Artificial intelligence-based degenerative cervical spinal disease image analysis method and system

Abstract

The invention discloses an artificial intelligence-based degenerative cervical spinal disease image analysis method and system, relates to the technical field of artificial intelligence, and solves the problems that in the prior art, image report standards are not uniform, image analysis is rough, patient anxiety is increased, medical resources are wasted, obvious bias exists between observers in DCM judgment, and the image report standards are not uniform. And the efficiency of image analysis of the degenerative cervical spinal disease is relatively low. Obtaining a segmentation label through the cervical vertebra anatomy segmentation model; obtaining a spinal canal stenosis prediction grade of which the segmentation label is an intervertebral canal through an intervertebral canal stenosis quantitative grading model; obtaining a spinal cord compression result of which the segmentation tag is a cervical spinal cord through a spinal cord compression classification model; the analysis report of the cervical vertebra nuclear magnetic image is obtained through the cervical vertebra image analysis model, the image report is standardized, and the improved segmentation model, the classification model and the cervical vertebra image analysis model are used for processing, so that the analysis accuracy is improved, and the efficiency of degenerative cervical spinal disease image analysis is improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and specifically relates to an imaging analysis method and system for degenerative cervical spondylomyelopathy based on artificial intelligence. Background Art

[0002] Degenerative cervical spondylomyelopathy (DCM) is a chronic and progressive degenerative disease of the cervical spine and the most common cause of chronic spinal cord injury in adults. Such diseases mainly include cervical spondylotic myelopathy, ossification of the posterior longitudinal ligament, ossification of the ligamentum flavum, and intervertebral disc degeneration. With the increase of age, almost everyone will show imaging manifestations of cervical spine degeneration, and early diagnosis is particularly important for preventing further spinal cord injury. Although clinical symptoms and signs are also important criteria for diagnosing DCM, the mechanisms of human behavioral adaptation and neural plasticity may mask the disease deterioration. Therefore, an objective criterion for imaging to evaluate the severity of spinal cord compression and spinal canal stenosis is very meaningful.

[0003] At present, the diagnosis and treatment level of degenerative cervical spondylomyelopathy in community medical institutions is limited, and the differences in the professional cognition of this disease among physicians in each unit and the imaging report standards provided by imaging physicians have led to patients' lack of correct understanding of DCM, and often there will be situations of missed diagnosis or overdiagnosis. Rough imaging analysis often exacerbates patients' anxiety and at the same time leads to waste of medical resources and unnecessary burden on social medical insurance costs. At the same time, for large-sample data research, the inter-observer bias in the evaluation of DCM is very large, which is not conducive to the development of multi-center big data clinical research. Therefore, it is particularly important to develop a standardized and accurate deep learning algorithm to analyze cervical magnetic resonance images to assist clinical and imaging physicians in diagnosing cervical spinal canal stenosis caused by DCM and issuing a standardized imaging analysis report. Summary of the Invention

[0004] This application aims to solve at least one of the technical problems existing in the prior art; for this purpose, this application proposes an imaging analysis method and system for degenerative cervical spondylomyelopathy based on artificial intelligence, which is used to solve the technical problems of inconsistent imaging report standards in the prior art, rough imaging analysis increasing patients' anxiety and wasting medical resources, and significant inter-observer bias in DCM evaluation, resulting in low efficiency of imaging analysis of degenerative cervical spondylomyelopathy.

[0005] To achieve the above object, the first aspect of this application provides an imaging analysis system for degenerative cervical spondylomyelopathy based on artificial intelligence, including: a data acquisition module, a data analysis module, and a database;

[0006] The data acquisition module: obtains cervical magnetic resonance images through a data acquisition device;

[0007] The data analysis module: Inputs cervical magnetic resonance images into the cervical spine anatomical segmentation model to obtain segmentation labels. The cervical spine anatomical segmentation model is obtained by improving the nnU-Net architecture through the cross pseudo-supervision (CPS) framework; inputs the detection images with segmentation labels of the intervertebral canal into the intervertebral canal stenosis quantification and grading model to obtain the spinal canal stenosis prediction grade. The intervertebral canal stenosis quantification and grading model is constructed through an artificial intelligence model; inputs the detection images with segmentation labels of the cervical spinal cord into the spinal cord compression classification model to obtain the spinal cord compression result. The spinal cord compression classification model is constructed through the ResNet network; inputs several detection images corresponding to the segmentation labels, the spinal canal stenosis prediction grade, and the spinal cord compression result into the cervical spine image analysis model to obtain an analysis report. The cervical spine image analysis model is constructed through a large language model;

[0008] The database is used to store a number of historical data required for training the model.

