A method and system for automatically interpreting knee joint lesion characteristics based on medical images

Abstract

The present invention provides a method and system for automatically interpreting the characteristics of knee joint lesions based on medical images, which relates to the field of intelligent medical detection technology. The method includes: obtaining a medical image to be interpreted; filtering the medical image to obtain a filtered image; constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image; performing histogram equalization on the denoised image to obtain an enhanced image; based on deep learning technology, using the enhanced image to perform feature segmentation to obtain lesion characteristics; the lesion characteristics include: joint space, bone hyperplasia and sclerosis under the joint surface, osteophyte formation, bone cystic changes under the joint surface, joint deformity, etc. The present invention realizes the automatic interpretation of knee joint lesions through image processing and machine learning technology, and provides an efficient and accurate diagnostic tool.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical detection technology, and in particular to a method and system for automatically interpreting knee joint lesion characteristics based on medical images. Background Art

[0002] With the development of society and the change of people's lifestyle, the prevalence of knee joint diseases has increased year by year, becoming an important issue affecting people's quality of life and health. Common knee joint diseases include knee joint degeneration, osteoarthritis, rheumatoid arthritis, etc. These diseases cause great trouble to patients' daily activities. Early and accurate diagnosis and treatment of knee joint diseases are important ways to improve patients' prognosis and relieve pain, and medical imaging has become an important means of diagnosing knee joint lesions.

[0003] X-rays and CT images are commonly used clinical examination methods for knee joints, but the diagnostic efficiency and accuracy of these two types of images depend on the experience and professional level of clinicians. Doctors spend a long time interpreting images, and sometimes misjudgment may occur due to subjective factors, especially in the early stages of the lesion or when the imaging features are not obvious. In addition, the characteristics of knee joint lesions vary, including narrowing of the joint space, osteophyte formation, subarticular bone sclerosis, cystic changes, and joint deformities. These characteristics are complex and difficult to identify and analyze comprehensively and consistently through manual interpretation. Relying on manual interpretation can easily lead to missed diagnoses or misdiagnoses.

[0004] With the popularization of medical imaging equipment, a large amount of imaging data generated every year needs to be analyzed and interpreted, which brings tremendous pressure to clinicians. How to efficiently process and analyze these imaging data and improve diagnostic efficiency is an urgent problem to be solved.

[0005] In recent years, with the advancement of computer technology and the development of machine learning algorithms, the use of deep learning technology for medical image analysis has become a research hotspot. Deep learning models, especially convolutional neural networks (CNNs), have made significant progress in image classification, segmentation, and feature extraction. By training large-scale medical imaging data, deep learning models can automatically learn and identify complex lesion features, significantly improving the accuracy and efficiency of diagnosis.

[0006] The automated diagnosis system uses image processing technology and machine learning algorithms to pre-process, segment and extract features from medical images, and ultimately achieve automatic interpretation of lesions. This method can not only reduce the workload of doctors, but also provide objective and consistent diagnostic results, effectively assisting clinical decision-making.

[0007] The invention patent with application number CN202311341693.6 discloses a method for automatic recognition of medical image feature points based on an AI algorithm, including the following steps: 1) Obtaining medical images: Obtaining medical image data of a target object according to preset imaging information; 2) Rough segmentation of the target object: Based on the threshold of the target object, the medical image data of the target object is roughly segmented to obtain the regional information of the target object; 3) Extracting the boundary of the target object: On the basis of rough segmentation, the boundary information of the target object is segmented using an edge operator, and the smoothly segmented medical image data is obtained based on the seed growth algorithm; 4) Extracting point cloud information: Based on the medical image data after smooth segmentation, a three-dimensional model of the target object is created, and the three-dimensional model is converted into point cloud information; 5) Importing a neural network, inputting the point cloud information into a trained neural network model, and predicting the probability density map of the surface feature points of the target object through the accuracy test loss function. However, in the rough segmentation stage of the target object, the segmentation based on the preset threshold may result in inaccurate segmentation results for areas with small or fuzzy boundaries, affecting the subsequent extraction and recognition accuracy of feature points.

