A method for extracting trabecular bone features based on image analysis

Abstract

The present invention relates to the technical field of medical image processing, and discloses a method for extracting trabecular bone features based on image analysis, including: scanning bone tissue using an X-ray imaging device to obtain a two-dimensional projection image of the bone tissue; preprocessing the obtained two-dimensional projection image; performing segmentation processing on the image through an edge detection segmentation method to separate the trabecular bone from the image and obtain a trabecular bone region; constructing a deep learning model to perform morphological processing on the segmented trabecular bone region to refine the boundary of the trabecular bone region; performing multiple iterative feature extractions on the segmented trabecular bone region and adjusting the feature extraction parameters according to the results of the previous iteration; performing correlation analysis on the extracted trabecular bone features and displaying the distribution and change trend of the features through a data visualization method. The method of the present invention can comprehensively and accurately extract trabecular bone features, providing technical support for the diagnosis and research of orthopedic diseases such as osteoporosis.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and particularly to a method for extracting trabecular bone features based on image analysis. Background Art

[0002] The structural features of trabecular bone are of great significance for diagnosing orthopedic diseases such as osteoporosis. Accurately extracting the features of trabecular bone can provide an important basis for the early diagnosis and treatment of diseases. However, the existing methods for extracting trabecular bone features have certain limitations in processing complex images and extracting comprehensive features, and need to be further optimized to improve the accuracy and efficiency of feature extraction.

[0003] The existing methods for extracting trabecular bone features mainly rely on traditional image processing techniques. When these methods segment the trabecular bone region and extract features, they are often affected by image noise and background interference, resulting in low accuracy of feature extraction. In addition, when dealing with trabecular bone with different densities and complex structures, the existing methods are difficult to maintain a consistent segmentation effect, which affects the stability of feature extraction.

[0004] In trabecular bone image processing, image segmentation is a key step in feature extraction. When the existing segmentation methods process images with low contrast between trabecular bone and background, missegmentation easily occurs, resulting in a decrease in the accuracy of feature extraction. In addition, when dealing with a large amount of image data, the existing methods have low computational efficiency and are difficult to meet the needs of clinical rapid diagnosis.

[0005] In recent years, deep learning technology has made remarkable progress in the field of image processing, especially in image segmentation and feature extraction. However, when the existing deep learning models process trabecular bone images, there are still some problems, such as the model training requires a large amount of labeled data and is prone to overfitting. In addition, the existing models have deficiencies in processing multi-scale features and are difficult to capture the complex structural information of trabecular bone.

[0006] Therefore, the present invention proposes a method for extracting trabecular bone features based on image analysis, which has an improved deep learning model and can better adapt to the characteristics of trabecular bone images, improving the accuracy and robustness of feature extraction. Summary of the Invention

[0007] The purpose of the present invention is to propose a method for extracting trabecular bone features based on image analysis to solve the problems in the prior art.

[0008] To achieve the above purpose, the present invention adopts the following technical solution: A method for extracting trabecular bone features based on image analysis, comprising the following steps:

[0009] Step S1, using an X-ray imaging device to scan bone tissue to obtain a two-dimensional projection image of the bone tissue;

[0010] Step S2, preprocess the obtained two-dimensional projection image of bone tissue, including denoising, contrast enhancement, histogram equalization, and normalization processing;

[0011] Step S3, perform segmentation processing on the image through an edge detection segmentation method to separate the trabecular bone from the image and obtain the trabecular bone region;

[0012] Step S4, construct a deep learning model to perform morphological processing on the segmented trabecular bone region and refine the boundary of the trabecular bone region;

[0013] Step S5, perform multiple iterative feature extractions on the segmented trabecular bone region and adjust the feature extraction parameters according to the results of the previous iteration;

[0014] Step S6, perform correlation analysis on the extracted trabecular bone features and display the distribution and change trend of the features through a data visualization method;

[0015] Among them, in Step S5, the following sub-steps are also included:

[0016] S5-1, perform the first iterative feature extraction, and extract the morphological features of the trabecular bone from the segmented image. The morphological features include trabecular bone thickness, trabecular bone number, and trabecular bone separation degree. The specific formulas are as follows:

[0017] ;

[0018] ;

[0019] ;

[0020] Among them, TbTh represents trabecular bone thickness, TbN represents trabecular bone number, TbSp represents trabecular bone separation degree, is the area of the trabecular bone region, is the total length of the trabecular bone, is the area of the entire image, is the area of the background region;

[0021] S5-2, perform the first iterative feature extraction, and extract the structural features of the trabecular bone from the segmented image. The structural features include trabecular bone connection density and anisotropy degree. The specific formulas are as follows:

[0022] ;

[0023] ;

[0024] Among them, Xv represents the Euler characteristic of the trabecular bone structure, which is used to describe the connection density of the trabecular bone structure, and DA represents the degree of anisotropy. is the 0th Betti number, representing the number of connected regions. is the second-order Betti number, representing the number of holes. and are both eigenvalues of the inertia tensor of the trabecular bone structure. is the maximum eigenvalue, representing the distribution of trabecular bone in the main direction. is the second largest eigenvalue, representing the distribution of trabecular bone in the secondary direction.

