Spine image segmentation method and device based on vertebra positioning

Abstract

The invention provides a spine image segmentation method and device based on vertebra positioning. The method comprises the following steps: acquiring a spine medical image; inputting the spine medical image into the contour correction structure to obtain a correction mask; and inputting the correction mask and the spine medical image into the segmentation structure to obtain an image segmentation result. According to the method and the device, the problem that the vertebral bodies are easy to adhere or the sequence numbers of the vertebral bodies are not accurate in the segmentation result is avoided through the processing mode of generating the correction mask to correct the segmentation.

Description

Technical Field

[0001] The present application relates to the technical field of medical image recognition, and in particular to a method and device for spinal image segmentation based on vertebral positioning. Background Art

[0002] Currently, spinal image segmentation is performed using deep convolutional models, but the segmentation results are prone to problems such as vertebral adhesion or inaccurate vertebral numbers. Summary of the Invention

[0003] The problem to be solved by this application is that the current spinal segmentation results are prone to vertebral adhesion or inaccurate vertebral numbering.

[0004] To solve the above problems, the present application provides a first aspect of a spinal column image segmentation method based on vertebrae positioning, comprising:

[0005] Acquire medical images of the spine;

[0006] Inputting the spinal medical image into the contour correction structure to obtain a correction mask;

[0007] The corrected mask and the spinal medical image are input into the segmentation structure to obtain the image segmentation result.

[0008] A second aspect of the present application provides a spinal image segmentation device based on vertebrae positioning, comprising:

[0009] An image acquisition module, which is used to acquire spinal medical images;

[0010] An image correction module is used to input a spinal medical image into a contour correction structure to obtain a correction mask;

[0011] The image segmentation module is used to input the corrected mask and the spinal medical image into the segmentation structure to obtain the image segmentation result.

[0012] A third aspect of the present application provides an electronic device, comprising: a memory and a processor;

[0013] The memory is used to store programs;

[0014] The processor, coupled to the memory, is configured to execute the program to:

[0015] Acquire medical images of the spine;

[0016] Inputting the spinal medical image into the contour correction structure to obtain a correction mask;

[0017] The corrected mask and the spinal medical image are input into the segmentation structure to obtain the image segmentation result.

[0018] A fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the above-mentioned spinal image segmentation method based on vertebral positioning.

[0019] In the present application, the segmentation is corrected by generating a correction mask, thereby avoiding the problem of vertebral adhesion or inaccurate vertebral numbering in the segmentation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of a spinal column image segmentation method based on vertebral positioning according to an embodiment of the present application;

[0021] Figure 2 Schematic diagram of the contour correction structure of the spinal image segmentation method based on vertebra positioning according to an embodiment of the present application;

[0022] Figure 3 1. It is an architectural diagram of a segmentation structure of a spinal column image segmentation method based on vertebra positioning according to an embodiment of the present application;

[0023] Figure 4 1 is a structural block diagram of a spinal column image segmentation device based on vertebral positioning according to an embodiment of the present application;

[0024] Figure 5 2 is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] To make the above-mentioned objects, features, and advantages of the present application more clearly understood, specific embodiments of the present application are described in detail below with reference to the accompanying drawings. Although the accompanying drawings show exemplary embodiments of the present application, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0026] It should be noted that, unless otherwise specified, the technical or scientific terms used in this application should have the common meanings understood by those skilled in the art to which this application belongs.

[0027] To address the above problems, the present application provides a new spinal image segmentation solution based on vertebral positioning, which can segment the vertebrae after positioning, eliminating the problems of vertebral adhesion or inaccurate vertebral numbering in the spinal segmentation results.

