Aortic valve calcification degree evaluation method based on autonomous learning

Abstract

The invention provides an aortic valve calcification degree evaluation method based on autonomous learning, and relates to the technical field of medical evaluation, and the method comprises the steps: collecting medical image data containing an aortic valve; segmenting and identifying an aortic valve region in the medical image data based on a feature extraction algorithm of an adaptive pixel distance; performing quantitative analysis on the calcification degree of the segmented aortic valve region by using a pre-trained deep learning model; and outputting a calcification degree evaluation result of the aortic valve. Through the scheme of the invention, the accurate and efficient automatic evaluation of the aortic valve calcification degree can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of medical evaluation, and particularly to a method for evaluating the degree of aortic valve calcification based on autonomous learning. Background Art

[0002] For the method for evaluating the degree of aortic valve calcification based on autonomous learning, it is a method that utilizes machine learning technology, especially the autonomous learning algorithm in deep learning, to automatically evaluate the degree of aortic valve calcification. This method identifies and quantifies the calcified regions of the aortic valve in medical images by training a model, so as to provide quantitative evaluation results for clinicians to assist in diagnosis and treatment decisions. However, one of the main challenges faced by this method is how to extract the key features of the aortic valve region and how to achieve accurate and efficient automatic evaluation of the degree of aortic valve calcification, which includes how to improve the accuracy of the algorithm, reduce the misjudgment rate, and enhance the processing speed to better meet the clinical needs. Summary of the Invention

[0003] In view of this, the present invention provides a method for evaluating the degree of aortic valve calcification based on autonomous learning, which at least partially solves the problems existing in the prior art.

[0004] The method for evaluating the degree of aortic valve calcification based on autonomous learning includes:

[0005] Collect medical image data including the aortic valve;

[0006] Segment and identify the aortic valve region in the medical image data based on a machine learning algorithm;

[0007] Use a pre-trained deep learning model to perform quantitative analysis on the calcification degree of the segmented aortic valve region;

[0008] Output the evaluation result of the calcification degree of the aortic valve.

[0009] In a specific embodiment, the collection of medical image data including the aortic valve includes:

[0010] Obtain the original image data through a medical imaging device;

[0011] Preprocess the original image data to remove noise;

[0012] Based on image enhancement technology, improve the contrast of the aortic valve region;

[0013] Use image registration technology to align the enhanced image with a standard template.

[0014] In a specific embodiment, the improvement of the contrast of the aortic valve region based on image enhancement technology includes:

[0015] Apply histogram equalization technique to adjust the pixel intensity distribution;

[0016] Smooth the image based on a Gaussian filter to reduce high-frequency noise, where the standard deviation σ of the Gaussian filter needs to satisfy the condition σ > 0;

[0017] Adopt an adaptive threshold segmentation method to highlight the aortic valve structure;

[0018] Adjust the contrast according to the contrast enhancement coefficient CEC, where the calculation formula of CEC is CEC = (MaxI - MinI) / AvgI, and MaxI, MinI, and AvgI represent the maximum pixel value, the minimum pixel value, and the average pixel value respectively.

[0019] In a specific embodiment, the adopting an adaptive threshold segmentation method to highlight the aortic valve structure includes:

[0020] Determine the local threshold T, where T = Tavg + K * Tstd, Tavg is the local average gray level, Tstd is the local standard deviation, and K is a constant;

[0021] Perform binarization processing on the image based on the local threshold T;

[0022] Use morphological operations to remove small non-target areas;

[0023] Perform post-processing according to the shape characteristics of the aortic valve, where the shape characteristics include the area A and the perimeter P, ensuring that A and P satisfy the conditions A > A_min and P > P_min, and A_min and P_min are the preset minimum area threshold and minimum perimeter threshold respectively.

[0024] In a specific embodiment, the performing post-processing according to the shape characteristics of the aortic valve includes:

[0025] Perform connected component analysis on the binarized image;

[0026] Based on the shape characteristics, screen out the possible aortic valve areas, where the shape characteristics include the roundness R, and the calculation formula is R = 4 * π * A / P^2, A is the area, and P is the perimeter;

[0027] Retain the connected components that satisfy the condition R > R_min, and R_min is the preset minimum roundness threshold;

[0028] Perform thinning processing on the retained connected components to further optimize the boundary.

