A method and system for multi-spinal structure segmentation based on CT images

Abstract

The application provides a multi-spinal column structure segmentation method and system based on a CT image, and belongs to the technical field of medical image segmentation; a CT image to be segmented and a plurality of historical CT images are acquired and preprocessed; subsequently, the historical CT images are input into a deep neural network for training; the CT image to be segmented is input into the trained deep neural network, and the deep neural network is composed of a bone tissue segmentation network and a soft tissue enhancement network; the vertebral body features of the CT image are extracted by using the bone tissue segmentation network to obtain a vertebral body segmentation image; the soft tissue structure of the CT image is segmented by using the soft tissue enhancement network to obtain a disc segmentation image and a nerve segmentation image; and the vertebral body segmentation image, the disc segmentation image and the nerve segmentation image are integrated to obtain a complete lumbar vertebra structure segmentation image. The application can simultaneously and efficiently automatically segment the vertebral body, the disc and the nerve in the spinal column CT image.

Description

Technical Field

[0001] This invention belongs to the field of medical image segmentation technology, and in particular relates to a method and system for segmenting multiple spinal structures based on CT images. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the increasing aging population and changes in lifestyles and work patterns, the prevalence of lumbar spine diseases is gradually rising. Currently, clinical diagnosis and treatment of lumbar spine diseases still rely on doctors' analysis and judgment, which is not only time-consuming but also easily influenced by the doctor's subjective experience and condition, and can no longer meet current diagnostic and treatment needs. Therefore, developing an automated lumbar spine structure segmentation system is crucial.

[0004] The segmentation of lumbar spine anatomy is primarily based on MRI images because of MRI's superior imaging capabilities for soft tissues. However, MRI examinations take significantly longer than CT scans, meaning patients seeking MRI may have to wait hours or even days, undoubtedly causing inconvenience for outpatients. Furthermore, MRI scans for the same area are generally more expensive than CT scans, increasing the financial burden on patients. Moreover, some special patients cannot undergo MRI (such as those who have had pacemakers implanted or those who cannot maintain a supine position for extended periods due to pain). Therefore, after weighing the pros and cons, spinal surgeons prioritize the faster and relatively cheaper CT scan for outpatients with lumbar spine diseases; MRI is only performed if the CT scan reveals significant lesions in the lumbar spine. However, CT scans are not as effective at imaging soft tissues as MRI, and diagnosing lumbar spine diseases via CT often requires doctors with extensive clinical experience, placing higher demands on their skills and increasing their workload.

[0005] Therefore, there is an urgent need to establish a new CT image segmentation method to assist radiologists and spinal surgeons in quickly and accurately diagnosing lumbar spine diseases using only CT images. Furthermore, it should be possible to intuitively understand the location of intervertebral disc herniation, as well as the specific location and severity of spinal cord and nerve root compression in 3D mode. This would allow spinal surgeons to gain a more detailed understanding of the specific anatomical structure of each patient's lumbar spine before and after lumbar spine surgery, which would have significant guiding significance for both preoperative and postoperative assessments of lumbar spine surgery.

[0006] The inventors discovered that existing CT image segmentation methods still have some technical problems, such as:

[0007] (1) For example, a two-stage AIS spinal CT image multi-label segmentation method based on deep learning, as described in patent application number 202410283675.5, trains a convolutional neural network for vertebral localization by creating new labels for vertebral localization in the image dataset, and then uses this convolutional neural network to locate vertebrae in CT images. It is evident that existing CT image segmentation methods can only segment a single spinal structure (vertebra) and cannot simultaneously segment other adjacent structures.

[0008] (2) Compared with MRI images, CT images have lower soft tissue resolution, resulting in lower differentiation between neural structures and surrounding tissues. This may lead to reduced visibility of details in CT images, increased misidentification rate, and ultimately make it difficult to identify soft tissue areas.