[0008] Based on this, it is very necessary to develop an efficient and accurate automatic interpretation technology for knee joint lesion characteristics. Summary of the invention

[0009] In order to overcome the deficiencies of the prior art, the object of the present invention is to provide a method and system for automatically interpreting knee joint lesion characteristics based on medical images.

[0010] To achieve the above object, the present invention provides the following solutions:

[0011] A method for automatically interpreting characteristics of knee joint lesions based on medical images, comprising:

[0012] Obtain medical images to be interpreted;

[0013] Performing filtering on the medical image to obtain a filtered image;

[0014] Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising, to obtain a denoised image;

[0015] Performing histogram equalization processing on the denoised image to obtain an enhanced image;

[0016] Based on deep learning technology, the enhanced image is used for feature segmentation to obtain lesion characteristics; the lesion characteristics include: joint space, bone hyperplasia and sclerosis under the articular surface, osteophyte formation, bone cysts under the articular surface, joint deformities, etc.

[0017] Preferably, the medical images include X-rays and CT images.

[0018] Preferably, filtering the medical image to obtain a filtered image includes:

[0019] Performing wavelet decomposition on the medical image at multiple scales to obtain corresponding wavelet coefficients;

[0020] A filtering threshold is constructed according to the size and decomposition scale of the medical image; the formula of the filtering threshold is: Among them, λ j represents the filtering threshold at the jth decomposition scale, σ j represents the standard deviation of the noise at the jth decomposition scale, N j represents the signal length at the jth decomposition scale, where j represents the decomposition scale;

[0021] constructing an image filtering model using the filtering threshold;

[0022] The medical image is filtered using the image filtering model to obtain the filtered image.

[0023] Preferably, the formula of the image filtering model is: Wherein, sign is the sign function, a is the first preset coefficient, b is the second preset coefficient, ω j,k represents the kth wavelet coefficient at the jth decomposition scale, Represents the wavelet coefficients after filtering.

[0024] Preferably, constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising, to obtain a denoised image, comprises:

[0025] Use the image recognition model to detect the filtered image and obtain the proportion value under each image enhancement window;

[0026] When the proportion value in the image enhancement window is greater than a preset threshold, median denoising is performed on the image in the corresponding image enhancement window to obtain an image after median denoising;

[0027] The image enhancement window is slid until the entire filtered image is traversed to obtain a denoised image.

[0028] Preferably, the formula for median denoising is:

[0029] Where f(a, b) represents the pixel value of pixel (a, b) after median denoising, D is an adjustable parameter, x(a, b) represents the pixel value of pixel (a, b) in the image enhancement window, mean(a, b) represents the pixel median of the image enhancement window, σ x(a, b) represents the variance of pixel values ​​within the image enhancement window.

[0030] Preferably, based on deep learning technology, the enhanced image is used to perform feature segmentation to obtain lesion features, including:

[0031] Get the original YOLOv7tiny model;

[0032] In the ELAN structure in the backbone network of the YOLOv7tiny model, the original CBL ordinary convolution module with a convolution kernel size of 3 is replaced with a deformable convolution DCNv2, and the outputs of different levels are subjected to feature fusion through the Concat module and then introduced into the SimAM attention mechanism to obtain the improved DCNv2-SimAM-ELAN structure;

[0033] The improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the backbone network of the YOLOv7tiny model, and adds a SimAM attention mechanism after the first ELAN structure;

[0034] Replace the loss function from CIoU to WIoU to obtain an improved YOLOv7tiny network recognition model;

[0035] Inputting the enhanced image into an improved YOLOv7tiny network recognition model to obtain a knee joint area;

[0036] Feature extraction is performed on the knee joint region to obtain the pathological changes.

