Intelligent prediction method and system for lumbar disc herniation grade based on CT image

Abstract

The invention discloses an intelligent prediction method and system for the lumbar disc herniation level based on a CT image, and the method comprises the steps: carrying out the multi-dimensional image reconstruction processing according to the current original CT image data, and obtaining a reconstructed CT reconstruction image; performing dynamic region segmentation according to the reconstructed CT reconstruction image, and segmenting the lumbar intervertebral disc and surrounding tissues thereof through an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; performing feature extraction processing according to the lumbar intervertebral disc structure region, and extracting multi-dimensional features of the lumbar intervertebral disc structure region by using a deep learning network to obtain a multi-dimensional feature vector; and according to the multi-dimensional feature vector, performing protrusion level prediction by using an integrated learning model, and performing a voting mechanism or weighted averaging by combining results of a plurality of classifiers to obtain a final lumbar disc protrusion level prediction result. According to the embodiment of the invention, the accuracy and efficiency of lumbar disc herniation prediction can be effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular to an intelligent prediction method and system for lumbar disc herniation grade based on CT images. Background Art

[0002] With the advancement of imaging technology, computed tomography (CT) has become an important tool for the clinical diagnosis of lumbar disc herniation. Through CT images, medical professionals can intuitively observe the status of the lumbar disc and its surrounding structures, thereby formulating a reasonable treatment plan for the patient. However, the interpretation of CT images still relies on the experience and knowledge of professional doctors, which is subjective and time-sensitive, and is easily affected by individual differences and imaging conditions. These factors have led to a lack of uniformity in the criteria for determining the grade of lumbar disc herniation, and there are certain errors in clinical diagnosis.

[0003] Currently, the evaluation of lumbar disc herniation usually adopts manual analysis methods based on image features. Doctors make judgments by observing the tissue morphology, structural changes, relative position, etc. in CT images. This process is not only time-consuming and laborious, but also requires a high level of subjective judgment ability of doctors, which can easily lead to misdiagnosis or missed diagnosis. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent prediction method and system for lumbar disc herniation grade based on CT images, so as to solve the deficiencies in the prior art and effectively improve the accuracy and efficiency of lumbar disc herniation prediction.

[0005] An embodiment of the present application provides an intelligent prediction method for lumbar disc herniation grade based on CT images, the method comprising: According to the current original CT image data of the patient, multi-dimensional image reconstruction processing is performed to improve the image resolution and contrast, and a reconstructed CT reconstructed image is obtained; According to the reconstructed CT images, dynamic region segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain the segmented lumbar intervertebral disc structure region; According to the segmented lumbar disc structure area, feature extraction processing is performed, and the multi-dimensional features of the lumbar disc structure area are extracted using a deep learning network to obtain a multi-dimensional feature vector; According to the extracted multi-dimensional feature vector, the ensemble learning model is used to predict the herniation grade. By combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation grade prediction result.

[0006] Optionally, performing multi-dimensional image reconstruction processing according to the current original CT image data of the patient to improve the image resolution and contrast to obtain a reconstructed CT reconstructed image includes: The historical CT image data set is preliminarily reconstructed using the filtered back-projection method to obtain a preliminary reconstructed image; Taking the preliminary reconstructed image as input, a convolutional neural network is used to train the specific noise and artifacts of the preliminary reconstructed image, wherein the convolutional neural network optimizes network parameters to achieve noise suppression and contrast enhancement by comparing the real preset high-quality image with the preliminary reconstructed image; The patient's current original CT image data is input into the trained convolutional neural network, and the reconstructed CT image is output.

[0007] Optionally, the dynamic region segmentation is performed based on the reconstructed CT reconstruction image, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain the segmented lumbar intervertebral disc structure region, including: The reconstructed CT image is divided into a number of overlapping local area image blocks, and the size of each local area image block is adaptively adjusted according to the resolution of the image to ensure that important structural information is contained; For each local region image block, a grayscale histogram of the local region image block is calculated to obtain local grayscale distribution characteristics, and a threshold of each local region image block is dynamically determined; The determined threshold is used as the boundary to perform region growing and obtain the segmented lumbar disc structure region.

[0008] Optionally, the segmented lumbar disc structure region is subjected to feature extraction processing, and a deep learning network is used to extract multidimensional features of the lumbar disc structure region to obtain a multidimensional feature vector, including: Inputting the segmented lumbar intervertebral disc structural area data into a pre-trained deep learning model to extract morphological features at different levels, wherein the deep learning model can reflect the morphological features of the lumbar intervertebral disc in the feature map extracted by the convolution layer; Applying texture analysis on the extracted feature map to extract texture features to capture subtle texture changes in the lumbar disc structure, thereby reflecting the health status or pathological changes of the tissue; The morphological features and the texture features are integrated to generate a multi-dimensional feature vector to comprehensively describe the structural features of the lumbar intervertebral disc.

[0009] Another embodiment of the present application provides an intelligent prediction system for lumbar disc herniation grade based on CT images, the system comprising: A reconstruction module is used to perform multi-dimensional image reconstruction processing based on the patient's current original CT image data to improve the image resolution and contrast and obtain a reconstructed CT reconstructed image; A segmentation module is used to perform dynamic region segmentation based on the reconstructed CT reconstruction image, segment the lumbar intervertebral disc and its surrounding tissues through an adaptive segmentation algorithm, and obtain a segmented lumbar intervertebral disc structure region; An extraction module is used to perform feature extraction processing based on the segmented lumbar disc structure area, and use a deep learning network to extract multi-dimensional features of the lumbar disc structure area to obtain a multi-dimensional feature vector; The prediction module is used to predict the protrusion grade using an integrated learning model based on the extracted multi-dimensional feature vector. By combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation grade prediction result.

