sacrum as key vertebrae, the specific form of the key vertebra region is a mask, and the specific form of the vertebral positioning point is the vertebral center point and the vertebral centroid.

[0118] This embodiment includes three stages: the first stage is coarse segmentation of the spine and identification of key vertebrae; the second stage is vertebral localization; and the third stage is fine binary segmentation and labeling of vertebrae.

[0119] In the first stage of this embodiment, the original spinal image is resampled to a smaller size for coarse segmentation of the spine, and key vertebrae masks are labeled with tags such as the first cervical vertebra, second cervical vertebra, thoracic vertebra, and sacrum to complete key vertebrae identification. In the second stage, the spinal region is cropped based on the coarse segmentation results of the first stage, the vertebral center points of all vertebrae in the spinal image are detected, and the vertebral center points are corrected and automatically labeled using the results of the first stage. The vertebral category corresponding to the vertebral center point can be determined through automatic labeling. In the third stage, using the vertebral center points from the second stage, image sub-blocks including single complete vertebrae are cropped, and fine binary segmentation is performed on the image sub-blocks, distinguishing only between vertebrae and background, to obtain the mask of a single vertebra. Finally, based on the vertebral category corresponding to the vertebral center point, the mask of the single vertebra is labeled and reconstructed onto the original spinal image.

[0120] It should be noted that the first stage of coarse segmentation of the spine involves key vertebrae segmentation. Since only a portion of the vertebrae are considered key vertebrae, coarse segmentation of the spine does not require segmenting all vertebrae in the spinal image; only the key vertebrae need to be segmented. However, in the third stage, all vertebrae in the spinal image need to be segmented. Therefore, compared to the third stage of vertebrae segmentation, the first stage of key vertebrae segmentation constitutes coarse segmentation of the spine, while the third stage constitutes fine segmentation of the vertebrae.

[0121] The following combination Figure 5 The above stages are described below:

[0122] Phase 1:

[0123] The original spinal image is resampled to obtain a smaller original spinal image. With the first cervical vertebra, second cervical vertebra, thoracic vertebra and sacrum as key vertebrae, the resampled original spinal image is subjected to coarse spinal segmentation and key vertebra identification to obtain the mask of key vertebra (i.e. key vertebra segmentation result) and the key vertebra category corresponding to each mask (i.e. key vertebra identification result).

[0124] Based on the mask of key vertebrae, the original spinal image is cropped to obtain the spinal region image.

[0125] Phase Two:

[0126] Vertebrae are located in images of the spinal region to obtain a sequence of vertebrae location points;

[0127] Based on the location of the vertebral positioning points in the spinal region image, the masks of key vertebrae obtained in the first stage, and the key vertebrae categories corresponding to each mask, the vertebral positioning points are identified to determine the vertebrae category corresponding to each vertebral positioning point.

[0128] Phase Three:

[0129] The image of the spinal region is cropped to obtain multiple image sub-blocks; each image sub-block contains a single complete vertebra.

[0130] Based on image sub-blocks and corresponding Gaussian probability maps, vertebral segmentation (i.e., single vertebra segmentation) is performed to obtain a mask for a single vertebra.

[0131] The vertebral category corresponding to the vertebral location point determined in the second stage is used as the label for the corresponding single vertebral mask obtained in the third stage. The single vertebral mask obtained in the third stage is then reconstructed into the spinal region image, and the spinal region image is then reconstructed into the original spinal image.

[0132] In the second stage, based on the coarse segmentation results of the spine in the first stage, the spinal region is extracted for vertebral localization to obtain the vertebral center point sequence, and the vertebral center point localization results are corrected and automatically labeled using the results of the first stage.

[0133] In the vertebral segmentation stage, the center point of the vertebra in the second stage is used to cut out a single vertebral region, and a fine binary segmentation is performed that distinguishes only the vertebra and the background. Finally, the segmentation results are labeled according to the labeling results of the center point and reconstructed onto the whole image.

