Three-dimensional CT image reconstruction method and system for lumbar disease diagnosis

Abstract

The invention provides a three-dimensional CT image reconstruction method and system for lumbar disease diagnosis, and belongs to the technical field of three-dimensional medical image reconstruction. Training the deep neural network by acquiring a plurality of two-dimensional CT images; inputting a to-be-segmented two-dimensional CT image into the trained deep neural network for segmentation, and taking the segmented image as a pre-reconstructed image; and performing three-dimensional CT image reconstruction on the pre-reconstructed image based on a VTK rendering method so as to realize visualization of the three-dimensional CT image. According to the method, the vertebrae, the intervertebral disc and the neural structure in the lumbar vertebra CT image can be accurately and efficiently segmented at the same time, the segmented CT image is subjected to three-dimensional reconstruction to realize image visualization, and powerful support is provided for lumbar vertebra disease diagnosis, preoperative planning and intraoperative guidance.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional medical image reconstruction, and in particular relates to a three-dimensional CT image reconstruction method and system for lumbar disease diagnosis. Background Art

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

[0003] With the increasingly serious problem of population aging and the changes in modern life and work patterns, the incidence of spinal diseases has increased significantly, accounting for more than 40%; among them, lumbar diseases are particularly prominent, which not only erode the patient's nervous system function, but also greatly limit the ability to do daily activities, thereby reducing the overall quality of life. Common lumbar diseases include lumbar spinal stenosis (LSS), lumbar disc herniation (IDH) and lumbar deformity. LSS is the primary reason for patients over 65 years old to undergo spinal surgery. It is mainly due to the stenosis of the spinal canal or nerve root canal caused by bone hyperplasia or fibrous tissue hyperplasia, which in turn stimulates or compresses the spinal nerve roots and cauda equina, causing a series of clinical symptoms; IDH is caused by the protrusion of intervertebral disc tissue, compressing adjacent nerve roots, often manifested as low back pain and radiating pain in the lower limbs; and lumbar deformity is often accompanied by scoliosis or spondylolisthesis, which not only affects the beauty of the body, but is also likely to cause activity disorders and seriously damage the patient's quality of life. Given the complexity and comprehensiveness of lumbar diseases, they often involve multiple lesions of the vertebral body, intervertebral disc and nervous system, so the diagnosis process becomes particularly complicated.

[0004] The inventors have found that the existing methods for diagnosing lumbar diseases have some technical problems, such as:

[0005] (1) In the actual clinical field, the diagnosis of lumbar spine diseases is currently mainly based on the CT images taken, and the diagnosis method still relies entirely on the doctor's personal experience; however, current medical equipment mainly provides two-dimensional images of the patient's internal anatomical structure, which requires doctors to construct the three-dimensional shape, size and spatial relationship of the lesion with the surrounding tissues in their minds based on these two-dimensional image sequences, and then make a diagnosis. Therefore, doctors can only diagnose the disease based on this single CT image, and it is difficult to detect relatively subtle lesions in the CT image, which is time-consuming and laborious.

[0006] (2) Some existing technologies that can reconstruct two-dimensional medical images into three-dimensional images can only directly map the original two-dimensional medical images into three-dimensional images, but cannot perform simultaneous segmentation of multiple structures on the original two-dimensional medical images themselves; therefore, even if they are mapped into three-dimensional medical images, it is still difficult to accurately identify the various parts represented in the image. Summary of the invention

[0007] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a three-dimensional CT image reconstruction method and system for lumbar disease diagnosis, which can accurately and efficiently segment the vertebrae, intervertebral discs and neural structures in lumbar CT images at the same time, and realize image visualization by performing three-dimensional reconstruction of the segmented CT images, thereby providing strong support for lumbar disease diagnosis, preoperative planning and intraoperative guidance.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] A first aspect of the present invention provides a three-dimensional CT image reconstruction method for lumbar disease diagnosis.