[0010] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when running.

[0011] Yet another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described above.

[0012] Compared with the prior art, the present invention provides an intelligent prediction method for lumbar disc herniation grade based on CT images. According to the current original CT image data, multi-dimensional image reconstruction processing is performed to obtain a reconstructed CT reconstructed image; according to the reconstructed CT reconstructed image, dynamic region segmentation is performed, and the lumbar disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain a segmented lumbar disc structure region; according to the lumbar disc structure region, feature extraction processing is performed, and a deep learning network is used to extract multi-dimensional features of the lumbar disc structure region to obtain a multi-dimensional feature vector; according to the multi-dimensional feature vector, an integrated learning model is used to predict the herniation grade, and a voting mechanism or weighted average is performed by combining the results of multiple classifiers to obtain the final lumbar disc herniation grade prediction result, thereby effectively improving the accuracy and efficiency of lumbar disc herniation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A hardware structure block diagram of a computer terminal for an intelligent prediction method of lumbar disc herniation grade based on CT images provided in an embodiment of the present invention; Figure 2 A schematic flow chart of an intelligent prediction method for lumbar disc herniation grade based on CT images provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an intelligent prediction system for lumbar disc herniation grade based on CT images provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but should not be construed as limiting the present invention.

[0015] The embodiment of the present invention firstly provides an intelligent prediction method for the grade of lumbar disc herniation based on CT images. The method can be applied to electronic devices, such as computer terminals, specifically ordinary computers, etc.

[0016] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal for an intelligent prediction method of lumbar disc herniation grade based on CT images provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any intelligent prediction method for lumbar disc herniation grade based on CT images.

[0018] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0019] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any intelligent prediction method for the grade of lumbar disc herniation based on CT images.

[0020] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0021] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0022] See also Figure 2 The embodiment of the present invention provides an intelligent prediction method for lumbar disc herniation grade based on CT images, which may include the following steps: S201, performing multi-dimensional image reconstruction processing according to the current original CT image data of the patient to improve the image resolution and contrast, and obtain a reconstructed CT reconstructed image; The present invention provides an intelligent prediction method for lumbar disc herniation grade based on CT images, which aims to improve the prediction accuracy of lumbar disc herniation grade by comprehensively processing and analyzing the patient's CT image data. The method first performs multi-dimensional reconstruction of the original CT image to improve the resolution and contrast of the image, thereby making the subsequent regional segmentation and feature extraction process more accurate. Through the adaptive segmentation algorithm, the system can dynamically identify and extract the lumbar disc and its surrounding tissues, thereby obtaining clear structural areas, and then combine the deep learning network to extract the multi-dimensional feature vectors of these areas. Finally, an integrated learning model is used to integrate the prediction results of multiple classifiers to achieve accurate judgment of the lumbar disc herniation grade.

[0023] This intelligent prediction method has important clinical application value. By improving the quality and resolution of CT images, doctors can obtain clearer and more detailed information about the lumbar structure, thereby making more scientific diagnosis and treatment decisions. In addition, the use of deep learning and integrated learning technology has achieved automated prediction of the grade of lumbar disc herniation, reducing the errors and subjectivity caused by manual analysis. This not only improves the accuracy and efficiency of the prediction, but also saves doctors a lot of time, allowing them to focus on more complex cases and provide stronger support for patients' rehabilitation.

[0024] Specifically, the filtered back-projection method can be used to preliminarily reconstruct the historical CT image data set to obtain a preliminary reconstructed image; In this step, the historical CT image data set is processed by the filtered back projection (FBP) algorithm to obtain a preliminary reconstructed image. This process involves processing the projection data collected from different angles through a filter and then back-projecting it. The final image generated can better reflect the internal structure of the scanned object. This step is the basis of the entire image processing process. Through preliminary reconstruction, doctors can obtain preliminary image information of the lumbar spine, laying the foundation for subsequent noise suppression and contrast enhancement. Improving the quality and information content of the image is the key to the accuracy of subsequent analysis and feature extraction.

[0025] In this step, a large amount of CT projection data related to the lumbar spine area needs to be collected first. These data usually come from multiple scans, each using a different angle to ensure that comprehensive anatomical information is obtained. Then, an appropriate filtered back-projection algorithm is selected, where the selection of filters is crucial. For example, the Ram-Lak filter or Shepp-Logan filter can be used to pre-process the data to remove high-frequency noise that may be generated during data acquisition. This process ensures the clarity of the reconstructed image and makes subsequent analysis more reliable.

[0026] Next, the data is back-projected. The processed projection data is back-projected layer by layer, and each projection data is back-projected onto the image plane at different angles of the object. The projections of all angles are accumulated to synthesize the reconstructed image. This process requires calculating the accumulated value of each pixel to form a preliminary reconstructed image. Linear interpolation is usually used to fill in the blank areas between different projection angles to ensure the integrity and continuity of the reconstructed image.

[0027] Finally, after the reconstruction is completed, the quality of the preliminary reconstructed image needs to be evaluated. Image quality evaluation indicators such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) can be used to evaluate the quality of the image. If there is obvious noise or artifacts in the reconstructed image, it may be necessary to return to the preliminary steps, adjust the filtering parameters or modify the reconstruction strategy to ensure that the generated image is suitable for subsequent noise suppression processing.

