Quantization method, system and equipment of lumbar vertebra nuclear magnetic resonance image and medium

Abstract

The embodiment of the invention discloses a quantification method, system, equipment and medium for a lumbar vertebra nuclear magnetic resonance image, the system obtains an external MRI image set needing to be processed, then the system inputs the MRI image set to be processed into a VM-UNet model, the VM-UNet model outputs segmentation mask data of the classified MRI image, and the segmentation mask data are stored in the VM-UNet model. Then the system carries out ROI extraction, signal and texture analysis and morphological processing, and finally generates a quantitative analysis report which is provided for doctors to diagnose lumbar diseases. Through automatic segmentation and quantitative analysis, the system can quickly and efficiently process a large-scale MRI image set, the diagnosis efficiency of a doctor is remarkably improved, meanwhile, an accurate analysis result is provided, and the doctor is assisted to make accurate diagnosis in complex cases. The system provides a quantitative analysis report, greatly reduces the workload of doctors in analyzing MRI images, and particularly reduces errors and time cost in manual measurement.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a quantification method, system, equipment and medium for lumbar magnetic resonance imaging. Background Art

[0002] Lumbar diseases include: intervertebral disc herniation, spinal stenosis, spondylolisthesis, intervertebral foraminal stenosis, muscle degeneration, etc. In clinical medicine, the traditional way of diagnosis is to use magnetic resonance imaging (MRI images), which uses strong magnetic fields and radio frequency pulses to generate detailed tissue images inside the human body. Doctors then judge whether the above lumbar diseases exist based on the MRI images.

[0003] The traditional judgment method relies on the doctor's personal experience and knowledge level, has a certain degree of subjectivity, and is easily affected by factors such as the doctor's condition on the day, fatigue level, diagnostic habits, etc. Especially when judging complex anatomical structures or minor lesions, subjective bias may lead to misdiagnosis or missed diagnosis.

[0004] Secondly, when doctors manually measure and quantify certain anatomical structures in MRI images (such as the area of ​​intervertebral disc herniation, the degree of spinal canal stenosis, and the width of the intervertebral foramen), it is difficult to ensure accuracy. Especially when dealing with smaller lesions or irregular shapes, manual measurement will produce large errors.

[0005] In the process of processing MRI images with the existing VM-UNet technology, VM-UNet technology cannot achieve good results in processing the details of MRI images. Although VM-UNet technology can segment the relevant anatomical structures of the lumbar spine, it is impossible to quantify the specific details of each lesion in the MRI image, and doctors need to use tools for further analysis. As a result, doctors need to further analyze the morphological changes of the lesions, which will still cause errors in the analysis process. Especially in the process of large-scale image processing, a large amount of analysis data will affect the workload and time cost of doctors, and it is not convenient to assist doctors in processing MRI images. Summary of the invention

[0006] Based on this, it is necessary to propose a quantification method, system, equipment and medium for lumbar magnetic resonance images to address the above problems.

[0007] A method for quantifying a lumbar vertebrae nuclear magnetic resonance image, the method comprising:

[0008] Obtaining a set of MRI images to be processed;

[0009] Inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI images;

[0010] Extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set;

[0011] Performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data;

[0012] Performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions;

[0013] A quantitative analysis report is generated based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in making a diagnosis.

[0014] In at least one embodiment of the present application, the specific steps of inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputting the segmentation mask data of the MRI image include:

[0015] The VM-UNet model extracts features from the MRI image set to be processed through an encoder, and then restores the resolution of the image through a decoder, generates and outputs segmentation mask data of the classified MRI image, and the anatomical structures include: intervertebral disc, dural sac, facet joints, psoas major muscle, erector vertebrae and multifidus muscle.

[0016] In at least one embodiment of the present application, the specific steps of extracting the MRI image set to be processed based on the segmentation mask data of the classified MRI image to obtain the extracted classified ROI region image set include:

[0017] Marking each MRI image in the MRI image set to be processed according to the segmentation mask data of the classified MRI image to generate a marked MRI image set;

[0018] Feature extraction is performed on each MRI image in the labeled MRI image set to obtain an extracted ROI region image set, and the extracted ROI region image set is classified to obtain a classified ROI region image set, wherein the classified ROI region image set includes: an intervertebral disc ROI region image set, a dural sac ROI region image set, a facet joint ROI region image set, a psoas major muscle ROI region image set, a vertical spine ROI region image set, and a multifidus muscle ROI region image set.

[0019] In at least one embodiment of the present application, the step of performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data comprises:

[0020] Convert each image in the classified ROI region image set into a grayscale image to generate a ROI region grayscale image set, calculate the signal intensity mean and signal intensity standard deviation of the ROI region in each image in the ROI region grayscale image set, and obtain all ROI region signal intensity feature data;

[0021] Texture feature extraction is performed on the classified ROI region grayscale image set to extract the contrast difference value, entropy value and uniformity value in each ROI region grayscale image to obtain all ROI region texture feature data.

[0022] In at least one embodiment of the present application, the step of performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions includes:

[0023] Use Canny edge detection to extract the ROI boundary of each ROI image in the classified ROI region image set, and obtain all ROI edge image sets and all ROI contour image sets;

[0024] Calculate the area of ​​each ROI region and the total length of the ROI region edge in all ROI edge image sets to obtain ROI region area data and ROI edge length data;

[0025] Calculating the local curvature and the overall curvature of each ROI region in the set of all ROI contour images to obtain ROI region curvature data;

[0026] The convexity of each ROI region in the ROI contour image set is calculated to obtain ROI region convexity data.

[0027] In at least one embodiment of the present application, the specific steps of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis include:

[0028] Screening out an intervertebral disc ROI region image from the classified ROI region image set to obtain an intervertebral disc ROI region image;

[0029] Filtering out the ROI edge image corresponding to the intervertebral disc ROI area image from all ROI edge images;

[0030] The herniation area is calculated based on the intervertebral disc ROI region image and the ROI edge image to generate an intervertebral disc herniation area value.

[0031] In at least one embodiment of the present application, the specific step of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also includes:

[0032] Screening out an intervertebral foramen ROI region image from the classified ROI region image set to obtain an intervertebral foramen ROI region image;

[0033] The left and right width values ​​of the intervertebral foramen ROI region image are calculated to obtain the intervertebral foramen width value.

[0034] In at least one embodiment of the present application, the specific step of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also includes:

[0035] Screening out a spinal canal ROI region image from the classified ROI region image set to obtain a spinal canal ROI region image;

[0036] The relative area of ​​the spinal canal and the median diameter of the spinal canal in the spinal canal ROI region image are calculated to obtain the spinal canal stenosis value.

[0037] In at least one embodiment of the present application, the specific step of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also includes:

[0038] Screening out a psoas major muscle ROI region image and a multifidus muscle ROI region image from the classified ROI region image set to obtain a muscle fat ROI region image;

[0039] The Otsu binarization algorithm is used to calculate the fat area ratio value in the muscle fat ROI region image to obtain the muscle fat infiltration value.

[0040] A quantification system for lumbar vertebrae nuclear magnetic resonance images, applied to any one of the above-mentioned quantification methods for lumbar vertebrae nuclear magnetic resonance images, the quantification system for lumbar vertebrae nuclear magnetic resonance images comprising:

[0041] An acquisition module, used for acquiring the MRI image to be processed;

[0042] The VM-UNet model is used to segment the MRI image set to be processed to obtain segmentation mask data of the classified MRI images;

[0043] A ROI extraction module extracts ROI from the MRI image set to be processed according to the segmentation mask data of the classified MRI image;

[0044] A signal intensity processing module performs signal intensity processing on the classified ROI area image set;

[0045] Texture feature analysis module, which performs texture analysis on the classified ROI area image set;

[0046] A morphological feature calculation module calculates morphological features of the classified ROI area image set;

[0047] The quantitative analysis report output module generates and outputs a quantitative analysis report based on the classified ROI area image set and the morphological data of all ROI areas to assist doctors in making a diagnosis;

[0048] The quantification system of the lumbar vertebrae nuclear magnetic resonance image executes the steps of the quantification method of the lumbar vertebrae nuclear magnetic resonance image described above.