[0028] The embodiment of the present application provides a spinal image segmentation method based on vertebra positioning, the specific scheme of the method is as follows: Figure 1-Figure 3As shown, the method can be performed by a spinal image segmentation device based on vertebra positioning, and the spinal image segmentation device based on vertebra positioning can be integrated into electronic devices such as computers, servers, computers, server clusters, and data centers. Figure 1 FIG. 1 is a flowchart of a spinal image segmentation method based on vertebral positioning according to an embodiment of the present application; wherein the spinal image segmentation method based on vertebral positioning includes:

[0029] S101, acquiring a spinal medical image;

[0030] In this application, it refers to 3D volume data of CT or MRI (such as DICOM sequence or NIfTI file), which contains the complete spinal region (cervical vertebrae, thoracic vertebrae, lumbar vertebrae, etc.) and has a sufficiently high resolution (such as CT slice thickness ≤ 1mm).

[0031] Preferably, the spinal medical image is preprocessed: the grayscale range of the image is adjusted (such as the Hounsfield unit of CT is truncated to [-1000, 2000]) to highlight the bones; and the long axis of the spine is ensured to be consistent with a certain coordinate axis of the image (usually the Z axis) to align the direction.

[0032] S102, inputting the spinal medical image into the contour correction structure to obtain a correction mask;

[0033] S103: Input the corrected mask and the spinal medical image into the segmentation structure to obtain an image segmentation result.

[0034] In this application, the contour correction structure and the segmentation structure are two parallel branches. The branch of the contour correction structure is used to generate the overall contour of each vertebra of the spine. The vertebrae of the overall contour are independent and not adhered to each other. The segmentation structure branch is used to accurately segment the spinal medical image based on the overall contour, thereby ensuring that the segmentation results are not adhered.

[0035] In the present application, the segmentation is corrected by generating a correction mask, thereby avoiding the problem of vertebral adhesion or inaccurate vertebral numbering in the segmentation result.

[0036] In this application, the precise segmentation of the spinal medical image is corrected by identifying and positioning the contours, thereby fundamentally dividing the different vertebrae, avoiding the problem of vertebral adhesion and the problem of inaccurate vertebral numbering (inaccurate vertebral numbering is due to the confusion of numbers caused by vertebral adhesion).

[0037] In one embodiment, combined Figure 2 As shown, the step S102 of inputting the spinal medical image into the contour correction structure to obtain a correction mask includes:

[0038] Downsampling the spinal medical image to obtain a thumbnail feature map;

[0039] Resize the thumbnail feature map to obtain a scaled feature map;

[0040] The scaled feature map is input into the segmentation module to obtain the modified mask.

[0041] In this application, the overall contour of the spine is retained through contour correction, and the amount of processed data is reduced while satisfying the positioning of the lumbar vertebrae and sacral vertebrae.

[0042] In the present application, the amount of data processed by the segmentation module is greatly reduced by downsampling the spinal medical image first and then performing the segmentation module.

[0043] In this application, the thumbnail feature image is reset to a scaled feature image by resizing, and the aspect ratio of the scaled feature image is the same as that of the spinal medical image. When resizing, an appropriate scaling ratio is calculated to ensure that pixels are added to the width or length of the thumbnail feature image to obtain the scaled feature image.

[0044] Preferably, in the present application, the scaled feature map is obtained by adding pixels on the left and right sides of the thumbnail feature map, or by adding pixels on the upper and lower sides of the thumbnail feature map shown; in this way, the added size of the thumbnail feature map can be minimized.

[0045] Preferably, the sizes added on both sides of the scaled feature map are the same, so that after the addition, the outline of the spine is located in the middle.

[0046] In this application, the segmentation module can be a U-Net network; the scaled feature map is input into the U-Net network to perform coarse segmentation, and the coarse segmentation result is obtained, which is the corrected mask. Among them, the training loss of the U-Net network can use Dice Loss.

[0047] In this application, before downsampling the spinal medical image to obtain the thumbnail feature map, the following steps are also included:

[0048] Obtain the boundary threshold between bone tissue and soft tissue in CT images;

[0049] The contour image is processed based on the demarcation threshold to obtain a demarcation image;

[0050] Perform a closing operation on the boundary image to obtain a filled boundary image;

[0051] Based on the connected domain analysis, the largest connected region in the boundary image is selected;

[0052] Calculate the minimum bounding box of the connected area as the boundary of the contour image.

