Lumbar transverse process positioning method, system, equipment and product

Abstract

The invention discloses a lumbar transverse process positioning method, system, equipment and product, and belongs to the technical field of lumbar transverse process localization, and the method comprises the steps: obtaining an original collection image, collected by a probe, of a lumbar region, carrying out the time alignment of the original collection image, obtaining a collection image sequence, carrying out the feature extraction and coding of the collection image sequence, and carrying out the positioning of the lumbar transverse process. The method comprises the following steps: acquiring a local feature prediction map by using a semantic segmentation network, performing up-sampling and global prediction on an acquired image sequence to obtain a global feature prediction map, fusing the local feature prediction map and the global feature prediction map to obtain a feature fusion map, and segmenting the feature fusion map by using the optimized semantic segmentation network. And marking the original acquired image based on the transverse process position and the upper and lower edge distance data to obtain a transverse process positioning marking graph. According to the method, personal errors can be reduced, the diagnosis accuracy and efficiency of doctors are remarkably improved, the transverse process identification efficiency and accuracy are improved, and the upper and lower edge distance data are automatically obtained.

Description

Technical Field

[0001] This invention belongs to the field of lumbar transverse process localization technology, specifically relating to a method, system, device, and product for lumbar transverse process localization. Background Technology

[0002] The transverse processes of the lumbar vertebrae are bony protrusions extending outward from both sides of the vertebral body. They are vertebral accessory structures, and their main function is to provide attachment points for muscles and ligaments. The length and shape of the transverse processes of different lumbar vertebrae vary. Current techniques for locating the transverse processes of the lumbar vertebrae mainly employ manual positioning or conventional image positioning methods, such as manual palpation based on the doctor's experience, or positioning using CT or ultrasound images.

[0003] However, all of the above methods have corresponding shortcomings, including: (1) For palpation based on the doctor's experience: due to the extreme reliance on diagnostic experience, the palpation error rate is extremely high for obese patients, anatomical variations, or revision surgeries; (2) For localization using CT images: for soft tissues with low resolution, using CT images to locate the transverse process of the lumbar vertebrae is prone to problems such as unclear boundaries; (3) For localization using ultrasound images: using only two-dimensional images for localization, it is impossible to automatically know the distance between the upper and lower edges of the transverse process.

[0004] Therefore, how to provide an effective technical solution to address the problems of high error rate, unclear boundaries, and inability to automatically determine the distance between the upper and lower edges of the transverse process in existing technologies has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device, and product for lumbar transverse process localization, in order to solve the above-mentioned problems existing in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for locating a transverse process, comprising: The original images of the lumbar spine region acquired by the probe are obtained. The original images of the lumbar spine region acquired by the probe are preprocessed to obtain preprocessed original images. The preprocessed original images are then time-aligned based on the time series to obtain the image sequence. The acquired image sequence is input into a pre-optimized feature extractor for feature extraction, resulting in a feature representation of the local feature space of the acquired image. This feature representation is then input into an encoder to obtain a local feature prediction map. The feature representation of the local feature space of the acquired image is upsampled to obtain an upsampled feature image. The upsampled feature image is then input into the global feature discriminator to obtain a global feature prediction map. The local feature prediction map and the global feature prediction map are fused to obtain a feature fusion map. The feature fusion map is then segmented using a pre-optimized semantic segmentation network to obtain multiple segmented images. Based on the segmented images, the transverse process position is obtained. The distance between the transverse process position and the upper and lower edge distance data is calculated. The original acquired image is then labeled based on the transverse process position and the upper and lower edge distance data to obtain a transverse process localization label map.

[0007] In one possible design, the raw images acquired by the probe are preprocessed to obtain preprocessed raw images, including: The original acquired image is denoised to obtain the denoised original acquired image. An image enhancement operation is performed on the denoised original acquired image to obtain the enhanced original acquired image; The enhanced original acquired image is normalized to obtain a normalized original acquired image, and the normalized original acquired image is then standardized to obtain a preprocessed original acquired image.

[0008] In one possible design, the process of optimizing the feature extractor includes: Obtain the target domain image training set and the source domain image training set, input the target domain image training set and the source domain image training set into a preset feature extractor for feature extraction, and obtain the feature representation of the local feature space of the target domain training image and the feature representation of the local feature space of the source domain training image; Based on the local feature discriminator, feature extraction is performed on the feature representations of the local feature space of the target domain training image and the feature representations of the local feature space of the source domain training image to obtain the local feature adversarial loss. Align the feature distributions of the source and target domains within a local space; Within the aligned local space, the adversarial loss of local features is backpropagated to a preset feature extractor to optimize the preset feature extractor, resulting in an optimized feature extractor.