[0010] A three-dimensional CT image reconstruction method for lumbar disease diagnosis, comprising:

[0011] Acquire multiple two-dimensional CT images and construct a training data set;

[0012] Training a deep neural network based on the training data set;

[0013] Input the two-dimensional CT image to be segmented into the trained deep neural network for two-dimensional CT image segmentation, and use the segmented two-dimensional CT image as a pre-reconstructed image;

[0014] The three-dimensional CT image is reconstructed for the pre-reconstructed image based on the VTK rendering method, specifically: a user interface is built, and a VTK rendering window is embedded in the user interface; the pre-reconstructed image is read based on a reader provided by VTK, and the pre-reconstructed image is converted into three-dimensional volume data based on an entity conversion method; the volume data is three-dimensionally reconstructed using a light transmission method, and the three-dimensionally reconstructed volume data is converted into a data format readable by an image pipeline based on a mapper and the data is rendered based on a VTK rendering window; a rendering loop is started using an interactor to display the rendered three-dimensional CT image, so as to realize visualization of the three-dimensional CT image.

[0015] Furthermore, the deep neural network is a dual network, that is, the deep neural network is composed of a Res-Net network and a V-Net network.

[0016] Furthermore, the two-dimensional CT image is segmented using a deep neural network, including: first, segmenting the vertebral structure of the two-dimensional CT image based on the Res-Net network; then, segmenting the intervertebral disc and neural structure of the two-dimensional CT image based on the V-Net network; then, integrating the obtained vertebral structure segmentation image and the intervertebral disc and neural structure segmentation images and recording them as a pre-reconstructed image.

[0017] Furthermore, the pre-reconstructed image is converted into three-dimensional volume data based on the entity conversion method. Specifically: first, the vtkMarchingCubes class of VTK is used to call the Marching Cubes algorithm, each voxel in the volume data is traversed, and the position of the isosurface is determined according to the sign change of the vertex value in the voxel; then, the generated isosurface data is mapped into a geometric representation using the mapper of VTK, and the obtained geometric representation is added to the renderer using the actor vtkActor; wherein the voxel is a three-dimensional pixel.

[0018] Furthermore, building a user interface includes: building an initial user interface based on the interface tool Qt Designer, and adding controls to the initial user interface; after the user interface is built, converting the format of the user interface in preparation for subsequent integration with the VTK rendering window.

[0019] Furthermore, the control includes buttons and labels; wherein the buttons include single-click to import CT data and single-click to start segmentation and reconstruction.

[0020] Furthermore, after the volume data is 3D reconstructed and before VTK rendering, normal calculation, smoothing and polygon reduction operations are sequentially performed on the volume data to ensure the quality of the volume data before rendering.

[0021] A second aspect of the present invention provides a three-dimensional CT image reconstruction system for diagnosing lumbar diseases.

[0022] A three-dimensional CT image reconstruction system for lumbar disease diagnosis, comprising:

[0023] The data acquisition module is configured to: acquire a plurality of two-dimensional CT images and construct a training data set;

[0024] A model training module is configured to: train a deep neural network based on the training data set;

[0025] The two-dimensional CT image segmentation module is configured to: input the two-dimensional CT image to be segmented into the trained deep neural network to perform two-dimensional CT image segmentation, and use the segmented two-dimensional CT image as a pre-reconstructed image;

[0026] The three-dimensional CT image reconstruction module is configured to: reconstruct the three-dimensional CT image of the pre-reconstructed image based on the VTK rendering method, specifically: build a user interface and embed a VTK rendering window for the user interface; read the pre-reconstructed image based on the reader provided by VTK, and convert the pre-reconstructed image into three-dimensional volume data based on the entity conversion method; use the light transmission method to perform three-dimensional reconstruction on the volume data, convert the three-dimensionally reconstructed volume data into a data format readable by the image pipeline based on the mapper and render the data based on the VTK rendering window; use the interactor to start the rendering cycle to display the rendered three-dimensional CT image to achieve visualization of the three-dimensional CT image.

[0027] 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 three-dimensional CT image reconstruction method for lumbar disease diagnosis as described in the first aspect of the present invention.

[0028] 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 when the processor executes the program, the steps in a three-dimensional CT image reconstruction method for diagnosing lumbar diseases as described in the first aspect of the present invention are implemented.