[0025] S5-3. Perform the first iteration of feature extraction to extract the texture features of trabecular bone from the segmented image. The texture features include wavelet transform features and trabecular bone score. The specific formulas are as follows:

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] Among them, is the wavelet coefficient. represents a two-dimensional image. is the value of the wavelet basis function at position (j, k). E is the wavelet energy, and H is the wavelet entropy. is the wavelet coefficient 's probability distribution. TBS represents the trabecular bone score, and V(e) is the experimental variogram, representing the gray-level change of the distance e function.

[0031] S5-4. According to the feature results extracted in the first iteration, adjust the feature extraction parameters.

[0032] S5-5. Perform the second iteration of feature extraction, repeat steps S5-1 to S5-3, and extract the trabecular bone features after adjusting the feature extraction parameters.

[0033] S5-6. Repeat steps S5-4 and S5-5 to perform multiple iterations of feature extraction until the standard deviation of the extraction results of the same feature in multiple iterations is less than the preset value, and the preset value is 0.005.

[0034] Furthermore, in step S1, the following sub-steps are also included:

[0035] S1-1. Set the scanning parameters of the X-ray imaging device and calibrate the scanning parameters using a standard calibration sample. The scanning parameters include voltage, current, and exposure time.

[0036] S1-2. Start the X-ray imaging device, scan the bone tissue, and obtain a two-dimensional projection image of the bone tissue.

[0037] S1-3. Store the scanned image data in DICOM format on a local server and back up the stored image data, which is stored on an external storage device.

[0038] Further, in step S2, the following sub-steps are also included:

[0039] S2-1. Set the parameters of the Gaussian filter and use convolution operation through the Gaussian filter to perform weighted averaging of each pixel in the two-dimensional projection image of the bone tissue with the pixels in its neighborhood to reduce the noise in the image. The parameters include standard deviation and filter window size.

[0040] S2-2. Adjust the gray value range of the denoised image through an adaptive contrast enhancement method to increase the contrast of the image and highlight the structural features of the trabeculae.

[0041] S2-3. Perform histogram equalization on the enhanced image to make the histogram of the image evenly distributed by adjusting the gray value distribution of the image.

[0042] S2-4. Normalize the image after histogram equalization and adjust the gray value range of the image to [0,1].

[0043] Further, in step S3, the following sub-steps are also included:

[0044] S3-1. Calculate the image gradient of the preprocessed image through an edge detection algorithm, identify the boundaries of the trabeculae, and generate an edge map. The edge detection algorithm includes the Canny algorithm and the Sobel algorithm.

[0045] S3-2. Perform binarization on the image after edge detection, and set the threshold through an adaptive threshold method to convert the edge map into a binary image to distinguish the trabecular region from the background region.

[0046] S3-3. Traverse each pixel in the binarized image through depth-first search to identify the connected trabecular regions.

[0047] S3-4. Mark the identified connected regions through a marking algorithm and assign a unique identifier.

[0048] S3-5. Calculate the area of each connected region, remove isolated noise points and small regions with an area smaller than a preset threshold, and separate the connected trabecular bone regions.

[0049] Further, in step S4, the following sub-steps are further included:

[0050] S4-1: Design the encoder path of a deep learning model based on the improved DeepLab v3+ neural network structure to extract the context information of the image. The encoder path includes a convolutional layer and an atrous convolutional layer.

[0051] S4-2: Design the decoder path of a deep learning model based on the improved DeepLab v3+ neural network structure to restore the image information. The decoder path includes an upsampling layer and a convolutional layer.

[0052] S4-3: Introduce a skip connection between the encoder and the decoder, and splice the feature maps in the encoder path with the feature maps in the decoder path.

[0053] S4-4: Introduce a spatial pyramid pooling ASPP module to capture feature information at different scales.

[0054] S4-5. Perform model training, use the training dataset to train the model until the Dice coefficient reaches a preset value, and the preset value is 0.9.

[0055] S4-6. Use the trained deep learning model to perform morphological processing on the segmented trabecular bone regions to refine the boundaries of the trabecular bone regions. The morphological processing includes dilation and erosion.

[0056] Further, in step S6, the following sub-steps are further included:

[0057] S6-1. Perform normalization processing on the extracted morphological features, structural features, and texture features of the trabecular bone, and scale the feature values to the range of [0,1].

[0058] S6-2. Use the Pearson correlation coefficient to calculate the correlation between the morphological features, structural features, and texture features of the trabecular bone, and evaluate the relationship between the morphological features, structural features, and texture features of the trabecular bone according to the value of the correlation coefficient.

[0059] S6-3. Display the morphological features, structural features, and texture features through data visualization methods. The data visualization methods include heat maps, scatter plots, histograms, line charts, bar charts, and box plots.

[0060] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0061] The present invention can more accurately identify and separate the trabecular bone region by using an X-ray imaging device to obtain high-resolution two-dimensional projection images and combining an improved deep learning model for image segmentation and feature extraction. Through multiple iterations of feature extraction and parameter adjustment, the stability and accuracy of the feature extraction results are ensured, thus providing a more reliable basis for the diagnosis of orthopedic diseases such as osteoporosis.