[0009] In one possible design, the process of optimizing the semantic segmentation network includes: Upsampling is performed on the source domain image training set to obtain an upsampled training image. The upsampled training image is then input into the global feature discriminator to obtain a global feature training prediction map. The source domain image training set is input into the semantic segmentation network for segmentation to obtain the first predicted value. The source domain image training set includes label values. The label values ​​and the first predicted value are input into a preset loss function to obtain the first loss value. The semantic segmentation network is updated using the first loss value to obtain the updated semantic segmentation network. The source domain training dataset is input into the updated semantic segmentation network again for segmentation to obtain the second predicted value. The label value and the second predicted value are then input into the preset loss function to obtain the second loss value. Repeat the above update steps until the preset iteration termination condition is met to obtain the trained semantic segmentation network. Obtain the source domain label map, and calculate the cross-entropy loss between the global feature training prediction map and the source domain label map to obtain the semantic segmentation loss value; The trained semantic segmentation network is optimized based on the semantic segmentation loss value to obtain the optimized semantic segmentation network.

[0010] In one possible design, the upsampled feature image is input into a global feature discriminator to obtain a global feature prediction map, including: Perform feature enhancement operations on the upsampled feature image to obtain the enhanced feature image; The enhanced feature image is input into the pre-optimized global feature discriminator to obtain the global feature prediction map.

[0011] In one possible design, the original acquired image is annotated based on the transverse process position and the distance between its upper and lower edges to obtain a transverse process positioning annotation map, including: The center point of the transverse process is calculated to obtain the coordinates of the transverse process center point; The segmented image is subjected to erosion, and the coordinates of the center point of the transverse protrusion after erosion are calculated to obtain the coordinates of the center point of the eroded transverse protrusion. The difference between the coordinates of the center point of the transverse protrusion and the coordinates of the corroded transverse protrusion is calculated to obtain the moving distance. The transverse process position is moved based on the moving distance to obtain the transverse process position label box; The transverse process location marker box and the distance data between the upper and lower edges are superimposed on the original acquired image to obtain the transverse process location marker map.

[0012] In one possible design, the distance to the transverse process is calculated to obtain the distance data between the upper and lower edges, including: An edge detection algorithm is used to detect the position of the transverse protrusion to obtain the transverse protrusion edge, which includes an upper edge and a lower edge; Calculate the Euclidean distance from each point on the upper edge to all points on the lower edge to obtain multiple distance data. Select the minimum distance data for each point on the upper edge as the initial distance data between the upper and lower edges. The average value of multiple initial upper and lower edge distance data is calculated to obtain the upper and lower edge distance data.

[0013] Secondly, the present invention provides a lumbar transverse process positioning system, comprising: The preprocessing module is used to acquire the original images of the lumbar spine region captured by the probe, preprocess the original images of the lumbar spine region captured by the probe to obtain the preprocessed original images, and perform time alignment operation on the preprocessed original images based on the time series to obtain the image sequence. The first prediction module is used to input the acquired image sequence into the pre-optimized feature extractor for feature extraction, obtain the feature representation of the local feature space of the acquired image, and input the feature representation of the local feature space of the acquired image into the encoder to obtain the local feature prediction map; The second prediction module is used to upsample the feature representation of the local feature space of the acquired image to obtain an upsampled feature image. The upsampled feature image is then input into the global feature discriminator to obtain a global feature prediction map. The fusion computing module is used to fuse the local feature prediction map and the global feature prediction map to obtain a feature fusion map. The feature fusion map is then segmented using a pre-optimized semantic segmentation network to obtain multiple segmented images. Based on the segmented images, the transverse process position is obtained. The distance between the transverse process position and the upper and lower edge distance data is calculated. Based on the transverse process position and the upper and lower edge distance data, the original acquired image is labeled to obtain a transverse process localization label map.

[0014] Thirdly, the present invention provides a computer device comprising a memory, a processor, and a transceiver connected in sequence and communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the lumbar transverse process localization method as described in the first aspect.

[0015] Fourthly, the present invention provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the lumbar transverse process localization method as described in the first aspect.