[0009] (3) Artifacts may be present in bones in CT images. These artifacts can affect the display of soft tissue structures such as intervertebral discs and nerves around the bones. Therefore, the presence of artifacts may cause neural networks to ignore some structural features during learning, which are important for 3D visualization and clinical diagnosis and treatment. Summary of the Invention

[0010] To overcome the shortcomings of the prior art, the present invention provides a multi-spine structure segmentation method and system based on CT images, which can efficiently and automatically segment vertebrae, intervertebral discs and nerves in spinal CT images simultaneously.

[0011] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0012] The first aspect of the present invention provides a method for segmenting multiple spinal structures based on CT images.

[0013] A multi-spine structure segmentation method based on CT images includes:

[0014] The CT image to be segmented and multiple historical CT images are acquired and respectively formed into a dataset to be segmented and a historical dataset; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spine diseases and spinal CT images of normal lumbar spine.

[0015] The dataset to be segmented and the historical dataset are preprocessed separately;

[0016] Multiple historical CT images from the preprocessed historical dataset are input into a deep neural network to train the deep neural network;

[0017] The CT images to be segmented in the dataset are input into a trained deep neural network, which consists of a bone tissue segmentation network and a soft tissue enhancement network. Specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT images to obtain vertebral body segmentation images. The soft tissue enhancement network is used to segment the soft tissue structures of the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images.

[0018] The obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image.

[0019] Furthermore, the preprocessing includes: cropping, slice selection, volume resampling, and normalization.

[0020] Furthermore, during the cropping process, all image slices are cropped to the same size; then, slice selection is performed on all images in the CT image dataset to remove regions of interest that do not exist in the horizontal plane; subsequently, the volume of all images in the image dataset is resampled to the median voxel spacing of the corresponding dataset.

[0021] Furthermore, the soft tissue enhancement network includes a bone tissue removal module, a grayscale self-adjustment module, and a soft tissue segmentation module.

[0022] Furthermore, the bone tissue removal module performs a zero-to-one operation on the vertebral body features extracted by the bone tissue segmentation network, and then adds them to the initial image to eliminate the influence of the vertebral body region in the CT image.

[0023] Furthermore, the grayscale self-adjustment module automatically adjusts and selects the optimal grayscale value in the CT image through a loss function to enhance the display of soft tissue in the image; then the enhanced soft tissue data is sent to the soft tissue segmentation module.

[0024] Furthermore, the soft tissue segmentation module includes: an encoding layer, a bottleneck layer, a decoding layer, and skip connections.

[0025] A second aspect of the present invention provides a multi-spine structure segmentation system based on CT images.

[0026] A multi-spinal structure segmentation system based on CT images, comprising:

[0027] The image acquisition module is configured to acquire a CT image to be segmented and multiple historical CT images and form a dataset to be segmented and a historical dataset, respectively; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spondylosis and spinal CT images of normal lumbar vertebrae;

[0028] The preprocessing module is configured to preprocess the dataset to be segmented and the historical dataset respectively;

[0029] The model training module is configured to input multiple historical CT images from the preprocessed historical dataset into a deep neural network and train the deep neural network.

[0030] The CT image segmentation module is configured to: input the CT images to be segmented from the dataset into a trained deep neural network, which consists of a bone tissue segmentation network and a soft tissue enhancement network; specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT images to obtain vertebral body segmentation images; the soft tissue enhancement network is used to segment the soft tissue structures of the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images; and the obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image.

[0031] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a multi-spine structure segmentation method based on CT images as described in the first aspect of the present invention.

[0032] The fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a multi-spine structure segmentation method based on CT images as described in the first aspect of the present invention.

[0033] The above one or more technical solutions have the following beneficial effects:

[0034] (1) In this invention, the CT image to be segmented is input into a trained deep neural network, which consists of a bone tissue segmentation network and a soft tissue enhancement network. Specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT image to obtain a vertebral body segmentation image. The soft tissue enhancement network is used to segment the soft tissue structure of the CT image based on the CT image and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images. Therefore, this invention can simultaneously segment the vertebral body, intervertebral disc, and nerve in a CT image, achieving simultaneous segmentation of multiple spinal structures.