[0062] During the feature extraction process of the present invention, not only the morphological features of trabecular bone are concerned, but also the structural features and texture features are extracted, which can more comprehensively evaluate the health status of trabecular bone. By performing morphological processing on the segmented trabecular bone region through a deep learning model, the boundary of the trabecular bone region is refined, further improving the accuracy of feature extraction.

[0063] The present invention can more accurately separate the trabecular bone region and effectively identify the boundary of trabecular bone by introducing an edge detection algorithm and a deep learning model. Through depth-first search and labeling algorithms, connected trabecular bone regions can be accurately identified and labeled, thereby improving the accuracy of image segmentation.

[0064] The present invention significantly improves the computational efficiency of feature extraction by optimizing the image processing process and introducing an efficient deep learning model. Combining adaptive contrast enhancement and histogram equalization methods can quickly improve the image quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0066] Figure 1 It is a flowchart of the method provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0067] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in conjunction with the accompanying drawings and preferred embodiments, describe in detail a method for extracting trabecular bone features based on image analysis according to the present invention, its specific implementation manner, structure, features and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art belonging to the technical field of the present invention.

[0069] The following embodiments are for illustrative purposes only and are not intended to limit the scope of the present invention.

[0070] The following specifically describes the specific solution of a trabecular bone feature extraction method provided by the present invention in conjunction with the accompanying drawings.

[0071] Please refer to Figure 1 , which shows the method flow chart of a trabecular bone feature extraction method provided by an embodiment of the present invention. The method includes the following steps:

[0072] The use process of the standard calibration sample includes placing the standard calibration sample in the scanning device, scanning according to the set scanning parameters, and adjusting the scanning parameters by analyzing the scanning results of the calibration sample to ensure that the performance of the device reaches the best state.

[0073] The use process of the X-ray imaging device includes

[0074] 1. Device startup: Start the X-ray imaging device according to the device operation manual to ensure that the device is in a normal working state;

[0075] 2. Place the sample: Place the bone tissue sample to be scanned in the scanning area of the scanning device to ensure that the position and orientation of the sample meet the scanning requirements;

[0076] 3. Start scanning: Start the scanning program, and the device scans the bone tissue according to the preset parameters (voltage, current, and exposure time) to obtain a two-dimensional projection image;

[0077] 4. Real-time monitoring: During the scanning process, monitor the image quality in real time to ensure that there is no obvious artifact or noise interference;

[0078] Storage format of image data: The scanned image data is stored in DICOM format. DICOM format is the standard format for medical image storage and transmission, with wide compatibility and scalability. Using DICOM format to store image data can ensure the integrity and accuracy of image information;

[0079] Storage of image data: Store the scanned image data on the local server. The local server can provide fast data access and processing capabilities, facilitating researchers and clinicians to view and analyze the image data in a timely manner;

[0080] Backup of image data: Back up the image data stored on the local server. The backup is stored on an external storage device (external hard drive, network attached storage NAS). The backup operation can ensure the security and reliability of the data and prevent data loss caused by device failure or human operation errors.

[0081] Step S2: Preprocess the obtained two-dimensional projection image of the bone tissue, including denoising, contrast enhancement, histogram equalization, and normalization processing;

[0082] Among them, in step S2, the following sub-steps are further included:

[0083] S2-1: Set the parameters of the Gaussian filter. Use the Gaussian filter to perform convolution operation to weight-average each pixel in the two-dimensional projection image of the bone tissue with the pixels in its neighborhood, reducing the noise in the image. The parameters include the standard deviation and the filter window size;

[0084] S2-2: Adjust the gray value range of the denoised image through the adaptive contrast enhancement method, increasing the contrast of the image and highlighting the structural features of the trabeculae;

[0085] S2-3: Perform histogram equalization on the enhanced image. By adjusting the gray value distribution of the image, make the histogram of the image evenly distributed;

[0086] S2-4: Perform normalization processing on the image after histogram equalization, adjusting the gray value range of the image to [0, 1].

[0087] It should be noted that the Gaussian filter is a commonly used image smoothing technique for reducing noise in the image. It achieves the denoising effect by weight-averaging each pixel with the pixels in its neighborhood. The weights of the Gaussian filter are determined by the Gaussian function, with the weight of the central pixel being the largest and the weights of the neighborhood pixels decreasing with the increase of the distance.

[0088] The standard deviation determines the smoothness of the Gaussian filter. A larger standard deviation will result in a stronger smoothing effect but will blur the details in the image. A smaller standard deviation retains more details but has a relatively weaker denoising effect.

[0089] The filter window size determines the number of pixels participating in the weighted average within the neighborhood. A larger window can better remove noise but has a higher computational cost. A smaller window has a higher computational efficiency but the denoising effect may not be ideal. The filter window size is an odd number.

[0090] The adaptive contrast enhancement method is a technique for dynamically adjusting the contrast of an image. It can automatically adjust the gray value range according to the local features of the image, thereby enhancing the contrast of the image and highlighting the structural features of interest; in the present invention, the adaptive contrast enhancement method is used to adjust the gray value range of the two-dimensional projection image of the bone tissue after denoising to increase the contrast of the image and highlight the structural features of the trabeculae.