[0053] In the present application, the grayscale threshold for distinguishing bone tissue from soft tissue may be determined by empirically setting a threshold (eg, 200HU).

[0054] In this application, the grayscale values of CT images are based on the HU scale, where bone tissue (such as vertebral bodies) typically ranges from 200 to 1000 HU, and soft tissue (muscle, fat) is usually -100 to 200 HU.

[0055] In the present application, the Otsu algorithm or histogram analysis is used to automatically determine the optimal threshold value, thereby being applicable to spinal images with different scanning conditions or contrast changes.

[0056] In this application, the demarcation image is generated based on the demarcation threshold. The CT image is converted into a binary mask to clearly distinguish bone tissue from the background. The voxel values in the CT image are compared with the demarcation threshold: if the voxel value is ≥ the threshold, it is marked as foreground (bone tissue, value 1); if the voxel value is < the threshold, it is marked as background (soft tissue / air, value 0).

[0057] In this application, a closing operation is performed on the demarcated image to repair breaks in the bone tissue area and smooth edges, ensuring the connectivity of the main spine. Closing operations include: Dilation: dilating the foreground area with a structuring element (such as a spherical or cubic core), filling small holes and connecting adjacent bone fragments; Erosion: contracting the dilated area with the same structuring element, restoring the original general shape while maintaining connectivity.

[0058] In the closing operation, the size of the structural element is usually 3×3×3 voxels to avoid incorrect merging of structures such as the spine and ribs due to being too large.

[0059] In this application, connected domain analysis screens the largest region, identifying the main spine in the image while excluding other bone structures or noise. The filled boundary image is scanned, and each connected foreground region is assigned a unique label. The number of voxels (volume) in each connected domain is calculated and sorted from largest to smallest. The largest connected domain is the spine region.

[0060] In this application, the minimum bounding box is calculated as the contour boundary to determine the spatial range of the cropped spinal medical image and focus on the target area. The bounding box calculation includes: counting the minimum and maximum coordinate values of the target connected domain along the X, Y, and Z axes respectively to determine the minimum bounding box range; expanding 5 to 10 mm (or 5 to 10 voxels) in each direction to avoid cutting into the edge of the spine during cropping; and mapping the bounding box back to the original CT image coordinate system for accurate cropping of the area to be segmented.

[0061] In this application, by cropping the spinal medical image, the amount of calculation is reduced and the target area is focused to avoid redundant background interference.

[0062] In one embodiment, combined Figure 3 As shown, the step S103 of inputting the corrected mask and the spinal medical image into the segmentation structure to obtain an image segmentation result includes:

[0063] Performing block processing on the spinal medical image to obtain a spinal block image;

[0064] Performing block processing on the corrected mask to obtain a mask block image;

[0065] Input the spine block image and mask block image into the improved Unet structure to obtain the block segmentation result;

[0066] Aggregation processing is performed on the block segmentation results to obtain the image segmentation result.

[0067] In this application, the spinal medical image and the corrected mask are divided into patches in the same way, and then each patch and its corresponding corrected mask patch are sent to the improved Unet structure for fine segmentation, and finally the patch fine segmentation results are spliced into a complete fine segmentation result.

[0068] In one embodiment, the spine block image and the mask block image have the same aspect ratio.

[0069] In this application, the correction mask is the segmentation result after downsampling of the spinal medical image, and the scaling ratio is retained by resizing. In this way, after the same patching, each patch of the spinal medical image and each patch of the correction mask still retain the corresponding scaling size.

[0070] In the present application, the scaling size is retained, thereby facilitating the correction of the spinal medical image by the correction mask.