[0049] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of a quantification method for lumbar magnetic resonance images.

[0050] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the steps of the above-mentioned method for quantifying lumbar magnetic resonance images:

[0051] Implementing the embodiments of the present invention will have the following beneficial effects:

[0052] The quantification method, system, device and medium of the lumbar magnetic resonance image in this embodiment, the system obtains an external MRI image set that needs to be processed, and then the system inputs the MRI image set to be processed into the VM-UNet model. The VM-UNet model outputs the segmentation mask data of the classified MRI image, and then the system performs ROI extraction, signal and texture analysis, and morphological processing, and finally generates a quantitative analysis report, which is provided to doctors for diagnosing lumbar diseases.

[0053] Through automatic segmentation and quantitative analysis, the system can quickly and efficiently process large-scale MRI image sets, significantly improving doctors' diagnostic efficiency while providing precise analysis results to assist doctors in making accurate diagnoses in complex cases.

[0054] The system provides quantitative analysis reports, which greatly reduces the workload of doctors when analyzing MRI images, especially reducing the errors and time costs in manual measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] in:

[0057] Figure 1 is a flow chart of a method for quantifying lumbar vertebrae magnetic resonance images in one embodiment;

[0058] Figure 2 is a flow chart of a method for quantifying lumbar vertebrae magnetic resonance images in another embodiment;

[0059] Figure 3 A flowchart of a method for quantifying lumbar vertebrae magnetic resonance images in yet another embodiment;

[0060] Figure 4 This is a full flow chart of the quantification method of lumbar spine magnetic resonance images;

[0061] Figure 5 is a structural block diagram of a quantification system for lumbar vertebrae magnetic resonance images in one embodiment;

[0062] Figure 6 is a structural block diagram of a computer device in one embodiment;

[0063] Explanation of reference numerals: 100, quantification system of lumbar MRI images; 110, acquisition module; 120, VM-UNet model; 130, ROI extraction module; 140, signal intensity processing module; 150, texture feature analysis module; 160, morphological feature calculation module; 170, quantitative analysis report output module. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] like Figure 1 - Figure 6 As shown, this embodiment provides a method for quantifying a lumbar vertebrae nuclear magnetic resonance image, and the method for quantifying a lumbar vertebrae nuclear magnetic resonance image includes:

[0066] S101, obtaining a set of MRI images to be processed;

[0067] S102, inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI images;

[0068] S103, extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set;

[0069] S104, performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data;

[0070] S105, performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions;

[0071] S106: Generate a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis.

[0072] Please refer to Figure 1-Figure 4 In this embodiment, the system obtains an external MRI image set that needs to be processed, and then the system inputs the MRI image set to be processed into the VM-UNet model. The VM-UNet model outputs the segmentation mask data of the classified MRI image, and then the system performs ROI extraction, signal and texture analysis, and morphological processing, and finally generates a quantitative analysis report, which is provided to doctors for diagnosing lumbar diseases.

[0073] Through automatic segmentation and quantitative analysis, the system can quickly and efficiently process large-scale MRI image sets, significantly improving doctors' diagnostic efficiency while providing precise analysis results to assist doctors in making accurate diagnoses in complex cases.

[0074] The system provides quantitative analysis reports, which greatly reduces the workload of doctors when analyzing MRI images, especially reducing the errors and time costs in manual measurement.

[0075] It should be noted that the MRI image set is an MRI image set obtained from an MRI scanner or an image archiving system (such as PACS).

[0076] The MRI image set is then converted to a three-channel format of 512X512 pixels. The brightness and contrast of the image are adjusted using the average and standard deviation method for each channel to ensure that the input data has a similar distribution and range. Linear scaling is used to normalize the pixel values ​​of the image to the range of [0,1] to obtain the MRI image set to be processed.

[0077] The MRI image enters the VM-UNet model, whose encoder extracts multi-level features in the image, the decoder restores the resolution of the image, and outputs the segmentation mask data of the classified MRI image (segmentation mask for each anatomical structure).

[0078] The VM-UNet model segments and classifies complex lumbar anatomical structures (such as intervertebral discs, dura mater sacs, articular processes, muscles, etc.) and labels each anatomical structure as an independent region.

[0079] VM-UNet can automatically complete the accurate segmentation of the lumbar anatomical structure without human intervention, overcoming the errors and subjectivity caused by traditional manual segmentation.

[0080] It reduces the workload of doctors' manual labeling and significantly improves the efficiency of large-scale image processing, making it particularly suitable for practical clinical applications.

[0081] The system uses the segmentation mask data of the classified MRI images to extract each anatomical structure (such as intervertebral disc, dural sac, etc.) in the MRI image set to be processed, excludes background or non-relevant areas, and generates an independent ROI image set.

[0082] In at least one embodiment of the present application, the specific steps of inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputting the segmentation mask data of the MRI image include:

[0083] The VM-UNet model extracts features from the MRI image set to be processed through an encoder, and then restores the resolution of the image through a decoder, generates and outputs segmentation mask data of the classified MRI image, and the anatomical structures include: intervertebral disc, dural sac, facet joints, psoas major muscle, erector vertebrae and multifidus muscle.

[0084] Please refer to Figure 1-Figure 4 In this embodiment, the VM-UNet model first performs feature extraction on the input MRI image set to be processed, and the VM-UNet encoder converts the original two-dimensional MRI image into feature maps to extract the multi-level structural information in the image.

[0085] The encoder gradually abstracts the pixel information of the MRI image into high-dimensional features through a series of convolutional layers and downsampling operations.

[0086] During the encoding process, the spatial resolution gradually decreases and the abstraction level of the features gradually increases, and the encoder is able to extract global and local information in the anatomical structure of the lumbar spine.

[0087] Through multi-level convolution and downsampling operations, the encoder is able to capture the complex lumbar anatomical structure, including the subtle features of tissues such as the intervertebral disc and dural sac.

[0088] The encoder effectively enhances the segmentation capability of the model through feature extraction and improves the recognition accuracy of complex anatomical structures, especially when processing tissues with rich details, the accuracy is significantly improved.

[0089] The decoder part of the VM-UNet model is used to restore the resolution of the image. Through deconvolution operations and upsampling, the high-dimensional feature map extracted by the encoder is gradually restored to the resolution of the original image and mapped back to the spatial position.

[0090] The decoder uses deconvolution layers and upsampling techniques to gradually restore low-resolution feature maps to high-resolution segmented images. The decoder realigns the features with spatial positions to ensure that the segmentation results have accurate spatial distribution.

[0091] Through skip connections, the low-level features of the encoder are combined with the high-level features of the decoder to retain more detailed information, avoid the loss of spatial information in high-level features, and ensure that the segmented anatomical structure has clear boundaries. Through skip connections, the decoder can retain more image details and edge information, especially for structures with irregular morphology or blurred boundaries (such as intervertebral discs or articular processes).

[0092] The decoder can restore the original resolution of the image, making the segmented MRI images easier to analyze and diagnose, and ensuring the integrity of the spatial information.

[0093] The VM-UNet model finally generates the segmentation mask data of the classified MRI images by extracting features through the encoder and restoring resolution through the decoder. These segmentation mask data of the classified MRI images are binary images that identify the location and boundaries of each anatomical structure in the MRI image.

[0094] The segmentation mask enables each key anatomical structure to be automatically and accurately identified and segmented, reducing the workload of doctors' manual annotation and improving segmentation accuracy and consistency.

[0095] By automatically generating segmentation masks, the workload of doctors can be greatly reduced, especially when processing large-scale MRI images. The segmentation masks provide strong support for subsequent clinical analysis.

[0096] During the training process of the VM-UNet model, the system distinguishes different anatomical structures based on manually annotated data and trains the model for accurate segmentation. The final output mask data identifies each key anatomical structure separately.

[0097] The segmentation results can help doctors clearly identify the morphological changes and degree of lesions of different anatomical structures (such as intervertebral disc herniation, spinal canal stenosis, etc.), and provide MRI images with clear structures for clinical diagnosis.

