Automatic positioning and typing method for ossification of posterior longitudinal ligament of cervical vertebra

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to an automatic positioning and typing method for cervical posterior longitudinal ligament ossification, which is characterized in that ossification focus spatial distribution characteristics of CT images and spinal cord morphological parameters of MRI images are synchronously analyzed based on CT and MRI images, a vertebral body-ossification-spinal cord spatial relationship is explicitly modeled through anatomical structure topological constraint, and a cervical vertebra-ossification-spinal cord spatial relationship is obtained. Diffuse calcification artifacts, heterogeneity signal interference and multi-modal data registration deviation are overcome, accurate positioning and typing diagnosis of the ossification are achieved, multi-dimensional feature fusion analysis is achieved, the limitation of traditional single iconography index diagnosis is broken through, space continuity and clinical parameters are integrated through a hybrid classifier, and the accuracy of diagnosis is improved. The accuracy and robustness of typing judgment are improved, a structured diagnosis report is automatically generated in the whole process, and the clinical decision-making efficiency and the standardization level are remarkably improved; the principle is scientific and reliable, and the ossification continuity index, the spinal canal invasion rate and the spinal cord compression grading parameters are automatically calculated according to the typing standard.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing, and particularly relates to an automatic positioning and typing method for ossification of the posterior longitudinal ligament of the cervical spine based on CT (Computed Tomography) and MRI (Magnetic Resonance Imaging) images, which accurately locates and classifies osteophytes for diagnosis. Background Art

[0002] With the aging of the population and the aggravation of spinal degenerative diseases, the incidence of ossification of the posterior longitudinal ligament of the cervical spine (OPLL) has been increasing year by year. The spinal stenosis and nerve compression symptoms caused by it seriously threaten the quality of life of patients. In clinical diagnosis and treatment, the combined application of CT and MRI imaging technologies provides key support for the accurate analysis of OPLL, and the medical image processing technology based on artificial intelligence is gradually breaking through the efficiency bottleneck of the traditional diagnosis mode.

[0003] In the imaging evaluation of OPLL, CT scan, with its high spatial resolution and the advantage of bone tissue imaging, has become the preferred image for the morphological analysis of osteophytes. CT is a medical imaging technology that uses a precise X-ray beam and a high-sensitivity detector to scan the human body layer by layer. Through computer processing of the scanned data, high-resolution images of the internal cross-section, coronal plane or sagittal plane of the body are generated. CT images reflect the degree of X-ray absorption of organs and tissues in different grayscales, with high density resolution, and can clearly display soft tissue and bone structures. Through multi-planar observation of CT, clinicians can clearly identify the longitudinal extension range and cross-sectional occupancy rate of osteophytes, and then quantify the degree of spinal stenosis. MRI, with its excellent soft tissue contrast, can accurately show the compression and deformation of the spinal cord, providing an important basis for the evaluation of nerve function damage. MRI uses the principle of nuclear magnetic resonance. Based on the different attenuations of the released energy in different structural environments within the substance, the electromagnetic waves emitted are detected by applying an external gradient magnetic field, and the positions and types of the atomic nuclei constituting this object are obtained, and then the internal structure image of the object is drawn accordingly. The complementary imaging characteristics of CT and MRI at the bone-soft tissue interface jointly establish the imaging basis for the diagnosis and treatment of OPLL.

[0004] Current AI (Artificial Intelligence) assisted diagnosis technology focuses on three core aspects: 1) Firstly, in the field of medical image segmentation, deep learning methods construct a hierarchical segmentation model of vertebral body - ossification - spinal cord to achieve precise separation of calcified ligaments and the posterior margin of the vertebral body in CT images, as well as spatial decoupling of the spinal cord and the vertebral body in MRI images; 2) Secondly, in terms of ossification detection, based on a three - dimensional convolutional neural network, parameters such as the continuity, thickness, and spinal canal encroachment rate of ossification foci are automatically extracted from CT images. Meanwhile, combined with the signal changes in the T2 - weighted image of MRI, the inflammatory reaction stage during the ossification process of the ligament is evaluated; 3) Finally, at the level of classification diagnosis, by fusing the ossification morphological characteristics of CT and the spinal cord compression parameters of MRI, the AI system performs intelligent classification according to the classification criteria (focal type, segmental type, continuous type, mixed type) and the spinal canal encroachment rate grading standard to assist in formulating the surgical approach plan.