[0035] (2) In segmenting CT images, this invention extracts vertebral body features from the CT image using a bone tissue segmentation network, and then simultaneously inputs these vertebral body features and the CT image to be segmented into a soft tissue enhancement network. The bone tissue removal module, grayscale self-adjustment module, and soft tissue segmentation module within the soft tissue enhancement network achieve feature enhancement and extraction in the soft tissue structure. Therefore, this invention can reduce the self-recognition rate of the network model and enhance the recognition effect on soft tissue regions.

[0036] (3) The present invention uses a 3D U-Net network as the backbone network, which can effectively reduce artifacts. Therefore, the neural network can pay attention to more detailed structural features during learning, and enhance the segmentation effect of multiple spinal structures in CT images.

[0037] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0039] Figure 1 This is a flowchart of a multi-spine structure segmentation method based on CT images in Embodiment 1 of the present invention.

[0040] Figure 2 This is a general framework diagram of a multi-spine structure segmentation method based on CT images in Embodiment 1 of the present invention.

[0041] Figure 3 This is a framework diagram of the bone tissue segmentation network in Embodiment 1 of the present invention.

[0042] Figure 4 This is a framework diagram of the soft tissue enhancement network in Embodiment 1 of the present invention.

[0043] Figure 5 This is a framework structure diagram of the grayscale adaptive module in Embodiment 1 of the present invention.

[0044] Figure 6 This is a schematic diagram illustrating the qualitative comparison of segmentation results in Embodiment 1 of the present invention. Detailed Implementation

[0045] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0046] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0047] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0048] Example 1

[0049] This embodiment discloses a method for segmenting multiple spinal structures based on CT images.

[0050] like Figure 1 As shown, a multi-spine structure segmentation method based on CT images includes:

[0051] Step S1: Obtain the CT image to be segmented and multiple historical CT images, and form the dataset to be segmented and the historical dataset respectively; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spondylosis and spinal CT images of normal lumbar vertebrae;

[0052] Step S2: Preprocess the dataset to be segmented and the historical dataset respectively;

[0053] Step S3: Input multiple historical CT images from the preprocessed historical dataset into a deep neural network to train the deep neural network;

[0054] Step S4: Input the CT images to be segmented from the dataset into a trained deep neural network. The deep neural network consists of a bone tissue segmentation network and a soft tissue enhancement network. Specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT images to obtain vertebral body segmentation images. The soft tissue enhancement network is used to segment the soft tissue structures of the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images. The obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image.

[0055] Based on the above steps, this invention can efficiently and automatically segment vertebral bodies, intervertebral discs, and nerves simultaneously in spinal CT images. To facilitate understanding of the technical solution of this invention, the above steps are further explained and illustrated, specifically:

[0056] Step S1: Obtain the CT image to be segmented and multiple historical CT images, and form the dataset to be segmented and the historical dataset respectively; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spondylosis and spinal CT images of normal lumbar vertebrae.

[0057] The historical dataset used in this embodiment was collected from Qilu Hospital of Shandong University (Qingdao Campus), and all data in the dataset (historical CT images) has been approved by the hospital's data ethics committee. The historical dataset includes 200 cases of lumbar spine CT plain scan data (from 200 different subjects); moreover, the voxel spacing of these CT images is 0.3467 × 0.3467 × 0.999 mm. 3 .

[0058] Step S2: Preprocess the dataset to be segmented and the historical dataset separately.

[0059] Preprocessing is performed on the dataset to be segmented and the historical dataset separately. Specifically, for each image in the dataset to be segmented and the historical dataset, preprocessing operations such as cropping, slice selection, volume resampling, and normalization are performed sequentially.