[0009] Through the above steps, this application standardizes the imaging report, uses the improved segmentation model, cooperates with the classification model and the cervical spine image analysis model for processing, and gives an imaging analysis report, providing an efficient and convenient imaging auxiliary analysis tool, thereby helping doctors accurately diagnose degenerative cervical spondylomyelopathy and improving the accuracy of the analysis; grades the spinal canal stenosis and spinal cord compression to assist users in better evaluating the severity of intervertebral disc degeneration.

[0010] Furthermore, the cervical spine anatomical segmentation model is obtained by improving the nnU-Net architecture through the cross pseudo-supervision (CPS) framework, including:

[0011] Obtain a number of historical cervical magnetic resonance images and their corresponding segmentation labels;

[0012] Divide a number of historical cervical magnetic resonance images and their corresponding segmentation labels into training data, validation data, and test data; perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0013] Select the improved nnU-Net architecture as the basic model;

[0014] Train the basic model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0015] Verify the pre-trained model on the test set, and finally obtain the cervical spine anatomical segmentation model with the input of cervical magnetic resonance images and the output of segmentation labels.

[0016] Furthermore, the basic model is a neural network architecture based on U-Net and residual blocks, and is configured and trained through the nnU-Net framework, specifically including an encoder, a decoder, a skip connection module, and a feature combination module.

[0017] Furthermore, based on the basic model, the forced spacing setting, optimizer, and preprocessing process are optimized, including:

[0018] Forced spacing setting: In the preprocessing stage, by modifying the spacing parameters of the original cervical spine MRI images, the patch size input into the model can be constrained to an optimal value;

[0019] Optimizer: The Ranger optimizer is used to better adapt to the nnU-Net model improved by the semi-supervised learning idea of the CPS framework;

[0020] Preprocessing process: Advanced data augmentation methods such as contrast adjustment and noise addition are added.

[0021] Furthermore, the loss function adopted by the Ranger optimizer is obtained from the Dice loss and cross-entropy loss, including:

[0022] Obtain the Dice loss L Dice and the cross-entropy loss L Cross-Entropy ;

[0023] Calculate the loss function L adopted by the Ranger optimizer through the formula;

[0024] L = L Dice + L Cross-Entropy + λ(L CPSlabeled + L CPSunlabeled ); where λ represents the weight, λ ∈ (0, 1); L CPSlabeled and L CPSunlabeled respectively represent the losses of labeled data and unlabeled data.

[0025] Furthermore, the intervertebral canal stenosis quantification and grading model is constructed through an artificial intelligence model, including:

[0026] Obtain a number of historical detection images with the segmentation label of the intervertebral canal and their corresponding intervertebral canal stenosis grades;

[0027] Divide a number of historical detection images and their corresponding intervertebral canal stenosis grades into training data, validation data, and test data; perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0028] Select an artificial intelligence model as the basic model;

[0029] Train the basic model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0030] By validating the pre-trained model on the test set, a quantitative grading model for intervertebral canal stenosis is finally obtained, with the input being the detection image with the segmentation label of the intervertebral canal and the output being the predicted grade of spinal canal stenosis.

[0031] Further, the spinal cord compression classification model is constructed through the ResNet network, including:

[0032] Obtain a number of historical detection images with the segmentation label of the cervical spinal cord and their corresponding spinal cord compression results;

[0033] Divide a number of historical detection images and their corresponding spinal cord compression results into training data, validation data, and test data; perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0034] Select the ResNet network model as the basic model;

[0035] Train the basic model through the training set and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0036] By validating the pre-trained model on the test set, a spinal cord compression classification model is finally obtained, with the input being the detection image with the segmentation label of the cervical spinal cord and the output being the spinal cord compression result.

[0037] Further, the ResNet network model adopts at least N layers of ResNet network model to perform morphological analysis on the detection image with the segmentation label of the cervical spinal cord through multi-layer feature extraction and multi-scale feature fusion to obtain the spinal cord compression result of whether there is compression, as well as the compression location and degree; where N is a positive integer.