[0037] Preferably, performing feature extraction on the knee joint region to obtain the pathological features includes:

[0038] measuring the width of the joint space in the knee joint region using a morphological processing method;

[0039] The edge detection algorithm is combined with preset morphological features to automatically identify and count the number and location of osteophytes;

[0040] Extracting the hardening features in the knee joint area through gray level co-occurrence matrix and secondary texture analysis;

[0041] Using the density characteristics of the knee joint area in the CT image, the cystic area is automatically identified and segmented;

[0042] The three-dimensional structure of the joint is extracted from the knee joint area of ​​the CT image using three-dimensional reconstruction technology, and the degree of deformity of the joint surface is calculated.

[0043] An automatic interpretation system for knee joint lesion characteristics based on medical images, comprising:

[0044] An image acquisition unit, used for acquiring medical images to be interpreted;

[0045] A preprocessing unit, used for filtering the medical image to obtain a filtered image;

[0046] An image denoising unit is used to construct an image enhancement window, and use the image enhancement window to traverse the filtered image to perform denoising, so as to obtain a denoised image;

[0047] An image enhancement unit is used to perform histogram equalization processing on the denoised image to obtain an enhanced image;

[0048] The segmentation unit is used to perform feature segmentation on the enhanced image based on deep learning technology to obtain lesion features; the lesion features include: joint space, bone hyperplasia and sclerosis under the articular surface, osteophyte formation, bone cysts under the articular surface, joint deformities, etc.

[0049] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0050] The present invention provides a method and system for automatically interpreting the characteristics of knee joint lesions based on medical images, the method comprising: obtaining a medical image to be interpreted; filtering the medical image to obtain a filtered image; constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image; performing histogram equalization on the denoised image to obtain an enhanced image; based on deep learning technology, using the enhanced image to perform feature segmentation to obtain lesion characteristics; the lesion characteristics include: joint space, bone hyperplasia and sclerosis under the joint surface, osteophyte formation, bone cystic changes under the joint surface, joint deformity, etc. The present invention realizes the automatic interpretation of knee joint lesions through image processing and machine learning technology, and provides an efficient and accurate diagnostic tool. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] 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 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 labor.

[0052] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0053] Figure 2 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] 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.

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a method for automatically interpreting knee joint lesion characteristics based on medical images, comprising:

[0057] Step 100: Obtaining a medical image to be interpreted;

[0058] Step 200: Filter the medical image to obtain a filtered image;

[0059] Step 300: constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising, thereby obtaining a denoised image;

[0060] Step 400: performing histogram equalization processing on the denoised image to obtain an enhanced image;

[0061] Step 500: Based on deep learning technology, feature segmentation is performed using the enhanced image to obtain lesion features; the lesion features include: joint space, bone hyperplasia and sclerosis under the articular surface, osteophyte formation, bone cysts under the articular surface, joint deformities, etc.

[0062] Preferably, the medical images include X-rays and CT images.

[0063] Specifically, this embodiment uses an X-ray machine to obtain anteroposterior and lateral X-ray films of the knee joint and saves them in DICOM format. This embodiment also uses a spiral CT scanner to perform CT scan on the knee joint to obtain a three-dimensional CT image and save it in DICOM format.

[0064] Preferably, filtering the medical image to obtain a filtered image includes:

[0065] Performing wavelet decomposition on the medical image at multiple scales to obtain corresponding wavelet coefficients;

[0066] A filtering threshold is constructed according to the size and decomposition scale of the medical image; the formula of the filtering threshold is: Among them, λ j represents the filtering threshold at the jth decomposition scale, σ j represents the standard deviation of the noise at the jth decomposition scale, N j represents the signal length at the jth decomposition scale, where j represents the decomposition scale;

[0067] constructing an image filtering model using the filtering threshold;

[0068] The medical image is filtered using the image filtering model to obtain the filtered image.

[0069] Preferably, the formula of the image filtering model is: Wherein, sign is the sign function, a is the first preset coefficient, b is the second preset coefficient, ω j,k represents the kth wavelet coefficient at the jth decomposition scale, Represents the wavelet coefficients after filtering.