[0029] In a specific embodiment, the retaining the connected components that satisfy the condition R > R_min further includes:

[0030] Calculate the roundness R of each connected component;

[0031] Judge whether R is greater than R_min. If not, exclude this connected region;

[0032] If so, further check whether the area A of this connected region is greater than A_min. If so, retain this connected region;

[0033] If not, exclude this connected region.

[0034] In a specific embodiment, the segmentation and recognition of the aortic valve region in the medical image data based on the feature extraction algorithm of adaptive pixel distance includes:

[0035] Use image preprocessing technology to remove noise and enhance contrast;

[0036] Extract key features of the aortic valve region based on the feature extraction algorithm of adaptive pixel distance;

[0037] Apply a machine learning model to classify the extracted features to identify the aortic valve region;

[0038] Post-process the recognition result to optimize the segmentation boundary.

[0039] In a specific embodiment, the extraction of key features of the aortic valve region based on the feature extraction algorithm of adaptive pixel distance includes:

[0040] Select different directions;

[0041] For each pixel (x, y), according to the selected direction and the adaptive pixel distance d adaptive (x, y) calculate the gray-level co-occurrence matrix P(i, j);

[0042] Extract key texture features through the gray-level co-occurrence matrix P(i, j);

[0043] Among them,

[0044] d0 is the basic pixel distance, |κ| is the normalized local curvature, and G is the local gradient amplitude.

[0045] In a specific embodiment, the extraction of key texture features through the gray-level co-occurrence matrix P(i, j) includes:

[0046] Based on the gray-level co-occurrence matrix, calculate the contrast C = ∑ i,j P(i, j)*(i - j) 2 ;

[0047] If the contrast C is greater than the threshold T, it is considered that the texture feature is obvious.

[0048] Through the technical solution of the present invention, the following technical effects can be achieved, including:

[0049] Improve feature discrimination: In regions with lesions such as aortic valve stenosis and calcification, texture features may exhibit abnormal contrast and gray-scale distribution due to the lesions. A fixed pixel distance may not be able to fully capture the local features of these lesions, thereby reducing the accuracy of disease detection. In the lesion area, the adaptive pixel distance can be adjusted according to information such as local gradient and curvature, enabling the GLCM to more sensitively reflect the texture changes of the diseased tissue.

[0050] Reduce noise interference: Medical images often suffer from noise interference (such as ultrasound speckle noise, CT artifacts, MRI motion artifacts). A fixed pixel distance may cause error accumulation in the noise area, affecting feature stability. The adaptive pixel distance can adaptively adjust the distance based on the gradient magnitude, enabling feature calculation to avoid high-noise areas and improving stability.

[0051] Improve the accuracy of aortic valve lesion detection: Due to the complex morphology of the aortic valve, the GLCM features calculated with a fixed pixel distance may have low feature discrimination in lesion detection tasks such as valve stenosis, thickening, and calcification. The adaptive pixel distance combined with machine learning can obtain better training results through clearer feature extraction, better adapt to individual differences, and improve diagnostic performance. Description of the Drawings

[0052] In the drawings, unless otherwise specified, the same reference numerals throughout the several views denote the same or similar components or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0053] Figure 1 is a flowchart of a method for evaluating the degree of aortic valve calcification based on self-learning;

[0054] Figure 2 is a flowchart of collecting medical image data containing the aortic valve;

[0055] Figure 3 is a flowchart of improving the contrast of the aortic valve region based on image enhancement technology;

[0056] Figure 4 is a flowchart of highlighting the aortic valve structure using an adaptive threshold segmentation method;

[0057] Figure 5 is a flowchart of post-processing based on the shape features of the aortic valve;

[0058] Figure 6 is a flowchart of retaining the connected components that satisfy the condition R > R_min;

[0059] Figure 7 It is a flowchart for segmenting and identifying the aortic valve region in medical image data based on machine learning algorithms;

[0060] Figure 8 It is a flowchart for extracting key features of the aortic valve region using a feature extraction algorithm based on adaptive pixel distance. Specific implementation manners

[0061] In the following, only some exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the drawings and the description are considered to be exemplary in nature rather than restrictive.