[0098] This system specifically segments and classifies anatomical structures related to lumbar spine diseases, ensuring that the model can effectively process imaging data of lumbar spine diseases, especially distinguishing different tissues under complex anatomical structures.

[0099] The segmented anatomical structures cover a variety of common lumbar disease areas, such as intervertebral disc degeneration, dural sac compression, muscle degeneration, etc., providing wide applicability and meeting the needs of clinical diagnosis.

[0100] The VM-UNet encoder extracts feature information from the image, and the decoder restores it to high resolution to generate classified segmentation mask data. The system accurately segments various anatomical structures (such as intervertebral discs, dural sacs, etc.) in MRI images and outputs segmentation masks. This series of processes ensures that the boundaries of each anatomical structure are clear and the data is accurate, which is suitable for subsequent quantitative analysis and clinical use.

[0101] Through the deep learning technology of the VM-UNet model, automatic high-precision image segmentation is achieved, which greatly improves processing efficiency and solves the errors and subjectivity that may be caused by manual segmentation.

[0102] By automatically generating a segmentation mask for each anatomical structure, doctors can quickly and accurately assess the pathological conditions of the lumbar spine structure, such as herniated disc, spinal stenosis, muscle degeneration, etc., to assist in clinical decision-making.

[0103] Automated segmentation masks significantly reduce doctors’ workload, especially when faced with large amounts of MRI data. They can significantly improve the efficiency of clinical processing and reduce the error rate.

[0104] It should be further explained that VM-UNet adopts an asymmetric design rather than asymmetric structure. The PatchEmbedding layer divides the input MRI image into non-overlapping blocks and maps its dimensions to a default value of 96. After normalization by LayerNormalization, the image is input to the encoder for feature extraction. The encoder consists of four stages, with block merging operations performed at the end of the first three stages to reduce the dimensionality of the input features and increase the number of channels. VSS modules are used in these stages to gradually increase the number of channels.

[0105] The decoder also consists of four stages, using block expansion operations to adjust the channels, height, and width of the features. VSS modules are also applied in these stages to gradually reduce the number of channels. After the decoder, the final projection layer restores the feature size to match the segmentation target through upsampling and projection, where the skip connection uses a simple addition operation to avoid additional parameters.

[0106] VMamba’s VSS module forms the core of VM-UNet and consists of two branches after Layer Normalization. The first branch is processed by a linear layer and an activation function, while the second branch includes a linear layer, a depthwise separable convolution, and an activation function, followed by a SS2D module to further extract features. After normalization and element-wise multiplication with the output of the first branch, the linear layer mixes the features and combines them with a residual connection to generate the output of the VSS module. SiLU is used as the default activation function. SS2D includes a scan expansion operation, an S6 module for feature extraction, and a scan merge operation to capture diverse features by scanning in four directions. The Mamba-based S6 module enhances S4 by adjusting parameters to retain relevant information.

[0107] It should be further explained that before obtaining the MRI images to be processed, all MRI images are converted to a unified format, such as PNG format. All MRI images are standardized to 512X512 pixels in a three-channel format. The brightness and contrast of the image are adjusted using the average and standard deviation method for each channel to ensure that the input data has a similar distribution and range. Linear scaling is used to normalize the pixel values ​​of the image to the range of [0,1]. Before input model training, in order to enhance the generalization and robustness of network training, data augmentation techniques such as mirror reflection, random rotation, random cropping and Gaussian noise injection are used. At the beginning of training, the hidden layer weights are randomly initialized, the batch size is set to 16, and the AdamW

[10] optimizer is used with an initial learning rate of 1e-3. The training cycle is set to 200.

[0108] The model training process was carried out in the environment of Python 3.8, Pytorch 1.10.0 and NVIDIA RTX 3080 GPU (24GB). The loss function uses the most basic Cross-Entropy and Dice loss (CeDiceloss) as the loss functions for multi-class segmentation tasks, such as the following loss function:

[0109]

[0110] L Dice =1-(2|P∩G|) / |P|+|G|;

[0111] Where N is the number of samples and C is the sample category, y i,c is the indicator variable, y i，c Indicates whether sample i belongs to category c; if y i，c =1 means it belongs to, y i，c =0 means it does not belong to; is the model's prediction of the probability that sample i belongs to category c.

[0112] |P| is the predicted value; |G| is the true value; λ 1 and λ 2 Represents the weight of the loss function, the default value is 1.

[0113] In the VM-UNet model, the evaluation is mainly based on the following aspects: precision, accuracy, F1 score and specificity to evaluate the multi-task classification performance of the model; the Dice coefficient is used to evaluate the overlap between the model prediction and the manual segmentation standard; the Hausdorff distance

[14] is used to evaluate the similarity between the target contour predicted by the model and the manual segmentation. Although the Dice coefficient has been recognized as a standard indicator for evaluating the segmentation performance of the model, the role of the Hausdorff distance in this evaluation is equally important because it represents the accuracy and directly affects the credibility of the subsequent automatic quantization. The algorithms for the Dice coefficient and the Hausdorff distance are as follows:

[0114] Dice = (2|P∩G|) / |P|+|G|;

[0115] H(P, G) = max{sup p∈P inf g∈G d(p,g),sup g∈G inf p∈P d(g,p)};

[0116] Among them, the Dice coefficient refers to the ratio of the intersection of the predicted area and the true labeled area to their average size. The Hausdorff distance is used to measure the maximum distance between two point sets. In the formula, P represents the point set of the predicted contour, G represents the point set of the true labeled contour, p∈P, g∈G, represents the distance from point p to point g. sup represents the supremum (the smallest upper bound), and inf represents the infimum (the largest lower bound).

[0117] In at least one embodiment of the present application, the specific steps of extracting the MRI image set to be processed based on the segmentation mask data of the classified MRI image to obtain the extracted classified ROI region image set include:

[0118] Marking each MRI image in the MRI image set to be processed according to the segmentation mask data of the classified MRI image to generate a marked MRI image set;

[0119] Feature extraction is performed on each MRI image in the labeled MRI image set to obtain an extracted ROI region image set, and the extracted ROI region image set is classified to obtain a classified ROI region image set, wherein the classified ROI region image set includes: an intervertebral disc ROI region image set, a dural sac ROI region image set, a facet joint ROI region image set, a psoas major muscle ROI region image set, a vertical spine ROI region image set, and a multifidus muscle ROI region image set.

[0120] Please refer to Figure 1-Figure 4 ,In this implementation, the anatomical structure in each MRI image is labeled based on the segmentation mask data generated by the VM-UNet model, and a set of labeled MRI ,images is generated.

[0121] Feature extraction is performed on the labeled MRI images to extract a set of extracted ROI area images (anatomical region of interest (ROI)), and the key anatomical region in each MRI image is extracted separately.

[0122] The extracted ROI region image set is classified, and each anatomical structure (intervertebral disc, dural sac, articular process, etc.) is divided into different image sets to form a classified ROI region image set.

[0123] Through automatic labeling, feature extraction and classification of MRI images, the entire processing process is automated, greatly improving the efficiency of image processing. Especially when faced with large-scale MRI data, the automated process reduces the workload of doctors.

[0124] The use of segmentation masks for marking and extraction ensures the accurate identification of each anatomical structure. The classified ROI image set can clearly show the independent characteristics of each structure, avoid confusion between different structures, and improve segmentation accuracy and consistency.

[0125] The classified ROI area image set directly focuses on specific anatomical structures, which facilitates doctors to perform quantitative analysis and clinical diagnosis, such as intervertebral disc lesions, joint degeneration, muscle degeneration, etc., greatly improving the accuracy of the analysis.

[0126] Through precise ROI extraction, background noise or other irrelevant information interference in the image is reduced, providing a clear data basis for subsequent signal intensity analysis, morphological analysis and other steps.

[0127] The system marks the corresponding anatomical structure regions in each MRI image based on the segmentation mask data of the classified MRI images. The pixel values ​​of these regions are updated to specific label values ​​corresponding to their masks, so that the anatomical structures in the MRI images are clearly identified, and a labeled MRI image set is obtained.