[0071] In the present application, the block processing of the spinal medical image and the aggregation processing of the segmentation results are reversible operations, so that the inverse operation of the block processing can be used to directly aggregate and ensure the aggregation effect.

[0072] In this application, by first dividing the image into blocks, then segmenting it, and finally aggregating it, the amount of data processed at a single time can be greatly reduced, and the loss of details due to reducing the image resolution can be avoided; the segmentation focuses on the anatomical structure within the local patch, avoiding complex background interference (such as rib overlap and metal artifacts).

[0073] In this application, by dividing into blocks and then segmenting, large-size images can be segmented in parallel to improve the segmentation speed.

[0074] Preferably, overlap cropping and weighted fusion are used for patch edge areas to avoid splicing artifacts. In this way, a 10% to 20% overlap area is retained between adjacent patches, and a weighted average (such as Gaussian weighting) is used for smooth transition, which can ensure the robustness of the inverse operation result aggregation.

[0075] In one embodiment, combined Figure 3 As shown, the spine block image and the mask block image are input into the improved Unet structure to obtain the block segmentation result, including:

[0076] Continuously downsampling the spinal block image to obtain multiple downsampled images;

[0077] Perform feature extraction on the mask block image to obtain a feature extraction map;

[0078] After fusing the downsampled image and the feature extraction image, continuous downsampling is performed to obtain multiple downsampled feature images;

[0079] The downsampled image and the downsampled feature map are upsampled to obtain the block segmentation result.

[0080] In one embodiment, combined Figure 3 As shown, after the down-sampling map and the feature extraction map are fused, continuous down-sampling processing is performed to obtain multiple down-sampling feature maps, including:

[0081] Get the minimum size downsampled image;

[0082] Input the minimum-sized downsampled image and the feature extraction image into the fusion attention module to obtain the fused feature map;

[0083] Downsampling the fused feature map to obtain a downsampled feature map;

[0084] The downsampled feature map is downsampled twice to obtain the downsampled feature map of the minimum size.

[0085] In one embodiment, combined Figure 3 As shown, the upsampling of the downsampled image and the downsampled feature image to obtain the block segmentation result includes:

[0086] Adaptively adjust the minimum size downsampled feature map to obtain the upsampled map;

[0087] Repeatedly upsample the upsampled image and fuse it with the adaptively adjusted downsampled image of the corresponding size to obtain multiple upsampled images;

[0088] The up-sampled image of the maximum size is subjected to convolution processing to obtain the block segmentation result.

[0089] In this application, different sizes are set from large to small as L4-L1, where L4 is the largest and L1 is the smallest.

[0090] Combine Figure 3 As shown, the size of the spine block image is L4, and through continuous preset operation processing, a down-sampling image of L4 size is obtained; the down-sampling image of L4 size is down-sampled and continuously preset operated to obtain a down-sampling image of L3 size, and the down-sampling image of L3 size is the minimum size down-sampling image.

[0091] Combine Figure 3 As shown in the figure, the size of the mask block image is L3, which is fused with the downsampled image of L3 size into the fusion attention module to obtain a fused feature map of L3 size.

[0092] Combine Figure 3 As shown, the fused feature map of size L3 is downsampled and continuously preset to obtain a downsampled feature map of size L2, and the downsampled feature map of size L2 is downsampled to obtain a downsampled feature map of size L1, which is the downsampled feature map of size L1.

[0093] It should be noted that, in this application, L1 and L2 sizes are downsampled feature maps, and L3 and L4 sizes are downsampled maps.

[0094] Combine Figure 3 As shown, the down-sampled feature map of size L1 is adaptively adjusted to obtain an up-sampled map of size L1; the down-sampled feature map of size L2 is adaptively adjusted and fused with the up-sampled up-sampled map of size L1 (the size after upsampling is L2) to obtain an up-sampled map of size L2; the down-sampled map of size L3 is adaptively adjusted and fused with the up-sampled up-sampled map of size L2 (the size after upsampling is L3) to obtain an up-sampled map of size L3; the down-sampled map of size L4 is adaptively adjusted and fused with the up-sampled up-sampled map of size L3 (the size after upsampling is L4) to obtain an up-sampled map of size L4; the up-sampled map of size L4 is convolved (not shown in the figure) to obtain the block segmentation result.