[0134] This embodiment provides a novel multi-stage spinal image segmentation and recognition method, transforming the complex problem of spinal segmentation and recognition into several simpler tasks. The first stage reduces the difficulty of recognition by simplifying spinal recognition to the segmentation of a few key vertebrae. The second stage simplifies spinal segmentation to single-class vertebral localization, requiring only a single-channel network output and not depending on the scan range or the number of vertebrae. The third stage transforms spinal segmentation into fine-grained binary segmentation of individual vertebrae and the background. This embodiment offers the following advantages:

[0135] (1) It has stronger scalability, higher precision, faster speed, and less memory consumption;

[0136] (2) Each stage is integrated with each other and the results are optimized for each other, thereby reducing the learning difficulty of a single stage and making the whole scheme more stable and efficient.

[0137] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0138] In one embodiment, such as Figure 6 As shown, a segmentation and recognition device for spine images is provided, comprising:

[0139] The key vertebra segmentation and recognition module 601 is used to segment and recognize key vertebrae in a spinal image, taking some vertebrae of the spine as key vertebrae, and to obtain key vertebra segmentation results and key vertebra recognition results.

[0140] Vertebrae localization module 602 is used to locate the vertebrae included in the spinal image and obtain a sequence of vertebrae localization points;

[0141] The positioning point marking module 603 is used to determine the vertebral category corresponding to each vertebral positioning point based on the position of the vertebral positioning point in the spinal image, the key vertebral segmentation result, and the key vertebral identification result.

[0142] The image cropping module 604 is used to take one of the vertebral positioning points in the vertebral positioning point sequence as the cropping reference point, determine the cropping boundary based on the vertebral positioning points that are adjacent to the cropping reference point, and crop out an image sub-block including a single complete vertebra from the spinal image.

[0143] The single vertebra segmentation and recognition module 605 is used to input the image sub-block and the Gaussian distribution probability map generated with the cropping reference point as the center into the vertebra segmentation model to obtain the segmentation result of a single vertebra, and to obtain the recognition result of a single vertebra according to the vertebra category corresponding to the vertebra positioning point used as the cropping reference point.

[0144] In one embodiment, the positioning point marking module 603 is further configured to determine the vertebral positioning points in the critical vertebral region based on the position of the vertebral positioning points in the spinal image and the critical vertebral segmentation results; and to obtain the vertebral category corresponding to the vertebral positioning points in the critical vertebral region and the vertebral category corresponding to the vertebral positioning points in the non-critical vertebral region based on the vertebral category corresponding to the vertebral category corresponding to the vertebral positioning points in the critical vertebral region and the vertebral category corresponding to the vertebral positioning points in the non-critical vertebral region, based on the vertebral category corresponding to ... positioning points in the non-critical vertebral region.

[0145] In one embodiment, the vertebrae that are key vertebrae include the upper cervical vertebrae; the positioning point marking module 603 is further configured to, if the key vertebrae region of the spinal image is determined to include the upper cervical vertebrae region based on the key vertebrae segmentation result and the key vertebrae recognition result, then, using the vertebrae positioning point corresponding to the upper cervical vertebrae region as a reference, determine the vertebrae category corresponding to the vertebrae positioning point after the vertebrae positioning point corresponding to the upper cervical vertebrae region according to the constraints of the physiological structure of the spine.

[0146] In one embodiment, the vertebrae that are key vertebrae include the thoracic vertebrae; the positioning point marking module 603 is further configured to, if the key vertebrae segmentation result and the key vertebrae recognition result determine that the key vertebrae region of the spinal image includes the thoracic vertebrae region, then the vertebrae positioning point whose upper adjacent vertebrae positioning point is not in the thoracic vertebrae region and whose lower adjacent vertebrae positioning point is in the thoracic vertebrae region is used as the vertebrae positioning point corresponding to the first thoracic vertebrae; based on the vertebrae positioning point corresponding to the first thoracic vertebrae, and according to the constraints of the physiological structure of the spine, determine the vertebrae category corresponding to the vertebrae positioning point after the vertebrae positioning point corresponding to the first thoracic vertebrae.

[0147] In one embodiment, the vertebrae that are key vertebrae include the thoracic vertebrae; the positioning point marking module 603 is further configured to, if the key vertebrae region of the spinal image is determined to include the thoracic vertebrae region based on the key vertebrae segmentation result and the key vertebrae recognition result, obtain the lumbar vertebrae region based on the region below the thoracic vertebrae region; take the vertebrae positioning point that is adjacent to both the vertebrae positioning point in the thoracic vertebrae region and the vertebrae positioning point in the lumbar vertebrae region as the vertebrae positioning point corresponding to the first lumbar vertebra; and, based on the vertebrae positioning point corresponding to the first lumbar vertebrae, determine the vertebrae category corresponding to the vertebrae positioning point after the vertebrae positioning point corresponding to the first thoracic vertebrae according to the constraints of the spinal physiological structure.