[0060] Step S2-1: During the cropping process, all image slices are cropped to the same size, that is, all image slices are cropped to 256*256, and each data (image slices cropped to 256*256) is cropped to the non-zero value area.

[0061] In practice, on the one hand, datasets may contain some extreme values, which may not be true characteristics of the data but rather noise or erroneous records. Cropping can limit these outliers to a certain range, preventing them from affecting model training or data analysis. On the other hand, combining cropping with normalization methods can keep the scales of various features consistent. Therefore, cropping is necessary to handle outliers and standardize the data. Furthermore, cropping based on step S2-1 can remove excessively bright or dark pixel values, making the remaining data more representative and usable, thus ensuring the quality of the image data. Further, cropping a 512*512 image slice into a 256*256 pixel image slice can significantly save GPU memory, thereby reducing training time.

[0062] Step S2-2: Perform slice selection on all images in the dataset to be segmented and the historical dataset, removing regions of interest that do not exist in the horizontal plane. First, find the slice range of the region of interest and expand the slice range; then, after removing too few slices, perform slice selection again.

[0063] Step S2-3: Resample the volumes of all images in the dataset to be segmented and the historical dataset to the median voxel spacing of the corresponding dataset to reduce inconsistencies between different volumes.

[0064] The SimpleITK library (an open-source database for medical image processing) was used to process all images in the dataset to be segmented and the historical datasets, and the voxel spacing of the dataset was unified by calculating the median voxel spacing. Specifically, each image was resampled according to the target voxel spacing to ensure that all images had a consistent spatial resolution. The process involved: first, obtaining the voxel spacing of the images; then, calculating the median voxel spacing of all images in the image dataset; next, resampling the images to the target voxel spacing using SimpleITK's ResampleImageFilter for interpolation resampling; subsequently, resampling was performed on each image in the dataset and the image file was read; finally, the resampled images were saved.

[0065] Step S2-4: Perform volumetric strength trimming through normalization. Specifically, the normalization formula is expressed as:

[0066]

[0067] Where I′ represents the normalized image intensity, I represents the cropped image intensity, and I... 2％ I represents the 2nd percentile of the intensity of the cropped image. 98％ I represents the intensity of the cropped image. 98％ Percentiles. Thus, the volumetric strength value is clipped to the percentile of the entire volumetric strength value range [2.0, 98.0].

[0068] Step S3: Input multiple historical CT images from the preprocessed historical dataset into a deep neural network to train the deep neural network.

[0069] Historical CT images are used as training data to train a deep neural network model. The entire training process is iterative. Before iterative training, the number of epochs, the initial learning rate, and the weight decay values ​​are pre-set. An epoch is a unit for measuring the model's training progress; it refers to the process of the model traversing the entire training data once. In other words, when one epoch is completed, it means that the model has traversed every CT image in the training set (historical dataset) at least once. When training the deep neural network, the historical dataset can be divided into multiple batches, each containing a certain number of CT images. In each iteration, the model processes the samples in one batch, calculates the loss and gradient based on these samples, and then updates the model's weights. This process is repeated until all batches are processed, thus completing one epoch. Furthermore, since a single epoch may not be sufficient for the model to fully learn the features of the data, this embodiment chooses to train the model for multiple epochs. However, too many epochs can also lead to overfitting, meaning the model becomes too complex and begins to memorize the noise and details of individual samples in the historical dataset, rather than learning the general patterns of the data. Therefore, it is necessary to choose an appropriate number of epochs for training the deep neural network.

[0070] Furthermore, in this embodiment, the number of epochs is set to 30, with 200 iterations per epoch; simultaneously, the initial learning rate is set to 0.0001, and the weight decay value is set to 0.0001. Training is terminated if the model performance does not improve after 30 epochs.