[0062] Next, refer to Figure 1 to describe the method for evaluating the degree of aortic valve calcification based on autonomous learning:

[0063] S101: By collecting medical image data containing the aortic valve, we first ensure that the data source for analysis is of high quality and reliable. This process typically involves using medical imaging techniques such as computed tomography (CT), magnetic resonance imaging (MRI), or echocardiography to obtain images of the heart region. For example, in a study, researchers may collect CT scan images of hundreds of patients from a hospital's database, which clearly show the position and structure of the aortic valve. To ensure the quality of the data, these images also need to be preprocessed, such as removing noise, enhancing contrast, and standardizing the image size. In addition, considering privacy protection and ethical requirements, all patient information must be anonymized to ensure that the patient's identity will not be disclosed. In this way, we can obtain a series of high-quality aortic valve images, providing a solid foundation for subsequent analysis.

[0064] S102: By segmenting and identifying the aortic valve region in the medical image data, we can accurately locate the position of the aortic valve and separate it from the background. This step usually relies on advanced machine learning algorithms, such as convolutional neural network (CNN) or U-Net, etc. We can extract key features of the aortic valve region, such as texture, using a feature extraction algorithm based on adaptive pixel distance, and then apply a machine learning model to classify the extracted features to identify the aortic valve region. Once the model is trained, it can be applied to new unlabeled images to automatically identify the position of the aortic valve. In this way, we can not only accurately identify the aortic valve region but also provide more accurate data for subsequent analysis of the degree of calcification.

[0065] S103: Use the pre-trained deep learning model to perform quantitative analysis on the segmented aortic valve region, and we can further evaluate the health status of the aortic valve. This step usually involves using a deep learning model specifically designed for calcification detection and quantification. For example, a deep learning model that has been trained on a large number of calcification samples can be used to evaluate the degree of calcification within the aortic valve region. These models can identify specific patterns of calcification and give a quantitative score based on their density and distribution. To improve the accuracy of the evaluation, the model may also consider other factors, such as changes in the thickness of the aortic valve. In addition, to ensure the generalization ability of the model, it is usually validated and adjusted on multiple different datasets. In this way, we can not only accurately evaluate the degree of aortic valve calcification, but also provide valuable diagnostic information for clinicians.

[0066] S104: Output the evaluation result of the aortic valve calcification degree. We finally summarize and present all the analysis results to clinicians or researchers. This step usually includes generating a detailed report, which contains the specific value of the aortic valve calcification degree, visualization images, and possible clinical suggestions, etc. For example, for each patient's aortic valve image, the system will generate a report, which details the calcification degree score, image annotation of the calcification area, and related risk assessment, etc. These reports can not only help doctors better understand the patient's condition, but also serve as an important basis for formulating treatment plans. In addition, for the convenience of communication and discussion among doctors, these reports can be shared through the electronic medical record system.

[0067] Through the above steps, the problem of "how to achieve accurate and efficient automatic evaluation of the aortic valve calcification degree" can be solved. The whole process starts from data acquisition, uses machine learning algorithms to segment and identify the aortic valve region, then uses a deep learning model to perform quantitative analysis of the calcification degree, and finally outputs a detailed evaluation result. This method not only improves the accuracy of the evaluation, but also greatly reduces the workload of doctors, helps to detect disease signs earlier, and thus improves the treatment effect of patients.

[0068] Next, refer to Figure 2 Describe the specific steps of collecting medical image data containing the aortic valve in the aortic valve calcification degree evaluation method based on self-learning:

[0069] S201: Obtain the original image data through a medical imaging device. This process usually involves using advanced medical imaging technologies such as computed tomography (CT) or magnetic resonance imaging (MRI) to capture high-resolution images of the patient's heart region. These devices can provide detailed anatomical structure information, especially for the details of the aortic valve and its surrounding tissues.

[0070] S202: Preprocess the original image data to remove noise. After obtaining the original image, preprocessing is required. This step is mainly to reduce or eliminate various noises generated during the imaging process, such as random noise, stripe artifacts, etc. Preprocessing techniques can include filtering algorithms, such as median filtering or Gaussian filtering, to smooth the image and reduce the impact of noise.

[0071] S203: Improve the contrast of the aortic valve region based on image enhancement techniques. To more clearly identify and analyze the structure of the aortic valve, it is necessary to further enhance the contrast in the image. This can be achieved through a variety of image enhancement techniques, such as histogram equalization, local contrast enhancement, or using specific filters to highlight the characteristics of the aortic valve region.