[0070] Specifically, when this embodiment uses wavelet denoising, the value of the wavelet threshold has an important influence on the final denoising effect. The wavelet threshold should be able to separate the wavelet coefficients of the noise and the original image signal to the greatest extent. As the number of wavelet decomposition layers increases, the wavelet coefficient area where the noise is located should gradually decrease. Most of the wavelet coefficients decomposed at a higher level come from the original image signal. Therefore, this embodiment constructs a self-transformation filtering threshold based on the decomposition scale.

[0071] Preferably, constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising, to obtain a denoised image, comprises:

[0072] Use the image recognition model to detect the filtered image and obtain the proportion value under each image enhancement window;

[0073] When the proportion value in the image enhancement window is greater than a preset threshold, median denoising is performed on the image in the corresponding image enhancement window to obtain an image after median denoising;

[0074] The image enhancement window is slid until the entire filtered image is traversed to obtain a denoised image.

[0075] Specifically, the original filtering algorithm, such as the mean filtering algorithm, performs mean processing on the pixel points in each neighborhood (i.e., the denoising window) on the image (regardless of whether there are noise points), so the processed image will be distorted. The present invention can find the characteristic noise points on the image by utilizing the image enhancement window, and then process the corresponding characteristic noise points. It can smooth out the noise points in the image while retaining the original pixel information on the medical image as much as possible.

[0076] Preferably, the formula for median denoising is:

[0077] Where f(a, b) represents the pixel value of pixel (a, b) after median denoising, D is an adjustable parameter, x(a, b) represents the pixel value of pixel (a, b) in the image enhancement window, mean(a, b) represents the pixel median of the image enhancement window, σ x (a, b) represents the variance of pixel values ​​within the image enhancement window.

[0078] Furthermore, this embodiment performs median denoising on the image in the corresponding window based on the image enhancement window, which can alleviate the problem that the existing denoising method (such as mean filter denoising) causes certain features in the image to disappear, so that the original pixel information of the image can be retained to the greatest extent. In addition, the present invention uses histogram equalization to process medical images, which can improve the image interpretation effect.

[0079] Exemplarily, after image denoising, this embodiment also performs registration on the X-ray film and the CT image to ensure accurate alignment between the different images.

[0080] Preferably, based on deep learning technology, the enhanced image is used to perform feature segmentation to obtain lesion features, including:

[0081] Get the original YOLOv7tiny model;

[0082] In the ELAN structure in the backbone network of the YOLOv7tiny model, the original CBL ordinary convolution module with a convolution kernel size of 3 is replaced with a deformable convolution DCNv2, and the outputs of different levels are subjected to feature fusion through the Concat module and then introduced into the SimAM attention mechanism to obtain the improved DCNv2-SimAM-ELAN structure;

[0083] The improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the backbone network of the YOLOv7tiny model, and adds a SimAM attention mechanism after the first ELAN structure;

[0084] Replace the loss function from CIoU to WIoU to obtain an improved YOLOv7tiny network recognition model;

[0085] Inputting the enhanced image into an improved YOLOv7tiny network recognition model to obtain a knee joint area;

[0086] Feature extraction is performed on the knee joint region to obtain the pathological changes.

[0087] Specifically, in this embodiment, the structure of the improved YOLOv7tiny network is mainly the backbone network (Backbone) part of the YOLOv7tiny network, wherein the backbone network in the YOLOv7tiny network has a total of four ELAN structures. Based on the YOLOv7tiny model, this embodiment designs a new ELAN structure. Specifically, in the original ELAN structure in the YOLOv7tiny backbone network, the original CBL ordinary convolution module (Conv3×3) with a convolution kernel size of 3 is replaced with a deformable convolution DCNv2 (Deformable ConvolutionNetwork v2), so as to more effectively capture the morphology and changes of the knee joint area in the enhanced image. After the features output at different levels are fused through the Concat module, the SimAM attention mechanism is introduced to enhance the feature extraction and feature integration capabilities. The improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the original YOLOv7tiny backbone network, and outputs ELAN structures with feature maps of 180×80, 40×40, and 20×20, respectively. The SimAM attention mechanism is added after the first ELAN structure (the ELAN structure with an output feature map of 160×160). At the same time, the loss function of the model is replaced from CIoU (Complete-Intersection over Union) to WIoU (Wise-Intersection over Union), so that the model focuses on high-quality bounding boxes, improves model accuracy and accelerates convergence.