[0095] Preferably, the continuous preset operation processing in the downsampling process is a continuous Resnet processing.

[0096] In this application, the fusion attention module can be pixel superposition, so that the pixel values of the segmented area in the mask block image (pixels outside the segmented area are 0) can be used to accumulate the downsampled image, increase the pixel values of the segmented area in the downsampled image, and achieve the purpose of increasing the weight.

[0097] In another embodiment, the fusion attention module can accumulate the mask block image and the downsampled image (add pixel values), and then perform spatial attention mechanism processing and channel attention mechanism processing to obtain a fused feature map. In this way, spatial features and channel features can be extracted to improve the recognition characteristics of the segmentation contour.

[0098] In one embodiment, the step S102 of cropping the spinal medical image to obtain an image of the region to be segmented includes:

[0099] Performing contour extraction processing on the spinal medical image to obtain a contour image;

[0100] Crop the contour image to obtain the boundary of the contour image;

[0101] The boundary of the contour image is inversely mapped to the spinal medical image according to the contour extraction to obtain the image of the area to be segmented.

[0102] In this application, spinal medical images are cropped to reduce the amount of calculation and focus on the target area to avoid redundant background interference.

[0103] Preferably, adaptive adjustment is performed after obtaining the contour image to avoid excessive cropping and loss of vertebrae.

[0104] Preferably, the fracture contours are connected by a closing operation (dilation followed by erosion) to eliminate isolated edges caused by soft tissue or noise.

[0105] In this application, the minimum bounding rectangle of the contour is calculated, the contour area within the bounding box is retained, and irrelevant background is removed. The bounding box of the contour image is mapped back to the original spinal medical image to maintain the same spatial resolution.

[0106] In this application, a high-resolution image of the area to be segmented is output, which only includes the main body of the spine and adjacent tissues.

[0107] Preferably, the processing of the adaptive adjustment module includes:

[0108] Divide the downsampled image into blocks to obtain independent blocks;

[0109] For each independent block, obtain a first neighboring block and a second neighboring block with different spacings;

[0110] generating a first feature block based on the independent block and the first neighborhood block;

[0111] generating a second feature block based on the independent block and the second neighborhood block;

[0112] Performing feature compression on the first feature block and the second feature block to obtain a compressed block;

[0113] Iterate over all independent blocks and generate adjusted downsampled maps based on the resulting compressed blocks.

[0114] In this application, the downsampled image is divided into blocks (it should be noted that the block division in this step is not the same as the aforementioned Patch), that is, the downsampled image is divided into corresponding image blocks through a checkerboard; wherein, the image block can be pixel level (that is, each pixel is an image block) or other levels, and the specific division shall be based on the actual processing situation.

[0115] In this application, a sliding window or a fixed step size is used to divide the image into blocks of the same size.

[0116] It should be noted that if the downsampled image is a three-dimensional image, a surface is selected for chessboard division, and each grid is a strip grid with a lot of depth (the depth is the depth of the three-dimensional image), and the strip grid is an image block.

[0117] In this application, an image block is selected as an independent block. The adjacent image blocks above, below, to the left, and to the right of the independent block are the first neighboring blocks. The image blocks above, below, to the left, and to the right of the independent block, separated by one grid, are the second neighboring blocks. The first and second neighboring blocks have different spacing from the independent block.

[0118] In this application, the neighborhood information of each independent block is extracted to capture the local structure.

[0119] In the present application, the first feature block is generated to generate a local feature representation using an independent block and its first neighborhood block. Specifically, the independent block and the first neighborhood block are processed by a convolution layer and an attention layer to obtain the first feature block.