[0028] Taking the preliminary reconstructed image as input, a convolutional neural network is used to train the specific noise and artifacts of the preliminary reconstructed image, wherein the convolutional neural network optimizes network parameters to achieve noise suppression and contrast enhancement by comparing the real preset high-quality image with the preliminary reconstructed image; In this step, the convolutional neural network (CNN) model is used to train the preliminary reconstructed image in order to identify and eliminate noise and artifacts in the image. By comparing with the real high-quality image, the CNN will continuously adjust its internal parameters, optimize the processing results, and improve the quality of the image. Training the convolutional neural network can significantly improve the clarity and contrast of the image, making subsequent image analysis and feature extraction more accurate. The image information after removing noise and artifacts is purer, which will effectively improve the accuracy of subsequent classification and prediction models.

[0029] In this step, we first need to prepare a training dataset containing real high-quality CT images and the corresponding preliminary reconstructed images. In order to improve the performance of the model, the dataset should cover a variety of cases so that the network can learn various noise characteristics and their corresponding coping strategies. After the dataset is labeled, data enhancement techniques (such as random rotation, flipping, and scaling) are used to increase the diversity of samples and further improve the generalization ability of the model.

[0030] Next, build a convolutional neural network model. When designing the network architecture, you can use a deep learning structure that combines multiple convolutional layers and pooling layers, such as U-Net or ResNet, which have performed well in the field of medical image processing. The convolutional layer is used to extract image features, while the pooling layer is used to gradually reduce the dimension of the feature map and reduce the computational complexity. An activation function (such as ReLU) can be added after each convolutional layer to introduce nonlinear changes, thereby improving the expressiveness of the model. Finally, the output layer of the model should use a suitable activation function (such as Sigmoid or Softmax) to perform binary classification of noise and artifacts.

[0031] Finally, the model is trained through cross-validation technology, and the difference between the model's output effect and the real high-quality image is evaluated using loss functions (such as mean square error), and the network parameters are continuously optimized. The key steps in the training process include back propagation, which updates the weight value by calculating the gradient to gradually improve the learning effect of the model. After multiple rounds of training and tuning, a convolutional neural network model that can effectively suppress noise and enhance image contrast is finally obtained, providing strong support for the subsequent processing of patient images.

[0032] The patient's current original CT image data is input into the trained convolutional neural network, and the reconstructed CT image is output.

[0033] In this step, the trained convolutional neural network is applied to the patient's original CT image data. After inputting this data into the network, the system will generate processed reconstructed images to optimize image quality. By inputting the patient's CT images into the trained CNN, personalized image processing can be achieved, significantly improving the quality and usability of the images. The successful implementation of this process will provide valuable high-quality data for subsequent analysis and diagnosis, supporting clinical decision-making.

[0034] In this step, when the patient's current raw CT image data is prepared, it first needs to be properly preprocessed to ensure that the input data meets the requirements of the convolutional neural network. This includes adjusting the size and resolution of the image, usually scaling the image to a size consistent with the training phase. For example, if the training model uses a 256x256 input image, then all new raw CT images also need to be adjusted to the same size. In addition, for some models, grayscale value normalization is also necessary, which helps to improve the training effect of the neural network while keeping the image data in the same range, thereby speeding up convergence.

[0035] Next, the processed original CT image is input into the trained convolutional neural network for processing. At this point, the network will use its learned features to identify and suppress noise and artifacts in the input image. Each layer of the network will extract features, and through the forward propagation mechanism, these features are processed step by step, and finally the final image result is generated through the output layer. This process must not only be carried out in real time, but also ensure that the output image can retain the original anatomical structure information as much as possible to avoid the loss of important features.

[0036] Finally, the output reconstructed image needs to be evaluated and verified for quality, including comparing the difference between the reconstructed image and the real high-quality image, to ensure that the processing effect meets expectations. If the quality of the output image still needs to be improved, further optimization of the convolutional neural network parameters or different types of model fusion strategies can be considered to enhance the quality of the final image. Ultimately, through this process, doctors can obtain clear, high-contrast CT reconstructed images, providing effective support for the next step of diagnosis and treatment.

[0037] S202, performing dynamic region segmentation based on the reconstructed CT reconstruction image, segmenting the lumbar intervertebral disc and its surrounding tissues by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; According to the reconstructed CT images, dynamic regional segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented using an adaptive segmentation algorithm. This process analyzes the grayscale distribution in the image and dynamically adjusts the segmentation threshold to ensure that the lumbar intervertebral disc structure and its adjacent tissues can be accurately identified. Specifically, the algorithm first divides the entire image into local areas, and then dynamically determines the appropriate segmentation threshold based on the characteristics of each area (such as the grayscale histogram), and finally achieves accurate segmentation of the lumbar intervertebral disc and its surrounding tissues through the method of regional growing.

[0038] The process of dynamic region segmentation is of great significance in intelligent prediction methods. Accurate segmentation of the lumbar disc and its surrounding tissues can significantly improve the accuracy of subsequent feature extraction and help the deep learning model obtain more reliable feature information. This lays the foundation for the intelligent prediction of the grade of lumbar disc herniation, improves the efficiency and accuracy of diagnosis, and helps doctors make more scientific and reasonable treatment plans.

[0039] Specifically, the reconstructed CT image can be divided into a number of overlapping local area image blocks, and the size of each local area image block is adaptively adjusted according to the resolution of the image to ensure that important structural information is contained; This process aims to cut the entire image into multiple local areas so that each area can be analyzed in depth separately. By setting the local areas as overlapping image blocks, the details of the lumbar intervertebral disc and its surrounding tissues can be captured more comprehensively, ensuring that important features are not lost due to cutting. The division of local area image blocks optimizes the accuracy of subsequent analysis and makes the segmentation of each area more flexible and dynamic. This design not only improves the accuracy of segmentation, but also helps to enhance the robustness of image processing, allowing the algorithm to adapt to the physiological characteristics of different patients.