[0029] One or more of the above technical solutions have the following beneficial effects:

[0030] (1) The present invention inputs the two-dimensional CT image to be segmented into a deep neural network for two-dimensional CT image segmentation, and uses the segmented two-dimensional CT image as a pre-reconstructed image; based on the VTK rendering method, the pre-reconstructed image is reconstructed into a three-dimensional CT image to achieve visualization of the three-dimensional CT image. Therefore, the present invention can perform three-dimensional reconstruction based on the two-dimensional CT image, which not only reduces the workload of spinal surgeons, but also allows doctors to freely switch perspectives and levels through three-dimensional visualization technology, and gain in-depth insights into the patient's internal structure. It is no longer necessary to imagine in vain to accurately grasp the condition and anatomical details, providing strong support for low back pain diagnosis, preoperative planning and intraoperative guidance.

[0031] (2) The present invention performs two-dimensional CT image segmentation based on a deep neural network, that is, based on the Res-Net network and the V-Net network, the vertebral structure, intervertebral disc, and neural structure of the two-dimensional CT image are segmented respectively. Therefore, the present invention can segment the three spinal structures (intervertebral disc + vertebra + neural structure) in one step, and then perform three-dimensional reconstruction on this basis, so that each part in the three-dimensional CT image can be clearly distinguished, which can significantly improve the positioning accuracy of the pain source.

[0032] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0034] Figure 1 This is a flow chart of a three-dimensional CT image reconstruction method for lumbar disease diagnosis in Example 1 of the present invention.

[0035] Figure 2 This is an example diagram of a two-dimensional CT image in Embodiment 1 of the present invention.

[0036] Figure 3 4 is a structural diagram of a deep neural network in Embodiment 1 of the present invention.

[0037] Figure 4 This is a flow chart of a method for reconstructing a three-dimensional CT image from a pre-reconstructed image in Embodiment 1 of the present invention.

[0038] Figure 5 This is a schematic diagram of the user interface before importing a two-dimensional CT image in Embodiment 1 of the present invention.

[0039] Figure 6 This is a schematic diagram of the user interface after importing a two-dimensional CT image in Embodiment 1 of the present invention.

[0040] Figure 7 This is a schematic diagram of the user interface after three-dimensional CT image reconstruction is performed in the first embodiment of the present invention.

[0041] Figure 8 Schematic diagram of data interaction relationship in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0043] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0044] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0045] Embodiment 1

[0046] This embodiment discloses a three-dimensional CT image reconstruction method for diagnosing lumbar diseases.

[0047] like Figure 1 As shown, a three-dimensional CT image reconstruction method for lumbar disease diagnosis includes:

[0048] Step S1, acquiring multiple two-dimensional CT images and constructing a training data set;

[0049] Step S2: training a deep neural network based on the training data set;

[0050] Step S3, inputting the two-dimensional CT image to be segmented into the trained deep neural network to perform two-dimensional CT image segmentation, and using the segmented two-dimensional CT image as a pre-reconstructed image;

[0051] Step S4, reconstructing the three-dimensional CT image of the pre-reconstructed image based on the VTK rendering method, specifically: building a user interface, and embedding a VTK rendering window into the user interface; reading the pre-reconstructed image based on the reader provided by VTK, and converting the pre-reconstructed image into three-dimensional volume data based on the entity conversion method; reconstructing the volume data in three dimensions using the light transmission method, converting the three-dimensionally reconstructed volume data into a data format readable by the image pipeline based on the mapper, and rendering the data based on the VTK rendering window; using the interactor to start the rendering loop to display the rendered three-dimensional CT image, so as to realize the visualization of the three-dimensional CT image.

[0052] Based on the above process, the present invention can accurately and efficiently segment the vertebrae, intervertebral discs and neural structures in lumbar CT images at the same time, and realize image visualization by three-dimensionally reconstructing the segmented CT images, providing strong support for lumbar disease diagnosis, preoperative planning and intraoperative guidance. To facilitate the understanding of the technical solution of the present invention, the specific implementation methods of each step in the technical solution of the present invention are further explained and illustrated below.

[0053] Step S1, obtaining multiple two-dimensional CT images and constructing a training data set.