[0005] The existing automatic classification technologies mainly develop along two technical paths: One is that the deep regression network directly predicts the ossification type. However, it lacks the display of ossification segmentation and has little clinical guiding significance; The other is the end - to - end segmentation model based on the deep learning architecture, which synchronously extracts the spatial distribution characteristics of ossification in CT and the spinal cord morphological parameters of MRI, and makes a classification decision by combining the spatial topological relationship of vertebral body - ossification - spinal cord. Although it can achieve three - dimensional visualization of ossification, when encountering diffuse calcification, vertebral edge osteophyte interference, or metal implant artifacts, there are still problems such as blurred ossification boundaries and deviation in spinal cord contour recognition. In addition, the heterogeneous signal characteristics during the ossification process of the ligament also pose challenges to the quantitative analysis of MRI images, which has become the key direction for the optimization of current AI algorithms.

[0006] Therefore, it is necessary to break through the limitations of traditional imaging diagnosis, research and design an automatic positioning and classification method for ossification of the posterior longitudinal ligament of the cervical spine. Based on a multi - task deep learning network, it synchronously analyzes the spatial distribution characteristics of ossification foci in CT images and the spinal cord morphological parameters in MRI images, and explicitly models the spatial relationship of vertebral body - ossification - spinal cord through anatomical structure topological constraints, effectively overcoming diffuse calcification artifacts, heterogeneous signal interference, and multi - modal data registration deviation, and realizing precise positioning and classification diagnosis of ossification. Summary of the Invention

[0007] The purpose of the present invention is to overcome the shortcomings of the existing technology, research and design an automatic positioning and classification method for ossification of the posterior longitudinal ligament of the cervical spine. By constructing a deep learning architecture, the ossification map of CT images and the spinal cord compression deformation field of MRI images are respectively extracted, and the ossification continuity index, spinal canal encroachment rate, and spinal cord compression grading parameters are automatically calculated according to the classification criteria, and an intelligent classification report is output.

[0008] To achieve the above - mentioned purpose, the automatic positioning and classification method for ossification of the posterior longitudinal ligament of the cervical spine involved in the present invention includes the following steps:

[0009] S1. Multimodal Dataset Construction

[0010] Collect a special OPLL dataset containing cervical CT images and MRI images, and convert the CT image and MRI image data to obtain a multimodal dataset;

[0011] For the CT and MRI sagittal images, respectively select 5 - 10 clear images adjacent to the mid - sagittal plane to effectively avoid the influence of image overlap and blur on the analysis results;

[0012] For the CT axial images, focus on extracting the cervical region images, and by randomly removing redundant images, while ensuring the image sample size, reduce the risk of mutual interference between images.

[0013] S2. Data Processing

[0014] Based on the collected multimodal dataset, construct a three - level anatomical structure topological map of the vertebral body - ossification - spinal cord, label the contours of the vertebral body and the ossification behind the vertebral body for the CT images, and synchronously label the deformed regions of the vertebral body and the spinal cord for the MRI images to generate a labeled mask;

[0015] For the CT sagittal images, complete the vertebral body labeling within the range of cervical vertebrae C2 - C7 segments and thoracic vertebrae T1 - T2 segments. According to anatomical features, divide the left and right vertebral body shapes into independent categories, and classify the ossifications separately. Finally, form 17 orderly arranged labeled categories to provide a structured data basis for subsequent ossification localization analysis;

[0016] For the CT axial images, label four anatomical structure categories of the spinal canal, vertebral body, lamina, and ossification to effectively support the subsequent quantitative analysis of the spinal canal stenosis rate;

[0017] For the MRI sagittal images, adopt a unified labeling strategy for the cervical vertebrae C2 - C7 segments and thoracic vertebrae T1 - T2 segments, only label the left vertebral body as a single category, and independently label the spinal cord region to accurately evaluate the spinal cord compression state and its spatial distribution characteristics.

[0018] S3. Deep Learning Network Architecture Design

[0019] Construct a multi - task segmentation network. At the same time, perform feature reduction through a double - layer fusion module, add a multi - layer supervised loss function to strengthen the spatial continuity of the vertebral body, ossification, and spinal cord;

[0020] Among them, the process of constructing the multi - task segmentation network is as follows:

[0021] The CT images and MRI images are respectively input into the deep neural network (Cer-Net). After extracting the low-level image features through a layer of convolution operation, they are input into the U-shaped architecture encoder and decoder. The encoder uses a multi-scale residual downsampling module to extract features from the medical images, and a feature fusion module is added at the skip connection to enhance the expression of cervical spine semantic information. The decoder uses a residual upsampling fusion module;

[0022] First, four different feature mapping heads are set based on the feature maps of multiple layers of the encoder. All four feature mapping heads include convolution, batch normalization, and Relu activation function. The convolution kernel results output feature maps of the original size and , , , sizes respectively, and loss calculation is performed with the real results;

[0023] Then, multi-layer supervision is carried out;

[0024] Finally, through a fully connected layer, a feature map with the same number of layers as the segmentation result category is obtained, and the multi-layer results are merged into one image to obtain the segmentation result map.