[0040] At this stage, the resolution and overall size of the reconstructed CT image need to be analyzed first. According to these parameters, a dynamic segmentation algorithm is designed to determine the size of each local area image block. Generally speaking, the size of the local area image block can be set to 64x64 pixels or 128x128 pixels, and the specific size can be adjusted according to the complexity of the image and the required detail information. In order to achieve the design of overlapping areas, the extraction process of each image block will follow a rule, that is, after extracting an image block, slide to the right and downward by a certain pixel distance (for example, 20 to 30 pixels), so as to ensure that there is an overlap between adjacent image blocks, so as to facilitate information sharing during subsequent segmentation.

[0041] When performing the segmentation, relevant functions in the image processing library can be used to ensure that the extraction process is fast and error-free. At the same time, considering the morphological complexity of the lumbar disc structure, boundary conditions can be set during the segmentation process to avoid cutting at the edges of important structures. After each local area image block is extracted, it needs to be stored in a data structure, such as a list or array, for subsequent processing. In addition, in order to be able to check the rationality of the segmentation in the entire image, marks can be added to the image to show the number of each image block and its coverage area. Such a segmentation step will provide sufficient data support for the subsequent dynamic area segmentation to ensure that no important structural information is missed.

[0042] For each local region image block, a grayscale histogram of the local region image block is calculated to obtain local grayscale distribution characteristics, and a threshold of each local region image block is dynamically determined; This step identifies the grayscale distribution characteristics of different structures in the image by calculating the grayscale histogram of the local area. This distribution information is crucial for determining the appropriate segmentation threshold, because different tissues usually show different grayscale values ​​on CT images. Dynamically determining the threshold can improve the adaptability and accuracy of segmentation, accurately capture different types of tissues and lesion areas, and ensure the reliability of subsequent segmentation results. This process also enables the algorithm to effectively cope with the differences in image characteristics of different patients and promote the realization of personalized medicine.

[0043] At this stage, for each local area image block, it is necessary to first extract the grayscale value of each pixel in the area. The grayscale value range of each local image block is usually from 0 to 255. A one-dimensional array or dictionary can be used to count the frequency of pixels at each grayscale level. In specific implementation, each pixel in the local area is looped through and the corresponding frequency is updated according to the numerical value of its grayscale value. After these processes are completed, a complete grayscale histogram will be generated, reflecting the grayscale distribution characteristics of the local area.

[0044] The next step is to analyze the histogram and calculate the statistical characteristics such as the overall brightness mean and standard deviation of the local image block. This information not only helps to understand the contrast and brightness distribution of the area, but also provides a basis for the subsequent setting of dynamic thresholds. For example, methods such as the Otsu algorithm can be used to evaluate the optimal threshold. In this process, it is recommended to use efficient processing libraries (such as NumPy) to speed up the statistics and analysis of grayscale values, especially when processing a large number of image blocks, which can effectively improve the calculation speed.

[0045] Finally, in order to facilitate subsequent analysis and threshold selection, the histogram can be visualized. This can show the frequency of each grayscale value through a bar chart or line chart, which is convenient for researchers to intuitively understand the characteristics of the image block. This visualization can not only be used to assist in selecting the threshold, but also provide rich information support for subsequent dynamic region segmentation. Such implementation steps ensure that the grayscale features of each local region are fully mined, providing a solid foundation for subsequent region growth.

[0046] The determined threshold is used as the boundary to perform region growing and obtain the segmented lumbar disc structure region.

[0047] After obtaining the dynamically determined thresholds, the region growing algorithm is constructed based on these thresholds, with the goal of gradually expanding the segmented region from the initial seed point until the set growth conditions are met. This method can effectively extract the lumbar intervertebral disc and its surrounding tissue structures while avoiding segmentation errors caused by noise or artifacts. The application of the region growing method can ensure the consistency and integrity of the segmentation results, especially when dealing with complex tissue structures. By merging similar regions, it helps to better identify the target area, further improve the effectiveness and accuracy of the segmentation, and lay a solid foundation for subsequent feature extraction.

[0048] In this step, the seed points of each local area image block are first set according to the optimal threshold calculated by the grayscale histogram determined in the second step. These seed points should be located at pixels whose grayscale values ​​are close to the set threshold, and the central area of ​​the image block is usually selected as the initial position. Next, the algorithm will start the process of region growing according to the selected threshold. In specific implementation, the neighboring pixels of the seed point are first checked. If the grayscale values ​​of these pixels are similar to the grayscale values ​​of the seed point (within a preset tolerance range), these neighboring pixels are added to the current segmented area.

[0049] The process of region growing is implemented through iteration. In each iteration, the algorithm checks the boundary pixels of the current segmented region, checks the adjacent pixels of these pixels, and determines whether the similarity conditions are met to continue to expand the segmented region. In order to ensure the effectiveness of the segmentation, stop conditions are usually set, such as the maximum number of iterations and the minimum growing area, to avoid excessive growth or the formation of incoherent regions. In order to improve processing efficiency, queues or stacks can be used to maintain the pixels to be processed when organizing the data structure to ensure that the algorithm can quickly access the required data when performing region growing.

[0050] After completing the region growing, the algorithm will post-process the segmentation results to remove small noise and artifacts. For example, morphological processing can be used to smooth the segmentation boundaries to avoid cracks and irregular shapes in the segmented area. In addition, the program compares the final lumbar disc structure area with the original CT image to ensure the accuracy and consistency of the segmentation effect, thus laying a solid foundation for subsequent feature extraction. This series of steps ensures that the final segmented area is clear and complete, providing reliable input data for subsequent deep learning feature extraction.