[0088] Preferably, performing feature extraction on the knee joint region to obtain the pathological features includes:

[0089] measuring the width of the joint space in the knee joint region using a morphological processing method;

[0090] The edge detection algorithm is combined with preset morphological features to automatically identify and count the number and location of osteophytes;

[0091] Extracting the hardening features in the knee joint area through gray level co-occurrence matrix and secondary texture analysis;

[0092] Using the density characteristics of the knee joint area in the CT image, the cystic area is automatically identified and segmented;

[0093] The three-dimensional structure of the joint is extracted from the knee joint area of ​​the CT image using three-dimensional reconstruction technology, and the degree of deformity of the joint surface is calculated.

[0094] Specifically, in this embodiment, the morphological processing method is used to measure the width of the joint gap in the knee joint area, which is specifically achieved by performing image binarization in the following manner:

[0095] 1. Binarize the medical image of the region of interest and convert it into a black and white image. This can be done by setting a threshold and setting the pixels above the threshold to white and the pixels below the threshold to black.

[0096] 2. Use morphological processing methods, that is, through expansion, corrosion, opening operation, closing operation, etc., to process the binary image. The specific steps include:

[0097] Using a structural element (such as a rectangle or a circle) to dilate a binary image can fill gaps and increase the size of the region. Using a structural element to erode a binary image can remove small areas in the image and refine the edges. The erosion operation is specifically divided into:

[0098] The operation of first erosion and then expansion eliminates small noise points and retains large connected areas, smoothing the edges;

[0099] The operation of dilation followed by erosion fills small holes and preserves large connected areas to smooth the contour.

[0100] 3. Detecting the width of the joint gap: This embodiment detects width information in the joint gap region based on the image after morphological processing. The width of the joint gap can be estimated by measuring the width of the connected area in the gap region or the distance from edge to edge.

[0101] 4. Calculate and record the measurement results. This embodiment converts pixels to actual lengths based on the actual measured joint gap width and combines the pixel scale information on the image to obtain the actual width value of the joint gap. The measurement results are recorded for subsequent analysis and diagnosis.

[0102] Furthermore, this embodiment uses an edge detection algorithm combined with preset morphological features to automatically identify and count the number and location of osteophytes, specifically including:

[0103] 1. This embodiment uses a classic edge detection algorithm, such as Sobel, Canny or Prewitt algorithm, to perform edge detection on the preprocessed medical image in order to accurately identify the edge information around the osteophyte. Morphological processing methods, such as dilation, corrosion, opening operation, closing operation, etc., are applied to the edge-detected image to remove noise, connect broken edges, smooth edges, etc., to improve the detection accuracy of osteophytes.

[0104] 2. Define the morphological features of osteophytes. This embodiment provides preset morphological features of osteophytes, such as size, shape, etc., to filter out non-osteophyte structures or noise. These features can be defined as parameters for morphological processing to identify areas that meet osteophyte characteristics.

[0105] 3. Osteophyte detection and statistics. This embodiment uses defined morphological features and morphologically processed images to identify and mark each region that meets the osteophyte characteristics to detect the location and number of osteophytes. This embodiment uses methods such as connected region analysis and region growing algorithm to identify and count the number and location information of osteophytes in medical images.

[0106] 4. Measure and record osteophyte information. This embodiment calculates the size, location and other characteristics of the detected osteophytes and records the results for subsequent analysis and diagnosis.