[0016] The beneficial effects of this invention are as follows: (1) This invention discloses a method, system, device and product for lumbar transverse process localization. This invention extracts and predicts features from the acquired image sequence to obtain a local feature prediction map and a global feature prediction map. The local feature prediction map and the global feature prediction map are fused to obtain a feature fusion map. An optimized semantic segmentation network is used to segment the feature fusion map and to identify the segmented image to obtain the transverse process position. This invention processes the images acquired by the probe. Compared with the traditional localization method that relies on the doctor's experience and manual palpation, this invention can significantly and indirectly improve the doctor's diagnostic accuracy and efficiency, reduce human error, and assist the doctor in making corresponding diagnostic suggestions. (2) This invention preprocesses the original images of the lumbar region acquired by the probe to enhance the contrast of the original images, eliminate the influence of noise on the images, and uses an optimized semantic segmentation network to segment the images, which can improve the recognition efficiency and accuracy of the transverse process, highlight the edge line of the transverse process, and reduce computational redundancy. (3) The present invention annotates the original acquired image based on the transverse process position and the distance between the upper and lower edges to obtain the transverse process positioning annotation map, which can be clearly and intuitively distinguished in the image acquired by the probe. Furthermore, by calculating the distance between the transverse process position, the distance between the upper and lower edges can be automatically obtained, which is convenient for application and promotion. Attached Figure Description

[0017] Figure 1 A flowchart of a transverse process localization method provided in an embodiment of the present invention; Figure 2 A block diagram of a transverse protrusion positioning system provided in an embodiment of the present invention; Figure 3 This is a structural diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0019] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0020] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0021] Example: like Figure 1 As shown, the first aspect of this embodiment provides a transverse process localization method, which can be executed, but is not limited to, by a computer device or virtual machine with certain computing resources, such as a personal computer or smartphone, or by a virtual machine; the transverse process localization method includes, but is not limited to, the following steps: S1. Acquire the original images of the lumbar spine region acquired by the probe, preprocess the original images of the lumbar spine region acquired by the probe to obtain the preprocessed original images, and perform time alignment operation on the preprocessed original images based on the time series to obtain the acquired image sequence; It should be noted that the probe mentioned in this embodiment is an ultrasonic probe, and the original acquired image is an ultrasonic image.

[0022] Specifically, in step S1, the original acquired image is preprocessed to obtain a preprocessed original acquired image, including: S11. Denoise the original acquired image obtained by the probe to obtain the denoised original acquired image; S12. Perform image enhancement operation on the denoised original acquired image to obtain the enhanced original acquired image; S13. Normalize the enhanced original acquired image to obtain a normalized original acquired image, and then standardize the normalized original acquired image to obtain a preprocessed original acquired image.

[0023] It should be noted that ultrasound images contain a lot of noise, so they need to be denoised to eliminate the impact of noise on subsequent operations. Denoising is not limited to using adaptive median filtering or Bayesian nonlocal mean optimization algorithms, etc., and this embodiment does not make specific limitations. At the same time, image enhancement operations are performed on the denoised original image to enhance image contrast, and normalization and standardization are performed to convert the image pixel values ​​into uniform scalars to eliminate the influence of dimensional differences, which facilitates subsequent image processing.

[0024] S2. Input the acquired image sequence into the pre-optimized feature extractor for feature extraction to obtain the feature representation of the local feature space of the acquired image. Input the feature representation of the local feature space of the acquired image into the encoder to obtain the local feature prediction map. Specifically, in step S2, the process of optimizing the feature extractor includes: S21. Obtain the target domain image training set and the source domain image training set, input the target domain image training set and the source domain image training set into a preset feature extractor for feature extraction, and obtain the feature representation of the local feature space of the target domain training image and the feature representation of the local feature space of the source domain training image. S22. Based on the local feature discriminator, feature extraction is performed on the feature representations of the local feature space of the target domain training image and the feature representations of the local feature space of the source domain training image to obtain the local feature adversarial loss; S23. Align the feature distributions of the source and target domains within the local space; S24. In the aligned local space, the local feature adversarial loss is backpropagated to the preset feature extractor to optimize the preset feature extractor and obtain the optimized feature extractor.

[0025] It should be noted that in this embodiment, the source domain image training set is training data containing label values, while the target domain image training set is training data without label values. By reducing the difference between the target domain image and the source domain image, the feature extractor can be optimized.

[0026] S3. Upsample the feature representation of the local feature space of the acquired image to obtain an upsampled feature image. Input the upsampled feature image into the global feature discriminator to obtain the global feature prediction map. Specifically, in step S3, the upsampled feature image is input into the global feature discriminator to obtain the global feature prediction map, including: S31. Perform feature enhancement operation on the upsampled feature image to obtain the enhanced feature image; S32. Input the enhanced feature image into the pre-optimized global feature discriminator to obtain the global feature prediction map.