[0120] In this application, the specific structure and specific parameters of the convolutional layer and the attention layer can be obtained according to the training data or determined according to the actual situation.

[0121] It should be noted that, in this application, there are four first neighborhood blocks and multiple first feature blocks.

[0122] In this application, the independent block and the first neighborhood block are processed by the convolution layer and the attention layer to obtain the first feature block. The specific process is: the independent block and the four neighborhood blocks are spliced together to form a multi-channel input, and the convolution layer is used to extract features from the spliced block; the self-attention mechanism or the channel attention mechanism is used to enhance important features, calculate the attention weight, and weight the convolution layer output to enhance important features; the output of the attention layer is split into multiple feature blocks, each feature block corresponds to the processing results of the independent block and at least one neighborhood block.

[0123] In this application, a second feature block is generated to generate a broader local feature representation using an independent block and its second neighborhood block. The specific generation process is the same as that of the first feature block, except that the parameters of the convolution layer and the attention layer are different.

[0124] In this application, the generated feature blocks are compressed into a more compact representation to reduce the amount of computation and retain key information. Feature compression is performed using pooling operations (such as maximum pooling or average pooling) or fully connected layers.

[0125] In this way, through compression, the plurality of first characteristic blocks and second characteristic blocks are compressed into a compressed block, which corresponds to the size and position of the independent block and is used to replace the independent block. All image blocks are replaced by the compressed block to obtain a processed spinal medical image.

[0126] In the present application, each image block of the spinal medical image is traversed in a traversal manner to obtain a corresponding compressed block.

[0127] In this application, for image blocks / independent blocks near edges, their first and second neighboring blocks are incomplete. In this case, they are completed by copying the first and second neighboring blocks at relatively similar positions. For example, if the first neighboring block above the independent block does not exist, the first neighboring block below is copied and used as the block above.

[0128] In this application, the processing accuracy of the adjacent image blocks is greatly improved by completion.

[0129] In this application, the similarity relationship between local regions is captured through an adaptive adjustment module, thereby enhancing feature representation.

[0130] An embodiment of the present application provides a spinal image segmentation device based on vertebral positioning, which is used to execute a spinal image segmentation method based on vertebral positioning described above in the present application. The spinal image segmentation device based on vertebral positioning is described in detail below.

[0131] like Figure 4 As shown, the spinal column image segmentation device based on vertebra positioning includes:

[0132] An image acquisition module 101 is used to acquire spinal medical images;

[0133] An image correction module 102 is used to input a spinal medical image into a contour correction structure to obtain a correction mask;

[0134] The image segmentation module 103 is used to input the corrected mask and the spinal medical image into the segmentation structure to obtain an image segmentation result.

[0135] In one embodiment, the image correction module 102 is further configured to:

[0136] Downsampling is performed on the spinal medical image to obtain a thumbnail feature map; the thumbnail feature map is resized to obtain a scaled feature map; and the scaled feature map is input into a segmentation module to obtain the modified mask.

[0137] In one embodiment, the image segmentation module 103 is further configured to:

[0138] The spinal medical image is divided into blocks to obtain a spinal block image; the corrected mask is divided into blocks to obtain a mask block image; the spinal block image and the mask block image are input into the improved Unet structure to obtain a block segmentation result; the block segmentation result is aggregated to obtain the image segmentation result.

[0139] In one embodiment, the spine block image and the mask block image have the same aspect ratio.

[0140] In one embodiment, the image segmentation module 103 is further configured to:

[0141] The spine block image is continuously downsampled to obtain multiple downsampled images; the mask block image is feature extracted to obtain a feature extraction image; the downsampled image and the feature extraction image are fused and continuously downsampled to obtain multiple downsampled feature images; the downsampled image and the downsampled feature image are upsampled to obtain the block segmentation result.