[0071] Step S4: Input the CT image to be segmented into the trained deep neural network. For example... Figure 2 As shown, the deep neural network consists of a bone tissue segmentation network and a soft tissue enhancement network. Specifically, the bone tissue segmentation network extracts vertebral body features from CT images to obtain vertebral body segmentation images. The soft tissue enhancement network segments the soft tissue structures from the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images. The obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image. Specifically, step S4 can be implemented through the following process:

[0072] Step S4-1: Input the original 3D CT volume to be segmented into the Bone Tissue Segmentation Network (BTS-Net). After simple bone tissue segmentation, relatively accurate vertebral body features are obtained. The BTS-Net can divide the obtained vertebral body features into two parts (the two vertebral body features are completely identical). One part is output as the final result (i.e., the vertebral body image), and the other part is input into the Soft Tissue Enhancement Network (STE-Net).

[0073] like Figure 3 As shown, due to U-Net's excellent performance in medical image segmentation, this embodiment uses 3D U-Net as the backbone network, primarily to obtain more accurate vertebral body features. Because 3D U-Net can operate in three dimensions, it can better capture and preserve spatial information in the image, while also better handling object boundaries, reducing artifacts and breaks in the segmentation results. This helps improve the accuracy and stability of segmentation, especially in CT images with complex structures or noise. Furthermore, accurate segmentation results will facilitate subsequent soft tissue enhancement.

[0074] Step S4-2: The soft tissue enhancement network includes a bone tissue removal module, a grayscale self-adjustment module, and a soft tissue segmentation module. The original 3D CT volume is input into the soft tissue enhancement network (STE-Net).

[0075] To more effectively segment low-resolution soft tissue in CT image data, this embodiment proposes a soft tissue enhancement network (STE-Net). STE-Net consists of three parts: a bone tissue removal module, a grayscale self-adjusting module, and a soft tissue segmentation network. In this embodiment, vertebral body features obtained from BTS-Net and the original CT image are first fed into the bone tissue removal module, making the grayscale value of the vertebral body region negative. Then, the grayscale self-adjusting module (GSAM) automatically adjusts and selects the optimal grayscale value in the CT image using a loss function, thereby enhancing the display of soft tissue in the image. The enhanced soft tissue data is then fed into the backbone network U-Net, outputting the segmentation results for nerves and intervertebral discs. Finally, the segmentation results from BTS-Net and STE-Net are fused to obtain a complete segmentation. This can be achieved through the following process:

[0076] Step S4-2-1: Input the original 3D CT volume into the soft tissue enhancement network (STE-Net), and together with the input vertebral body feature information, clear the vertebral body region through the bone tissue removal module. This can reduce the complexity and difficulty of the task, while saving computational resources and improving efficiency.

[0077] Specifically, to eliminate the influence of the vertebral body region during soft tissue segmentation, this embodiment introduces a bone tissue removal module. This module performs a zero-to-one operation on the vertebral body features extracted by the bone tissue segmentation network, and then adds them to the initial image to eliminate the influence of the vertebral body region in the CT image, thereby improving the network's segmentation of nerves and intervertebral discs. Bone tissue removal can be described as follows:

[0078] x b =BTE-Net(x);

[0079] x e =x + x·arg(x) b -1;

[0080] Where x is the original input, x b For the output of the BTS-Net network, arg() and x e These are the outputs of the argmax function and the bone tissue removal module, respectively.

[0081] Step S4-2-2: Use the Gray self-adjusting module (GSAM) to automatically convert the bone window into a soft tissue window to discover more details of the soft tissue structure, thereby enhancing the network's ability to learn about soft tissue and improving the accuracy of the lumbar spine segmentation task.