[0027] S4. The local feature prediction map and the global feature prediction map are fused to obtain a feature fusion map. The feature fusion map is segmented using a pre-optimized semantic segmentation network to obtain multiple segmented images. Based on the segmented images, the transverse process position is obtained. The distance between the transverse process position and the upper and lower edge distance data is calculated. The original acquired image is labeled based on the transverse process position and the upper and lower edge distance data to obtain a transverse process localization label map.

[0028] Specifically, the optimization process for the semantic segmentation network includes: upsampling the source domain image training set to obtain an upsampled training image; inputting the upsampled training image into a global feature discriminator to obtain a global feature training prediction map; inputting the source domain image training set into the semantic segmentation network for segmentation to obtain a first predicted value, wherein the source domain image training set includes label values; inputting the label values ​​and the first predicted value into a preset loss function to obtain a first loss value; updating the semantic segmentation network using the first loss value to obtain an updated semantic segmentation network; inputting the source domain training dataset into the updated semantic segmentation network again for segmentation to obtain a second predicted value; inputting the label values ​​and the second predicted value into a preset loss function to obtain a second loss value; repeating the above update steps until a preset iteration termination condition is reached to obtain a trained semantic segmentation network; obtaining a source domain label map; calculating the cross-entropy loss between the global feature training prediction map and the source domain label map to obtain a semantic segmentation loss value; and optimizing the trained semantic segmentation network based on the semantic segmentation loss value to obtain an optimized semantic segmentation network.

[0029] In this embodiment, the default loss function is the binary cross-entropy loss function. Those skilled in the art can choose the loss function according to their needs, and no specific limitation is made here.

[0030] Furthermore, upsampling is performed on the source domain image training set to obtain upsampled training images. This includes: extracting features from the source domain image training set using a pre-optimized feature extractor to obtain feature representations of the local feature space of the source domain training images; and upsampling these feature representations to obtain upsampled training images. The upsampling operation increases the image resolution.

[0031] Specifically, in step S4, the original acquired image is annotated based on the transverse process position and the distance between its upper and lower edges to obtain a transverse process positioning annotation map, including: S41. Calculate the center point of the transverse process to obtain the coordinates of the center point of the transverse process; S42. Perform an erosion operation on the segmented image, calculate the coordinates of the center point of the transverse protrusion after the erosion operation, and obtain the coordinates of the center point of the eroded transverse protrusion. S43. Calculate the difference between the coordinates of the center point of the transverse protrusion and the coordinates of the corroded transverse protrusion to obtain the moving distance; S44. Move the transverse process position based on the moving distance to obtain the transverse process position annotation box; S45. Overlay the transverse process position marking box and the distance data between the upper and lower edges with the original acquired image to obtain the transverse process positioning marking map.

[0032] Specifically, in step S4, distance calculation is performed on the overlaid acquired images to obtain the upper and lower edge distance data, including: S46. Use an edge detection algorithm to detect the position of the transverse protrusion and obtain the transverse protrusion edge, wherein the transverse protrusion edge includes an upper edge and a lower edge; S47. Calculate the Euclidean distance from each point on the upper edge to all points on the lower edge, obtain multiple distance data, and select the minimum distance data of each point on the upper edge as the initial upper and lower edge distance data; S48. Calculate the average value of multiple initial upper and lower edge distance data to obtain the upper and lower edge distance data.

[0033] In one possible design, the transverse process position is obtained based on the segmented image. This can be achieved using either an object detection algorithm or a method based on anatomical prior knowledge to identify the lumbar vertebrae in the segmented image, thereby obtaining the position of the lumbar transverse process. Specifically, object detection algorithms can include the YOLO series, SSD (Single Shot MultiBox Detector), or Mask R-CNN. These algorithms are existing technologies and will not be elaborated upon here.

[0034] It should be noted that after obtaining the transverse process positioning map, the transverse process positioning map is visualized.

[0035] Based on the aforementioned publicly available information, by pre-optimizing the feature extractor and semantic segmentation network to extract and segment features from the original acquired images, organs in the acquired images can be accurately identified, and the location of the transverse process can be determined based on organ identification. This can significantly improve the diagnostic efficiency of doctors and the accuracy of their operations during surgery, without requiring doctors to have extensive relevant experience. It assists doctors in making appropriate treatment decisions. At the same time, the distance between the upper and lower edges of the transverse process is calculated, and the distance between the upper and lower edges and the location of the transverse process are annotated in the original acquired images. The annotated images are then visualized, providing an intuitive and clear understanding of the relevant data of the transverse process, which is convenient for application and promotion.