[0051] S203, performing feature extraction processing based on the segmented lumbar intervertebral disc structure region, using a deep learning network to extract multi-dimensional features of the lumbar intervertebral disc structure region, and obtaining a multi-dimensional feature vector; In this step, based on the segmented lumbar disc structure area, a deep learning network is used to extract the multidimensional features of the area and generate a multidimensional feature vector. This process allows the model to extract representative features at different levels by inputting the segmented image data into a pre-trained deep learning model. These features include not only morphological features, such as the shape, size, and position of the lumbar disc, but also texture features that can reflect the health status and pathological changes of the tissue. This method can achieve accurate quantification of the state of the lumbar disc and provide strong feature support for subsequent herniation grade prediction.

[0052] The core significance of this step is to systematically and quantitatively extract the multi-dimensional features of the lumbar disc through deep learning technology, so that it has high reliability and accuracy in the subsequent prediction of the protrusion grade. Compared with the traditional manual feature extraction method, deep learning can automatically identify and learn complex patterns and rules, improving the efficiency and effectiveness of feature extraction. In addition, combined with multi-dimensional features, it can provide more comprehensive information to help hospitals or doctors make more scientific judgments in clinical decision-making.

[0053] Specifically, the segmented lumbar intervertebral disc structural area data can be input into a pre-trained deep learning model to extract morphological features at different levels, wherein the deep learning model can reflect the morphological features of the lumbar intervertebral disc in the feature map extracted by the convolution layer; In this process, the segmented lumbar disc structure area data will be input into a fully trained deep learning model, usually a convolutional neural network (CNN). The hierarchical structure of CNN enables it to effectively extract the hierarchical features of the image. In the initial convolutional layer, the features extracted by the model may be basic morphological structures such as edges and corners. As the number of network layers increases, the extracted features gradually become more complex, and can capture the overall morphology of the lumbar disc and its detailed features.

[0054] The significance of this step is to accurately quantify the morphological characteristics of the lumbar intervertebral disc with the help of deep learning technology, laying the foundation for subsequent feature analysis and protrusion grade judgment. These morphological characteristics not only provide intuitive information about the structure, but also help identify the progression of the lesion, thereby providing data support for the diagnosis and treatment of spinal diseases.

[0055] When implementing this step, first ensure that the segmented lumbar disc structure area is standardized to meet the input requirements of the deep learning model. For example, use bilinear interpolation to resample the image and adjust the image size to the fixed size required by the model, such as 224×224 pixels. Next, convert the image data into a format suitable for model input, usually a three-dimensional array containing the number of channels, such as 1 for grayscale images. In order to improve the training efficiency of the model, the image can be normalized at the input stage to scale the pixel values ​​to the range of 0 to 1, which helps to speed up the convergence of the model.

[0056] The processed image is then input into the trained convolutional neural network (CNN). During the forward propagation of the model, CNN gradually extracts features through a series of convolutional layers and pooling layers. In each convolutional layer, different convolution kernels can detect different types of features (such as edges, corners, etc.), while the pooling layer can gradually reduce the dimension of the feature map and extract important high-order features. The feature map generated by each convolutional layer fully reflects the morphological and structural characteristics of the lumbar intervertebral disc, providing rich information for subsequent feature analysis.

[0057] Finally, the extracted feature maps are merged and flattened to form a one-dimensional feature vector for subsequent feature integration and analysis. This feature vector will fully reflect the geometric characteristics of the lumbar disc and lay the foundation for subsequent texture analysis and classification tasks. Through this implementation, the model effectively captures complex morphological features and provides a reliable basis for subsequent disease judgment.

[0058] Applying texture analysis on the extracted feature map to extract texture features to capture subtle texture changes in the lumbar disc structure, thereby reflecting the health status or pathological changes of the tissue; In this step, texture analysis is performed on the extracted feature map to capture and quantify the microstructural changes of the lumbar disc. This step may use a variety of texture feature extraction methods, such as gray level co-occurrence matrix (GLCM), local binary pattern (LBP), etc. These methods can analyze the subtle texture information in the image and reveal changes in tissue health and potential pathological conditions.

[0059] By extracting texture features, the health of the lumbar disc can be quantitatively evaluated. This not only helps to diagnose early pathological changes, but also provides doctors with more reliable data support for more scientific treatment decisions. At the same time, changes in texture features are usually related to the severity of clinical symptoms, which helps to develop personalized treatment plans.

[0060] In this step, texture features are first extracted from the feature map obtained in the previous stage. One or more texture analysis methods can be selected, such as gray-level co-occurrence matrix (GLCM), local binary pattern (LBP) or texture spectrum analysis. Taking GLCM as an example, this method can capture the texture features of an image by calculating the co-occurrence relationship of the grayscale values ​​of pixels in the image. In the implementation process, a specific distance and direction must be specified as needed to generate a gray-level co-occurrence matrix, which can quantify the relationship between pixels.

[0061] Subsequently, a series of texture features, such as contrast, energy, homogeneity, and entropy, are extracted from the generated gray-level co-occurrence matrix. These features can provide an effective quantitative description of the subtle changes in the structure of the lumbar disc, reflecting changes in its health status. In order to ensure that the extracted texture features are representative, the texture features can be calculated independently for different regions or segmented fragments, and certain statistical methods (such as taking the mean) can be maintained to improve the robustness of the features.