[0082] like Figure 5 As shown, the grayscale self-adjustment module automatically adjusts and selects the optimal grayscale value in the CT image using a loss function to enhance the display of soft tissue in the image; then, the enhanced soft tissue data is sent to the soft tissue segmentation module. Specifically, the grayscale self-adjustment module adjusts the window width w in the CT image... w and window position w c This is to achieve the selection of an appropriate grayscale value. Among them, the window width w w The window level w determines the range of grayscale values ​​covered by the window. c This corresponds to the CT value for centering the window. In this embodiment, the initial window width w is set empirically. w and window position w c The values ​​are (40, 350); during network training, the window width w in the CT image is adjusted by learning two parameters, α and β. w and window position w c .like Figure 4 As shown, the original grayscale values ​​of the 3D CT image (x) are used as input, and the data after adjusting the grayscale value distribution is used as output data (x′) to improve the contrast of soft tissue structures and enhance the network's learning ability for neural and intervertebral disc structures. Specifically, firstly, a 1×1×1 convolution kernel with parameter θ is used to convolve the original input x to generate a c-channel feature map f, i.e.:

[0083] f = Conv(x,θ);

[0084] Where θ represents the convolution parameters, and Conv() represents the convolution operation.

[0085] Then, the feature map f is sent to two parallel branches, each learning adaptive parameters α and β to adjust the grayscale values ​​w and c of the window width and window level, respectively. Within each branch, a 1×1×1 convolution is performed on the feature map f to generate a 1-channel feature map, followed by global average pooling (GAP) and compression of the values ​​to the [0,1] range using a sigmoid activation function. The adaptive parameters α and β are expressed as follows:

[0086] α=Sigmoid(GAP(Conv(f,θ w )))+t;

[0087] β=Sigmoid(GAP(Conv(f,θ c )))+t;

[0088] Where Sigmoid() represents the activation function, GAP() represents the global average pooling operation, and θ w θ represents the convolution parameters, where α is the adaptive parameter. c Let β represent the convolution parameters of the adaptive parameter β, and t represent the hyperparameters.

[0089] This embodiment adds a hyperparameter t (usually set to 0.5) to ensure that the values ​​of α and β are within the range [0.5, 1.5], thus avoiding grayscale value ranges unsuitable for segmentation tasks. The learned parameters α and β are then used to modify the original w. w and w c Scale the image to obtain the adjusted w′ w and w′ c ,Right now:

[0090] w′ w =α·w w ;

[0091] w′ c =β·w c ;

[0092] Among them, w w The grayscale value representing the window width, w c The grayscale value representing the window level, w′ w w′ represents the grayscale value of the adjusted window width. c This represents the grayscale value of the adjusted window level.

[0093] Finally, the deep neural network performs windowing operations on the original input x, including normalization and thresholding. The normalization operation is based on the adjusted w′. w and w′ c Pixel values ​​are mapped to the range [0,1], and then thresholded to limit the values ​​to [0,255], resulting in the grayscale-adjusted image x′, i.e.:

[0094] y = Normalization(x) = (xw′) c ) / (w′ w -w′ c );

[0095] x′=Threshold(y)=min(1,max(0,y))·255;

[0096] Where y represents the intermediate process quantity after normalization, Normalization() represents the normalization operation; x′ represents the image after grayscale adjustment, Threshold() represents the threshold operation.

[0097] Step S4-2-3: Automatically segment soft tissue structures in the soft tissue segmentation module to obtain segmented images of intervertebral discs and nerves.

[0098] The soft tissue segmentation module includes: an encoding layer, a bottleneck layer, a decoding layer, and skip connections.

[0099] The Soft Tissue Segmentation Network is considered a variant of 3D U-Net, retaining a similar structure and functionality but differing in implementation and hierarchical details. Specifically, in the encoding phase, compared to the standard 3D U-Net, this embodiment introduces normalization, corrected linear units, and max-pooling layers sequentially after 3×3×3 convolutions; the normalization technique is BatchNorm3d, the corrected linear units are ReLU, and each dimension of the max-pooling layer uses 1×2×2 max-pooling. This embodiment uses normalization and a smaller pooling stride to improve network stability and training speed; furthermore, the smaller pooling stride enhances network stability and convergence speed while preserving more spatial details. After the encoding phase, the next layer is a bottleneck layer, used to map high-dimensional features to a low-dimensional space, thereby reducing computational cost while maintaining feature expressiveness. In the final layer, 1×1×1 convolutions are used to reduce the number of output channels to the number of labels; in this embodiment, the number of classes is 3.