[0038] Further, the cervical vertebra image analysis model is constructed through a large language model, including:

[0039] Obtain a number of historical detection images corresponding to the segmentation label, historical predicted grades of spinal canal stenosis, historical spinal cord compression results, and their corresponding historical analysis reports;

[0040] Divide a number of historical detection images corresponding to the segmentation label, historical predicted grades of spinal canal stenosis, historical spinal cord compression results, and their corresponding historical analysis reports into training data, validation data, and test data; perform data preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0041] Select the large language network model as the basic model;

[0042] Train the basic model through the training set and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0043] By validating the pre-trained model on the test set, a cervical spine image analysis model is finally obtained, with the input being several detection images corresponding to the segmentation labels, the predicted grade of spinal canal stenosis, and the results of spinal cord compression, and the output being an analysis report.

[0044] Another aspect of the present invention provides an artificial intelligence-based imaging analysis method for degenerative cervical spondylomyelopathy, including:

[0045] S0: Obtain cervical spine MRI images;

[0046] S1: Input the cervical spine MRI images into the cervical spine anatomical segmentation model to obtain segmentation labels;

[0047] S2: Input the detection images with the segmentation label of the intervertebral canal into the intervertebral canal stenosis quantification and grading model to obtain the predicted grade of spinal canal stenosis;

[0048] S3: Input the detection images with the segmentation label of the cervical spinal cord into the spinal cord compression classification model to obtain the results of spinal cord compression;

[0049] S4: Input several detection images corresponding to the segmentation labels, the predicted grade of spinal canal stenosis, and the results of spinal cord compression into the cervical spine image analysis model to obtain an analysis report.

[0050] Compared with the prior art, the beneficial effects of the present application are:

[0051] 1. In the present application, the cervical spine MRI images are input into the cervical spine anatomical segmentation model to obtain segmentation labels; the detection images with the segmentation label of the intervertebral canal are input into the intervertebral canal stenosis quantification and grading model to obtain the predicted grade of spinal canal stenosis; the detection images with the segmentation label of the cervical spinal cord are input into the spinal cord compression classification model to obtain the results of spinal cord compression; several detection images corresponding to the segmentation labels, the predicted grade of spinal canal stenosis, and the results of spinal cord compression are input into the cervical spine image analysis model to obtain an analysis report, standardizing the imaging report, using an improved segmentation model, cooperating with a classification model and a cervical spine image analysis model for processing, improving the accuracy of analysis, and thus enhancing the efficiency of imaging analysis for degenerative cervical spondylomyelopathy.

[0052] 2. In the present application, the nnU-Net network improved by a semi-supervised deep learning framework based on the CPS idea is adopted, and a specific optimizer is adapted. The segmentation and recognition accuracy of the model is improved through the combination of the model and the optimizer, and the performance of the model in the case of limited annotation is improved based on the semi-supervised training method.

[0053] 3. This application generates an analysis report by using a pre-trained large language model. It can utilize the basic knowledge and question-and-answer capabilities of the large language model to expand services, provide personalized analysis and guidance for nuclear magnetic resonance images to users, and at the same time achieve standardized analysis results, providing support for large-sample data research and multi-center clinical research. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0055] Figure 1 Schematic diagram of the principle of an artificial intelligence-based imaging analysis system for degenerative cervical spondylomyelopathy of the present application;

[0056] Figure 2 Flowchart of an artificial intelligence-based imaging analysis method for degenerative cervical spondylomyelopathy of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The following will clearly and completely describe the technical solutions of the present application in combination with the embodiments. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts belong to the scope of protection of the present application.

[0058] Please refer to Figure 1 , the first aspect embodiment of the present application provides an artificial intelligence-based imaging analysis system for degenerative cervical spondylomyelopathy, including: a data acquisition module, a data analysis module, and a database;

[0059] Data acquisition module: Obtain cervical magnetic resonance images through data acquisition devices; the data acquisition devices include various sensors, etc.;