[0036] like Figure 2 As shown, the second aspect of this embodiment provides a lumbar transverse process positioning system, including: The preprocessing module is used to acquire the original images of the lumbar spine region captured by the probe, preprocess the original images of the lumbar spine region captured by the probe to obtain the preprocessed original images, and perform time alignment operation on the preprocessed original images based on the time series to obtain the image sequence. The first prediction module is used to input the acquired image sequence into the pre-optimized feature extractor for feature extraction, obtain the feature representation of the local feature space of the acquired image, and input the feature representation of the local feature space of the acquired image into the encoder to obtain the local feature prediction map; The second prediction module is used to upsample the feature representation of the local feature space of the acquired image to obtain an upsampled feature image. The upsampled feature image is then input into the global feature discriminator to obtain a global feature prediction map. The fusion computing module is used to fuse the local feature prediction map and the global feature prediction map to obtain a feature fusion map. The feature fusion map is then segmented using a pre-optimized semantic segmentation network to obtain multiple segmented images. Based on the segmented images, the transverse process position is obtained. The distance between the transverse process position and the upper and lower edge distance data is calculated. Based on the transverse process position and the upper and lower edge distance data, the original acquired image is labeled to obtain a transverse process localization label map.

[0037] The working process, working details and technical effects of the lumbar transverse process positioning system provided in the second aspect of this embodiment can be found in the lumbar transverse process positioning method described in the first aspect, and will not be repeated here.

[0038] like Figure 3 As shown, the third aspect of this embodiment provides a computer device, including a memory, a processor, and a transceiver connected in sequence for communication. The memory stores a computer program, the transceiver sends and receives messages, and the processor reads the computer program and executes the lumbar transverse process localization method as described in the first aspect. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; the processor may include, but is not limited to, a microprocessor of the STM32F105 series. Furthermore, the computer device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0039] The working process, working details and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment can be found in the lumbar transverse process localization method described in the first aspect, and will not be repeated here.

[0040] The fourth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, are used to implement the lumbar transverse process localization method as described in the first aspect.

[0041] The working process, working details and technical effects of the aforementioned computer program product provided in this embodiment can be found in the lumbar transverse process localization method described in the first aspect, and will not be repeated here.

[0042] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of locating a transverse process of a lumbar vertebrae, comprising: The method comprises the following steps: Obtaining the original acquisition image of the lumbar region collected by the probe, preprocessing the original acquisition image of the lumbar region collected by the probe to obtain the preprocessed original acquisition image, and performing time alignment on the preprocessed original acquisition image based on a time sequence to obtain an acquisition image sequence; The acquisition image sequence is input into the pre-optimized feature extractor for feature extraction to obtain the feature representation of the acquisition image local feature space, and the feature representation of the acquisition image local feature space is input into the encoder to obtain the local feature prediction map; The feature representation of the acquisition image local feature space is upsampled to obtain an upsampled feature image, and the upsampled feature image is input into the global feature discriminator to obtain a global feature prediction map; The local feature prediction map and the global feature prediction map are fused to obtain a feature fusion map, the feature fusion map is segmented using a pre-optimized semantic segmentation network to obtain a plurality of segmented images, the segmented images are used for recognition to obtain a transverse process position, the transverse process position is subjected to distance calculation to obtain upper and lower edge distance data, and the original acquisition image is labeled based on the transverse process position and the upper and lower edge distance data to obtain a transverse process positioning label map.

2. The method of claim 1, wherein, The original acquisition image collected by the probe is preprocessed to obtain the preprocessed original acquisition image, which comprises: The original acquisition image collected by the probe is denoised to obtain a denoised original acquisition image; The denoised original acquisition image is subjected to an image enhancement operation to obtain an enhanced original acquisition image; The enhanced original acquisition image is subjected to normalization processing to obtain a normalized original acquisition image, and the normalized original acquisition image is subjected to standardization processing to obtain the preprocessed original acquisition image.