[0072] S204: Align the enhanced image with a standard template using image registration techniques. The last step is to ensure that the enhanced image is accurately aligned with a known standard template for subsequent analysis and comparison. Image registration techniques can use rigid registration or non-rigid registration methods, adjusting the position, rotation angle, and possible deformation of the image as needed to make the structure of the aortic valve precisely match the corresponding position in the template.

[0073] Next, refer to Figure 3 Describe the specific steps of enhancing the contrast of the aortic valve region based on image enhancement techniques in the method for evaluating the degree of aortic valve calcification based on self-learning:

[0074] S301: Adjust the pixel intensity distribution by applying histogram equalization techniques. This process aims to optimize the overall contrast of the image, making the details of the aortic valve region more clearly visible. Specifically, histogram equalization remaps the pixel values in the image to ensure that the pixel values are evenly distributed across the entire dynamic range, thereby achieving the effect of enhancing the contrast.

[0075] S302: Smooth the image based on a Gaussian filter to reduce high-frequency noise. In this step, a Gaussian filter with a standard deviation σ greater than zero is selected to perform a convolution operation on the image to remove random noise in the image while keeping the basic structure of the image unchanged. Gaussian filters are widely used in the image preprocessing stage due to their smoothing characteristics and can effectively reduce the impact of high-frequency noise introduced during the imaging process on subsequent analysis.

[0076] S303: Highlight the aortic valve structure by adopting an adaptive threshold segmentation method. This method automatically determines the threshold based on the change of local pixel values, thereby separating the aortic valve region from the background. Adaptive threshold segmentation can better adapt to the brightness changes in different regions of the image, ensuring accurate identification of the target region even under uneven illumination.

[0077] S304: Adjust the contrast according to the Contrast Enhancement Coefficient (CEC). First, calculate the maximum pixel value MaxI, the minimum pixel value MinI, and the average pixel value AvgI of the image. Then, use the formula CEC = (MaxI – MinI) / AvgI to calculate the CEC. Finally, further adjust the contrast of the image according to the calculated CEC value to ensure that the details in the aortic valve area are presented optimally.

[0078] Next, refer to Figure 4 Describe the specific steps of using the adaptive threshold segmentation method to highlight the aortic valve structure in the method for evaluating the degree of aortic valve calcification based on self-learning:

[0079] S401: Determine the local threshold T: First, calculate the average gray value Tavg and the standard deviation Tstd within the local neighborhood of each pixel point in the image. Then, determine the local threshold T according to the formula T = Tavg + K * Tstd, where K is a preset constant used to adjust the sensitivity of the threshold. This process can ensure relatively accurate segmentation results in images with different lighting conditions or different gray distributions.

[0080] S402: Binarize the image based on the local threshold T: Using the determined local threshold T above, divide the pixel points in the image into two categories: pixel points higher than T are marked as the foreground, usually assigned a value of 1, and pixel points lower than T are marked as the background, usually assigned a value of 0. This can effectively highlight the aortic valve structure while suppressing the influence of background noise.

[0081] S403: Use morphological operations to remove small non-target areas: To further improve the segmentation accuracy, morphological operations such as opening or closing operations can be used to remove those small areas that are not related to the aortic valve. These operations help eliminate noise interference in the image and make the contour of the aortic valve clearer.

[0082] S404: Perform post-processing according to the shape characteristics of the aortic valve: Finally, perform post-processing according to the shape characteristics of the aortic valve, such as the area A and the perimeter P, to ensure that the segmented target area actually represents the aortic valve. Specifically, only when the area of the segmented area is greater than the preset minimum area threshold A_min and the perimeter is greater than the preset minimum perimeter threshold P_min is it considered a valid aortic valve area. This step helps exclude those segmentation results that do not conform to the size characteristics of the aortic valve, thereby improving the accuracy of the overall evaluation.

[0083] Through the above steps, the aortic valve structure can be effectively highlighted and accurately identified from medical images, providing a reliable basis for further evaluation of the degree of calcification.

[0084] Next, refer toFigure 5 Describe the specific steps for post - processing based on the shape characteristics of the aortic valve:

[0085] S501: Process the binarized image through connected - component analysis. At this stage, the system analyzes the image data after binarization, identifies and marks all continuous pixel regions, which represent potential aortic valve structures. Connected - component analysis helps us distinguish different objects and provides a basis for subsequent shape - feature extraction.