[0107] Furthermore, the steps of extracting hardening features in this embodiment are as follows:

[0108] 1. Calculate the grayscale co-occurrence matrix of the extracted knee joint area. The grayscale co-occurrence matrix is ​​a matrix used for image grayscale distribution and texture features, including statistical information on the grayscale value versus frequency of occurrence. This embodiment calculates various features of the grayscale co-occurrence matrix, such as energy, contrast, inertia, correlation, etc., according to different grayscale levels and directions to describe the texture features of the knee joint area.

[0109] 2. Perform secondary texture analysis based on the calculated gray-level co-occurrence matrix. Secondary texture analysis is to obtain richer texture feature information by performing mathematical operations on the gray-level co-occurrence matrix. Secondary texture feature extraction methods, such as entropy, moment of inertia, and distance measurement of the gray-level co-occurrence matrix, are used to further describe the texture features of the knee joint area.

[0110] Based on the results of gray-level co-occurrence matrix and secondary texture analysis, texture features related to hardening characteristics in the knee joint area are extracted. These features can include uniformity of grayscale distribution, roughness of texture, complexity of structure, etc., which are used to describe the hardening situation in the knee joint.

[0111] 3. Analyze and record the results. This embodiment analyzes and quantifies the extracted sclerosis characteristics so that doctors can make further diagnosis and treatment decisions. The extracted sclerosis characteristic information is recorded for subsequent research and analysis.

[0112] Furthermore, the specific steps of automatically identifying and segmenting the cystic region by using the density features of the knee joint region in the CT image are as follows:

[0113] 1. Calculate density features in the knee joint area. Different tissues and structures in CT images have different density values, and density features are described by statistical information of pixel gray values. For example, this embodiment calculates density feature parameters such as average density, standard deviation of density distribution, and density histogram to distinguish cystic areas from surrounding normal tissues.

[0114] 2. According to the calculated density feature parameters, the cystic region is automatically identified by using threshold segmentation or machine learning algorithm. In this embodiment, the cystic region is segmented from the surrounding tissues according to the difference in density features.

[0115] 3. Perform region growing or edge detection on the identified cystic region to further refine and accurately locate the edge of the cystic region.

[0116] 4. Verify and optimize the cystic degeneration area obtained by automatic identification and segmentation. In this embodiment, the accuracy of the segmentation result is verified by manual annotation, and adjustment and optimization are performed as needed.

[0117] 5. Analyze and record the size, shape, location and other information of the identified cystic area. The information of the cystic area can provide a reference for doctors for further diagnosis and treatment plan formulation.

[0118] Finally, this embodiment uses three-dimensional reconstruction technology to extract the three-dimensional structure of the joint from the knee joint area of ​​the CT image, and the specific steps of calculating the degree of deformity of the joint surface are as follows:

[0119] 1. Through voxelization and surface reconstruction technology, the CT images of the knee joint area are reconstructed in three dimensions, and the two-dimensional image data is converted into a three-dimensional model, thereby obtaining a three-dimensional structural model of the knee joint.

[0120] 2. Based on the three-dimensional reconstruction, the present embodiment performs edge detection and surface connection operations to obtain the accurate boundary of the joint surface.

[0121] 3. Calculate the degree of deformity of the joint surface through 3D model analysis software or programming tools. The specific steps are as follows:

[0122] The curvature information of the joint surface, including the principal curvature, mean curvature, etc., is calculated to evaluate the curvature of the joint surface.

[0123] Analyze the topology of the joint surface to detect any uneven or shaped parts.

[0124] The calculated joint surface is compared and matched with the standard structure to evaluate the difference between the two and reflect the degree of joint deformity.

[0125] Measuring angles and distances between specific points on the joint surface is used to quantify the degree of deformity.

[0126] 4. Visualize the calculated joint deformity results. This embodiment can generate a color map of the three-dimensional model, mark the deformity area, etc. Write a report to describe the deformity of the joint surface and provide quantitative evaluation results to provide reference information for doctors.