[0025] The process of feature restoration by the double-layer fusion module is as follows:

[0026] In the U-shaped architecture encoder, the upper and lower layer feature fusion modules integrate multi-scale information through cross-layer connection. Let the feature output by the ( )-th layer of the encoder be , and its next layer (the ( )-th layer) obtains the feature through downsampling;

[0027] The mathematical process of fusion and skip connection is as follows:

[0028] First, the high-level feature is upsampled

[0029] For , perform upsampling (such as transposed convolution or bilinear interpolation) to restore its spatial resolution: , where represents the upsampling operation;

[0030] Then, the feature fusion mechanism

[0031] Combines with the upsampled through splicing fusion:

[0032] ;

[0033] Followed by Convolution to adjust the number of channels:

[0034] , where is the convolution kernel, and

[0035] Finally, the skip connection is concatenated with the encoder

[0036] The th layer of the decoder obtains features through upsampling , and concatenates it with the after fusion along the channel dimension: ;

[0037] The concatenated features are processed by the convolution module of the decoder:

[0038] , where includes convolution, normalization, and activation functions for refining features;

[0039] By fusing shallow (high-resolution details) and deep (high-level semantics) features of the encoder, it improves the performance of the decoder in target localization and boundary recovery;

[0040] The process of multi-level supervision is as follows:

[0041] The decoder part of the U-shaped architecture introduces supervision signals on feature maps at different scales through a multi-level supervision mechanism to enhance feature expression ability and optimize gradient propagation. The output of each layer of the decoder is connected to an independent feature map head, and supervision is performed through the joint loss calculation of multi-scale prediction and the ground truth label;

[0042] Feature map head structure

[0043] Each feature map head consists of a convolution layer + batch normalization + ReLU activation function. Finally, the number of channels and resolution are adjusted through convolution, and prediction results at different scales are output. Let the output feature map of the th layer of the decoder be , and the corresponding operations of the feature map head include:

[0044] Convolution operation: Use convolution to extract features and maintain the spatial resolution:

[0045] ;

[0046] Batch normalization and activation: = ;

[0047] Resolution adjustment convolution: Generate the target number of channels through convolution and introduce transposed convolution to adjust the resolution: The output size is the original image ((s = 0, 1, 2, 3) corresponding to the original size, 1 / 2, 1 / 4, 1 / 8 respectively), implemented through convolution or upsampling with a step size of ;

[0048] Multi-scale prediction and loss calculation

[0049] The resolutions of the feature maps output by the four feature map heads are respectively , , , , and perform multi-scale supervision with the ground truth label:

[0050] Downsample the ground truth label, perform average pooling on the ground truth label to generate multi-scale labels: = ;

[0051] Multi-scale loss function: Use CELoss and DiceLoss to jointly optimize all scales: , where , are the loss weights for each scale (usually set to 0.5);

[0052] During the process of training the deep neural network, data augmentation is performed using random flipping, random scaling, and contrast variation. The Adam optimizer is used for training, the learning rate is set to 0.001, the number of batch samples is 8, and a total of 400 training rounds are performed.

[0053] S4, Precise segmentation of vertebral bodies and ossification sites

[0054] Train a deep neural network through a multi-modal dataset, and extract the vertebral body and ossification morphology features of CT images and the vertebral body morphology features of MRI images through a U-shaped architecture encoder;

[0055] Input any original CT image and MRI image into the trained deep neural network to obtain the predicted CT sagittal vertebral body and ossification segmentation, CT axial vertebral body and ossification segmentation, and MRI sagittal vertebral body and spinal cord segmentation results;

[0056] The specific process is as follows:

[0057] First, perform multi-modal data preprocessing on the original CT image and MRI image, including standardizing the spatial resolution adjustment, normalizing the intensity of heterogeneous images, and cross-modal registration based on anatomical landmarks to ensure the alignment of anatomical structures in different imaging sequences in space;

[0058] Then, extract features. For CT images, focus on analyzing the density gradient features of the bony structure, capture the calcified boundaries of the vertebral body contours and the morphological features of ossified substances through multi-scale convolution operations. At the same time, extract the cross-sectional spinal canal morphology by combining the annular structure characteristics of the axial tomographic images; for MRI images, mainly analyze the sagittal soft tissue contrast features, and use the signal intensity differences between sequences to enhance the distinguishability between the spinal cord and the vertebral body boundary.

[0059] Secondly, after the input images are feature-encoded, they enter the multi-task inference module. By establishing cross-modal feature correlation mappings, binary masks of the CT sagittal vertebral body and ossified substances, reconstructed contours of the CT axial vertebral body and ossified substances, and anatomical demarcations between the MRI sagittal vertebral body and the spinal cord are generated synchronously.