[0091] Histogram equalization is a commonly used image enhancement technique. By adjusting the gray value distribution of the image to make the histogram of the image evenly distributed, the contrast of the image can be further improved, making the details in the image clearer.

[0092] Normalization adjusts the gray value range of the image to [0, 1]. The unified gray value range can reduce the differences between different images and improve the robustness of the algorithm. The smaller gray value range after normalization can improve the efficiency of subsequent calculations. Especially when involving deep learning models, normalization is a necessary preprocessing step.

[0093] Step S3, perform segmentation processing on the image through an edge detection segmentation method to separate the trabecular bone from the image and obtain the trabecular bone region.

[0094] Among them, in step S3, the following sub-steps are also included:

[0095] S3-1, calculate the gradient of the preprocessed image through an edge detection algorithm, identify the boundaries of the trabecular bone, and generate an edge map. The edge detection algorithms include the Canny algorithm and the Sobel algorithm.

[0096] S3-2, perform binarization processing on the image after edge detection, and set the threshold through an adaptive threshold method to convert the edge map into a binary image to distinguish the trabecular bone region from the background region.

[0097] S3-3, traverse each pixel in the binarized image through depth-first search to identify the connected trabecular bone regions.

[0098] Step S1, use an X-ray imaging device to scan the bone tissue to obtain a two-dimensional projection image of the bone tissue.

[0099] Among them, in step S1, the following sub-steps are also included:

[0100] S1-1, set the scanning parameters of the X-ray imaging device and calibrate the scanning parameters using a standard calibration sample. The scanning parameters include voltage, current, and exposure time.

[0101] S1-2, start the X-ray imaging device, scan the bone tissue, and obtain a two-dimensional projection image of the bone tissue.

[0102] S1-3, store the scanned image data in DICOM format on the local server and back up the stored image data, and the backup is stored on an external storage device.

[0103] It should be noted that voltage affects the penetration ability of X-rays. A higher voltage can increase the penetration ability of X-rays and is applicable to thicker or denser bone tissues. However, an excessively high voltage may lead to a decrease in image contrast. The voltage range is between 50 - 150 kV, and the specific value needs to be adjusted according to the density and thickness of the bone tissue.

[0104] Current determines the intensity of X-rays. A higher current can improve the signal-to-noise ratio of the image, but it will also increase the radiation dose. On the premise of ensuring image quality, a lower current value should be selected as much as possible. The current range is between 10 - 100 mA.

[0105] Exposure time affects the brightness and contrast of the image. A longer exposure time can increase the brightness of the image, but it will also increase noise. The appropriate exposure time needs to be selected according to the density of the bone tissue and the required image quality, usually between 10 - 100 ms.

[0106] Calibration sample: Use a standard calibration sample for calibration. The calibration sample has known density and structural characteristics. Through calibration, the scanning parameters can be adjusted to ensure that the performance of the device reaches the best state.

[0107] S3-4, Mark the identified connected regions through a marking algorithm and assign unique identifiers;

[0108] S3-5, Calculate the area of each connected region, remove isolated noise points and small regions with an area smaller than the preset threshold, and separate the connected trabecular bone regions.

[0109] It should be noted that edge detection is an important step in image processing, which is used to identify regions with significant changes in gray values in the image. These regions correspond to the boundaries of objects. In trabecular bone images, edge detection can help identify the boundaries between trabecular bone and the background.

[0110] Edge detection algorithms identify boundaries by calculating the gradient of the image and generate an edge map. The edge map is a binary image, where edge pixels are marked as 1 and non-edge pixels are marked as 0.

[0111] The Canny algorithm is a classic edge detection algorithm that detects edges through multiple steps, including Gaussian filtering for denoising, calculating the gradient magnitude and direction, non-maximum suppression, and double-threshold detection. It can detect clear edges and has a good noise suppression effect.

[0112] The Sobel algorithm is an edge detection algorithm based on gradient calculation that detects edges by calculating the gradients of the image in the horizontal and vertical directions; the Sobel operator is sensitive to horizontal and vertical edges in the image, is simple to calculate, and is suitable for fast edge detection.

[0113] Binarization processing is to convert an image into a binary image, where the pixel values are only 0 and 1, which can simplify the image processing process and facilitate subsequent connected region recognition and marking.

[0114] The adaptive threshold method can dynamically set the threshold according to the local features of the image instead of using a global fixed threshold, which can better adapt to the illumination changes and noise effects in the image.

[0115] Distinguish the trabecular bone area from the background area: By setting a threshold, the trabecular bone area can be distinguished from the background area.

[0116] Depth-First Search (DFS) is a traversal algorithm used to identify connected regions in a binary image. In a trabecular bone image, DFS can identify the connected regions of trabecular bone, providing a basis for subsequent marking and analysis.