[0128] By using the segmentation mask generated by VM-UNet, each anatomical structure in the MRI image can be accurately located and labeled. The error caused by manual labeling is reduced, especially in complex anatomical structures (such as intervertebral discs, dural sacs, etc.), and the system can accurately distinguish the boundaries of different structures.

[0129] The automated labeling process ensures consistency across all images, reduces physician manual operation time, and ensures accuracy and consistency in large batches of image processing.

[0130] The system uses image processing algorithms to extract key features from the labeled MRI image set. These features include grayscale information, boundary information, etc. in the image.

[0131] Based on the feature extraction results, the system cuts out specific ROIs from the labeled MRI images. ROIs are anatomical regions of interest, such as intervertebral discs, dural sacs, articular processes, etc., to ensure that the extracted regions focus on key structures and exclude irrelevant background.

[0132] Through ROI extraction, the system focuses the analysis on the target anatomical structure and excludes irrelevant background information, which helps the accuracy of subsequent signal analysis and morphological analysis.

[0133] After extracting the ROI, subsequent image processing only needs to focus on the extracted area, which greatly reduces the amount of calculation and improves processing efficiency.

[0134] The system classifies different anatomical structures (such as intervertebral disc, dural sac, etc.) into independent categories based on the features and segmentation mask data of each ROI, and each category corresponds to a different set of ROI area images.

[0135] The classified ROI area image set includes:

[0136] Intervertebral disc ROI area image set: used to analyze intervertebral disc degeneration, herniation and other problems.

[0137] Dural sac ROI region image set: used to evaluate spinal cord and nerve root compression.

[0138] Facet joint ROI region image set: used to analyze joint degeneration.

[0139] Psoas, erector spinae, and multifidus ROI image set: used to evaluate problems such as muscle degeneration or fatty infiltration.

[0140] By classifying different anatomical regions, the system can more clearly display the various anatomical structures in the MRI image, facilitating subsequent independent analysis.

[0141] The classified ROI area image set directly corresponds to the specific lesion, which can provide doctors with more targeted diagnostic support and facilitate further analysis of the morphology and functional status of each structure.

[0142] In at least one embodiment of the present application, the step of performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data comprises:

[0143] S201, converting each image in the classified ROI region image set into a grayscale image to generate a ROI region grayscale image set, calculating the signal intensity mean and signal intensity standard deviation of the ROI region in each image of the ROI region grayscale image set, and obtaining all ROI region signal intensity feature data;

[0144] S202, performing texture feature extraction on the classified ROI region grayscale image set to extract contrast difference value, entropy value and uniformity value in each ROI region grayscale image to obtain all ROI region texture feature data.

[0145] Please refer to Figure 1-Figure 4 ,In this implementation, the classified ROI area image is converted into a gray scale image, the data structure is simplified, and the signal intensity and texture information are retained.

[0146] The mean and standard deviation of the signal intensity of each ROI area are calculated to quantify the tissue density and signal distribution in the area, assisting doctors in determining whether there is a lesion.

[0147] By extracting contrast difference values, entropy values, and uniformity values, we can analyze subtle structural changes in tissues, quantify tissue complexity and heterogeneity, and reveal pathological changes.

[0148] By quantitatively analyzing signal intensity and texture features, doctors can obtain more objective data and reduce reliance on visual judgment, especially when detecting lesions such as intervertebral disc degeneration, spinal canal stenosis, and muscle degeneration.

[0149] The microstructural changes captured by texture feature extraction (such as contrast, entropy, and uniformity) can reveal lesions or structural changes that are difficult to detect with the naked eye, especially in early lesions or minor injuries, and can provide additional diagnostic support for doctors.

[0150] Through automated calculation of signal intensity and texture features, the system is able to generate standardized data, reducing subjective errors in measurement and diagnosis by doctors, and helping to ensure consistency, especially when processing large-scale images.

[0151] It should be noted that the extracted region of interest (ROI) image set of each anatomical structure (such as intervertebral disc, joint, muscle, etc.) is loaded and converted into a grayscale image.

[0152] Each color or multi-channel ROI image is converted into a grayscale image, and each pixel in the image is assigned a grayscale value (ranging from 0 to 255) according to its brightness. The grayscale image retains important brightness information while removing color information, which facilitates subsequent signal intensity calculation and texture feature analysis.

[0153] By graying the images, the processing of complex pixel information in MRI images is simplified, allowing the system to perform signal intensity analysis and texture feature extraction more efficiently.

[0154] Although the image has been grayscaled, it still retains key brightness and texture information and can accurately reflect the density and morphological changes of tissues, making it particularly suitable for the analysis of soft tissue lesions.

[0155] The system scans the pixel values ​​of the grayscale image of each ROI area and calculates the average signal intensity in the entire ROI area. The mean signal intensity represents the overall brightness level of the anatomical structure and is usually used to evaluate tissue density or pathological conditions (such as intervertebral disc degeneration).

[0156] At the same time, the system also calculates the standard deviation of signal intensity in each ROI area to reflect the degree of discreteness of signal intensity, helping to identify heterogeneity or irregular changes within the tissue, such as detecting fibrosis or lesions within soft tissue.

[0157] Signal strength analysis reduces the subjectivity of doctors' reliance on visual judgment, making the judgment more scientific and accurate. Especially in the diagnosis of lumbar spine diseases, signal strength can reveal potential pathological changes and provide reference data for doctors' clinical judgment.

[0158] Contrast difference value: Calculates the brightness difference between adjacent pixels in the image, reflecting the boundary clarity and internal detail changes of the tissue. For example, the contrast difference value of the intervertebral disc boundary can reveal whether there is a protruding lesion.

[0159] Entropy: Entropy is a measure of the complexity or disorder of an image. A higher entropy value usually means a more complex tissue structure, which may correspond to abnormal or diseased tissue, while a lower entropy value indicates a more consistent and regular tissue structure.

[0160] Homogeneity value: Homogeneity reflects the consistency of the grayscale values ​​of pixels in the image. A higher homogeneity means that the tissue structure is relatively uniform, while a lower homogeneity may indicate the presence of uneven changes in the tissue, such as fibrosis or fat infiltration.

[0161] Texture feature analysis can deeply capture subtle differences within tissues and reveal structural changes that cannot be discovered by traditional signal intensity analysis. For example, texture features can reveal problems such as joint degeneration, muscle fibrosis, or intervertebral disc damage.

[0162] By extracting texture features (contrast, entropy and uniformity), doctors are provided with quantitative diagnostic support data to facilitate the analysis of tissue complexity and the degree of lesions.

[0163] It should be noted that the signal intensity mean and signal intensity standard deviation are as follows: the classified ROI area image set is converted into a grayscale image, the signal distribution of the intervertebral disc area is extracted and stored in an array, and the following quantitative indicators are output after analysis: signal intensity mean and signal intensity standard deviation.

[0164] In at least one embodiment of the present application, the step of performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions includes:

[0165] S301, using Canny edge detection to extract the ROI boundary of each ROI image in the classified ROI region image set, and obtaining all ROI edge image sets and all ROI contour image sets;

[0166] S302, calculating the area of ​​each ROI region and the total length of the ROI region edge in all ROI edge image sets to obtain ROI region area data and ROI edge length data;

[0167] S303, calculating the local curvature and the overall curvature of each ROI region in the set of all ROI contour images to obtain ROI region curvature data;

[0168] S304, calculating the convexity of each ROI region in the set of all ROI contour images to obtain ROI region convexity data.

[0169] Please refer to Figure 1-Figure 4 In this embodiment, first, the Canny edge detection algorithm is used to extract the edge of each classified ROI image to generate a ROI edge image set and a contour image set.

[0170] Next, the system calculates the area and total edge length of the edge image of each ROI region to obtain the area data and edge length data of each anatomical structure.

[0171] The system then calculates the local and global curvature of the contour image of each ROI to quantify the morphological characteristics of the anatomical structure. In particular, the curvature provides detailed analysis data when evaluating the degree of boundary curvature.