[0148] In one embodiment, the vertebrae that are key vertebrae include the sacrum; the positioning point marking module 603 is further configured to, if the key vertebrae region of the spinal image is determined to include the sacrum region based on the key vertebrae segmentation result and the key vertebrae recognition result, then, using the vertebrae positioning point corresponding to the sacrum as a reference, determine the vertebrae category corresponding to the vertebrae positioning point above the vertebrae positioning point corresponding to the sacrum region based on the constraints of the spinal physiological structure.

[0149] In one embodiment, the device further includes a localization point removal module, used to determine non-spinal regions in the spinal image based on the key vertebral segmentation results; and to remove vertebral localization points located in the non-spinal regions from the vertebral localization point sequence based on the position of the vertebral localization points in the spinal image.

[0150] In one embodiment, the device further includes a positioning point adding module, configured to: if, based on the key vertebra segmentation result and the key vertebra identification result, it is determined that the spinal image includes a sacral region; if, based on the position of the vertebral positioning points in the spinal image, it is determined that there is no vertebral positioning point located in the sacral region in the vertebral positioning point sequence, then the centroid of the sacral region is used as the vertebral positioning point corresponding to the sacrum, and the vertebral positioning point corresponding to the sacrum is added to the vertebral positioning point sequence.

[0151] Specific limitations regarding the spinal image segmentation and recognition device can be found in the limitations of the spinal image segmentation and recognition method described above, and will not be repeated here. Each module in the aforementioned spinal image segmentation and recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0152] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores segmentation and recognition data for spinal images. The network interface communicates with external terminals via a network connection. The computer device also includes input / output interfaces (I / O interfaces), which are connection circuits between the processor and external devices for exchanging information; they are connected to the processor via a bus. When the computer program is executed by the processor, it implements a method for segmenting and recognizing spinal images.

[0153] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0154] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the various method embodiments described above.

[0155] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the various method embodiments described above.

[0156] In one embodiment, a computer program product is provided having a computer program stored thereon, the computer program being executed by a processor of the steps described in the various method embodiments above.

[0157] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0158] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0159] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0160] The above embodiments are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for segmenting and recognizing spine images, characterized in that, The method includes: Using some vertebrae of the spine as key vertebrae, the key vertebrae are segmented and identified in the spinal image to obtain the key vertebrae segmentation results and key vertebrae identification results. The vertebrae included in the spinal image are located to obtain a sequence of vertebral location points; Based on the location of the vertebral positioning points in the spinal image, the segmentation results of the key vertebrae, and the identification results of the key vertebrae, the vertebral category corresponding to each vertebral positioning point is determined, including: determining the vertebral positioning points in the key vertebral region based on the location of the vertebral positioning points in the spinal image and the segmentation results of the key vertebrae; and obtaining the vertebral category corresponding to the vertebral positioning points in the key vertebral region and the vertebral category corresponding to the vertebral positioning points in the non-key vertebral region based on the vertebral category corresponding to the key vertebral region determined by the key vertebral identification results and the constraints of the spinal physiological structure. Using one of the vertebral positioning points in the vertebral positioning point sequence as the cropping reference point, and based on the vertebral positioning points adjacent to the cropping reference point, the cropping boundary is determined, and an image sub-block including a single complete vertebra is cropped from the spinal image. The image sub-blocks and the Gaussian distribution probability map generated with the cropping reference point as the center are input into the vertebral segmentation model to obtain the segmentation result of a single vertebra, and the recognition result of a single vertebra is obtained according to the vertebral category corresponding to the vertebral positioning point that serves as the cropping reference point. Wherein, when the vertebrae that are considered key vertebrae include the upper cervical vertebrae, the vertebrae category corresponding to the key vertebrae region determined based on the key vertebrae identification results and the constraints of the spinal physiological structure, to obtain the vertebrae category corresponding to the vertebrae positioning point in the key vertebrae region and the vertebrae category corresponding to the vertebrae positioning point in the non-key vertebrae region, includes: if, based on the key vertebrae segmentation results and key vertebrae identification results, it is determined that the key vertebrae region of the spinal image includes the upper cervical vertebrae region, then, using the vertebrae positioning point corresponding to the upper cervical vertebrae region as a reference, according to the constraints of the spinal physiological structure, the vertebrae category corresponding to the vertebrae positioning point after the vertebrae positioning point in the upper cervical vertebrae region is determined.