[0117] Depth-First Search includes the following steps:

[0118] 1. Initialization: Start from the top-left pixel of the binarized image and traverse the image pixel by pixel;

[0119] 2. Search process: For each unvisited pixel, if its value is 1 (i.e., it belongs to the trabecular bone area), start a depth-first search from this pixel and mark all connected pixels;

[0120] 3. Recursive search: For the current pixel, check its 8 neighboring pixels (up, down, left, right, upper-left, lower-left, upper-right, lower-right). If the neighboring pixel value is 1 and it has not been visited, add this neighboring pixel to the search queue and continue to search its neighboring pixels until all connected pixels have been visited;

[0121] The marking algorithm is an algorithm used to identify and mark connected regions in an image. By assigning a unique identifier (label) to each connected region, the pixels in the image are divided into different regions.

[0122] The marking algorithm includes the following steps:

[0123] 1. Marking process: During the depth-first search process, assign a unique identifier to each newly discovered connected region and mark all pixels within the connected region with this identifier;

[0124] 2. Marking matrix: Create a marking matrix with the same size as the binarized image to store the identifier of the connected region to which each pixel belongs;

[0125] Area calculation: Calculate the area of each connected region, that is, the number of pixels within the region, which can be achieved by counting the number of times each identifier appears in the marking matrix;

[0126] Noise removal: Isolated noise points and small regions with an area smaller than a preset threshold are removed. These regions may be generated due to noise or inaccurate image segmentation. The preset threshold can be determined according to the minimum size of the trabecular bone.

[0127] Step S4: Construct a deep learning model to perform morphological processing on the segmented trabecular bone region and refine the boundary of the trabecular bone region.

[0128] Among them, in step S4, the following sub-steps are further included:

[0129] S4-1: Design the encoder path of the deep learning model based on the improved DeepLab v3+ neural network structure to extract the context information of the image. The encoder path includes a convolutional layer and an atrous convolutional layer.

[0130] S4-2: Design the decoder path of the deep learning model based on the improved DeepLab v3+ neural network structure to restore the image information. The decoder path includes an upsampling layer and a convolutional layer.

[0131] S4-3: Introduce a skip connection between the encoder and the decoder to splice the feature maps in the encoder path and the feature maps in the decoder path.

[0132] S4-4: Introduce the Atrous Spatial Pyramid Pooling (ASPP) module to capture feature information at different scales.

[0133] S4-5: Perform model training. Use the training dataset to train the model until the Dice coefficient reaches a preset value, and the preset value is 0.9.

[0134] S4-6: Use the trained deep learning model to perform morphological processing on the segmented trabecular bone region and refine the boundary of the trabecular bone region. The morphological processing includes dilation and erosion.

[0135] It should be noted that extracting context information: The main role of the encoder path is to extract rich context information from the input image. Through multiple convolutional operations, the model can learn the local and global features of the image, providing a basis for subsequent segmentation tasks.

[0136] Convolutional layer: By sliding the convolutional kernel on the input feature map, local features are extracted. The convolutional operation can capture edge and texture information in the image.

[0137] Atrous convolutional layer: Atrous convolution is a special convolutional operation. By introducing holes (i.e., skipping some pixels) in the convolutional kernel, the receptive field of the convolutional kernel is enlarged, so as to capture a larger range of context information without increasing the computational amount. Atrous convolution is particularly suitable for tasks that require a large receptive field.

[0138] Restoring Image Information: The main function of the decoder path is to gradually restore the high-level features extracted by the encoder path to the resolution of the original image. Through upsampling operations, the model can generate a segmentation result with the same resolution as the input image;

[0139] Upsampling Layers: Used to increase the spatial resolution of the feature maps. Upsampling methods include nearest neighbor interpolation, bilinear interpolation, and transposed convolution;

[0140] Convolution Layers: In the decoder path, convolution layers are used to refine the upsampled feature maps, removing the artifacts that may be introduced during the upsampling process and making the segmentation result more accurate.

[0141] Skip connections are a method of directly transferring low-level features in the encoder path to the decoder path. Through skip connections, the model can retain more detailed information, thus better restoring details when restoring the image resolution.

[0142] Skip connections can combine the detailed features in the encoder path with the high-level features in the decoder path, thereby improving the segmentation accuracy, especially for the segmentation of small objects and complex structures.

[0143] Atrous Spatial Pyramid Pooling (ASPP) module is a structure for capturing multi-scale features. Through parallel atrous convolution layers and pooling layers, it extracts feature information at different scales. The ASPP module can effectively process objects at different scales and improve the robustness of the model.

[0144] During the training process, the model optimizes the parameters by minimizing the difference between the predicted segmentation result and the ground truth annotation (using the Dice loss) and is trained using the Adam optimization algorithm.

[0145] The training dataset is the two-dimensional projection images of the labeled trabecular bone regions, which are used to provide the dataset for training the deep learning model to ensure that the model can learn the features of the trabecular bone regions.

[0146] The Dice coefficient is a commonly used metric for measuring segmentation accuracy, which calculates the similarity between the predicted segmentation result and the ground truth annotation.

[0147] Morphological processing is a processing method based on the shape of the image. Through dilation and erosion operations, the boundaries of the trabecular bone regions can be refined, making the segmentation result more accurate.