[0054] In this embodiment, two-dimensional CT images are collected from the Spine data set, and the collected two-dimensional CT images are used to construct a training data set. Furthermore, the Spine data set used is collected from Qilu Hospital of Shandong University (Qingdao Campus), and the use of the Spine data set has been approved by the hospital data ethics committee; among them, the Spine data set contains 200 cases of lumbar CT plain scan data, and, in order to better avoid homogeneity problems, these 200 cases of lumbar CT plain scan data come from 200 different subjects.

[0055] In order to ensure the accuracy of subsequent model training, before inputting the training data set into the deep neural network to train the network model, it is necessary to perform operations such as unifying the voxel spacing, image size, and slice number range in sequence. Specifically, the voxel spacing of each CT image in the training data set is set to 0.3467×0.3467×0.999mm 3 ; At the same time, for all lumbar CT plain scan data in the training data set, the image size is set to 512*512 pixels, and the number of slices is set to between 216 and 326. It should be noted that the same patient may suffer from multiple concurrent spinal diseases / disorders. In order to better reflect the image content, all lumbar CT plain scan data in the training data set are first manually drawn by junior experts; then, they are corrected by experienced spinal surgeons using ITK-SNAP1, and the outlined mask corrected by the spinal surgeon is used as the actual situation of the spinal CT image; ITK-SNAP1 is an open source medical image analysis software. Therefore, if Figure 2 As shown, each subject has a lumbar plain scan CT image and a mask corresponding to the lumbar plain scan CT image. In the actual implementation process, the lumbar plain scan CT image is used as the original image, and the mask image is used as the label to represent the area of ​​interest to the deep neural network model, so that the deep neural network can learn to pay more attention to these areas of interest; specifically, each vertebra, IVD (intervertebral disc) and nerve in the mask image is assigned a unique label; thus, the training data set contains four different categories, namely: background, nerve, intervertebral disc and vertebral body. Through label design, the deep neural network model can better pay attention to the three main areas of vertebrae, IVD (intervertebral disc) and nerves.

[0056] Step S2: training the deep neural network based on the obtained training data set.

[0057] Before training the deep neural network, the training data set needs to be divided. Specifically, 5-fold cross validation is used to divide the training data set into a pre-training set and a test set. For example, if there are 200 data in the training data set, the 200 data are randomly divided into 5 groups of data with the same number according to the ratio of 4:1, where each group contains data of 40 subjects. In the specific implementation steps, four groups are selected as pre-training sets, and the remaining group is used as a test set; first, the deep neural network is trained based on the obtained pre-training set; then, the model performance of the trained deep neural network is verified based on the obtained test set.

[0058] Furthermore, if Figure 3 As shown, the deep neural network used in this embodiment is a dual network, that is, the deep neural network is composed of a Res-Net network and a V-Net network. When the dual network of the deep neural network is trained and verified based on the pre-training set and the test set, the whole process is implemented by iteration. Specifically, the adaptive moment estimation (AdamW) with weight decay is used as the optimizer. Before the iterative training of the model, the number of epochs, the initial learning rate, and the weight decay value are pre-set; wherein, epoch is a unit for measuring the progress of model training, which refers to the process in which the entire training data is traversed by the model once. In other words, when an epoch is completed, it means that the model has traversed each CT image data in the training set (historical data set) at least once. When starting to train the deep neural network, the historical data set can be divided into multiple small batches, each batch containing a certain number of CT images. In each iteration, the model processes samples in a batch, calculates loss and gradient based on these samples, and then updates the weight of the model; this process is repeated until all batches are processed, that is, an epoch is completed. In addition, since a single epoch may not be enough for the model to fully learn the characteristics of the data, this embodiment chooses to train the model for multiple epochs; at the same time, too many epochs may also lead to overfitting, that is, the model is too complex, so that the model begins to remember the noise and details of each sample in the historical data set, rather than learning the general rules of the data. Therefore, it is necessary to select an appropriate number of epochs for deep neural network training.

[0059] Furthermore, in this embodiment, the number of epochs is set to 30, and each epoch is iterated 200 times; at the same time, the initial learning rate is set to 0.0001 and the weight decay value is set to 0.0001. If the model performance has not improved after 30 epochs, the training is terminated.