[0127] Optionally, after extracting the lesion features, this embodiment will conduct a comprehensive analysis of the extracted lesion features, and use machine learning methods such as support vector machine (SVM) or random forest for classification and interpretation. And based on the analysis results, a knee joint lesion feature report is automatically generated, including a detailed description of the lesion features and the interpretation results.

[0128] Corresponding to the above method, such as Figure 2 As shown, the present invention also provides a system for automatically interpreting knee joint lesion characteristics based on medical images, comprising:

[0129] An image acquisition unit, used for acquiring medical images to be interpreted;

[0130] A preprocessing unit, used for filtering the medical image to obtain a filtered image;

[0131] An image denoising unit is used to construct an image enhancement window, and use the image enhancement window to traverse the filtered image to perform denoising, so as to obtain a denoised image;

[0132] An image enhancement unit is used to perform histogram equalization processing on the denoised image to obtain an enhanced image;

[0133] The segmentation unit is used to perform feature segmentation using the enhanced image based on deep learning technology to obtain pathological features; the pathological features include: joint space, bone hyperplasia and sclerosis under the joint surface, osteophyte formation, bone cystic changes under the joint surface, joint deformity, etc. The beneficial effects of the present invention are as follows:

[0134] The present invention uses X-rays and CT images, and through image processing and machine learning algorithms, realizes the automatic recognition and interpretation of pathological features (including joint space, osteophytes, subarticular bone sclerosis, cystic changes, and joint deformities), thereby improving diagnostic efficiency and reducing misdiagnosis and missed diagnosis. The development of this method helps promote the development of knee joint disease diagnosis technology and improve the treatment effect and quality of life of patients.

[0135] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0136] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for automatically interpreting knee joint lesion characteristics based on medical images, characterized in that: include: Obtain medical images to be interpreted; Performing filtering on the medical image to obtain a filtered image; Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising, to obtain a denoised image; Performing histogram equalization processing on the denoised image to obtain an enhanced image; Based on deep learning technology, the enhanced image is used for feature segmentation to obtain lesion features; the lesion features include: joint space, bone hyperplasia and sclerosis under the articular surface, osteophyte formation, bone cystic changes under the articular surface, and joint deformity; Based on deep learning technology, the enhanced image is used for feature segmentation to obtain lesion features, including: Get the original YOLOv7tiny model; In the ELAN structure in the backbone network of the YOLOv7tiny model, the original CBL ordinary convolution module with a convolution kernel size of 3 is replaced with a deformable convolution DCNv2, and the outputs of different levels are subjected to feature fusion through the Concat module and then introduced into the SimAM attention mechanism to obtain the improved DCNv2-SimAM-ELAN structure; The improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the backbone network of the YOLOv7tiny model, and adds a SimAM attention mechanism after the first ELAN structure; Replace the loss function from CIoU to WIoU to obtain an improved YOLOv7tiny network recognition model; Inputting the enhanced image into an improved YOLOv7tiny network recognition model to obtain a knee joint area; Extracting features from the knee joint region to obtain the pathological changes; Extracting features from the knee joint region to obtain the pathological features includes: measuring the width of the joint space in the knee joint region using a morphological processing method; The edge detection algorithm is combined with preset morphological features to automatically identify and count the number and location of osteophytes; Extracting the hardening features in the knee joint area through gray level co-occurrence matrix and secondary texture analysis; Using the density characteristics of the knee joint area in the CT image, the cystic area is automatically identified and segmented; The three-dimensional structure of the joint is extracted from the knee joint area of ​​the CT image using three-dimensional reconstruction technology, and the degree of deformity of the joint surface is calculated.

2. The method for automatically interpreting knee joint lesion characteristics based on medical images according to claim 1, characterized in that: The medical images include X-rays and CT images.