[0148] Erosion operation is an image processing method based on a structuring element, which is used to remove small objects and boundary pixels in the image. By sliding the structuring element over the image and taking the minimum value of all pixel values covered by the structuring element, the erosion effect is achieved.

[0149] The dilation operation is the opposite of the erosion operation and is used to fill small holes in the image and connect broken regions. By sliding the structuring element over the image and taking the maximum value of all pixel values covered by the structuring element, the dilation effect is achieved.

[0150] Step S5: Perform multiple iterations of feature extraction on the segmented trabecular bone region and adjust the feature extraction parameters according to the results of the previous iteration.

[0151] In step S5, the following sub-steps are further included:

[0152] S5-1: Perform the first iteration of feature extraction, and extract the morphological features of the trabecular bone from the segmented image. The morphological features include trabecular bone thickness, trabecular bone number, and trabecular bone separation. The specific formulas are as follows:

[0153] ;

[0154] ;

[0155] ;

[0156] Where, TbTh represents trabecular bone thickness, TbN represents trabecular bone number, TbSp represents trabecular bone separation, is the area of the trabecular bone region, is the total length of the trabecular bone, is the area of the entire image, is the area of the background region;

[0157] S5-2: Perform the first iteration of feature extraction, and extract the structural features of the trabecular bone from the segmented image. The structural features include trabecular bone connection density and anisotropy. The specific formulas are as follows:

[0158] ;

[0159] ;

[0160] Where, Xv represents the Euler characteristic of the trabecular bone structure, which is used to describe the connection density of the trabecular bone structure, and DA represents the anisotropy, is the 0th Betti number, representing the number of connected regions, is the 2nd Betti number, representing the number of holes, and are both eigenvalues of the inertia tensor of the trabecular bone structure, is the maximum eigenvalue, representing the distribution of the trabecular bone in the main direction, is the second largest eigenvalue, representing the distribution of the trabecular bone in the secondary direction;

[0161] S5-3. Perform the first iteration of feature extraction to extract the trabecular texture features from the segmented image. The texture features include wavelet transform features and trabecular bone score. The specific formulas are as follows:

[0162] ;

[0163] ;

[0164] ;

[0165] ;

[0166] Among them, is the wavelet coefficient, represents a two-dimensional image, is the value of the wavelet basis function at position (j, k). E is the wavelet energy, H is the wavelet entropy, is the wavelet coefficient 's probability distribution. TBS represents the trabecular bone score, and V(e) is the experimental variogram, representing the gray level change of the function at distance e;

[0167] S5-4. According to the feature results extracted in the first iteration, adjust the feature extraction parameters;

[0168] S5-5. Perform the second iteration of feature extraction, repeat steps S5-1 to S5-3 to extract the trabecular features after adjusting the feature extraction parameters;

[0169] S5-6. Repeat steps S5-4 and S5-5 to perform multiple iterations of feature extraction until the standard deviation of the extraction results of the same feature in multiple iterations is less than the preset value, and the preset value is 0.005.

[0170] It should be noted that the trabecular bone thickness: represents the average thickness of the trabecular bone, reflecting the thickness of the trabecular bone, and determines the trabecular bone thickness by measuring the average width of the trabecular bone area.

[0171] The trabecular bone number: represents the number of trabecular bones per unit area, reflecting the density of the trabecular bones, and determines the trabecular bone density by counting the number of trabecular bones per unit area;

[0172] The trabecular bone separation: represents the average distance between trabecular bones, reflecting the distribution density of trabecular bones, and determines the separation degree of trabecular bones by measuring the average distance between trabecular bones;

[0173] The trabecular bone connectivity density: represents the connectivity density of the trabecular bone structure, reflecting the connection degree between trabecular bones, and obtains the Euler characteristic by calculating the 0th and 2nd Betti numbers of the trabecular bone structure;

[0174] Degree of anisotropy: It represents the degree of anisotropy of the trabecular bone structure, reflects the arrangement directionality of the trabecular bone, calculates the eigenvalues of the trabecular bone structure by computing the inertia tensor, and calculates the degree of anisotropy based on the eigenvalues of the inertia tensor;

[0175] Wavelet transform features: Extract the texture information of the image through wavelet transform, reflect the texture details of the trabecular bone, perform wavelet transform on the segmented trabecular bone region to extract wavelet coefficients; calculate the wavelet energy through the sum of squares of the wavelet coefficients; calculate the wavelet entropy through the probability distribution of the wavelet coefficients;

[0176] Trabecular bone score: Evaluate the texture uniformity of the trabecular bone through the experimental variogram, reflect the texture quality of the trabecular bone, and calculate the trabecular bone score through the experimental variogram;

[0177] Adjust parameters: For morphological feature extraction, the threshold parameters for image segmentation can be adjusted, including global threshold and adaptive threshold, to better separate the trabecular bone region and the background region; the morphological operation parameters can be adjusted, including the size of the structuring element and the number of operations, to better refine the trabecular bone boundary; the feature calculation parameters can be adjusted, including the step size for trabecular bone thickness calculation, the grid resolution for trabecular bone number statistics, and the neighborhood range for trabecular bone separation degree calculation, to improve the accuracy and precision of feature calculation.