[0060] Thirdly, optimize the results. Adopt a cascaded post-processing strategy, eliminate small noises based on morphological operations, correct the segmentation boundaries through edge continuity detection, and combine with the true segmentation images of the vertebral body for structural verification to ensure that each segmentation result conforms to the physiological structural characteristics of the cervical vertebrae.

[0061] Finally, generate quantitative measurement parameters for each structure, including clinical key indicators such as the sagittal diameter of the spinal canal, the volume of ossified substances, and the spinal cord compression index, providing a reliable anatomical basis for subsequent visualization reconstruction and surgical path planning.

[0062] S5. CT Ossification Location Analysis

[0063] Based on the segmented CT sagittal data, by locating the spatial distribution of ossification, extract the morphological features of the ossified area, calculate its extension length along the longitudinal axis of the spinal canal, and dynamically evaluate the ossification boundary in combination with the vertebral body reference plane. For uneven calcification or artifact interference, use an adaptive threshold to optimize boundary recognition and output the core ossification location parameters.

[0064] The specific process is as follows:

[0065] Perform morphological analysis on the ossified area R, calculate the extension length, and project along the longitudinal axis of the spinal canal (set as the y-axis): ;

[0066] Adaptive threshold boundary optimization. For uneven calcification or artifact interference, design a local adaptive threshold: , where is the local window gray mean value, is the standard deviation, and k is an empirical coefficient (usually taken as 0.5 - 1.5).

[0067] S6. MRI Spinal Cord Compression Analysis

[0068] In the MRI modality branch, use computer vision methods to capture the morphological variation features of the spinal cord, construct a spinal cord compression deformation field by combining the changes in cerebrospinal fluid signal intensity, and calculate the spinal cord angle.

[0069] The specific process is as follows: Automatically detect the anterior and posterior edges of the spinal cord on the sagittal plane image, fit the anterior and posterior edge curves by the least squares method, and define the spinal cord angle as the included angle between the tangent vectors of the upper and lower marker points:

[0070] , where , are the tangent vectors taken 5 mm above and below the compression point respectively.

[0071] S7. Measurement of spinal canal parameters

[0072] Analyze the CT axial images, calculate the ratio of the thickness of the ossified mass to the anteroposterior diameter of the spinal canal, and obtain the spinal canal stenosis rate;

[0073] The specific process is as follows: Locate the pedicle level on the CT axial image, and select the cross-section with the narrowest spinal canal as the measurement plane;

[0074] Maximum thickness (T) of the ossified mass: The vertical distance from the anterior edge to the posterior edge of the ossification focus along the sagittal direction of the spinal canal;

[0075] Anteroposterior diameter (D) of the spinal canal: The shortest distance from the posterior edge of the vertebral body to the anterior edge of the lamina in the same plane;

[0076] Calculation formula for spinal canal stenosis rate: S = (T / D) × 100%;

[0077] When there are multi-segmental lesions, take the maximum value of the S values of each segment as the final evaluation index.

[0078] S8. Intelligent classification decision

[0079] According to the classification criteria, analyze the spatial continuity characteristics of the ossification focus through deep learning classification methods, and combine the spinal canal stenosis rate threshold and the spinal cord compression situation to automatically distinguish focal type, segmental type, continuous type and mixed type, and generate a comprehensive diagnostic report;

[0080] The specific process is as follows:

[0081] First, establish an analysis model for the distribution characteristics of the ossification focus, reconstruct the cervical sagittal and axial images through a deep learning network, focus on extracting the spatial position relationship and continuous distribution characteristics of the ossification focus in the vertebral body sequence, use a convolutional neural network to identify the longitudinal extension range of the ossification focus, accurately divide the number of vertebral segments involved in the focus, and at the same time, detect the interval distance between adjacent ossification foci to lay a spatial analysis foundation for subsequent classification;

[0082] Then, a dynamic evaluation module for spinal canal morphological parameters is constructed. Based on the stenosis rate threshold standard set by clinical guidelines, a dynamic evaluation model for the effective volume of the spinal canal is established, focusing on monitoring the changes in the stenosis degree of key segments, establishing an associated mapping with spinal cord morphological changes, introducing an evaluation mechanism for the spinal cord compression state, and forming a compression evaluation system from macroscopic morphology to microscopic structure;

[0083] Finally, an intelligent decision-making system with multi-dimensional feature fusion is established. By constructing a hierarchical classification model, feature fusion of spatial continuity features, spinal canal stenosis degree, and spinal cord compression parameters is performed to accurately distinguish focal, segmental, continuous, and mixed types. The diagnostic report generation module adopts a structured output framework to integrate key imaging feature parameters, classification basis, and clinical suggestions to form a standardized diagnostic document that conforms to the diagnosis and treatment specifications.