3. The method for automatically interpreting knee joint lesion characteristics based on medical images according to claim 1, characterized in that: Performing filtering on the medical image to obtain a filtered image includes: Performing wavelet decomposition on the medical image at multiple scales to obtain corresponding wavelet coefficients; A filtering threshold is constructed according to the size and decomposition scale of the medical image; the formula of the filtering threshold is: Among them, λ j represents the filtering threshold at the jth decomposition scale, σ j represents the standard deviation of the noise at the jth decomposition scale, N j represents the signal length at the jth decomposition scale, where j represents the decomposition scale; constructing an image filtering model using the filtering threshold; The medical image is filtered using the image filtering model to obtain the filtered image.

4. The method for automatically interpreting knee joint lesion characteristics based on medical images according to claim 3, characterized in that: The formula of the image filtering model is: Wherein, sign is the sign function, a is the first preset coefficient, b is the second preset coefficient, ω j,k represents the kth wavelet coefficient at the jth decomposition scale, Represents the wavelet coefficients after filtering.

5. The method for automatically interpreting knee joint lesion characteristics based on medical images according to claim 1, characterized in that: Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising, to obtain a denoised image, including: Use the image recognition model to detect the filtered image and obtain the proportion value under each image enhancement window; When the proportion value in the image enhancement window is greater than a preset threshold, median denoising is performed on the image in the corresponding image enhancement window to obtain an image after median denoising; The image enhancement window is slid until the entire filtered image is traversed to obtain a denoised image.

6. The method for automatically interpreting knee joint lesion characteristics based on medical images according to claim 5, characterized in that: The formula for median denoising is: Where f(a, b) represents the pixel value of pixel (a, b) after median denoising, D is an adjustable parameter, x(a, b) represents the pixel value of pixel (a, b) in the image enhancement window, mean(a, b) represents the pixel median of the image enhancement window, σ x (a, b) represents the variance of pixel values ​​within the image enhancement window.

7. An automatic interpretation system for knee joint lesion characteristics based on medical images, characterized in that: include: An image acquisition unit, used for acquiring medical images to be interpreted; A preprocessing unit, used for filtering the medical image to obtain a filtered image; An image denoising unit is used to construct an image enhancement window, and use the image enhancement window to traverse the filtered image to perform denoising, so as to obtain a denoised image; An image enhancement unit is used to perform histogram equalization processing on the denoised image to obtain an enhanced image; A segmentation unit is used to perform feature segmentation using the enhanced image based on deep learning technology to obtain pathological features; the pathological features include: joint space, bone hyperplasia and sclerosis under the joint surface, osteophyte formation, bone cystic changes under the joint surface, and joint deformity; Based on deep learning technology, the enhanced image is used for feature segmentation to obtain lesion features, including: Get the original YOLOv7tiny model; In the ELAN structure in the backbone network of the YOLOv7tiny model, the original CBL ordinary convolution module with a convolution kernel size of 3 is replaced with a deformable convolution DCNv2, and the outputs of different levels are subjected to feature fusion through the Concat module and then introduced into the SimAM attention mechanism to obtain the improved DCNv2-SimAM-ELAN structure; The improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the backbone network of the YOLOv7tiny model, and adds a SimAM attention mechanism after the first ELAN structure; Replace the loss function from CIoU to WIoU to obtain an improved YOLOv7tiny network recognition model; Inputting the enhanced image into an improved YOLOv7tiny network recognition model to obtain a knee joint area; Extracting features from the knee joint region to obtain the pathological changes; Extracting features from the knee joint region to obtain the pathological features includes: measuring the width of the joint space in the knee joint region using a morphological processing method; The edge detection algorithm is combined with preset morphological features to automatically identify and count the number and location of osteophytes; Extracting the hardening features in the knee joint area through gray level co-occurrence matrix and secondary texture analysis; Using the density characteristics of the knee joint area in the CT image, the cystic area is automatically identified and segmented; The three-dimensional structure of the joint is extracted from the knee joint area of ​​the CT image using three-dimensional reconstruction technology, and the degree of deformity of the joint surface is calculated.

Images