[0060] Data analysis module: Input the cervical spine MRI images into the cervical spine anatomical segmentation model to obtain segmentation labels. The cervical spine anatomical segmentation model is obtained by improving the nnU-Net architecture through the Cross Pseudo Supervision (CPS) framework. The segmentation labels refer to the names after segmenting the cervical spine MRI images, including cervical vertebrae, intervertebral discs, cervical spinal cord, cerebrospinal fluid, paravertebral muscle tissues, etc.; Input the detection images with the segmentation label of intervertebral canal into the intervertebral canal stenosis quantification and grading model to obtain the spinal canal stenosis prediction grade. The intervertebral canal stenosis quantification and grading model is constructed through an artificial intelligence model. The spinal canal stenosis prediction grade refers to the degree of intervertebral canal stenosis, which can be divided into four grades: 1, 2, 3, and 4 in this embodiment; Input the detection images with the segmentation label of cervical spinal cord into the spinal cord compression classification model to obtain the spinal cord compression result. The spinal cord compression classification model is constructed through the ResNet network. The spinal cord compression results include two types: with compression and without compression. With compression includes three types: anterior compression, posterior compression, and anterior-posterior compression; Input the several detection images corresponding to the segmentation labels, the spinal canal stenosis prediction grade, and the spinal cord compression result into the cervical spine image analysis model to obtain an analysis report. The cervical spine image analysis model is constructed through a large language model. The analysis report is a comprehensive analysis of the cervical spine MRI images;

[0061] The database is used to store a number of historical data required for training the model.

[0062] The cervical spine anatomical segmentation model in this embodiment is obtained by improving the nnU-Net architecture through the Cross Pseudo Supervision (CPS) framework, including:

[0063] Obtain a number of historical cervical spine MRI images and their corresponding segmentation labels;

[0064] Divide the number of historical cervical spine MRI images and their corresponding segmentation labels into training data, validation data, and test data; Perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set; The ratio among the training set, the test set, and the validation set is 7:2:1;

[0065] Select the improved nnU-Net architecture as the basic model;

[0066] Train the basic model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0067] Verify the pre-trained model on the test set, and finally obtain the cervical spine anatomical segmentation model with the cervical spine MRI images as the input and the segmentation labels as the output.

[0068] The basic model in this embodiment is a neural network architecture based on U-Net and residual blocks, and is configured and trained through the nnU-Net framework, specifically including an encoder, a decoder, a skip connection module, and a feature combination module.

[0069] In this embodiment, based on the basic model, the forced spacing setting, optimizer, and preprocessing process are optimized, including:

[0070] Forced spacing setting: In the preprocessing stage, by modifying the spacing parameters of the original cervical MRI images, the patch size input into the model can be constrained to an optimal value; by adjusting the spacing parameters, the input patch size is neither too large nor too small, reaching a balance point, ensuring that the input patch size is constrained to the optimal value, reducing the computational burden of the model in the inference stage, not only optimizing the inference speed, improving the segmentation efficiency, but also reducing resource consumption;

[0071] Optimizer: The Ranger optimizer is used to better adapt to the nnU-Net model improved by the semi-supervised learning idea of the CPS framework;

[0072] Preprocessing process: Advanced data augmentation methods such as contrast adjustment and noise addition are added. In this embodiment, by adding advanced data augmentation methods such as contrast adjustment and noise addition, the adaptability of the model to different types of images is further improved. At the same time, for input images with different resolutions, a multi-scale processing method is adopted to improve the detection ability of the model for lesion areas of different sizes.

[0073] The loss function adopted by the Ranger optimizer in this embodiment is obtained from the Dice loss and the cross-entropy loss, including:

[0074] Obtain the Dice loss L Dice and the cross-entropy loss L Cross-Entropy ; where the Dice loss and the cross-entropy loss are configured by the default of nnU-Net and have been proven to be robust in medical image segmentation tasks;

[0075] Calculate the loss function L adopted by the Ranger optimizer through the formula;

[0076] L = L Dice + L Cross-Entropy + λ(L CPSlabeled + L CPSunlabeled ); where λ represents the weight, λ ∈ (0, 1), and the specific value is set according to experience; L CPSlabeled and L CPSunlabeled respectively represent the losses of labeled data and unlabeled data.

[0077] Through the above steps, this embodiment integrates a semi-supervised deep learning architecture with the CPS concept to optimize the nnU-Net network and configures an adapted optimizer. Through the synergistic effect of this model and the optimizer, the segmentation and recognition accuracy of the model are enhanced. At the same time, with the help of the semi-supervised training strategy, this solution effectively improves the performance of the model in the case of limited labeled data.