[0178] For structural feature extraction, the neighborhood size and connection weight for connection density calculation can be adjusted to more accurately reflect the connection degree between trabecular bones; the parameters for anisotropy degree calculation can be adjusted to more accurately reflect the arrangement directionality of trabecular bones; the parameters for Euler characteristic calculation can be adjusted to more accurately describe the connection density of the trabecular bone structure.

[0179] For texture feature extraction, the scale parameter, wavelet basis function, and direction parameter of wavelet transform can be adjusted to better capture the texture features of trabecular bones; the texture feature calculation parameters can be adjusted, including the weight for wavelet energy calculation, the normalization method for wavelet entropy calculation, and the calculation method for texture score, to more accurately evaluate the texture quality of trabecular bones.

[0180] Comparison result: Compare the feature results extracted in the second iteration with those in the first iteration to evaluate the effect of parameter adjustment, compare the standard deviation of the feature values, and check if there is any improvement.

[0181] Calculate the standard deviation: After each iteration, calculate the standard deviation of the extraction results of the same feature in multiple iterations, and check if the calculated standard deviation is less than the preset value of 0.005. If it is satisfied, stop the iteration; if not, continue with the next iteration.

[0182] Step S6, perform a correlation analysis on the extracted trabecular bone features, and display the distribution and change trend of the features through data visualization methods;

[0183] Among them, in step S6, the following sub-steps are further included:

[0184] S6-1, perform normalization processing on the extracted morphological features, structural features, and texture features of trabecular bone, and scale the feature values to the range [0, 1];

[0185] S6-2, use the Pearson correlation coefficient to calculate the correlation between the morphological features, structural features, and texture features of trabecular bone, and evaluate the relationship between the morphological features, structural features, and texture features of trabecular bone according to the value of the correlation coefficient;

[0186] S6-3, display the morphological features, structural features, and texture features through data visualization methods. The data visualization methods include heat maps, scatter plots, histograms, line charts, bar charts, and box plots.

[0187] It should be noted that the normalization processing scales all feature values to the same range ([0, 1]). The dimensions and ranges of different features may be different. Normalization can eliminate this difference and improve the efficiency of subsequent calculations.

[0188] The Pearson correlation coefficient is used to evaluate the linear correlation between two features, and its value range is between [-1, 1], where 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no correlation.

[0189] Data visualization can intuitively display the distribution and change trend of features, helping researchers and clinicians better understand the data. Through visualization, the relationships between features, outliers, and the laws of data distribution can be quickly identified.

[0190] Heat map: Display the correlation matrix between features through a heat map, and intuitively represent the intensity of the correlation between features;

[0191] Scatter plot: Display the relationship between two features through a scatter plot, and the distribution and correlation between features can be intuitively observed;

[0192] Histogram: Display the distribution of a single feature through a histogram, and the frequency distribution of the feature can be intuitively observed;

[0193] Line chart: Display the change trend of features over time or samples through a line chart, which is suitable for displaying dynamic data;

[0194] Bar chart: Display the difference in feature values between different categories or groups through a bar chart, which is suitable for comparative analysis;

[0195] Box plot: The statistical distribution of features is shown through a box plot, including the median, quartiles, and outliers, which is suitable for showing the distribution of data.

[0196] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for extracting trabecular bone features based on image analysis, characterized in that, The method includes: Step S1: Scanning bone tissue using an X-ray imaging device to obtain a two-dimensional projection image of the bone tissue; Step S2: Preprocessing the obtained two-dimensional projection image of the bone tissue, including denoising, contrast enhancement, histogram equalization, and normalization; Step S3: Performing segmentation processing on the image through an edge detection segmentation method to separate the trabecular bone from the image and obtain a trabecular bone region; Step S4: Constructing a deep learning model to perform morphological processing on the segmented trabecular bone region and refining the boundary of the trabecular bone region; Step S5: Performing multiple iterative feature extractions on the segmented trabecular bone region and adjusting the feature extraction parameters according to the results of the previous iteration; Step S6: Performing correlation analysis on the extracted trabecular bone features and displaying the distribution and change trend of the features through a data visualization method; Among them, in step S5, the following sub-steps are further included: S5-1: Performing the first iterative feature extraction, extracting the morphological features of the trabecular bone from the segmented image. The morphological features include trabecular bone thickness, trabecular bone number, and trabecular bone separation degree. The specific formulas are as follows: ； ； ； Among them, TbTh represents the trabecular bone thickness, TbN represents the trabecular bone number, and TbSp represents the trabecular bone separation degree. is the area of the trabecular bone region, is the total length of the trabecular bone, is the area of the entire image, is the area of the background region; S5-2: Performing the first iterative feature extraction, extracting the structural features of the trabecular bone from the segmented image. The structural features include trabecular bone connection density and anisotropy degree. The specific formulas are as follows: ； ； Among them, Xv represents the Euler characteristic of the trabecular bone structure, which is used to describe the connection density of the trabecular bone structure, and DA represents the degree of anisotropy. is the 0th Betti number, representing the number of connected regions. is the second-order Betti number, representing the number of holes. and are both eigenvalues of the inertia tensor of the trabecular bone structure. is the maximum eigenvalue, representing the distribution of trabecular bone in the main direction. is the second-largest eigenvalue, representing the distribution of trabecular bone in the secondary direction. S5-3: Performing the first iterative feature extraction, extracting the texture features of the trabecular bone from the segmented image. The texture features include wavelet transform features and trabecular bone score. The specific formulas are as follows: ； ； ； ； Among them, is a wavelet coefficient, represents a two-dimensional image, is the value of the wavelet basis function at position (j, k), E is the wavelet energy, H is the wavelet entropy, is the wavelet coefficient of the probability distribution, TBS represents the trabecular bone score, V(e) is the experimental variogram, indicating the gray-level change of the function at distance e; S5-4: Adjusting the feature extraction parameters according to the feature results extracted in the first iteration; S5-5: Performing the second iterative feature extraction, repeating steps S5-1 to S5-3, and extracting the trabecular bone features after adjusting the feature extraction parameters; S5-6: Repeating steps S5-4 and S5-5 to perform multiple iterative feature extractions until the standard deviation of the extraction results of the same feature in multiple iterations is less than a preset value, and the preset value is 0.