[0084] Compared with the prior art, the present invention synchronously analyzes the spatial distribution characteristics of ossification foci in CT images and the spinal cord morphological parameters in MRI images based on a multi-task deep learning network, explicitly models the spatial relationship between the vertebral body, ossification, and spinal cord through anatomical structure topological constraints, overcomes diffuse calcification artifacts, heterogeneous signal interference, and multi-modal data registration deviation, realizes accurate positioning and classification diagnosis of ossification, realizes multi-dimensional feature fusion analysis, breaks through the limitations of traditional single imaging index diagnosis, improves the accuracy and robustness of classification determination by integrating spatial continuity and clinical parameters through a hybrid classifier, and automatically generates a structured diagnostic report in the whole process, significantly improving the clinical decision-making efficiency and standardization level; its principle is scientific and reliable. By constructing a deep learning architecture, the ossification map of CT images and the spinal cord compression deformation field of MRI images are respectively extracted, and the ossification continuity index, spinal canal encroachment rate, and spinal cord compression grading parameters are automatically calculated according to the classification standard, and an intelligent classification report that conforms to clinical guidelines is output. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 It is a process flow chart of the present invention.

[0086] Figure 2 It is a schematic diagram of the segmentation information of the vertebral body, ossification, and spinal cord involved in the present invention.

[0087] Figure 3 It is an example diagram of the CT ossification position analysis involved in the present invention. Among them, a is the ossification of the C2-6 segments of the cervical vertebra, b is the ossification of the C2-3 and C4-6 segments of the cervical vertebra, and c is the ossification of the C5-6 intervertebral disc of the cervical vertebra.

[0088] Figure 4 It is an example diagram of the MRI spinal cord compression analysis involved in the present invention. Among them, a is the spinal cord angle of 102°, b is the spinal cord angle of 97°, and c is the spinal cord angle of 113°.

[0089] Figure 5This is an example diagram for measuring the spinal canal stenosis rate related to the present invention. Among them, a represents a spinal canal stenosis rate of 36.6%, b represents a spinal canal stenosis rate of 41.0%, and c represents a spinal canal stenosis rate of 45.1%.

[0090] Figure 6 This is a schematic diagram of the diagnosis result related to the present invention. Detailed implementation manners

[0091] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation manners.

[0092] Example 1:

[0093] An automatic positioning and typing method for ossification of the posterior longitudinal ligament of the cervical spine according to this example includes the following steps:

[0094] S1. Construction of multi-modal data set

[0095] Collect cervical CT images and MRI images, analyze and screen the original DICOM format;

[0096] In the sagittal image, select 5-10 images near the mid-sagittal position and export them in nii format;

[0097] In the axial image, randomly select 30-35 images in the C2-C7 interval and export them in nii format for subsequent annotation using the ITK-SNAP software;

[0098] S2. Data processing

[0099] Annotate 700 cervical CT sagittal images, 2600 cervical CT axial images, and 600 cervical MRI sagittal images provided by the hospital's spinal surgery department. The vertebral body positions, ossifications, and clarity in each cervical image are different. As Figure 2 shown, each cervical CT sagittal image is marked with 17 annotations, each cervical CT axial image is marked with 4 annotations, and each cervical MRI sagittal image is marked with 2 annotations;

[0100] S3. Design of deep learning network architecture

[0101] In the U-shaped architecture encoder, the upper and lower layer feature fusion integrates multi-scale information through cross-layer connections. Let the feature output by the ( ) layer of the encoder be , and its next layer (the ( ) layer) obtains the feature through downsampling;

[0102] S31. Upsampling of high-level features

[0103] For Perform upsampling to restore its spatial resolution: ;

[0104] S32. Feature fusion mechanism

[0105] The and the upsampled are fused by concatenation: , followed by a convolution to adjust the number of channels: ;

[0106] S33. Skip connection and encoder concatenation

[0107] The feature obtained by upsampling at the th layer of the decoder is concatenated with the fused along the channel dimension: , and the concatenated feature is processed by the decoder convolution module: ;

[0108] S34. Feature map head structure

[0109] Adjust the number of channels and resolution through convolution, and output prediction results at different scales. Let the output feature map of the th layer of the decoder be , and the corresponding feature map head operations include:

[0110] Convolution operation: Use convolution to extract features and maintain the spatial resolution:

[0111] ;

[0112] Batch normalization and activation: = ;

[0113] Resolution adjustment convolution: Generate the target number of channels through convolution, and introduce transposed convolution (TransposedConv) to adjust the resolution: The output size is the original image ((s = 0, 1, 2, 3) corresponds to the original size, 1 / 2, 1 / 4, 1 / 8 respectively), and is implemented through convolution or upsampling with a stride of ;

[0114] S35. Multi-scale prediction and loss calculation

[0115] The resolutions of the feature maps output by the four feature map heads are respectively , , , , perform multi-scale supervision with the true labels:

[0116] Downsample the true labels and perform average pooling on the true labels to generate multi-scale labels: = ;

[0117] Multi-scale loss function: Use CELoss and DiceLoss to jointly optimize all scales: ;

[0118] During the process of training the deep neural network, data augmentation is performed using random flipping, random scaling, and contrast change. The Adam optimizer is used for training, the learning rate is set to 0.001, the number of batch samples is 8, and a total of 400 rounds of training are performed.