[0100] Step S4-2-4: Integrate the obtained segmented images of the vertebral body, intervertebral disc, and nerve to obtain a complete lumbar spine structure segmentation result.

[0101] In this embodiment, the integration operation is implemented based on the `cat` operation within the `torch` package in Python. For example... Figure 6The diagram shown illustrates a qualitative comparison of the segmentation results of this invention. As can be seen from the diagram, this embodiment achieves automated multi-class segmentation of the vertebral body, intervertebral discs, and nerves based on CT images by employing a cascaded structure that first segments bone tissue and then soft tissue. Specifically: to obtain more accurate vertebral body features, a lightweight bone tissue segmentation network (BTS-Net) similar to 3D U-Net is innovatively proposed for simple bone tissue segmentation of the original CT data; to achieve automated enhancement and extraction of soft tissue features, a soft tissue enhancement network (STE-Net) is innovatively proposed; STE-Net first removes the vertebral body region from the vertebral body characteristics through a bone tissue removal module; then, it enhances the network's learning ability on soft tissue through a grayscale self-adjustment module (GSAM), improving the accuracy of lumbar spine anatomical segmentation tasks; finally, it segments the intervertebral discs and nerves through a soft tissue segmentation module. Because the Gray-Scale Adaptive Module (GASM) can automatically adjust the window width and window level of CT data to a suitable soft tissue window, it improves the contrast of soft tissue structures and enhances the ability of the novel segmentation network to identify and segment soft tissues. Furthermore, the simultaneous application of the Bone Removal Module (EBT) and the Gray-Scale Adaptive Module (GASM) produces a complementary effect, achieving optimal segmentation results under various segmentation conditions. In summary, the novel segmentation network proposed in this invention significantly outperforms other classical methods in segmenting vertebral bodies, intervertebral discs, and nerves, and this invention can simultaneously and efficiently segment vertebral bodies, intervertebral discs, and nerves in spinal CT images.

[0102] Based on the above process, the segmentation algorithm of this invention can more quickly and effectively segment vertebrae, nerves, and intervertebral discs from spinal CT medical images. With the assistance of the novel segmentation network, doctors can quickly and accurately diagnose lumbar spine diseases using only CT images. Furthermore, the segmentation accuracy is higher. This multi-spine structure segmentation method based on CT images allows doctors to have a sufficiently clear understanding of the relative location of the lumbar spine disease area before surgery, thus assisting in clinical disease diagnosis.

[0103] Example 2

[0104] This embodiment discloses a multi-spine structure segmentation system based on CT images.

[0105] A multi-spinal structure segmentation system based on CT images, comprising:

[0106] The image acquisition module is configured to acquire a CT image to be segmented and multiple historical CT images and form a dataset to be segmented and a historical dataset, respectively; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spine disease and spinal CT images of normal lumbar spine.

[0107] The preprocessing module is configured to preprocess the dataset to be segmented and the historical dataset respectively;

[0108] The model training module is configured to input multiple historical CT images from the preprocessed historical dataset into a deep neural network and train the deep neural network.

[0109] The CT image segmentation module is configured to: input the CT images to be segmented from the dataset into a trained deep neural network, which consists of a bone tissue segmentation network and a soft tissue enhancement network; specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT images to obtain vertebral body segmentation images; the soft tissue enhancement network is used to segment the soft tissue structures of the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images; and the obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image.

[0110] Example 3

[0111] The purpose of this embodiment is to provide a computer-readable storage medium.

[0112] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of a multi-spinal structure segmentation method based on CT images as described in Embodiment 1 of this disclosure.