2. The method according to claim 1, characterized in that, When the vertebrae considered as critical vertebrae include the thoracic vertebrae, the vertebral category corresponding to the critical vertebral region determined based on the critical vertebral identification result, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the critical vertebral region and the vertebral category corresponding to the vertebral location point in the non-critical vertebral region, including: If, based on the key vertebral segmentation results and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the thoracic vertebral region, then the vertebral positioning point whose upper adjacent vertebral positioning point is not in the thoracic vertebral region and whose lower adjacent vertebral positioning point is in the thoracic vertebral region is taken as the vertebral positioning point corresponding to the first thoracic vertebra. Using the vertebral positioning point corresponding to the first thoracic vertebra as a reference, and based on the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the first thoracic vertebra is determined.

3. The method according to claim 1, characterized in that, When the vertebrae considered as critical vertebrae include the thoracic vertebrae, the vertebral category corresponding to the critical vertebral region determined based on the critical vertebral identification result, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the critical vertebral region and the vertebral category corresponding to the vertebral location point in the non-critical vertebral region, including: If, based on the key vertebrae segmentation results and key vertebrae identification results, it is determined that the key vertebrae region of the spinal image includes the thoracic vertebrae region, then the lumbar vertebrae region is obtained based on the region below the thoracic vertebrae region. The vertebral positioning point that is adjacent to both the vertebral positioning point in the thoracic region and the vertebral positioning point in the lumbar region is taken as the vertebral positioning point corresponding to the first lumbar vertebra. Based on the vertebral positioning point corresponding to the first lumbar vertebra, and according to the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the first thoracic vertebra is determined.

4. The method according to claim 1, characterized in that, When the sacrum is included as a critical vertebra, the vertebral category corresponding to the critical vertebral region determined based on the critical vertebral identification result and the constraints of the spinal physiological structure are used to obtain the vertebral category corresponding to the vertebral positioning point in the critical vertebral region and the vertebral category corresponding to the vertebral positioning point in the non-critical vertebral region, including: If, based on the key vertebral segmentation and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the sacral region, then, using the vertebral positioning point corresponding to the sacral region as a reference, and according to the constraints of the spinal physiological structure, the vertebral category corresponding to the vertebral positioning point above the vertebral positioning point in the sacral region is determined.

5. The method according to any one of claims 1 to 4, characterized in that, Before determining the vertebral category corresponding to each vertebral location point based on the position of the vertebral location point in the spinal image, the key vertebral segmentation result, and the key vertebral identification result, the method further includes: Based on the segmentation results of the key vertebrae, the non-spinal regions in the spinal image are determined; Based on the position of the vertebral positioning points in the spinal image, vertebral positioning points located in the non-spinal region are removed from the vertebral positioning point sequence.

6. The method according to any one of claims 1 to 4, characterized in that, In cases where the sacrum is included among the key vertebrae, before determining the vertebrae category corresponding to each vertebrae location point based on the position of the vertebrae location points in the spinal image, the key vertebrae segmentation results, and the key vertebrae identification results, the method further includes: If, based on the key vertebrae segmentation results and the key vertebrae identification results, it is determined that the spinal image includes the sacral region; If, based on the position of the vertebral positioning point in the spinal image, it is determined that there is no vertebral positioning point located in the sacral region in the vertebral positioning point sequence, then the centroid of the sacral region is taken as the vertebral positioning point corresponding to the sacrum, and the vertebral positioning point corresponding to the sacrum is added to the vertebral positioning point sequence.