[0062] After the texture feature extraction is completed, these feature vectors are combined with the previously extracted morphological features to form a comprehensive feature descriptor. This process not only helps to find subtle changes in lumbar disc lesions, but also provides a rich and multi-dimensional information basis for subsequent classification and prediction, which helps to improve the accuracy and reliability of intelligent prediction.

[0063] The morphological features and the texture features are integrated to generate a multi-dimensional feature vector to comprehensively describe the structural features of the lumbar intervertebral disc.

[0064] This step integrates the morphological features extracted from the first two steps with the texture features to achieve a comprehensive feature description. By combining features from different sources and dimensions, an overall feature vector can be formed. This multidimensional feature vector can more comprehensively reflect the structural characteristics and status of the lumbar intervertebral disc.

[0065] The generation of multi-dimensional feature vectors can provide more comprehensive and detailed feature information for subsequent protrusion level prediction. This form of comprehensive features can ensure that the deep learning model takes into account changes in morphology and texture when performing classification and prediction, thereby improving the prediction accuracy and reliability of the model and providing solid data support for the diagnosis and evaluation of spinal disc lesions.

[0066] In actual operation, the previously extracted morphological features and texture features need to be sorted out first. To ensure that the features can be effectively combined, the extracted features are usually standardized to be in the same numerical range (such as 0 to 1). This step helps to eliminate the scale differences between features, thereby enhancing the effect of feature fusion. Standardization can be achieved through linear transformation, that is, scaling using the maximum and minimum values ​​of the features to ensure that all features are under a unified standard.

[0067] Next, a simple concatenation method can be used to combine the morphological features and the texture features. In practice, the two sets of features can be directly connected on a one-dimensional vector to form a longer feature vector. For example, if the dimension of the morphological features is n1 and the dimension of the texture features is n2, then the dimension of the resulting feature vector will be (n1+n2). In addition, data dimensionality reduction methods, such as principal component analysis (PCA), can be considered to reduce the feature dimension to ensure a balance between the computational efficiency and effect of the model.

[0068] Finally, the generated multidimensional feature vector is used in the subsequent classification model, especially in the input of the ensemble learning model, and the results of multiple classifiers are combined through voting mechanism or weighted average method to predict the grade of lumbar disc herniation. This step not only integrates morphological and texture information, but also effectively improves the robustness and prediction performance of the model, ensuring that the intelligent prediction method has the reliability and effectiveness of clinical application.

[0069] S204, based on the extracted multi-dimensional feature vector, an integrated learning model is used to predict the herniation grade, and the final lumbar disc herniation grade prediction result is obtained by combining the results of multiple classifiers and performing a voting mechanism or weighted averaging.

[0070] In this step, the multidimensional feature vector extracted from the CT image is used to predict the grade of lumbar disc herniation by applying an integrated learning model. In the specific implementation, the multidimensional feature vector is input into multiple different types of classifiers, such as logistic regression, support vector machine (SVM), random forest, etc. These classifiers analyze the input feature vector independently and output the prediction results according to their own algorithm rules. In order to obtain a more reliable prediction, the results of these classifiers are combined by a voting mechanism and a weighted average method. The voting mechanism means that each classifier votes on the result and finally selects the grade with the most votes as the prediction result; while the weighted average assigns different weights to the prediction results of each classifier according to its performance on the training set, so that the classifiers with better performance are given priority when merging predictions. This process not only improves the robustness of the model, but also effectively reduces the error that may be caused by a single classifier.

[0071] The core significance of this step is to optimize the prediction results of lumbar disc herniation grade through ensemble learning technology. Ensemble learning uses the information of multiple classifiers and fully integrates the advantages of different models, which can reduce the limitations and misjudgment rate of a single model in specific situations. Through voting and weighted average mechanisms, the accuracy and stability of predictions can be greatly improved, helping doctors to more effectively identify the pathological status of lumbar discs, thereby providing a scientific basis for clinical decision-making. Especially in complex cases, the ensemble learning model can combine the knowledge of each classifier to form a more comprehensive judgment and improve the quality of diagnosis and treatment of patients.

[0072] First, when building an integrated learning model, select a variety of classifiers suitable for predicting the grade of lumbar disc herniation. You can consider using random forest, support vector machine (SVM) and gradient boosting machine (GBM). Random forest can effectively reduce the risk of overfitting and handle high-dimensional data by constructing multiple decision trees and determining the final prediction based on the voting results. SVM is suitable for processing nonlinear features and uses kernel functions to map to high-dimensional space to achieve more accurate classification. Gradient boosting machine improves the predictive ability of the model through step-by-step optimization, by successively generating weak classifiers and combining them to form strong classifiers. These classifiers will train the extracted multidimensional feature vectors separately during the training phase to ensure that each model can independently learn the features and patterns in the data.

[0073] Secondly, the diversity and balance of the data set must be ensured during the training process so that each classifier can learn different types of features. For example, the original data set can be randomly divided into a training set and a test set. The training set is used to train the model, while the test set is used to verify the performance of the model and ensure that different levels of lumbar disc herniation cases are covered. The use of cross-validation technology can improve the generalization ability of the model and reduce the error caused by data division. In this process, it is also crucial to tune the hyperparameters of each classifier. For example, for SVM, choosing the appropriate kernel function and regularization parameters can significantly improve the performance of the model. This stage will lay a solid foundation for subsequent integrated predictions.