[0060] Step S3: input the two-dimensional CT image to be segmented into the trained deep neural network for two-dimensional CT image segmentation, and use the segmented two-dimensional CT image as a pre-reconstructed image.

[0061] First, the vertebral structure of the two-dimensional CT image is segmented based on the Res-Net network. Then, the intervertebral disc and neural structure of the two-dimensional CT image are segmented based on the V-Net network. Subsequently, the obtained vertebral structure segmentation image and the intervertebral disc and neural structure segmentation images are integrated and recorded as the pre-reconstructed image.

[0062] Furthermore, the Res-Net network is a deep convolutional neural network with residual connections, and the vertebral structure of the two-dimensional CT image is segmented based on the Res-Net network. Specifically: a Res-Net network with a layer number of ResNet101 is selected as the basic network for segmenting the vertebral structure, and on the basis of the Res-Net network, a full convolution layer is added for pixel-level classification; among them, U-Net or a network with a similar structure can be considered to be used, and Res-Net is used as the encoder part to extract image features; the decoder part can use upsampling and deconvolution operations to gradually restore the spatial resolution of the image and generate vertebral segmentation results.

[0063] Furthermore, the V-Net network adopts an encoder-decoder structure with multiple three-dimensional convolutional layers, pooling layers, upsampling layers and skip connections; the encoder part is used to gradually reduce the resolution of the CT image and extract high-level features; the decoder part is used to gradually restore the resolution of the CT image and generate segmentation results. In the specific implementation process, first, the three-dimensional convolutional layer in the V-Net network is replaced with a two-dimensional convolutional layer to adapt to the processing of two-dimensional CT images; and the intervertebral disc and neural structure (nerve root) in the CT image are segmented through the set pooling layer, upsampling layer and skip connection.

[0064] Step S4, reconstructing the three-dimensional CT image of the pre-reconstructed image based on the VTK rendering method, specifically: building a user interface, and embedding a VTK rendering window into the user interface; reading the pre-reconstructed image based on the reader provided by VTK, and converting the pre-reconstructed image into three-dimensional volume data based on the entity conversion method; reconstructing the volume data in three dimensions using the light transmission method, converting the three-dimensionally reconstructed volume data into a data format readable by the image pipeline based on the mapper, and rendering the data based on the VTK rendering window; using the interactor to start the rendering loop to display the rendered three-dimensional CT image, so as to realize the visualization of the three-dimensional CT image.

[0065] like Figure 4 As shown, the reconstruction process of a three-dimensional CT image requires volume data preprocessing, surface reconstruction and rendering, reconstruction of surface objects, and three-dimensional reconstruction. These processes are completed based on the coordination of PyQt5 and VTK rendering. The basic idea is: process the segmented image data and extract information to build a three-dimensional model; then, perform surface rendering and projection to achieve multi-view display. In VTK, the geometric shape of the CT image is drawn through a graphic model, and then the drawn graphics are displayed. Furthermore, several basic object elements involved in the drawing process of this embodiment include:

[0066] ①vtkMapper: A mapper that acts as a bridge between the visualization process and the graphics model. vtkMapper defines the mapping relationship between basic graphics elements and rendering data. Its derived classes map through LookupTable and generate corresponding vtkActor graphics elements. Each vtkMapper can be associated with one or more vtkActors.

[0067] ②vtkActor: represents the graphic entity to be drawn, which is analogous to an actor on the stage; the spatial position, rendering effect and other properties of vtkActor can be flexibly customized by adjusting parameters.

[0068] ③vtkRenderer: As the core of scene management, it is similar to a stage. It is responsible for integrating the properties of drawing objects, light sources, cameras and other elements, and embedding them into the renderer window vtkRenderWindow to display the complete scene.

[0069] ④vtkRenderWindow: Renderer window, the core component used to control the display window, like a theater stage, supports the combined display of multiple vtkRenderWindow, and provides functions such as stereoscopic visual effect setting and window size adjustment.

[0070] ⑤vtkProperty: Focuses on enhancing the realism of visual objects. By setting the physical properties of objects (such as material, transparency, etc.) through vtkProperty, three-dimensional data can present more realistic visual effects.