[0078] The intervertebral canal stenosis quantification and grading model in this embodiment is constructed through an artificial intelligence model, including:

[0079] Obtain a number of historical detection images with the segmentation label of the intervertebral canal and their corresponding intervertebral canal stenosis grades;

[0080] Divide a number of historical detection images and their corresponding intervertebral canal stenosis grades into training data, validation data, and test data; perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set; the ratio among the training set, the test set, and the validation set is 7:2:1;

[0081] Select an artificial intelligence model as the basic model; the artificial intelligence model includes a convolutional neural network model, etc.;

[0082] Train the basic model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0083] Verify the pre-trained model on the test set, and finally obtain an intervertebral canal stenosis quantification and grading model with the detection image with the segmentation label of the intervertebral canal as the input and the predicted grade of spinal canal stenosis as the output.

[0084] The intervertebral canal stenosis grade in this embodiment is obtained by multiple disciplinary experts annotating the cervical spine MRI images. The annotation standard of this embodiment is: at least 2 chief orthopedic surgeons and at least 1 chief radiologist determine the stenosis grade of the corresponding segments of the cervical spinal canal C2 / 3 - C7 / T1 within the viewing window of the same model machine. Select the grading judgment result with more than 2 / 3 experts in agreement as the gold standard. For the judgment results that are all inconsistent, re-determine after a 1-month "washout period" until the standard is met; the entire image viewing process is based on the window operation, and there is no mutual interference in judgment among the viewers. During the annotation process of the cervical spine sagittal view, there are also corresponding cross-sections to assist the viewers. Before interpreting the cervical spine MRI images, the image data has been desensitized.

[0085] The spinal cord compression classification model in this embodiment is constructed through a ResNet network, including:

[0086] Obtain a number of historical detection images with the segmentation label of the cervical spinal cord and their corresponding spinal cord compression results; the spinal cord compression results are the compression results comprehensively determined by relevant experts based on the cerebrospinal fluid area and curvature;

[0087] Divide a number of historical detection images and their corresponding spinal cord compression results into training data, validation data, and test data; perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set; the ratio among the training set, the test set, and the validation set is 7:2:1;

[0088] Select the ResNet network model as the base model;

[0089] Train the base model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0090] Verify the pre-trained model on the test set, and finally obtain a spinal cord compression classification model with the input being the detection image with the segmentation label of the cervical spinal cord and the output being the spinal cord compression result.

[0091] Through the above steps, this embodiment not only identifies tissues and cervical spinal degenerative changes in cervical magnetic resonance images, but also can judge the level of spinal stenosis, the compression condition and the compression location, achieving a more comprehensive analysis effect of images using artificial intelligence and improving the efficiency of image analysis of degenerative cervical spondylomyelopathy.

[0092] The ResNet network model in this embodiment adopts at least N layers of ResNet network model to perform morphological analysis on the detection image with the segmentation label of the cervical spinal cord through multi-layer feature extraction and multi-scale feature fusion to obtain the spinal cord compression result of whether there is compression, as well as the compression location and compression degree; where N is a positive integer, and in this embodiment, N is set to 30.

[0093] The cervical spine image analysis model in this embodiment is constructed through a large language model, including:

[0094] Obtain a number of historical detection images corresponding to the segmentation label, historical spinal stenosis prediction levels, historical spinal cord compression results and their corresponding historical analysis reports;

[0095] Divide a number of historical detection images corresponding to the segmentation label, historical spinal stenosis prediction levels, historical spinal cord compression results and their corresponding historical analysis reports into training data, validation data and test data; perform data preprocessing on the training data, validation data and test data to obtain a training set, a validation set and a test set; the ratio among the training set, the test set and the validation set is 7:2:1;

[0096] Select a large language network model as the base model; the large language model includes models such as Llama-3-8B model;

[0097] Train the base model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0098] Verify the pre-trained model on the test set, and finally obtain a cervical spine image analysis model with the input being a number of detection images corresponding to the segmentation label, the spinal stenosis prediction level and the spinal cord compression result, and the output being an analysis report.