3. The method of claim 1, wherein, The process of optimizing the feature extractor comprises: Obtaining a target domain image training set and a source domain image training set, inputting the target domain image training set and the source domain image training set into a pre-set feature extractor for feature extraction to obtain feature representation of a target domain training image local feature space and feature representation of a source domain training image local feature space; Based on the local feature discriminator, the feature representation of the target domain training image local feature space and the feature representation of the source domain training image local feature space are subjected to feature extraction to obtain a local feature adversarial loss; Aligning the feature distribution of the source domain and the target domain in the local space; In the aligned local space, the local feature adversarial loss is backpropagated to the pre-set feature extractor to optimize the pre-set feature extractor to obtain the optimized feature extractor.

4. The method of claim 3, wherein, The process of optimizing the semantic segmentation network comprises: Based on the source domain image training set, an upsampled training image is obtained, the upsampled training image is input into the global feature discriminator to obtain a global feature training prediction map; The source domain image training set is input into the semantic segmentation network for segmentation to obtain a first prediction value, the source domain image training set comprises a label value, the label value and the first prediction value are input into a pre-set loss function to obtain a first loss value; The semantic segmentation network is updated using the first loss value to obtain an updated semantic segmentation network; The source domain training data set is input into the updated semantic segmentation network again for segmentation to obtain a second prediction value, and the label value and the second prediction value are input into a preset loss function to obtain a second loss value; The updating step is repeated until a preset iteration termination condition is reached to obtain the trained semantic segmentation network; A semantic segmentation loss value is obtained by performing cross-entropy loss calculation on the global feature training prediction map and the source domain marker map; The trained semantic segmentation network is optimized based on the semantic segmentation loss value to obtain an optimized semantic segmentation network.

5. The method of claim 1, wherein, The up-sampling feature image is input into the global feature discriminator to obtain a global feature prediction map, including: The up-sampling feature image is subjected to a feature enhancement operation to obtain an enhanced feature image; The enhanced feature image is input into the pre-optimized global feature discriminator to obtain the global feature prediction map.

6. The method of claim 1, wherein, The original collection image is labeled based on the transverse process position and the upper and lower edge distance data to obtain a transverse process positioning label map, including: The center point of the transverse process position is calculated to obtain a transverse process center point coordinate; The segmented image is subjected to an erosion operation, and the center point coordinate of the transverse process position after the erosion operation is calculated to obtain an eroded transverse process center point coordinate; The transverse process center point coordinate and the eroded transverse process center point coordinate are subjected to difference calculation to obtain a movement distance; The transverse process position is moved based on the movement distance to obtain a transverse process position label frame; The transverse process position label frame and the upper and lower edge distance data are superimposed on the original collection image to obtain the transverse process positioning label map.

7. The method of claim 1, wherein, The upper and lower edge distance data are obtained by calculating the distance of the transverse process position, including: The transverse process position is detected using an edge detection algorithm to obtain a transverse process edge, the transverse process edge including an upper edge and a lower edge; The Euclidean distance of each point in the upper edge to all points in the lower edge is calculated to obtain a plurality of distance data, and the minimum distance data of each point in the upper edge is selected as initial upper and lower edge distance data; The plurality of initial upper and lower edge distance data are subjected to average value calculation to obtain the upper and lower edge distance data.

8. A transverse process locator system for use in carrying out the method of any one of claims 1 to 7, characterized by, including: The preprocessing module is configured to obtain an original collection image of a lumbar region collected by a probe, pre-process the original collection image of the lumbar region collected by the probe to obtain a pre-processed original collection image, and perform time alignment on the pre-processed original collection image based on a time sequence to obtain a collection image sequence; The first prediction module is configured to input the collection image sequence into a pre-optimized feature extractor to extract features, obtain feature representation of a local feature space of the collection image, and input the feature representation of the local feature space of the collection image into an encoder to obtain a local feature prediction map; The second prediction module is configured to up-sample the feature representation of the local feature space of the collection image to obtain an up-sampling feature image, and input the up-sampling feature image into a global feature discriminator to obtain a global feature prediction map; The fusion computing module is configured to fuse the local feature prediction map and the global feature prediction map to obtain a feature fusion map, segment the feature fusion map using a pre-optimized semantic segmentation network to obtain a plurality of segmented images, recognize based on the segmented images to obtain a transverse process position, calculate a distance of the transverse process position to obtain upper and lower edge distance data, and label an original acquisition image based on the transverse process position and the upper and lower edge distance data to obtain a transverse process positioning label map.

9. A computer device, comprising: The computer program or the instructions realize the lumbar transverse process positioning method as claimed in any one of claims 1-7 when executed by a computer.

10. A computer program product comprising computer programs or instructions, characterized in that, The computer program or the instructions realize the lumbar transverse process positioning method as claimed in any one of claims 1-7 when executed by a computer.

Images