[0071] ⑥vtkCamera: plays the role of observer in the visualization process, manages parameters such as view position and focus, and supports view manipulation such as rotation, translation, and scrolling to optimize the user experience.

[0072] ⑦vtkLight: Responsible for scene lighting. The vtkLight class can be used to finely control the light source state (such as switch), color, intensity, etc., to enhance the layering and realism of the rendering effect. The created light source can be directly added to vtkRenderer for use.

[0073] ⑧vtkRenderWindowInteractor: Enhance the interactive experience of the rendering window, such as adjusting the vtkCamera perspective, selecting Actor objects, processing user-defined messages, etc., to provide users with an intuitive operation interface. ) This can be achieved by the following steps:

[0074] Step S4-1, build a user interface and embed a VTK rendering window into the user interface.

[0075] The initial user interface is built based on the interface tool Qt Designer, and controls are added to the initial user interface. After the user interface is built, the format of the user interface is converted in preparation for subsequent integration with the VTK rendering window.

[0076] like Figure 5 As shown, the user interface before importing the two-dimensional CT image only has basic designs such as buttons and labels for data interaction with the user; among them, the buttons in the user interface designed in this embodiment include "click to import CT data" and "click to start segmentation and reconstruction" buttons; the labels can be used to display names or data, for example, the segmentation accuracy, segmentation and reconstruction time, etc.

[0077] Step S4-2: Read the pre-reconstructed image based on the reader vtkReader provided by VTK, wherein the user interface after reading the two-dimensional CT image is as follows: Figure 6 As shown; the pre-reconstructed image is converted into three-dimensional volume data based on the entity transformation method.

[0078] Furthermore, the pre-reconstructed image is converted into three-dimensional volume data based on the entity conversion method, including: first, using the vtkMarchingCubes class of VTK to call the Marching Cubes algorithm, traversing each voxel in the volume data, and determining the position of the isosurface according to the sign change of the vertex value in the voxel; then, using the mapper vtkMapper of VTK to map the generated isosurface data into a geometric representation, and using the actor vtkActor to add the obtained geometric representation to the renderer; wherein the voxel is a three-dimensional pixel.

[0079] Step S4-3, use the light transmission method to perform 3D reconstruction on the volume data, convert the 3D reconstructed volume data into a data format readable by the image pipeline based on the vtkMapper mapper, and render the data based on the VTK rendering window; use the interactor to start the rendering cycle to display the rendered 3D CT image, realize the visualization of the 3D CT image, and the user interface after the 3D CT image reconstruction is as follows Figure 7 shown.

[0080] After 3D reconstruction of the volume data and before VTK rendering, it is necessary to perform normal calculation, smoothing and polygon reduction operations on the volume data in sequence to ensure the quality of the volume data before rendering.

[0081] First, normal calculation is performed. Normal calculation is to calculate the normal of each triangle in the three-dimensional volume data through the points in the neighborhood, that is:

[0082] n = (P2-P1)(P3-P1);

[0083] Where n represents the normal value formed by the three vertices in the triangle, and P1, P2 and P3 represent the first, second and third vertices in the triangle respectively. Then, smoothing is performed. Smoothing is achieved by using the Laplace smoothing method, that is:

[0084]

[0085] Among them, P j represents the i-th vertex in the triangle, and n represents the vertex P j The number of adjacent nodes. Describe the i-th vertex in the triangle after Laplace smoothing. Finally, the number of polygon faces is reduced based on the Decimation method.

[0086] Step S4-4: Synchronize the segmentation results and the 3D reconstruction results to the user interface for display. Figure 8 As shown, the data interaction relationship in this embodiment uses a three-layer (bottom layer, middle layer, upper layer) architecture design, that is, the human-computer interaction design is realized through the presentation layer, business logic layer and bottom layer algorithm library. Medical personnel send user requests in the human-computer interaction interface; the business logic layer receives the user request, executes the corresponding task by calling the bottom layer related algorithm library and the trained model, and then displays the response result to the presentation layer.