[0099] Please refer toFigure 2 , another embodiment of the present application provides an artificial intelligence-based imaging analysis method for degenerative cervical spondylomyelopathy, including:

[0100] S0: Obtain cervical magnetic resonance imaging;

[0101] S1: Input the cervical magnetic resonance imaging into a cervical spine anatomical segmentation model to obtain segmentation labels;

[0102] S2: Input the detection image with the segmentation label of the intervertebral canal into an intervertebral canal stenosis quantification and grading model to obtain a spinal canal stenosis prediction grade;

[0103] S3: Input the detection image with the segmentation label of the cervical spinal cord into a spinal cord compression classification model to obtain a spinal cord compression result;

[0104] S4: Input several detection images corresponding to the segmentation labels, the spinal canal stenosis prediction grade, and the spinal cord compression result into a cervical spine image analysis model to obtain an analysis report.

[0105] Some of the data in the above formula are calculated by removing the dimension and taking its numerical value. The formula is obtained by software simulation of a large amount of collected data to get a formula closest to the actual situation; the preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained by simulation of a large amount of data.

[0106] The working principle of the present application: By obtaining cervical magnetic resonance imaging; inputting the cervical magnetic resonance imaging into a cervical spine anatomical segmentation model to obtain segmentation labels; inputting the detection image with the segmentation label of the intervertebral canal into an intervertebral canal stenosis quantification and grading model to obtain a spinal canal stenosis prediction grade; inputting the detection image with the segmentation label of the cervical spinal cord into a spinal cord compression classification model to obtain a spinal cord compression result; inputting several detection images corresponding to the segmentation labels, the spinal canal stenosis prediction grade, and the spinal cord compression result into a cervical spine image analysis model to obtain an analysis report, standardizing the imaging report, using an improved segmentation model, cooperating with a classification model and a cervical spine image analysis model for processing, improving the accuracy of the analysis, thereby enhancing the efficiency of imaging analysis of degenerative cervical spondylomyelopathy, avoiding the problems of inconsistent imaging report standards in the prior art, rough imaging analysis increasing patient anxiety and wasting medical resources, and significant inter-observer bias in DCM evaluation, resulting in low efficiency of imaging analysis of degenerative cervical spondylomyelopathy.

[0107] The above embodiments are only used to illustrate the technical method of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present application.

Claims

1. An artificial intelligence-based degenerative cervical myelopathy image analysis system, characterized in that: include: Data acquisition module and data analysis module; The data acquisition module is used to acquire cervical spine magnetic resonance images through data acquisition equipment; The data analysis module: inputs the cervical vertebrae MRI image into the cervical vertebrae anatomical segmentation model to obtain a segmentation label; inputs the detection image with the segmentation label of the intervertebral canal into the intervertebral canal stenosis quantitative grading model to obtain the spinal canal stenosis prediction grade; inputs the detection image with the segmentation label of the cervical spinal cord into the spinal cord compression classification model to obtain the spinal cord compression result; inputs several detection images corresponding to the segmentation label, the spinal canal stenosis prediction grade and the spinal cord compression result into the cervical vertebrae image analysis model to obtain an analysis report; wherein, the cervical vertebrae anatomical segmentation model is obtained by improving the nnU-Net architecture through the cross pseudo-supervised CPS framework, the intervertebral canal stenosis quantitative grading model is constructed by an artificial intelligence model, the spinal cord compression classification model is constructed by a ResNet network, and the cervical vertebrae image analysis model is constructed by a large language model.

2. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 1, characterized in that: The cervical anatomical segmentation model is obtained by improving the nnU-Net architecture through the cross pseudo-supervision CPS framework, including: Obtain several historical cervical MRI images and their corresponding segmentation labels; Divide several historical cervical spine MRI images and their corresponding segmentation labels into training data, verification data and test data; perform image preprocessing on the training data, verification data and test data to obtain a training set, a verification set and a test set; Choose the improved nnU-Net architecture as the base model; Train the basic model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain the pre-trained model; By verifying the pre-trained model on the test set, we finally obtained a cervical anatomical segmentation model with cervical MRI images as input and segmentation labels as output.

3. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 2, characterized in that: The basic model is a neural network architecture based on U-Net and residual blocks, and is configured and trained through the nnU-Net framework, specifically including an encoder, a decoder, a skip connection module and a feature combination module.

4. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 3, characterized in that: Based on the basic model, the forced spacing settings, optimizer and preprocessing process are optimized, including: Forced spacing setting: In the preprocessing stage, the spacing parameters of the original cervical spine MRI images are modified so that the patch size input into the model can be constrained to an optimal value; Optimizer: The Ranger optimizer is used to better adapt to the nnU-Net model improved by the semi-supervised learning idea of ​​the CPS framework; Preprocessing: Advanced data enhancement methods such as contrast adjustment and noise addition are added.

5. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 4, characterized in that: The loss function used by the Ranger optimizer is obtained from Dice loss and cross entropy loss, including: Get Dice loss L Dice and the cross entropy loss L Cross-Entropy ; The loss function L used by the Ranger optimizer is calculated using the formula; L=L Dice +L Cross-Entropy +λ(L CSPlabeled +L CSPunlabeled ), where λ represents the weight, λ∈(0,1); L CSPlabeled and L CSPunlabeled denote the losses of labeled data and unlabeled data, respectively.

6. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 1, characterized in that: The intervertebral canal stenosis quantitative grading model is constructed by an artificial intelligence model, including: Acquire a number of historical detection images whose segmentation labels are intervertebral canals and their corresponding intervertebral canal stenosis grades; Dividing a number of historical detection images and their corresponding intervertebral canal stenosis grades into training data, verification data and test data; performing image preprocessing on the training data, verification data and test data to obtain a training set, a verification set and a test set; Select an AI model as the base model; Train the basic model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain the pre-trained model; By verifying the pre-trained model on the test set, we finally obtained a quantitative grading model for intervertebral canal stenosis whose input was the detection image with the segmentation label of the intervertebral canal and whose output was the predicted grade of spinal canal stenosis.

7. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 1, characterized in that: The spinal cord compression classification model is constructed through a ResNet network, including: Obtaining several historical detection images with segmentation labels of cervical spinal cord and their corresponding spinal cord compression results; Divide a number of historical detection images and their corresponding spinal cord compression results into training data, verification data and test data; perform image preprocessing on the training data, verification data and test data to obtain a training set, a verification set and a test set; Select the ResNet network model as the basic model; Train the basic model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain the pre-trained model; By verifying the pre-trained model on the test set, we finally obtained a spinal cord compression classification model whose input is the detection image of the cervical spinal cord with the segmentation label and whose output is the spinal cord compression result.

8. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 7, characterized in that: The ResNet network model uses a ResNet network model of at least N layers to perform morphological analysis on the detection image with the segmentation label of the cervical spinal cord through multi-layer feature extraction and multi-scale feature fusion to obtain the spinal cord compression results of whether there is compression, the compression position and the compression degree; wherein N is a positive integer.

9. The artificial intelligence-based degenerative cervical myelopathy image analysis system according to claim 1, characterized in that: The cervical vertebra image analysis model is constructed through a large language model, including: Obtaining several historical detection images corresponding to the segmentation labels, historical spinal canal stenosis prediction levels, historical spinal cord compression results, and corresponding historical analysis reports; Divide several historical detection images, historical spinal stenosis prediction grades, historical spinal cord compression results and corresponding historical analysis reports corresponding to the segmentation labels into training data, verification data and test data; perform data preprocessing on the training data, verification data and test data to obtain a training set, a verification set and a test set; Select the large language network model as the base model; Train the basic model with the training set, and adjust the learning rate and hyperparameters on the validation set to obtain the pre-trained model; By verifying the pre-trained model on the test set, we finally obtained a cervical spine image analysis model whose input is a number of test images corresponding to the segmentation labels, the predicted level of spinal stenosis and the results of spinal cord compression, and the output is an analysis report.

10. An artificial intelligence-based degenerative cervical myelopathy image analysis method, applied to an artificial intelligence-based degenerative cervical myelopathy image analysis system according to any one of claims 1 to 9, characterized in that: include: S0: Obtain cervical spine MRI images; S1: Input the cervical spine MRI image into the cervical spine anatomical segmentation model to obtain the segmentation label; S2: Input the detection image with the segmentation label of intervertebral canal into the intervertebral canal stenosis quantitative grading model to obtain the predicted grade of spinal canal stenosis; S3: The root inputs the detection image with the segmentation label of cervical spinal cord into the spinal cord compression classification model to obtain the spinal cord compression result; S4: Inputting a number of detection images corresponding to the segmentation labels, the predicted level of spinal stenosis and the spinal cord compression results into the cervical image analysis model to obtain an analysis report.

Images