[0172] Finally, the system evaluates the morphological integrity and degree of abnormality of the anatomical structure by calculating the convexity of each ROI area.

[0173] Through Canny edge detection and calculation of morphological features (area, perimeter, curvature, convexity), the system can provide accurate morphological data for each anatomical structure. These data provide doctors with quantitative morphological evaluation indicators to help them identify subtle morphological changes and lesions.

[0174] Changes in morphological characteristics (especially curvature and convexity) can reveal the pathological state of tissues, such as the degree of intervertebral disc herniation, joint degeneration, muscle atrophy, etc.

[0175] By automatically calculating the area, perimeter, curvature and convexity, subjective errors in manual measurement are reduced. Especially in the case of irregular morphology, the system can extract morphological features more accurately, ensure the consistency of analysis, reduce subjective errors, and provide strong support for clinical treatment decisions.

[0176] The system calculates the gradient change of the image through Canny edge detection and accurately identifies the boundaries in the image. The system applies this algorithm to each classified ROI image and automatically extracts the edge information of these areas.

[0177] Based on the edge detection results, the system generates edge images for each ROI, and further forms a complete set of contour images by closing these edges. It can reduce noise interference while maintaining the clarity of boundary information, ensuring the accurate extraction of ROI area boundaries, which is especially suitable for anatomical structures with complex morphology.

[0178] Extracting edges and contours is the basis for subsequent morphological feature analysis (such as area, curvature, convexity, etc.). Accurate boundary information ensures the accuracy of morphological data.

[0179] The area of ​​each ROI region edge image is calculated to obtain the area data of each ROI region.

[0180] The area of ​​the enclosed region in the ROI edge image is calculated using the pixel counting method or the integration method to generate accurate area data for each anatomical structure.

[0181] The total edge length of each ROI region is calculated to obtain the perimeter (or edge length) data of each ROI region.

[0182] By calculating the total length of edge pixels, the total edge length of each ROI area, that is, the perimeter of the anatomical structure, is determined.

[0183] Area and perimeter are important morphological features of anatomical structures that can help doctors quantify the size and morphological changes of tissues. For example, the area change of an intervertebral disc may indicate its protrusion, and the edge length of the articular process may reflect the degree of joint degeneration.

[0184] By calculating the area and edge length, doctors can more clearly assess the size, range, and morphological characteristics of the lesion area. Especially when detecting intervertebral disc herniation or soft tissue degeneration, these data can provide accurate quantitative basis.

[0185] The system uses a curvature calculation method (such as the curvature vector method or the second-order derivative method) to calculate the local curvature of each ROI contour and identify the most obvious curved part of the contour, such as the location of a herniated disc.

[0186] The curvature of the entire contour is integrated to calculate the overall curvature of the ROI area, which helps to evaluate the global morphological characteristics of the anatomical structure.

[0187] Changes in curvature can reveal abnormal morphology of the anatomical structure. For example, an increase in local curvature of the intervertebral disc may indicate herniation or degeneration, and a change in the overall curvature may indicate morphological distortion.

[0188] Local and global curvature data provide doctors with quantitative basis for morphological changes, especially in the early identification of small lesions or abnormal morphological changes, curvature analysis can provide more detailed information.

[0189] The system uses the convex hull algorithm to first generate the convex hull of each ROI area, and then calculate the area difference or boundary difference between the ROI area and the convex hull to obtain the convexity value of the area. The higher the convexity, the more irregular the shape of the area.

[0190] For example, the convexity of the intervertebral disc can reflect its degree of herniation, and the convexity of the articular process or soft tissue can help identify morphological deformities.

[0191] Changes in convexity can help doctors identify anatomical distortions or abnormal protrusions. Convexity is an important quantitative indicator, especially when evaluating morphological abnormalities such as disc herniation and joint degeneration.

[0192] The combination of convexity with morphological features such as area and curvature can provide doctors with a more comprehensive analysis of the anatomical structure and assist in accurate diagnosis.

[0193] It should be noted that the area and perimeter are: the contour features of the ROI are extracted using the Canny edge detection algorithm. The method for calculating the contour area is to use the Green formula (also known as the contour integral) to calculate the area inside the contour. For the calculation of the perimeter, the lengths of all line segments of the contour are added together.

[0194] Gaussian filtering is used to remove image noise and enhance the smoothness of the image. The Gaussian filtering formula is:

[0195]

[0196] Among them, σ is the standard deviation of the Gaussian function, and * represents the convolution operation. Next, the gradient calculation is performed:

[0197] G x =S x *G(x,y),G y =S y *G(x,y);

[0198]

[0199] Θ(x, y) = tan -1 (G y / G x );

[0200] Among them, G x is the horizontal gradient of the image (calculated using the Sobel operator), G y is the vertical gradient of the image (calculated using the Sobel operator), M(x, y) is the calculated gradient magnitude, indicating the edge strength of each pixel. Θ(x, y) is the calculated gradient direction, indicating the direction of the edge.

[0201] Next, non-maximum suppression is performed. For each pixel (x, y), according to the value of its gradient direction Θ(x, y), it is quantized to 0°, 45°, 90°, and 135°. Compare the amplitude of the current pixel with the adjacent pixels in its gradient direction. If the current pixel is a local maximum, it is retained; otherwise, it is suppressed to 0.

[0202] Double threshold processing, setting two thresholds TH (high threshold) and TL (low threshold); strong edge: set to all pixels greater than TH; weak edge: set to pixels between TL and TH; non-edge: all pixels less than TL are set to 0; edge connection: if there is a strong edge pixel in the 8-neighborhood of a potential edge pixel, it will be retained as an edge; otherwise, it will be suppressed to 0.

[0203] The Canny edge detection algorithm is constructed to ensure the extraction of clear and continuous edge features. The Green formula (Green) is used to calculate the area of ​​the plane area. According to the closed area contour C of the ROI area image and its internal area R, Green is used for calculation:

[0204] A=∫ R dA=∫ C xdy-ydx;

[0205] Where A is the area of ​​region R, (x, y) is the coordinates of the point on the contour, dA is the area element, and the entire formula represents the integral within the region.

[0206] Curvature: The curvature of the contour is calculated using a difference method based on neighborhood points. Specifically, the curvature vector method is used to sample points on the contour, and its local curvature is calculated to estimate the overall curvature of the contour. The average of all local curvatures is taken as the curvature of the contour. The closer the curvature of the contour is to 1, the closer the entire contour is to a straight line, that is, the contour tends to be flat. The closer it is to 0, the more curved or complex the entire contour is. Figure 3 (A) shows a heat map of the change in curvature on the ROI contour image, from blue (smooth) to red (steep).

[0207] Curvature vector method formula:

[0208] κ i =(x′ i y″ i -y′ i x″ i ) / (x′ i 2 +y′ i 2 ) 3 / 2 ;

[0209] Among them, κ i For point (x i ,yi ). x′ i For point (x i ,y i ) can be calculated using the difference method, usually approximated by the coordinates of adjacent points:

[0210] x′ i ≈(x i+1 -x i-1 ) / 2Δt;

[0211] y′ i ≈(y i+1 -y i-1 ) / 2Δt;

[0212] Among them, y′ i For point (x i ,y i )’s first-order derivative; x″ i is the second-order derivative, which represents the change in curvature:

[0213] x″ i ≈(X i+1 -2x i +x i-1 ) / Δt 2 ;

[0214] y″ i is an approximation of the second-order derivative: y″ i ≈(y i+1 -2y i +y i-1 ) / Δt 2 ;

[0215] Δt is the parameter change between two adjacent sampling points, which is usually set to 1 or an equivalent distance.

[0216] For the curvature K of the entire contour, the average of all local curvatures can be taken, and the calculation formula for the overall curvature is as follows:

[0217]

[0218] Among them, K is the average curvature of the entire contour; N is the total number of sampling points on the contour.