[0086] S502: Screen out possible aortic valve regions based on shape features. Based on connected - component analysis, the system further calculates the roundness R of each connected component. The roundness is defined as R = 4π 2 *A / P 2 , where A represents the area of the connected component, and P is the perimeter of the connected component. In this way, the similarity between each connected component and an ideal circle can be quantified. Subsequently, the system will screen out the connected components that are most likely to represent the aortic valve according to a preset minimum roundness threshold R_min.

[0087] S503: Retain the connected components that satisfy the condition R > R_min. After calculating the roundness of each connected component, the system compares these roundness values with the preset minimum roundness threshold R_min. Only when the roundness of the connected component is greater than or equal to this threshold will it be retained as a possible aortic valve region. This step helps to exclude those connected components whose shapes do not conform to the characteristics of the aortic valve, thereby improving the accuracy of subsequent analysis.

[0088] S504: Refine the retained connected components to further optimize the boundary. Finally, to more precisely determine the boundary of the aortic valve, the system performs a refinement process on the retained connected components. This processing method can remove redundant pixels in the connected component, making the boundary clearer and more accurate. Through this series of steps, we can effectively extract the shape characteristics of the aortic valve from the original image and provide reliable data support for further assessment of the calcification degree.

[0089] Next, refer to Figure 6 Describe the specific steps for retaining connected components that meet specific conditions in the method for assessing the calcification degree of the aortic valve based on self - learning:

[0090] S601: First, for all connected components identified during the image - processing process, calculate the roundness R of each connected component. The roundness R is a quantitative index used to measure the proximity of the shape of the connected component to a circle. Generally, the roundness can be calculated from the perimeter and area of the connected component. For example, R can be defined as 4π times the area divided by the square of the perimeter.

[0091] S602: Next, compare the calculated roundness R with the preset minimum roundness threshold R_min. If the roundness R of a certain connected component is less than or equal to R_min, it is considered that the connected component does not meet the retention condition, so it is excluded; conversely, if R is greater than R_min, the area A of this connected component needs to be further checked.

[0092] S603: For those connected components with roundness R greater than R_min, further check whether their area A is greater than the preset minimum area threshold A_min. If the area A of the connected component is greater than A_min, it is considered that the connected component meets the retention condition, so it is retained; otherwise, if A is less than or equal to A_min, the connected component is excluded.

[0093] For example, in actual operation, assume that we have obtained an image containing multiple connected components through image segmentation technology. For each of these connected components, we first calculate its roundness R. Assume that R_min is set to 0.7, that is, only connected components with roundness greater than 0.7 will be considered, and A_min is set to 100 pixel^2, that is, only connected components with an area greater than 100 pixel^2 will be considered. If the calculated roundness R of a certain connected component is 0.8, then it will enter the next step of inspection; if its area A is 120 pixel^2, then the connected component will be retained because it meets both the roundness and area conditions. On the contrary, if the roundness R of another connected component is 0.6 or the area A is 90 pixel^2, then the connected component will be excluded because it does not meet the set threshold conditions.

[0094] Next, refer to Figure 7 Describe the specific steps of segmenting and recognizing the aortic valve region in the medical image data based on the feature extraction algorithm of adaptive pixel distance:

[0095] S701: Remove the noise in the medical image data and enhance the contrast by using image preprocessing technology. This process usually involves applying filters to reduce the random noise in the image and using techniques such as histogram equalization or local contrast enhancement to improve the distinguishability between the aortic valve region and other tissues. For example, a Gaussian filter can be used to smooth the image and reduce the influence of high-frequency noise; at the same time, the overall contrast of the image can be improved through adaptive histogram equalization technology, making the aortic valve structure more clearly visible.

[0096] S702: Extract the key features of the aortic valve region through the feature extraction algorithm based on adaptive pixel distance.

[0097] The Gray-Level Co-Occurrence Matrix (GLCM) is a commonly used method for texture feature extraction. It obtains texture information by calculating the spatial relationships between pixel gray values in an image. However, traditional GLCM uses a fixed pixel distance d, which may not be suitable for the complex morphology of the aortic valve region. Therefore, this application proposes an adaptive pixel distance method that adjusts the distance adaptively according to the anatomical structure, thereby improving the accuracy of aortic valve feature extraction.