[0113] Example 4

[0114] The purpose of this embodiment is to provide an electronic device.

[0115] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in a multi-spinal structure segmentation method based on CT images as described in Embodiment 1 of this disclosure.

[0116] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0117] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0118] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for segmenting multiple spinal structures based on CT images, characterized in that, include: The CT image to be segmented and multiple historical CT images are acquired and respectively formed into a dataset to be segmented and a historical dataset; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spine diseases and spinal CT images of normal lumbar spine. The dataset to be segmented and the historical dataset are preprocessed separately; Multiple historical CT images from the preprocessed historical dataset are input into a deep neural network to train the deep neural network; The CT images to be segmented in the dataset are input into a trained deep neural network, which consists of a bone tissue segmentation network and a soft tissue enhancement network. The soft tissue enhancement network includes a bone tissue removal module, a grayscale self-adjustment module, and a soft tissue segmentation module. Specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT images to obtain vertebral body segmentation images. The soft tissue enhancement network is used to segment the soft tissue structures of the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images. The obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image.

2. The method for segmenting multiple spinal structures based on CT images as described in claim 1, characterized in that, The preprocessing includes: cropping, slice selection, volume resampling, and normalization.

3. The method for segmenting multiple spinal structures based on CT images as described in claim 2, characterized in that, During the cropping process, all image slices are cropped to the same size; then, slice selection is performed on all images in the CT image dataset to remove regions of interest that do not contain the horizontal plane; subsequently, the volume of all images in the image dataset is resampled to the median voxel spacing of the corresponding dataset.

4. The method for segmenting multiple spinal structures based on CT images as described in claim 1, characterized in that, The bone tissue removal module performs a zero-to-one operation on the vertebral body features extracted by the bone tissue segmentation network, and then adds them to the initial image to eliminate the influence of the vertebral body region in the CT image.

5. The method for segmenting multiple spinal structures based on CT images as described in claim 1, characterized in that, The grayscale self-adjustment module automatically adjusts and selects the optimal grayscale value in the CT image through a loss function to enhance the display of soft tissue in the image; then the enhanced soft tissue data is sent to the soft tissue segmentation module.

6. The method for segmenting multiple spinal structures based on CT images as described in claim 1, characterized in that, The soft tissue segmentation module includes: an encoding layer, a bottleneck layer, a decoding layer, and skip connections.

7. A multi-spinal structure segmentation system based on CT images, characterized in that, include: The image acquisition module is configured to acquire a CT image to be segmented and multiple historical CT images and form a dataset to be segmented and a historical dataset, respectively; wherein, the multiple historical CT images include spinal CT images of patients with lumbar spine disease and spinal CT images of normal lumbar spine. The preprocessing module is configured to preprocess the dataset to be segmented and the historical dataset respectively; The model training module is configured to input multiple historical CT images from the preprocessed historical dataset into a deep neural network and train the deep neural network. The CT image segmentation module is configured to: input the CT images to be segmented from the dataset into a trained deep neural network, which consists of a bone tissue segmentation network and a soft tissue enhancement network; the soft tissue enhancement network includes a bone tissue removal module, a grayscale self-adjustment module, and a soft tissue segmentation module; specifically, the bone tissue segmentation network is used to extract vertebral body features from the CT images to obtain vertebral body segmentation images; the soft tissue enhancement network is used to segment the soft tissue structures of the CT images based on the CT images and the obtained vertebral body features to obtain intervertebral disc segmentation images and nerve segmentation images; the obtained vertebral body segmentation images, intervertebral disc segmentation images, and nerve segmentation images are integrated to obtain a complete lumbar spine structure segmentation image.

8. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by a processor, the program implements the steps in a multi-spinal structure segmentation method based on CT images as described in any one of claims 1-6.

9. An electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the multi-spine structure segmentation method based on CT images as described in any one of claims 1-6.

Images