[0219] In geometry, the convex hull is the smallest convex polygon of a contour point set, using the Graham scan algorithm: an algorithm based on polar angle sorting and stacking for computing the convex hull of a point set. The convexity of a contour is obtained by dividing the actual area of ​​the contour by the area of ​​its convex hull. The convexity value ranges from 0 to 1, with values ​​closer to 0 indicating a more convex contour and values ​​closer to 01 indicating a flatter contour.

[0220] First, get the convex hull vertices. If the convex hull vertices are arranged in order ((x 1 ,y 1 ), (x 2 ,y 2 ),...,(x n ,y n ), then the area (A) is calculated as follows:

[0221]

[0222] Area calculation of contour:

[0223]

[0224] Among them, A 实际 is the actual area of ​​the contour; m is the number of contour vertices; x i y i is the coordinate of the contour vertex (i), (i = 1, 2, ..., m); (x m+1 ,y m+1 ): defined as (x 1 ,y 1 ).

[0225] Convexity calculation formula: C = (A 实际 ) / (A 凸包 );

[0226] Among them, C is the convexity value of the profile, A 实际 is the actual area of ​​the contour; A 凸包 is the area of ​​the convex hull.

[0227] In at least one embodiment of the present application, the specific steps of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis include:

[0228] Screening out an intervertebral disc ROI region image from the classified ROI region image set to obtain an intervertebral disc ROI region image;

[0229] Filtering out the ROI edge image corresponding to the intervertebral disc ROI area image from all ROI edge images;

[0230] The herniation area is calculated based on the intervertebral disc ROI region image and the ROI edge image to generate an intervertebral disc herniation area value.

[0231] In this embodiment, the system automatically selects image segments corresponding to the intervertebral disc from the classified ROI region image set.

[0232] The system further selects the edge image corresponding to the intervertebral disc ROI image from the ROI edge image set, and automatically calculates the herniation area of ​​the intervertebral disc by combining the intervertebral disc ROI area image with its edge image, generating a quantitative herniation area value.

[0233] Through automatic screening and calculation, the system can efficiently process a large number of MRI images, reduce the doctor's manual operations, and ensure the consistency and accuracy of data processing. The automatic calculation of the herniation area provides doctors with quantitative evaluation data, which facilitates accurate judgment of the severity of intervertebral disc herniation and provides an objective basis for surgical decisions.

[0234] The system automatically generates herniation area data, avoiding subjective bias and errors in the manual measurement process, ensuring the consistency and accuracy of each calculation result, and reducing the risk of misdiagnosis and missed diagnosis.

[0235] Through quantitative analysis reports, the system provides strong support for doctors' diagnosis and treatment decisions. Especially when facing complex lesions, accurate data can significantly improve the quality of diagnosis and treatment effects.

[0236] The system has previously segmented and classified different anatomical structures (such as intervertebral discs, dural sacs, joints, etc.) in the MRI images through segmentation and classification operations. At this point, the system selects ROI area images related to the intervertebral disc.

[0237] Based on the segmentation mask, the system automatically identifies and extracts all image segments corresponding to the intervertebral disc, generates an intervertebral disc ROI area image set, and ensures the accuracy and consistency of image classification.

[0238] By automatically screening intervertebral disc images, the workload of doctors' manual labeling is reduced, ensuring that intervertebral disc-related data can be extracted quickly and accurately.

[0239] The automated screening process improves the efficiency of the system in processing large-scale data, ensures accurate classification and is suitable for large-scale MRI image processing, especially for batch processing needs in clinical practice.

[0240] The system has previously generated edge images of each anatomical structure using the Canny edge detection algorithm. At this point, the system selects image segments related to the intervertebral disc from these edge images, which will be used for subsequent area calculations.

[0241] The system matches the intervertebral disc ROI area image with its corresponding edge image to ensure that the extracted edge data is consistent with the boundary of the ROI area image.

[0242] The system first determines the outer contour of the intervertebral disc through the edge image, and calculates the area of ​​the herniated part by using the pixel integration method or the geometric method in combination with the protruding area in the intervertebral disc ROI area image.

[0243] The system automatically generates the herniated area data of the intervertebral disc and outputs the data into the quantitative analysis report. These data will provide doctors with accurate basis for lesion assessment, which is especially important when judging the degree of intervertebral disc herniation.

[0244] In at least one embodiment of the present application, the specific step of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also includes:

[0245] Screening out an intervertebral foramen ROI region image from the classified ROI region image set to obtain an intervertebral foramen ROI region image;

[0246] The left and right width values ​​of the intervertebral foramen ROI region image are calculated to obtain the intervertebral foramen width value.

[0247] Please refer to Figure 1-Figure 4 In this embodiment, the system automatically selects image segments related to the intervertebral foramen from the classified ROI area image set, and the system automatically identifies the left and right boundaries of the intervertebral foramen using an image analysis algorithm and calculates its width. Through precise boundary recognition and distance calculation, the system integrates the measured intervertebral foramen width data into a quantitative analysis report for doctors' reference to evaluate the degree of intervertebral foraminal stenosis.

[0248] Through automatic screening and measurement, the system can efficiently process a large number of MRI images, especially when faced with large-scale patient data. This automated process significantly improves diagnostic efficiency and reduces doctors' workload.

[0249] The width of the intervertebral foramen is a key indicator for diagnosing foraminal stenosis. Through the quantitative width value automatically generated by the system, doctors can more accurately judge the health status of the intervertebral foramen, especially in the early detection of foraminal stenosis. Quantitative data can help doctors make scientific diagnostic decisions.

[0250] It should be noted that the system automatically screens out the ROI area image of the intervertebral foramen through the previously generated segmentation mask. This process is based on the anatomical position and characteristics of the intervertebral foramen in the image to ensure that the extracted image segments are accurate.

[0251] The system identifies the two side edges of the intervertebral foramen according to the anatomical boundaries in the ROI image, automatically measures the distance between the left and right intervertebral foramina, and obtains the width value by calculating the shortest distance between the two side boundaries based on pixel-level edge detection.

[0252] It should be further explained that the Canny edge detection algorithm is first used to extract the contours of the intervertebral disc and FacetJoint. The nearest neighbor search algorithm is used for the point sets of the two contours. The nearest points between the left and right half contours are compared respectively, which become the key points for measuring the left and right intervertebral foramen diameters, named DL, DR, FL, FR, DL and FL, and the Euclidean distance between DR and FR are defined as the intervertebral foramen widths on the left and right sides. DC is the center of mass of the intervertebral disc, and the intervertebral disc point set between DL and DR is the key intervertebral disc contour; the FacetJoint point set between FL and FR is the key articular process joint contour, and the lowest point of this point set is taken as FM. Connect DL and DR to form a straight line DL-DR, connect DC and FM to form a straight line DC-FM, and the intersection of the straight line DC-FM and the key intervertebral disc contour is named DM.

[0253] The closed contour formed by the critical disc contour and DL-DR is defined as the relative herniation area of ​​the disc: The actual area of ​​the vertical line integral can be calculated by calculating the integral of the product of the perpendicular distance and the transverse distance from each point on the path to the curve, which can be achieved by parameterizing the path. Let the curve critical disc contour be parameterized as, where t changes from ta to tb (corresponding to the path from DL to DR).

[0254] The herniation area A can be calculated using the following integral formula:

[0255]

[0256] Where x is the vector product, r′(t) is the tangent vector of the curve at time t, and the vector (p DR -p DL ) is point p DR To point p DL direction, A represents the vector (p DR -p DL ) to the area of ​​the parallelogram formed by the vector r′(t), and the integral accumulates these areas to measure the actual herniation and depression area of ​​the intervertebral disc. The vertical line integral can be distinguished into positive and negative values ​​through vector calculation to obtain the herniation area, which can indicate the degree of herniation to a certain extent.

[0257] Similarly, the relative area of ​​the spinal canal can be calculated, and then the Euclidean distance between DM and FM can be defined as the median diameter of the spinal canal.

[0258] In at least one embodiment of the present application, the specific step of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also includes:

[0259] Screening out a spinal canal ROI region image from the classified ROI region image set to obtain a spinal canal ROI region image;

[0260] The relative area of ​​the spinal canal and the median diameter of the spinal canal in the spinal canal ROI region image are calculated to obtain the spinal canal stenosis value.