[0087] The upper layer is the presentation layer, which provides visualization of spinal CT data; users operate in the visualization interface and make user requests, and this layer displays the system response results to implement functions such as user interaction, CT image display, and data selection. The middle layer is the business logic layer, which provides the core functions of the system. This layer is mainly responsible for processing data and starts to perform corresponding tasks after receiving user requests. Then, the response results are provided to the presentation layer for display, which is used to implement functions such as medical image reading, image segmentation, reconstruction, and display. The bottom layer contains various algorithm libraries, underlying frameworks, and trained network models used to complete functions, and the middle layer calls them to complete business.

[0088] To verify the effect of the present invention, this embodiment is based on a unified experimental framework to comprehensively evaluate the performance differences between the algorithm of the present invention and five mainstream deep learning models (GraphCut, FCN, SegNet, U-Net and UNetR). The experimental framework strictly follows the same benchmark configuration. The experiments are all performed on PyTorch using Intel(R) Core(TM) CPU and NVIDIA GeForce RTX 3090 GPU with 24GB memory. The experimental results are shown in Table 1:

[0089] Table 1 Comparison of segmentation experiment results

[0090]

[0091] The experimental results show that the segmentation performance of the segmentation network of the present invention is significantly better than other classical methods. Therefore, on the one hand, the present invention can perform three-dimensional reconstruction based on two-dimensional CT images, which not only reduces the workload of spinal surgeons, but also allows doctors to freely switch perspectives and levels through three-dimensional visualization technology, and gain in-depth insights into the patient's internal structure. Doctors no longer need to imagine in vain to accurately grasp the condition and anatomical details, providing strong support for low back pain diagnosis, preoperative planning and intraoperative guidance. On the other hand, the present invention can segment three spinal structures (intervertebral disc + vertebra + neural structure) in one step, and then perform three-dimensional reconstruction on this basis, so that each part in the three-dimensional CT image can be clearly distinguished, which can significantly improve the positioning accuracy of the pain source.

[0092] Embodiment 2

[0093] This embodiment discloses a three-dimensional CT image reconstruction system for diagnosing lumbar diseases.

[0094] A three-dimensional CT image reconstruction system for lumbar disease diagnosis, comprising:

[0095] The data acquisition module is configured to: acquire a plurality of two-dimensional CT images and construct a training data set;

[0096] A model training module is configured to: train a deep neural network based on the training data set;

[0097] The two-dimensional CT image segmentation module is configured to: input the two-dimensional CT image to be segmented into the trained deep neural network to perform two-dimensional CT image segmentation, and use the segmented two-dimensional CT image as a pre-reconstructed image;

[0098] The three-dimensional CT image reconstruction module is configured to: reconstruct the three-dimensional CT image of the pre-reconstructed image based on the VTK rendering method, specifically: build a user interface and embed a VTK rendering window for the user interface; read the pre-reconstructed image based on the reader provided by VTK, and convert the pre-reconstructed image into three-dimensional volume data based on the entity conversion method; use the light transmission method to perform three-dimensional reconstruction on the volume data, convert the three-dimensionally reconstructed volume data into a data format readable by the image pipeline based on the mapper and render the data based on the VTK rendering window; use the interactor to start the rendering cycle to display the rendered three-dimensional CT image to achieve visualization of the three-dimensional CT image.

[0099] Embodiment 3

[0100] The purpose of this embodiment is to provide a computer-readable storage medium.

[0101] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a three-dimensional CT image reconstruction method for lumbar disease diagnosis as described in the first embodiment of the present disclosure.

[0102] Embodiment 4

[0103] The purpose of this embodiment is to provide an electronic device.

[0104] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a three-dimensional CT image reconstruction method for lumbar disease diagnosis as described in Embodiment 1 of the present disclosure are implemented.