[0142] In one embodiment, the image segmentation module 103 is further configured to:

[0143] Obtain a downsampled image of minimum size; input the downsampled image of minimum size and the feature extraction image into the fusion attention module to obtain a fused feature map; downsample the fused feature map to obtain a downsampled feature map; downsample the downsampled feature map twice to obtain a downsampled feature map of minimum size.

[0144] In one embodiment, the image segmentation module 103 is further configured to:

[0145] The down-sampled feature map of the smallest size is adaptively adjusted to obtain an up-sampled map; the up-sampled map is repeatedly up-sampled and fused with the down-sampled map of the corresponding size after adaptive adjustment to obtain multiple up-sampled maps; the up-sampled map of the largest size is convolved to obtain the block segmentation result.

[0146] The above-mentioned embodiment of the present application provides a spinal image segmentation device based on vertebral positioning and the embodiment of the present application provides a spinal image segmentation method based on vertebral positioning based on vertebral positioning based on the same inventive concept and has the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0147] The above describes the internal functions and structure of a spinal image segmentation device based on vertebral positioning, such as Figure 5 As shown, in practice, the spinal column image segmentation device based on vertebra positioning can be implemented as an electronic device, including: a memory 301 and a processor 303.

[0148] The memory 301 may be configured to store programs.

[0149] In addition, the memory 301 may also be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device, contact data, phone book data, messages, pictures, videos, etc.

[0150] The memory 301 can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0151] The processor 303 is coupled to the memory 301 and is configured to execute a program in the memory 301 to:

[0152] Acquire medical images of the spine;

[0153] Inputting the spinal medical image into the contour correction structure to obtain a correction mask;

[0154] The corrected mask and the spinal medical image are input into the segmentation structure to obtain the image segmentation result.

[0155] In one embodiment, the processor 303 is further configured to:

[0156] Downsampling is performed on the spinal medical image to obtain a thumbnail feature map; the thumbnail feature map is resized to obtain a scaled feature map; and the scaled feature map is input into a segmentation module to obtain the modified mask.

[0157] In one embodiment, the processor 303 is further configured to:

[0158] The spinal medical image is divided into blocks to obtain a spinal block image; the corrected mask is divided into blocks to obtain a mask block image; the spinal block image and the mask block image are input into the improved Unet structure to obtain a block segmentation result; the block segmentation result is aggregated to obtain the image segmentation result.

[0159] In one embodiment, the spine block image and the mask block image have the same aspect ratio.

[0160] In one embodiment, the processor 303 is further configured to:

[0161] The spine block image is continuously downsampled to obtain multiple downsampled images; the mask block image is feature extracted to obtain a feature extraction image; the downsampled image and the feature extraction image are fused and continuously downsampled to obtain multiple downsampled feature images; the downsampled image and the downsampled feature image are upsampled to obtain the block segmentation result.

[0162] In one embodiment, the processor 303 is further configured to:

[0163] Obtain a downsampled image of minimum size; input the downsampled image of minimum size and the feature extraction image into the fusion attention module to obtain a fused feature map; downsample the fused feature map to obtain a downsampled feature map; downsample the downsampled feature map twice to obtain a downsampled feature map of minimum size.

[0164] In one embodiment, the processor 303 is further configured to:

[0165] The down-sampled feature map of the smallest size is adaptively adjusted to obtain an up-sampled map; the up-sampled map is repeatedly up-sampled and fused with the down-sampled map of the corresponding size after adaptive adjustment to obtain multiple up-sampled maps; the up-sampled map of the largest size is convolved to obtain the block segmentation result.

[0166] In this application, Figure 5 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 5 Components shown.

[0167] The electronic device provided in this embodiment is based on the same inventive concept as the spinal image segmentation method based on vertebral positioning provided in an embodiment of the present application, and has the same beneficial effects as the method adopted, run or implemented by the application stored therein.