[0261] Please refer to Figure 1-Figure 4 In this embodiment, the system automatically selects image segments related to the spinal canal from the classified ROI image set, and calculates the relative area of ​​the spinal canal through morphological analysis of the spinal canal.

[0262] The system further calculates the median diameter of the spinal canal and, by identifying the distance between the anterior and posterior boundaries of the spinal canal, generates an accurate morphological index that reflects the cross-sectional changes of the spinal canal.

[0263] By combining the relative area with the median spinal diameter, the system generates a quantitative spinal stenosis value, which is used to evaluate the degree of spinal stenosis and serve as a basis for diagnosis and treatment.

[0264] Through automatic screening and calculation, the system can process a large number of MRI images quickly and efficiently, especially when faced with large-scale patient data, reducing the manual workload of doctors and improving diagnostic efficiency.

[0265] The relative area and median diameter of the spinal canal are key indicators for evaluating spinal stenosis. The spinal stenosis value generated by the system can provide doctors with accurate quantitative data, helping them to objectively evaluate the condition and avoid errors caused by relying solely on visual judgment.

[0266] The system automatically screens out the ROI area images related to the spinal canal based on the generated segmentation mask data. The system automatically extracts the ROI area images of the spinal canal by identifying and classifying the spinal canal areas in the MRI images.

[0267] The system first calculates the actual area of ​​the spinal canal based on the ROI area image of the spinal canal. By comparing it with the corresponding standard spinal canal area, the relative area is calculated. The relative area can reflect the size change of the spinal canal area, especially when the spinal canal is narrow, the relative area will be significantly reduced.

[0268] The median vertebral diameter is the anteroposterior diameter of the central axis of the spinal canal, which represents the central cross-sectional diameter of the spinal canal. The system automatically detects the anterior and posterior boundaries of the spinal canal and calculates the length of this diameter.

[0269] The system combines the relative area of ​​the spinal canal with the median diameter to generate a quantitative spinal stenosis value. This value can be directly used to assess the degree of spinal stenosis, providing doctors with accurate quantitative data to assist in diagnosis and treatment decisions.

[0270] In at least one embodiment of the present application, the specific step of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also includes:

[0271] Screening out a psoas major muscle ROI region image and a multifidus muscle ROI region image from the classified ROI region image set to obtain a muscle fat ROI region image;

[0272] The Otsu binarization algorithm is used to calculate the fat area ratio value in the muscle fat ROI region image to obtain the muscle fat infiltration value.

[0273] Please refer to Figure 1-Figure 4 In this embodiment, the system automatically selects image segments related to the psoas major and multifidus muscles from the classified ROI area image set.

[0274] The system uses the Otsu binarization algorithm to process the screened muscle fat ROI area images, automatically segment the fat area, and calculate the relative area ratio of fat.

[0275] Based on the calculation results of fat area, the system generates a quantitative value of muscle fat infiltration and integrates this value into the quantitative analysis report for the doctor's reference.

[0276] By automatically screening, segmenting and calculating fat infiltration, the system can efficiently process a large number of MRI images, significantly reducing the workload of doctors.

[0277] The muscle fat infiltration value is an important quantitative indicator for diagnosing muscle degeneration. Through automated calculation, the system provides objective and scientific quantitative data to help doctors make a more accurate assessment of the patient's muscle condition.

[0278] Through the Otsu binarization algorithm, fat and muscle can be automatically separated, reducing the subjectivity and inaccuracy in manual operations, ensuring the accuracy of fat area ratio calculations, and providing an important basis for doctors to diagnose diseases such as muscle degeneration and atrophy.

[0279] It should be noted that the system automatically identifies and screens out image regions related to the psoas major and multifidus muscles, and extracts ROI images related to the psoas major and multifidus muscles through automatic screening by the system.

[0280] The Otsu binarization algorithm was used to process the screened muscle fat ROI area images, automatically separate the fat and muscle parts, and calculate the proportion of the fat area.

[0281] The Otsu algorithm is an automatic threshold segmentation technique that separates muscle and fat tissue by determining an optimal grayscale threshold in the image.

[0282] The system finds the optimal threshold for distinguishing muscle and fat by analyzing the grayscale value distribution of the muscle-fat ROI area image.

[0283] Once the fat area is segmented, the system calculates the pixel area occupied by the fat and compares it with the total area of ​​the entire ROI area to obtain the relative area ratio value of the fat.

[0284] The system automatically generates a muscle fat infiltration value based on the previously calculated fat area ratio. This value directly reflects the content and distribution of fat in the muscle. Generally, muscle areas with higher fat infiltration indicate muscle degeneration or atrophy.

[0285] The muscle fat infiltration value will be integrated into the quantitative analysis report for doctors' reference, helping them to judge the patient's muscle health status and assist in diagnosis and treatment decisions.

[0286] It should be further explained that the muscle area, dural sac area, muscle fat infiltration, cauda equina density, cauda equina tightness, and cauda equina deviation are calculated according to the above-mentioned area calculation formula.

[0287] Muscle fat infiltration and cauda equina nerve density: The ROI area image first undergoes the following preprocessing steps: the resolution is enlarged by 4 times using the linear interpolation method, converted into a grayscale image, and then denoised using Gaussian filtering. Finally, the Otsu binarization algorithm is used to distinguish between fat tissue and muscle tissue, and then the resolution is enlarged (linear interpolation). The linear interpolation formula is:

[0288] I ′ (x ′ ,y ′ )=(1-a)(1-b)I(x,y)+a(1-b)I(x+1,y)+(1-a)bI(x,y+

[0289] 1)+abI(x+1,y+1);

[0290] Assume that the pixel value of the original image is I(x,y) and the pixel value of the target image is I ′ (x,y ′ ), where x ′ =4 and y ′ =4 (enlarge it 4 times), a = x ′ - x and b = y ′ -y. These parameters are between 0 and 1 and represent the weight of the interpolation.

[0291] It should be further explained that if the original image is in color (RGB image), it needs to be converted into a grayscale image.

[0292] Then it is denoised by Gaussian filtering, and finally calculated using the Otsu binarization algorithm:

[0293]

[0294] Among them, t * The optimal threshold is the inter-class variance corresponding to threshold t, \argmax t It means to select the threshold that maximizes the inter-class variance among all possible thresholds t. After the binary image, the muscle fat infiltration (cauda equina density) can be obtained by simply dividing the white (black) dot area by the ROI area.

[0295] After Otsu binarization, the ROI area image was extracted from black and white pixel arrays, and the average value of the reciprocal of the Euclidean distance between black points was calculated as the compactness index. The larger the value, the more compact the points are; conversely, the points are more dispersed. The compactness reflects the degree of dispersion of the cauda equina inside the dural sac; the central coordinates of the black and white points were calculated, and the horizontal and vertical Euclidean distances of the two coordinates were compared, representing the horizontal and vertical deviation of the cauda equina in the dural sac.

[0296] A lumbar vertebrae MRI image quantification system 100 is applied to any one of the above-mentioned lumbar vertebrae MRI image quantification methods, and the lumbar vertebrae MRI image quantification system 100 comprises:

[0297] The acquisition module 110 is used to acquire the MRI image to be processed; the VM-UNet model 120 is used to segment the MRI image set to be processed to obtain the segmentation mask data of the classified MRI image; the ROI extraction module 130 performs ROI extraction on the MRI image set to be processed according to the segmentation mask data of the classified MRI image; the signal intensity processing module 140 performs signal intensity processing on the classified ROI area image set; the texture feature analysis module 150 performs texture analysis on the classified ROI area image set; the morphological feature calculation module 160 performs morphological feature calculation on the classified ROI area image set; the quantitative analysis report output module 170 generates and outputs a quantitative analysis report based on the classified ROI area image set and the morphological data of all ROI areas to assist doctors in diagnosis; the lumbar vertebral magnetic resonance image The quantification system 100 performs the following steps: obtaining an MRI image set to be processed; inputting the MRI image set to be processed into a VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI image; extracting the MRI image set to be processed based on the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set; performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain signal intensity feature data of all ROI regions and texture feature data of all ROI regions; performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions; generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis.