[0074] Finally, in the actual prediction stage, the extracted multidimensional feature vectors are input into each classifier for independent prediction. Each classifier will output its judgment on the grade of lumbar disc herniation, for example, it may output the results of grade 3, grade 2 or grade 1. After obtaining the prediction of each classifier, two strategies are used to integrate these results: the first is a voting mechanism, in which the grade with the most votes is selected as the final prediction; the second is a weighted average, in which different weights are given to each classifier according to its performance on the validation set, so that the model with better performance is preferred when the results are merged. For example, if both random forest and support vector machine predict grade 3 herniation, while logistic regression only predicts grade 2 herniation, and the former two perform well on the validation set, the final predicted grade will be more likely to be grade 3 herniation. This integrated method ensures the robustness and accuracy of the prediction results, providing more reliable support for subsequent clinical decision-making.

[0075] It can be seen that, according to the current original CT image data, multi-dimensional image reconstruction processing is performed to obtain a reconstructed CT reconstructed image; according to the reconstructed CT reconstructed image, dynamic region segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; according to the lumbar intervertebral disc structure region, feature extraction processing is performed, and a deep learning network is used to extract multi-dimensional features of the lumbar intervertebral disc structure region to obtain a multi-dimensional feature vector; according to the multi-dimensional feature vector, an integrated learning model is used to predict the protrusion level, and by combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation level prediction result, which can effectively improve the accuracy and efficiency of lumbar disc herniation prediction.

[0076] Another embodiment of the present invention provides an intelligent prediction system for lumbar disc herniation grade based on CT images, see Figure 3 , the system may include: The reconstruction module 301 is used to perform multi-dimensional image reconstruction processing according to the current original CT image data of the patient to improve the image resolution and contrast and obtain a reconstructed CT reconstructed image; The segmentation module 302 is used to perform dynamic region segmentation based on the reconstructed CT reconstruction image, and segment the lumbar intervertebral disc and its surrounding tissues through an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; The extraction module 303 is used to perform feature extraction processing based on the segmented lumbar intervertebral disc structure area, and use a deep learning network to extract multi-dimensional features of the lumbar intervertebral disc structure area to obtain a multi-dimensional feature vector; The prediction module 304 is used to predict the protrusion grade using an integrated learning model based on the extracted multi-dimensional feature vector, and obtain the final lumbar disc herniation grade prediction result by combining the results of multiple classifiers and performing a voting mechanism or weighted averaging.

[0077] It can be seen that, according to the current original CT image data, multi-dimensional image reconstruction processing is performed to obtain a reconstructed CT reconstructed image; according to the reconstructed CT reconstructed image, dynamic region segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; according to the lumbar intervertebral disc structure region, feature extraction processing is performed, and a deep learning network is used to extract multi-dimensional features of the lumbar intervertebral disc structure region to obtain a multi-dimensional feature vector; according to the multi-dimensional feature vector, an integrated learning model is used to predict the protrusion level, and by combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation level prediction result, which can effectively improve the accuracy and efficiency of lumbar disc herniation prediction.

[0078] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0079] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps: S201, performing multi-dimensional image reconstruction processing according to the current original CT image data of the patient to improve the image resolution and contrast, and obtain a reconstructed CT reconstructed image; S202, performing dynamic region segmentation based on the reconstructed CT reconstruction image, segmenting the lumbar intervertebral disc and its surrounding tissues by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; S203, performing feature extraction processing based on the segmented lumbar intervertebral disc structure region, using a deep learning network to extract multi-dimensional features of the lumbar intervertebral disc structure region, and obtaining a multi-dimensional feature vector; S204, based on the extracted multi-dimensional feature vector, an integrated learning model is used to predict the herniation grade, and the final lumbar disc herniation grade prediction result is obtained by combining the results of multiple classifiers and performing a voting mechanism or weighted averaging.

[0080] It can be seen that, according to the current original CT image data, multi-dimensional image reconstruction processing is performed to obtain a reconstructed CT reconstructed image; according to the reconstructed CT reconstructed image, dynamic region segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; according to the lumbar intervertebral disc structure region, feature extraction processing is performed, and a deep learning network is used to extract multi-dimensional features of the lumbar intervertebral disc structure region to obtain a multi-dimensional feature vector; according to the multi-dimensional feature vector, an integrated learning model is used to predict the protrusion level, and by combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation level prediction result, which can effectively improve the accuracy and efficiency of lumbar disc herniation prediction.

[0081] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0082] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0083] Specifically, in this embodiment, the processor may be configured to perform the following steps through a computer program: S201, performing multi-dimensional image reconstruction processing according to the current original CT image data of the patient to improve the image resolution and contrast, and obtain a reconstructed CT reconstructed image; S202, performing dynamic region segmentation based on the reconstructed CT reconstruction image, segmenting the lumbar intervertebral disc and its surrounding tissues by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; S203, performing feature extraction processing based on the segmented lumbar intervertebral disc structure region, using a deep learning network to extract multi-dimensional features of the lumbar intervertebral disc structure region, and obtaining a multi-dimensional feature vector; S204, based on the extracted multi-dimensional feature vector, an integrated learning model is used to predict the herniation grade, and the final lumbar disc herniation grade prediction result is obtained by combining the results of multiple classifiers and performing a voting mechanism or weighted averaging.

[0084] It can be seen that, according to the current original CT image data, multi-dimensional image reconstruction processing is performed to obtain a reconstructed CT reconstructed image; according to the reconstructed CT reconstructed image, dynamic region segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain a segmented lumbar intervertebral disc structure region; according to the lumbar intervertebral disc structure region, feature extraction processing is performed, and a deep learning network is used to extract multi-dimensional features of the lumbar intervertebral disc structure region to obtain a multi-dimensional feature vector; according to the multi-dimensional feature vector, an integrated learning model is used to predict the protrusion level, and by combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation level prediction result, which can effectively improve the accuracy and efficiency of lumbar disc herniation prediction.