[0098] The adaptive pixel distance adjusts the pixel distance through local curvature and gradient, enabling more accurate capture of key texture features in the aortic valve region and improving the accuracy of lesion detection.

[0099] S703: Classify the extracted features using a machine learning model to identify the aortic valve region. At this stage, algorithms such as the Support Vector Machine (SVM), Random Forest (RF), or other supervised learning algorithms can be trained and supported. The previously extracted features are used as input variables to distinguish the aortic valve from other tissues. For example, a Support Vector Machine model can be trained with labeled training set data, which can classify pixels into aortic valve or non-aortic valve categories based on the feature vector.

[0100] S704: Post-process the recognition results to optimize the segmentation boundary. This step aims to refine the segmentation results and ensure that the aortic valve region is accurately defined. Morphological operations such as dilation and erosion can be used to eliminate small discontinuous regions, or methods such as Conditional Random Field (CRF) can be used to further optimize the boundary. For example, applying morphological closing operations to fill small holes within the aortic valve region while avoiding introducing new noise points, thereby obtaining a more accurate segmentation effect.

[0101] Next, refer to the appendix Figure 8 Describe the specific steps of extracting key features of the aortic valve region using the feature extraction algorithm based on adaptive pixel distance:

[0102] S801: Select different directions;

[0103] The calculation of the Gray-Level Co-Occurrence Matrix depends on the selection of directions. The four common directions are:

[0104] 0° (horizontal direction);

[0105] 45° (diagonal direction);

[0106] 90° (vertical direction);

[0107] 135° (anti-diagonal direction).

[0108] S802: For each pixel (x, y), calculate the Gray-Level Co-Occurrence Matrix P(i, j) according to the selected direction and the adaptive pixel distance d adaptive (x, y)

[0109] Specifically, first, according to the selected direction, the neighborhood pixels of the current pixel are determined. Assume that the gray value of the current pixel is i and the gray value of its neighborhood pixel is j, where

[0110] For the 0° direction, the neighborhood pixel is the pixel to the right of the current pixel;

[0111] For the 45° direction, the neighborhood pixel is the pixel above and to the right of the current pixel;

[0112] For the 90° direction, the neighborhood pixel is the pixel above the current pixel;

[0113] For the 135° direction, the neighborhood pixel is the pixel above and to the left of the current pixel.

[0114] In the calculation of the gray-level co-occurrence matrix (GLCM), the pixel distance (usually denoted as d) refers to the spatial distance between the current pixel and its neighborhood pixel. The pixel distance usually adopts fixed values (1, 2, 3,...). For example, d = 1 represents the relationship between the current pixel and its directly adjacent pixels (such as the right side, left side, upper side, lower side, etc.).

[0115] This application sets an adaptive pixel distance d based on local curvature and gradient change adaptive ,

[0116] where

[0117] d0 is the basic pixel distance, |κ| is the normalized local curvature, and G is the local gradient magnitude.

[0118] Among them, the local curvature κ reflects the degree of shape bending, which is often larger at the edges and lesion areas. The two-dimensional curvature calculation method is used:

[0119]

[0120] where:

[0121] I x ,I y are the image gradients (partial derivatives in the x and y directions),

[0122] I xx ,I yy ,I xy are the second-order derivatives,

[0123] ∈ is a small value (to prevent the denominator from being zero);

[0124] The larger the curvature (the bending area), the larger d adaptive increases to capture more information,

[0125] When the curvature is small (smooth region), d adaptive remains small, improving the local feature resolution.

[0126] Among them, the local gradient magnitude G reflects the gray-scale change rate and is high in regions with obvious edges or textures:

[0127]

[0128] When the gradient is large (at the edge), d adaptive increases, expanding the information capture range,

[0129] When the gradient is small (smooth region), d adaptive remains small, enhancing the local feature stability.

[0130] In the lesion area, the adaptive pixel distance can be adjusted according to local gradient, curvature and other information, enabling the GLCM to more sensitively reflect the texture changes of the diseased tissue; the adaptive pixel distance can adaptively adjust the distance based on the gradient magnitude, enabling the feature calculation to avoid high-noise regions and improving stability.

[0131]

[0132] Among them, d x = d adaptive cos(θ), d y = d adaptive sin(θ), where θ is the direction angle, and P(i, j) reflects the co-occurrence probability of gray levels i to j in the selected direction.