[0119] S4. Precise segmentation of the vertebral body and ossification sites

[0120] First, perform standardized spatial resolution adjustment, cross-modal image intensity normalization, and cross-modal registration based on anatomical landmarks on the original CT images and MRI images;

[0121] Then, extract the density gradient features of the bony structures in the CT images, capture the calcified boundaries of the vertebral body contours and the morphological features of the ossifications through multi-scale convolution operations. At the same time, extract the cross-sectional spinal canal morphology by combining the circular structure characteristics of the axial tomographic images;

[0122] Extract the sagittal soft tissue contrast features of the MRI images and use the signal intensity difference between sequences to enhance the distinguishability between the spinal cord and the vertebral body boundary;

[0123] Secondly, after the input images are feature-encoded, they enter the multi-task inference module. By establishing cross-modal feature correlation mapping, binary masks of the CT sagittal vertebral body and ossifications, reconstructed contours of the CT axial vertebral body and ossifications, and anatomical demarcation between the MRI sagittal vertebral body and the spinal cord are generated synchronously;

[0124] Thirdly, optimize the results, adopt a cascaded post-processing strategy, eliminate small noises based on morphological operations, correct the segmentation boundaries through edge continuity detection, and perform structure verification by combining with the true segmentation images of the vertebral body;

[0125] Finally, generate clinical key indicators including the sagittal diameter of the spinal canal, the volume of the ossification, and the spinal cord compression index.

[0126] S5. CT ossification location analysis

[0127] Perform morphological analysis on the ossification region R, as Figure 3 shown, calculate the extension length, project along the longitudinal axis of the spinal canal (set as the y-axis): ;

[0128] Adaptive threshold boundary optimization. For uneven calcification or artifact interference, a local adaptive threshold is designed: .

[0129] S6. MRI spinal cord compression analysis

[0130] Automatically detect the anterior and posterior margins of the spinal cord on sagittal images, and fit the anterior and posterior margin curves by the least squares method. As Figure 4 shown, define the spinal cord angle as the included angle between the tangent vectors of the upper and lower marked points:

[0131] .

[0132] S7. Spinal canal parameter measurement

[0133] Locate the pedicle layer on CT axial images, and select the cross-section with the narrowest spinal canal as the measurement plane. As Figure 5 shown.

[0134] S8. Intelligent classification decision

[0135] First, establish an analysis model for the distribution characteristics of ossification lesions. Reconstruct the sagittal and axial images of the cervical spine through a deep learning network, focus on extracting the spatial position relationship and continuous distribution characteristics of ossification lesions in the vertebral body sequence, use a convolutional neural network to identify the longitudinal extension range of ossification lesions, accurately divide the number of vertebral segments involved in the lesions, and at the same time, detect the interval distance between adjacent ossification lesions;

[0136] Then, construct a dynamic evaluation module for spinal canal morphological parameters. Based on the stenosis rate threshold standard set by clinical guidelines, establish a dynamic evaluation model for the effective volume of the spinal canal, focus on monitoring the change of stenosis degree in key segments, establish an associated mapping with spinal cord morphological changes, introduce an evaluation mechanism for spinal cord compression status, and form a compression evaluation system from macroscopic morphology to microscopic structure;

[0137] Finally, establish an intelligent decision-making system for multi-dimensional feature fusion. By constructing a hierarchical classification model, fuse the spatial continuity features, spinal canal stenosis degree and spinal cord compression parameters, perform accurate discrimination of focal type, segmental type, continuous type and mixed type, and the diagnostic report generation module adopts a structured output framework to integrate key image feature parameters, classification basis and clinical suggestions to form a standardized diagnostic document. As Figure 6 shown.