[0085] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. An intelligent prediction method for lumbar disc herniation grade based on CT images, characterized in that: The method comprises: According to the current original CT image data of the patient, multi-dimensional image reconstruction processing is performed to improve the image resolution and contrast, and a reconstructed CT reconstructed image is obtained; According to the reconstructed CT images, dynamic region segmentation is performed, and the lumbar intervertebral disc and its surrounding tissues are segmented by an adaptive segmentation algorithm to obtain the segmented lumbar intervertebral disc structure region; According to the segmented lumbar disc structure area, feature extraction processing is performed, and the multi-dimensional features of the lumbar disc structure area are extracted using a deep learning network to obtain a multi-dimensional feature vector; According to the extracted multi-dimensional feature vector, the ensemble learning model is used to predict the herniation grade. By combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation grade prediction result.

2. The method according to claim 1, characterized in that The multi-dimensional image reconstruction process is performed according to the current original CT image data of the patient to improve the image resolution and contrast to obtain a reconstructed CT reconstructed image, including: The historical CT image data set is preliminarily reconstructed using the filtered back-projection method to obtain a preliminary reconstructed image; Taking the preliminary reconstructed image as input, a convolutional neural network is used to train the specific noise and artifacts of the preliminary reconstructed image, wherein the convolutional neural network optimizes network parameters to achieve noise suppression and contrast enhancement by comparing the real preset high-quality image with the preliminary reconstructed image; The patient's current original CT image data is input into the trained convolutional neural network, and the reconstructed CT image is output.

3. The method according to claim 2, characterized in that The method of performing dynamic region segmentation based on the reconstructed CT images, segmenting the lumbar intervertebral disc and its surrounding tissues by an adaptive segmentation algorithm, and obtaining the segmented lumbar intervertebral disc structure region includes: The reconstructed CT image is divided into a number of overlapping local area image blocks, and the size of each local area image block is adaptively adjusted according to the resolution of the image to ensure that important structural information is contained; For each local region image block, a grayscale histogram of the local region image block is calculated to obtain local grayscale distribution characteristics, and a threshold of each local region image block is dynamically determined; The determined threshold is used as the boundary to perform region growing and obtain the segmented lumbar disc structure region.

4. The method according to claim 3, characterized in that The method performs feature extraction processing based on the segmented lumbar intervertebral disc structure area, and uses a deep learning network to extract multi-dimensional features of the lumbar intervertebral disc structure area to obtain a multi-dimensional feature vector, including: Inputting the segmented lumbar intervertebral disc structural area data into a pre-trained deep learning model to extract morphological features at different levels, wherein the deep learning model can reflect the morphological features of the lumbar intervertebral disc in the feature map extracted by the convolution layer; Applying texture analysis on the extracted feature map to extract texture features to capture subtle texture changes in the lumbar disc structure, thereby reflecting the health status or pathological changes of the tissue; The morphological features and the texture features are integrated to generate a multi-dimensional feature vector to comprehensively describe the structural features of the lumbar intervertebral disc.

5. An intelligent prediction system for lumbar disc herniation grade based on CT images, characterized in that: The system comprises: A reconstruction module is used to perform multi-dimensional image reconstruction processing based on the patient's current original CT image data to improve the image resolution and contrast and obtain a reconstructed CT reconstructed image; A segmentation module is used to perform dynamic region segmentation based on the reconstructed CT reconstruction image, segment the lumbar intervertebral disc and its surrounding tissues through an adaptive segmentation algorithm, and obtain a segmented lumbar intervertebral disc structure region; An extraction module is used to perform feature extraction processing based on the segmented lumbar disc structure area, and use a deep learning network to extract multi-dimensional features of the lumbar disc structure area to obtain a multi-dimensional feature vector; The prediction module is used to predict the protrusion grade using an integrated learning model based on the extracted multi-dimensional feature vector. By combining the results of multiple classifiers, a voting mechanism or weighted average is performed to obtain the final lumbar disc herniation grade prediction result.

6. The system according to claim 5, characterized in that The reconstruction module is specifically used for: The historical CT image data set is preliminarily reconstructed using the filtered back-projection method to obtain a preliminary reconstructed image; Taking the preliminary reconstructed image as input, a convolutional neural network is used to train the specific noise and artifacts of the preliminary reconstructed image, wherein the convolutional neural network optimizes network parameters to achieve noise suppression and contrast enhancement by comparing the real preset high-quality image with the preliminary reconstructed image; The patient's current original CT image data is input into the trained convolutional neural network, and the reconstructed CT image is output.

7. The system according to claim 6, characterized in that The segmentation module is specifically used for: The reconstructed CT image is divided into a number of overlapping local area image blocks, and the size of each local area image block is adaptively adjusted according to the resolution of the image to ensure that important structural information is contained; For each local region image block, a grayscale histogram of the local region image block is calculated to obtain local grayscale distribution characteristics, and a threshold of each local region image block is dynamically determined; The determined threshold is used as the boundary to perform region growing and obtain the segmented lumbar disc structure region.

8. The system according to claim 7, characterized in that The extraction module is specifically used for: Inputting the segmented lumbar intervertebral disc structural area data into a pre-trained deep learning model to extract morphological features at different levels, wherein the deep learning model can reflect the morphological features of the lumbar intervertebral disc in the feature map extracted by the convolution layer; Applying texture analysis on the extracted feature map to extract texture features to capture subtle texture changes in the lumbar disc structure, thereby reflecting the health status or pathological changes of the tissue; The morphological features and the texture features are integrated to generate a multi-dimensional feature vector to comprehensively describe the structural features of the lumbar intervertebral disc.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 4 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 4.

Images