005.

2. The method for extracting trabecular bone features based on image analysis according to claim 1, wherein: Among them, in step S1, the following sub-steps are further included: S1-1: Setting the scanning parameters of the X-ray imaging device and calibrating the scanning parameters using a standard calibration sample. The scanning parameters include voltage, current, and exposure time; S1-2: Starting the X-ray imaging device to scan the bone tissue and obtaining a two-dimensional projection image of the bone tissue; S1-3: Storing the scanned image data in DICOM format on a local server and backing up the stored image data. The backup is stored on an external storage device.

3. The method for extracting trabecular bone features based on image analysis according to claim 1, wherein: Among them, in step S2, the following sub-steps are further included: S2-1: Setting the parameters of the Gaussian filter, and using the Gaussian filter to perform convolution operation to perform weighted averaging on each pixel in the two-dimensional projection image of the bone tissue with the pixels in its neighborhood to reduce the noise in the image. The parameters include standard deviation and filter window size; S2-2, Adjust the gray value range of the denoised image through an adaptive contrast enhancement method to increase the image contrast and highlight the structural features of the trabecular bone; S2-3, Perform histogram equalization on the enhanced image. By adjusting the gray value distribution of the image, make the histogram of the image evenly distributed; S2-4, Normalize the image after histogram equalization and adjust the gray value range of the image to [0,1].

4. A method for extracting trabecular bone features based on image analysis according to claim 1, characterized in that: Wherein in step S3, the following sub-steps are further included: S3-1, Calculate the gradient of the preprocessed image through an edge detection algorithm, identify the boundaries of the trabecular bone, and generate an edge map. The edge detection algorithm includes the Canny algorithm and the Sobel algorithm; S3-2, Perform binarization processing on the image after edge detection. Set the threshold through an adaptive threshold method to convert the edge map into a binary image and distinguish the trabecular bone area from the background area; S3-3, Traverse each pixel in the binarized image through depth-first search to identify connected trabecular bone areas; S3-4, Mark the identified connected areas through a marking algorithm and assign unique identifiers; S3-5, Calculate the area of each connected area, remove isolated noise points and small area amounts with an area smaller than a preset threshold, and separate connected trabecular bone areas.

5. A method for extracting trabecular bone features based on image analysis according to claim 1, characterized in that: Wherein in step S4, the following sub-steps are further included: S4-1: Design the encoder path of a deep learning model based on an improved DeepLab v3+ neural network structure to extract the context information of the image. The encoder path includes convolutional layers and dilated convolutional layers; S4-2: Design the decoder path of a deep learning model based on an improved DeepLab v3+ neural network structure to restore image information. The decoder path includes upsampling layers and convolutional layers; S4-3: Introduce a skip connection between the encoder and the decoder to splice the feature maps in the encoder path and the feature maps in the decoder path; S4-4: Introduce a spatial pyramid pooling ASPP module to capture feature information at different scales; S4-5, Perform model training, use the training data set to train the model until the Dice coefficient reaches a preset value, and the preset value is 0.9; S4-6, Use the trained deep learning model to perform morphological processing on the segmented trabecular bone area to refine the boundaries of the trabecular bone area. The morphological processing includes dilation and erosion.

6. A method for extracting trabecular bone features based on image analysis according to claim 1, characterized in that: Wherein in step S6, the following sub-steps are further included: S6-1, Normalize the extracted morphological features, structural features, and texture features of the trabecular bone and scale the feature values to the range of [0,1]; S6-2. Calculate the correlations between the trabecular morphological features, structural features, and texture features using the Pearson correlation coefficient, and evaluate the relationships between the trabecular morphological features, structural features, and texture features according to the values of the correlation coefficients; S6-3. Display the morphological features, structural features, and texture features through data visualization methods, and the data visualization methods include heat maps, scatter plots, histograms, line charts, bar charts, and box plots.

Images