Claims

1. A method for automatically locating and classifying ossification of the posterior longitudinal ligament of the cervical spine, characterized in that: The following steps are involved: S1. Construction of multimodal dataset Collect cervical spine CT images and MRI images and convert them to obtain a multimodal dataset; S2. Data Processing Construct a topological map of the three-level anatomical structure of vertebral body-ossification-spinal cord, annotate the contours of the vertebral body and ossification at the posterior edge of the vertebral body on CT images, and synchronously annotate the compressed and deformed areas of the vertebral body and spinal cord on MRI images to generate an annotation mask; S3. Deep learning network architecture design Construct a multi-task segmentation network. At the same time, perform feature restoration through a double-layer fusion module and add a multi-layer supervision loss function. S4. Accurate segmentation of vertebral bodies and ossified areas The deep neural network is trained with multimodal datasets, and the vertebral and ossification morphological features of CT images and the vertebral morphological features of MRI images are extracted through a U-shaped architecture encoder. Input any original CT image and MRI image into the trained deep neural network to obtain the predicted CT sagittal vertebral body and ossification segmentation, CT axial vertebral body and ossification segmentation, and MRI sagittal vertebral body and spinal cord segmentation results; S5. CT ossification location analysis Based on the segmented CT sagittal data, the spatial distribution of ossification is located, the morphological characteristics of the ossification area are extracted, and its extension length along the longitudinal axis of the spinal canal is calculated. The ossification boundary is dynamically evaluated in combination with the vertebral reference plane. For uneven calcification or artifact interference, the adaptive threshold is used to optimize the boundary recognition and output the ossification core positioning parameters. S6, MRI spinal cord compression analysis In the MRI modality branch, computer vision methods are used to capture the morphological variation characteristics of the spinal cord, and the spinal cord compression deformation field is constructed in combination with the changes in cerebrospinal fluid signal intensity to calculate the spinal cord angle; S7. Measurement of spinal canal parameters The axial CT images were analyzed, and the ratio of the thickness of the ossified material to the anterior-posterior diameter of the spinal canal was calculated to obtain the spinal canal stenosis rate. S8, Intelligent classification decision According to the classification standards, the spatial continuity characteristics of the ossification lesions are analyzed through deep learning classification methods. Combined with the spinal canal stenosis rate threshold and spinal cord compression, automatic identification is performed to generate a comprehensive diagnosis report.

2. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: In S1, for CT and MRI sagittal images, 5–10 clear images adjacent to the midsagittal plane were selected, and for CT axial images, images of the cervical spine region were randomly extracted.

3. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: In S2, for sagittal CT images, vertebral annotation was completed within the C2-C7 segment of the cervical spine and the T1-T2 segment of the thoracic spine, the left and right vertebral morphologies were divided into independent categories, and the ossification was classified separately, forming 17 annotation categories; For CT axial images, the four anatomical structure categories of spinal canal, vertebral body, lamina and ossification were marked; For sagittal MRI images, the left vertebral bodies of the cervical C2-C7 segments and the thoracic T1-T2 segments were annotated as a single category, and the spinal cord region was annotated independently.

4. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: The process of building a multi-task segmentation network in S3 is: The CT images and MRI images are input into the deep neural network respectively. After a layer of convolution operation to extract the low-level image features, they are input into the U-shaped architecture encoder and decoder. The encoder uses a multi-scale residual downsampling module to extract the features in the medical images, and adds a feature fusion module at the jump connection. First, four different feature mapping heads are set based on the multi-layer feature maps of the encoder. The four feature mapping heads all include convolution, batch normalization, and ReLU activation functions. The convolution kernel results are output in original size and , , , The feature map of the size is calculated and the loss is calculated with the real result; Then, multiple layers of supervision are carried out; Finally, through a fully connected layer, a feature map with the same number of layers as the segmentation result category is obtained, and the multi-layer results are merged into one map to obtain the segmentation result map; The process of feature restoration by the double-layer fusion module is as follows: In the U-shaped architecture encoder, the upper and lower layers of the feature fusion module integrate multi-scale information through cross-layer connections. The output feature of the layer is , the next layer (the ) layer) obtains features by downsampling ; The mathematical process of fusion and skip connection is as follows: First, high-level features are upsampled right Upsample to restore spatial resolution: ,in Represents an upsampling operation; Then, the feature fusion mechanism Will After upsampling Fusion by splicing: ； Followed by Convolution adjusts the number of channels: ,in is the convolution kernel, is bias; Finally, the skip connection is concatenated with the encoder Decoder Layers get features by upsampling , and merge it with Concatenate along the channel dimension: ; The concatenated features are processed by the convolution module of the decoder: ,in Contains convolution, normalization and activation functions to refine features; The process of multi-layer supervision is: The U-shaped architecture decoder introduces supervision signals on feature maps of different scales through a multi-layer supervision mechanism to enhance feature expression capabilities and optimize gradient propagation. Each layer output of the decoder is connected to an independent feature mapping head and supervised by the joint loss calculation of multi-scale prediction and true label. Feature map header structure Each feature map head consists of a convolution layer + batch normalization + ReLU activation function. Finally, the number of channels and resolution are adjusted through convolution to output prediction results of different scales. The output feature map of the layer is , the corresponding feature map head operations include: Convolution operation: using Convolution extracts features and maintains spatial resolution: ； Batch Normalization and Activation: = ; Resolution adjustment convolution: through Convolution generates the target number of channels, and introduces transposed convolution to adjust the resolution: the output size is the original image ((s=0,1,2,3) correspond to the original size, 1 / 2, 1 / 4, 1 / 8 respectively), with a step size of Convolution or upsampling implementation; Multi-scale prediction and loss calculation The feature map resolutions output by the four feature mapping heads are , , , , with multi-scale supervision of the true labels: The real label is downsampled and the real label is Perform average pooling to generate multi-scale labels: = ; Multi-scale loss function: Using CELoss and DiceLoss, jointly optimize all scales: ,in , is the loss weight of each scale.

5. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: The specific process of S4 is: Firstly, the original CT images and MRI images were preprocessed for multimodal data, including standardized spatial resolution adjustment, heterogeneous image intensity normalization, and cross-modality registration based on anatomical landmarks. Then, the features are extracted. The density gradient characteristics of the bone structure are analyzed by CT images. The calcification boundary of the vertebral contour and the morphological characteristics of the ossification are captured through multi-scale convolution operations. At the same time, the cross-sectional spinal canal morphology is extracted by combining the annular structure characteristics of the axial tomographic image. The MRI image analyzes the sagittal soft tissue contrast characteristics, and the signal intensity difference between the sequences is used to enhance the distinction between the spinal cord and the vertebral boundary. Secondly, the input image enters the multi-task reasoning module after feature encoding, and by establishing a cross-modal feature association mapping, it simultaneously generates the binary mask of the vertebral body and ossification in the CT sagittal plane, the reconstructed contour of the vertebral body and ossification in the CT axial plane, and the anatomical boundary between the vertebral body and the spinal cord in the MRI sagittal plane. Thirdly, the results were optimized by adopting a cascade post-processing strategy, eliminating small noises based on morphological operations, correcting the segmentation boundaries through edge continuity detection, and combining the real segmentation images of the vertebra for structural verification; Finally, quantitative measurement parameters of each structure are generated.

6. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: The specific process of S5 is: The ossification area R was subjected to morphological analysis, extended length calculation, and projection along the longitudinal axis of the spinal canal: ; Adaptive threshold boundary optimization, design local adaptive threshold for uneven calcification or artifact interference: ,in, is the local window grayscale mean, is the standard deviation, and k is the empirical coefficient.

7. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: The specific process of S6 is: The anterior and posterior edges of the spinal cord were automatically detected on the sagittal image, and the anterior and posterior edge curves were fitted by the least squares method. The spinal cord angle was defined as the angle between the tangent vectors of the upper and lower marking points: ,in , The tangent vectors are taken 5 mm above and below the pressure point respectively.

8. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: In S7, the pedicle level was located on the axial CT image, and the narrowest cross section of the spinal canal was selected as the measurement plane; Maximum thickness of ossification (T): the vertical distance from the front edge to the back edge of the ossification focus along the sagittal direction of the spinal canal; The anteroposterior diameter of the spinal canal (D): the shortest distance from the posterior edge of the vertebral body to the anterior edge of the lamina in the same plane; The formula for calculating the spinal canal stenosis rate is: S=(T / D)×100%; When there are multi-segment lesions, the maximum S value of each segment is taken as the final evaluation index.

9. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 1, characterized in that: The specific process of S8 is: Firstly, an analysis model was established for the distribution characteristics of ossified lesions. The sagittal and axial images of the cervical spine were reconstructed through a deep learning network, focusing on extracting the spatial position relationship and continuous distribution characteristics of the ossified lesions in the vertebral sequence. A convolutional neural network was used to identify the longitudinal extension range of the ossified lesions, accurately divide the number of vertebral segments involved in the lesions, and detect the interval distance between adjacent ossified lesions. Then, a dynamic evaluation module for spinal canal morphological parameters was constructed. Based on the stenosis rate threshold standard set by clinical guidelines, a dynamic evaluation model for the effective volume of the spinal canal was established to monitor the changes in the degree of stenosis of key segments, establish correlation mapping with spinal cord morphological changes, introduce an evaluation mechanism for the compression state of the spinal cord, and form a compression evaluation system from macroscopic morphology to microscopic structure. Finally, an intelligent decision-making system with multi-dimensional feature fusion was established. By constructing a hierarchical classification model, the spatial continuity features, the degree of spinal canal stenosis and the spinal cord compression parameters were fused to distinguish the focal, segmental, continuous and mixed types. The diagnostic report generation module adopted a structured output framework to integrate key imaging feature parameters, classification basis and clinical recommendations to form a standardized diagnostic document.

10. The method for automatic positioning and classification of ossification of the posterior longitudinal ligament of the cervical spine according to claim 4, characterized in that: In the process of training deep neural networks, random flipping, random scaling, and contrast changes are used for data enhancement. The Adam optimizer is used for training, the learning rate is set to 0.001, the number of batch samples is 8, and a total of 400 rounds of training are performed.

Images