[0105] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part 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 to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0106] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0107] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it 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 on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A three-dimensional CT image reconstruction method for lumbar disease diagnosis, characterized in that: include: Acquire multiple two-dimensional CT images and construct a training data set; Training a deep neural network based on the training data set; Input the two-dimensional CT image to be segmented into the trained deep neural network for two-dimensional CT image segmentation, and use the segmented two-dimensional CT image as a pre-reconstructed image; Reconstructing the three-dimensional CT image of the pre-reconstructed image based on the VTK rendering method, specifically: building a user interface, and embedding a VTK rendering window into the user interface; reading the pre-reconstructed image based on a reader provided by VTK, and converting the pre-reconstructed image into three-dimensional volume data based on a solid conversion method; The volume data is three-dimensionally reconstructed using a ray transmission method, and the three-dimensionally reconstructed volume data is converted into a data format readable by an image pipeline based on a mapper, and the data is rendered based on a VTK rendering window; a rendering loop is started using an interactor to display the rendered three-dimensional CT image, thereby realizing visualization of the three-dimensional CT image.

2. A three-dimensional CT image reconstruction method for lumbar disease diagnosis according to claim 1, characterized in that: The deep neural network is a dual network, that is, the deep neural network is composed of a Res-Net network and a V-Net network.

3. A three-dimensional CT image reconstruction method for lumbar disease diagnosis according to any one of claims 1 to 2, characterized in that: The two-dimensional CT image is segmented using a deep neural network, including: first, segmenting the vertebral structure of the two-dimensional CT image based on the Res-Net network; then, segmenting the intervertebral disc and neural structure of the two-dimensional CT image based on the V-Net network; then, the obtained vertebral structure segmentation image and the intervertebral disc and neural structure segmentation images are integrated and recorded as a pre-reconstructed image.

4. A three-dimensional CT image reconstruction method for lumbar disease diagnosis according to claim 1, characterized in that: The pre-reconstructed image is converted into three-dimensional volume data based on the entity conversion method. Specifically: first, the vtkMarchingCubes class of VTK is used to call the Marching Cubes algorithm, each voxel in the volume data is traversed, and the position of the isosurface is determined according to the sign change of the vertex value in the voxel; then, the generated isosurface data is mapped into a geometric representation using the mapper of VTK, and the obtained geometric representation is added to the renderer using the actor vtkActor; wherein the voxel is a three-dimensional pixel.

5. A three-dimensional CT image reconstruction method for lumbar disease diagnosis according to claim 1, characterized in that: Build the user interface, including: build the initial user interface based on the interface tool Qt Designer, and add controls to the initial user interface; after the user interface is built, convert the format of the user interface in preparation for subsequent integration with the VTK rendering window.

6. A three-dimensional CT image reconstruction method for lumbar disease diagnosis according to claim 5, characterized in that: The control includes buttons and labels; wherein the buttons include single-click to import CT data and single-click to start segmentation and reconstruction.

7. A three-dimensional CT image reconstruction method for lumbar disease diagnosis according to claim 1, characterized in that: After the volume data is 3D reconstructed and before VTK rendering, normal calculation, smoothing and polygon reduction operations are sequentially performed on the volume data to ensure the quality of the volume data before rendering.

8. A three-dimensional CT image reconstruction system for lumbar disease diagnosis, characterized in that: include: The data acquisition module is configured to: acquire a plurality of two-dimensional CT images and construct a training data set; A model training module is configured to: train a deep neural network based on the training data set; The two-dimensional CT image segmentation module is configured to: input the two-dimensional CT image to be segmented into the trained deep neural network to perform two-dimensional CT image segmentation, and use the segmented two-dimensional CT image as a pre-reconstructed image; The three-dimensional CT image reconstruction module is configured to: perform three-dimensional CT image reconstruction on the pre-reconstructed image based on the VTK rendering method, specifically: build a user interface and embed a VTK rendering window into the user interface; read the pre-reconstructed image based on the reader provided by VTK, and convert the pre-reconstructed image into three-dimensional volume data based on the entity conversion method; The volume data is three-dimensionally reconstructed using a ray transmission method, and the three-dimensionally reconstructed volume data is converted into a data format readable by an image pipeline based on a mapper, and the data is rendered based on a VTK rendering window; a rendering loop is started using an interactor to display the rendered three-dimensional CT image, thereby realizing visualization of the three-dimensional CT image.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of a three-dimensional CT image reconstruction method for lumbar disease diagnosis as described in any one of claims 1 to 7 are implemented.

10. 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, the steps of a three-dimensional CT image reconstruction method for lumbar disease diagnosis as described in any one of claims 1-7 are implemented.

Images