[0168] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0169] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0170] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0171] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0172] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0173] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0174] The present application also provides a computer-readable storage medium corresponding to a spinal image segmentation method based on vertebral positioning provided in the aforementioned embodiment, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it will execute a spinal image segmentation method based on vertebral positioning provided in any of the aforementioned embodiments.

[0175] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0176] The computer-readable storage medium provided in the above-mentioned embodiment of the present application and the spinal image segmentation method based on vertebral positioning provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0177] It should be noted that, in the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this description.

[0178] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0179] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A spinal column image segmentation method based on vertebral positioning, characterized in that: include: Acquire medical images of the spine; Inputting the spinal medical image into the contour correction structure to obtain a correction mask; The corrected mask and the spinal medical image are input into the segmentation structure to obtain the image segmentation result.

2. The spinal column image segmentation method based on vertebral positioning according to claim 1, characterized in that: The step of inputting the spinal medical image into the contour correction structure to obtain a correction mask comprises: Downsampling the spinal medical image to obtain a thumbnail feature map; Resize the thumbnail feature map to obtain a scaled feature map; The scaled feature map is input into the segmentation module to obtain the modified mask.

3. The spinal column image segmentation method based on vertebral positioning according to claim 1 or 2, characterized in that: The step of inputting the corrected mask and the spinal medical image into the segmentation structure to obtain an image segmentation result includes: Performing block processing on the spinal medical image to obtain a spinal block image; Performing block processing on the corrected mask to obtain a mask block image; Input the spine block image and mask block image into the improved Unet structure to obtain the block segmentation result; Aggregation processing is performed on the block segmentation results to obtain the image segmentation result.

4. The spinal column image segmentation method based on vertebral positioning according to claim 3, characterized in that: The spine block image and the mask block image have the same aspect ratio.

5. The spinal column image segmentation method based on vertebral positioning according to claim 3, characterized in that: The spine block image and the mask block image are input into the improved Unet structure to obtain the block segmentation result, including: Continuously downsampling the spinal block image to obtain multiple downsampled images; Perform feature extraction on the mask block image to obtain a feature extraction map; After fusing the downsampled image and the feature extraction image, continuous downsampling is performed to obtain multiple downsampled feature images; The downsampled image and the downsampled feature map are upsampled to obtain the block segmentation result.

6. The spinal column image segmentation method based on vertebral positioning according to claim 5, characterized in that: After fusing the downsampled image and the feature extraction image, continuous downsampling is performed to obtain multiple downsampled feature images, including: Get the minimum size downsampled image; Input the minimum-sized downsampled image and the feature extraction image into the fusion attention module to obtain the fused feature map; Downsampling the fused feature map to obtain a downsampled feature map; The downsampled feature map is downsampled twice to obtain the downsampled feature map of the minimum size.

7. The spinal column image segmentation method based on vertebral positioning according to claim 5, characterized in that: The upsampling of the downsampled image and the downsampled feature image to obtain a block segmentation result includes: Adaptively adjust the minimum size downsampled feature map to obtain the upsampled map; Repeatedly upsample the upsampled image and fuse it with the adaptively adjusted downsampled image of the corresponding size to obtain multiple upsampled images; The up-sampled image of the maximum size is subjected to convolution processing to obtain the block segmentation result.

8. A spinal image segmentation device based on vertebra positioning, characterized in that: include: An image acquisition module, which is used to acquire spinal medical images; An image correction module is used to input a spinal medical image into a contour correction structure to obtain a correction mask; The image segmentation module is used to input the corrected mask and the spinal medical image into the segmentation structure to obtain the image segmentation result.

9. An electronic device, characterized in that: include: memory and processor; The memory is used to store programs; The processor, coupled to the memory, is configured to execute the program to: Acquire medical images of the spine; Inputting the spinal medical image into the contour correction structure to obtain a correction mask; The corrected mask and the spinal medical image are input into the segmentation structure to obtain the image segmentation result.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the spinal column image segmentation method based on vertebral positioning according to any one of claims 1 to 7.

Images