[0298] Please refer to Figure 5 In this embodiment, the quantification system 100 of the lumbar magnetic resonance image acquires the MRI image to be processed through the acquisition module 110, and then inputs the MRI image set to be processed into the VM-UNet model 120. The VM-UNet model 120 outputs the segmentation mask data of the classified MRI image, and then the ROI extraction module 130, the signal intensity processing module 140, the texture feature analysis module 150, and the morphological feature calculation module 160 perform ROI extraction, signal and texture analysis, and morphological processing. Finally, the quantitative analysis report output module 170 generates a quantitative analysis report and outputs the quantitative analysis report, which is provided to doctors for diagnosing lumbar diseases.

[0299] Through automatic segmentation and quantitative analysis, the quantification system of lumbar MRI images can quickly and efficiently process large-scale MRI image sets, significantly improving the diagnostic efficiency of doctors. At the same time, it provides accurate analysis results to assist doctors in making accurate diagnoses in complex cases.

[0300] The quantification system 100 for lumbar MRI images provides a quantitative analysis report, which greatly reduces the workload of doctors when analyzing MRI images, and especially reduces the errors and time costs in manual measurement.

[0301] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the following steps:

[0302] The quantification method of the lumbar magnetic resonance image includes: obtaining an MRI image set to be processed; inputting the MRI image set to be processed into a VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI image; extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set; performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain signal intensity feature data of all ROI regions and texture feature data of all ROI regions; performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions; generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis.

[0303] Figure 6 FIG. 1 shows an internal structure diagram of a computer device in an embodiment. The computer device may be a terminal or a server. Figure 6 As shown, the computer device includes a processor, a memory and a network interface connected through a system bus.

[0304] A computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0305] The quantification method of the lumbar magnetic resonance image includes: obtaining an MRI image set to be processed; inputting the MRI image set to be processed into a VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI image; extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set; performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain signal intensity feature data of all ROI regions and texture feature data of all ROI regions; performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions; generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis.

[0306] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0307] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0308] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A method for quantifying lumbar vertebrae magnetic resonance images, characterized in that: The quantification method of the lumbar vertebrae magnetic resonance image comprises: Obtaining a set of MRI images to be processed; Inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI images; Extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set; Performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data; Performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions; A quantitative analysis report is generated based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in making a diagnosis.

2. The method for quantifying lumbar vertebrae nuclear magnetic resonance images according to claim 1, characterized in that: The specific steps of inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputting the segmentation mask data of the MRI image include: The VM-UNet model extracts features from the MRI image set to be processed through an encoder, and then restores the resolution of the image through a decoder, generates and outputs segmentation mask data of the classified MRI image, and the anatomical structures include: intervertebral disc, dural sac, facet joints, psoas major muscle, erector vertebrae and multifidus muscle.

3. The method for quantifying lumbar vertebrae nuclear magnetic resonance images according to claim 1, characterized in that: The specific steps of extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain the extracted classified ROI region image set include: Marking each MRI image in the MRI image set to be processed according to the segmentation mask data of the classified MRI image to generate a marked MRI image set; Feature extraction is performed on each MRI image in the labeled MRI image set to obtain an extracted ROI region image set, and the extracted ROI region image set is classified to obtain a classified ROI region image set, wherein the classified ROI region image set includes: an intervertebral disc ROI region image set, a dural sac ROI region image set, a facet joint ROI region image set, a psoas major muscle ROI region image set, a vertical spine ROI region image set, and a multifidus muscle ROI region image set.

4. The method for quantifying lumbar vertebrae nuclear magnetic resonance images according to claim 1, characterized in that: The step of performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data comprises: Convert each image in the classified ROI region image set into a grayscale image to generate a ROI region grayscale image set, calculate the signal intensity mean and signal intensity standard deviation of the ROI region in each image in the ROI region grayscale image set, and obtain all ROI region signal intensity feature data; Texture feature extraction is performed on the classified ROI region grayscale image set to extract the contrast difference value, entropy value and uniformity value in each ROI region grayscale image to obtain all ROI region texture feature data.

5. The method for quantifying lumbar vertebrae nuclear magnetic resonance images according to claim 1, characterized in that: The step of performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions comprises: Use Canny edge detection to extract the ROI boundary of each ROI image in the classified ROI region image set, and obtain all ROI edge image sets and all ROI contour image sets; Calculate the area of ​​each ROI region and the total length of the ROI region edge in all ROI edge image sets to obtain ROI region area data and ROI edge length data; Calculating the local curvature and the overall curvature of each ROI region in the set of all ROI contour images to obtain ROI region curvature data; The convexity of each ROI region in the ROI contour image set is calculated to obtain ROI region convexity data.

6. The method for quantifying lumbar vertebrae magnetic resonance images according to claim 1, characterized in that: The specific steps of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis include: Screening out an intervertebral disc ROI region image from the classified ROI region image set to obtain an intervertebral disc ROI region image; Filtering out the ROI edge image corresponding to the intervertebral disc ROI area image from all ROI edge images; The herniation area is calculated based on the intervertebral disc ROI region image and the ROI edge image to generate an intervertebral disc herniation area value.

7. The method for quantifying lumbar vertebrae magnetic resonance images according to claim 1, characterized in that: The specific steps of generating a quantitative analysis report based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in diagnosis also include: Screening out an intervertebral foramen ROI region image from the classified ROI region image set to obtain an intervertebral foramen ROI region image; Calculating the left and right width values ​​in the intervertebral foramen ROI region image to obtain the intervertebral foramen width value; Screening out a spinal canal ROI region image from the classified ROI region image set to obtain a spinal canal ROI region image; Calculating the spinal canal relative area and the median diameter of the spinal canal in the spinal canal ROI region image to obtain the spinal canal stenosis value; Screening out a psoas major muscle ROI region image and a multifidus muscle ROI region image from the classified ROI region image set to obtain a muscle fat ROI region image; The Otsu binarization algorithm is used to calculate the fat area ratio value in the muscle fat ROI region image to obtain the muscle fat infiltration value.

8. A lumbar vertebrae nuclear magnetic resonance image quantification system, applied to the lumbar vertebrae nuclear magnetic resonance image quantification method as claimed in any one of claims 1 to 9, characterized in that: The quantification system of the lumbar vertebrae magnetic resonance image comprises: An acquisition module, used for acquiring the MRI image to be processed; The VM-UNet model is used to segment the MRI image set to be processed to obtain segmentation mask data of the classified MRI images; A ROI extraction module extracts ROI from the MRI image set to be processed according to the segmentation mask data of the classified MRI image; A signal intensity processing module performs signal intensity processing on the classified ROI area image set; Texture feature analysis module, which performs texture analysis on the classified ROI area image set; A morphological feature calculation module calculates morphological features of the classified ROI area image set; The quantitative analysis report output module generates and outputs a quantitative analysis report based on the classified ROI area image set and the morphological data of all ROI areas to assist doctors in making a diagnosis; The quantification system of the lumbar spine magnetic resonance image performs the following steps: Obtaining a set of MRI images to be processed; Inputting the MRI image set to be processed into the VM-UNet model, and the VM-UNet model outputs segmentation mask data of the classified MRI images; Extracting the MRI image set to be processed according to the segmentation mask data of the classified MRI image to obtain an extracted classified ROI region image set; Performing signal intensity processing and texture feature analysis on the classified ROI region image set to obtain all ROI region signal intensity feature data and all ROI region texture feature data; Performing morphological feature processing on the classified ROI region image set to obtain morphological data of all ROI regions; A quantitative analysis report is generated based on the classified ROI region image set and the morphological data of all ROI regions to assist doctors in making a diagnosis.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for quantifying a lumbar magnetic resonance image as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor executes the steps of the quantification method of the lumbar magnetic resonance image according to any one of claims 1 to 7.

Images