7. A segmentation and recognition device for spine images, characterized in that, The device includes: The key vertebra segmentation and recognition module is used to segment and recognize key vertebrae in spinal images, taking some vertebrae of the spine as key vertebrae, and to obtain key vertebra segmentation results and key vertebra recognition results. The vertebral localization module is used to locate the vertebrae included in the spinal image and obtain a sequence of vertebral localization points. The positioning point marking module is used to determine the vertebral category corresponding to each vertebral positioning point based on the position of the vertebral positioning point in the spinal image, the segmentation result of the key vertebra, and the identification result of the key vertebra; The image cropping module is used to take one of the vertebral positioning points in the vertebral positioning point sequence as the cropping reference point, determine the cropping boundary based on the vertebral positioning points adjacent to the cropping reference point, and crop out an image sub-block including a single complete vertebra from the spinal image. The single vertebra segmentation and recognition module is used to input the image sub-block and the Gaussian distribution probability map generated with the cropping reference point as the center into the vertebra segmentation model to obtain the segmentation result of a single vertebra, and to obtain the recognition result of a single vertebra according to the vertebra category corresponding to the vertebra positioning point that serves as the cropping reference point; Specifically, the process involves determining the vertebral category corresponding to each vertebral location point based on its position in the spinal image, the segmentation result of the key vertebrae, and the identification result of the key vertebrae. This includes: determining the vertebral location points located in the key vertebral region based on their position in the spinal image and the segmentation result of the key vertebrae; and obtaining the vertebral category corresponding to the vertebral location points located in the key vertebral region and the vertebral category corresponding to the vertebral location points located in the non-key vertebral region based on the vertebral category corresponding to the key vertebral region determined by the identification result of the key vertebrae and the constraints of the spinal physiological structure. When the vertebrae that are considered key vertebrae include the upper cervical vertebrae, the process of determining the vertebrae category corresponding to the key vertebrae region based on the key vertebrae identification results and the constraints of the spinal physiological structure to obtain the vertebrae category corresponding to the vertebrae positioning points in the key vertebrae region and the vertebrae category corresponding to the vertebrae positioning points in non-key vertebrae regions includes: if, based on the key vertebrae segmentation results and key vertebrae identification results, it is determined that the key vertebrae region of the spinal image includes the upper cervical vertebrae region, then, using the vertebrae positioning points corresponding to the upper cervical vertebrae region as a reference, and based on the constraints of the spinal physiological structure, the vertebrae category corresponding to the vertebrae positioning points after the vertebrae positioning points corresponding to the upper cervical vertebrae region is determined.

8. The apparatus according to claim 7, characterized in that, When the vertebrae considered as critical vertebrae include the thoracic vertebrae, the vertebral category corresponding to the critical vertebral region determined based on the critical vertebral identification result, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the critical vertebral region and the vertebral category corresponding to the vertebral location point in the non-critical vertebral region, including: If, based on the key vertebral segmentation results and key vertebral identification results, it is determined that the key vertebral region of the spinal image includes the thoracic vertebral region, then the vertebral positioning point whose upper adjacent vertebral positioning point is not in the thoracic vertebral region and whose lower adjacent vertebral positioning point is in the thoracic vertebral region is taken as the vertebral positioning point corresponding to the first thoracic vertebra. Using the vertebral positioning point corresponding to the first thoracic vertebra as a reference, and based on the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the first thoracic vertebra is determined.

9. The apparatus according to claim 7, characterized in that, When the vertebrae considered as critical vertebrae include the thoracic vertebrae, the vertebral category corresponding to the critical vertebral region determined based on the critical vertebral identification result, along with spinal physiological structure constraints, yields the vertebral category corresponding to the vertebral location point in the critical vertebral region and the vertebral category corresponding to the vertebral location point in the non-critical vertebral region, including: If, based on the key vertebrae segmentation results and key vertebrae identification results, it is determined that the key vertebrae region of the spinal image includes the thoracic vertebrae region, then the lumbar vertebrae region is obtained based on the region below the thoracic vertebrae region. The vertebral positioning point that is adjacent to both the vertebral positioning point in the thoracic region and the vertebral positioning point in the lumbar region is taken as the vertebral positioning point corresponding to the first lumbar vertebra. Based on the vertebral positioning point corresponding to the first lumbar vertebra, and according to the constraints of the physiological structure of the spine, the vertebral category corresponding to the vertebral positioning point after the first thoracic vertebra is determined.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

Images