[0133] S803: Extract key texture features through the gray-level co-occurrence matrix P(i, j);

[0134] After completing the calculation of the gray-level co-occurrence matrix based on the adaptive pixel distance, key features such as contrast, entropy, energy, and uniformity can be extracted from the GLCM to describe the texture information of the aortic valve region.

[0135] Optionally, after obtaining the gray-level co-occurrence matrix P(i, j), the contrast C is calculated based on this matrix next. The contrast is an index to measure the difference in gray values in an image, and its calculation formula is C = ∑ i,j P(i, j)*(i - j) 2 , that is, the contrast C is equal to the weighted sum of the squares of the products of all elements in the gray-level co-occurrence matrix and their corresponding row index i and column index j. Through the calculated contrast C value, it can be further determined whether the texture features in the aortic valve region are obvious. If the contrast C is greater than a pre-set threshold T, it is considered that the texture features in this region are obvious; otherwise, it is considered that the texture features are not obvious.

[0136] Optionally, in addition to contrast, other statistics can be calculated as auxiliary features to comprehensively evaluate the degree of aortic valve calcification. These statistics include but are not limited to energy and entropy. By calculating these statistics, the gray-level distribution characteristics of the aortic valve region can be more comprehensively understood, thereby assisting in evaluating the degree of aortic valve calcification.

[0137] Next, the specific steps for extracting key texture features through the gray-level co-occurrence matrix P(i,j) are described:

[0138] S901: Traverse the gray-level co-occurrence matrix P(i,j), where i and j represent the gray levels in the image respectively. The gray-level co-occurrence matrix P(i,j) records the occurrence probability of pixels with gray value i and pixels with gray value j at a specific direction and distance. This process ensures that the relationships between different gray levels in the image can be comprehensively considered.

[0139] S902: Quantify the difference between gray levels by calculating P(i,j)*(i - j) for each element i,j. 2 Specifically, for each element P(i,j) in the matrix, calculate the product of (i - j) and P(i,j). This step essentially measures the contribution degree of the difference between gray levels i and j. Among them, (i - j) 2 reflects the size of the difference between gray levels, while Pi,j reflects the probability of this difference occurring. 2

[0140] S903: Accumulate all the calculation results to obtain the contrast C. This means adding up all the values of P(i,j)*(i - j) calculated in S902, so as to obtain the contrast C of the entire gray-level co-occurrence matrix C = ∑ 2 P(i,j)*(i - j) i,j 2 , and this value reflects the overall difference between gray levels in the image.

[0141] S904: If the calculated contrast C is greater than the preset threshold T, it is considered that the texture feature is significant. Here, the threshold T is a preset value used to determine whether the contrast is large enough to indicate obvious texture features in the image. If the contrast C exceeds this threshold, it can be considered that the texture feature in the image is significant, which helps the subsequent accurate evaluation of the degree of aortic valve calcification.

[0142] Next, the specific steps for quantitatively analyzing the degree of calcification of the segmented aortic valve region using a pre-trained deep learning model are described:

[0143] S1001: Obtain the segmented aortic valve region and perform preprocessing;

[0144] Obtain the aortic valve region after segmenting the medical image data using a feature extraction algorithm based on adaptive pixel distance, ensuring accurate segmentation results and avoiding the introduction of noise or incorrect regions.

[0145] Normalize the image pixel values to a fixed range (such as [0, 1] or [-1, 1]); crop the segmented aortic valve region to a fixed size, or maintain the original size by padding; augment the training data (such as rotation, flipping, scaling, etc.) to improve the robustness of the model.

[0146] S1002: Load a pre-trained deep learning model;

[0147] Select a pre-trained model suitable for medical image analysis, such as: 2D models: ResNet, DenseNet, EfficientNet, etc. These models are usually pre-trained on large natural image datasets (such as ImageNet) and have powerful feature extraction capabilities.

[0148] According to the task requirements, modify the last layer of the pre-trained model: The degree of calcification scoring can be regarded as a regression task, and replace the last layer with a single output node.

[0149] S1003: Model training;

[0150] Form a training dataset from the segmented aortic valve region images and their corresponding degree of calcification labels (such as grading or scoring), and divide it into a training set, a validation set, and a test set;

[0151] Use the training set data to train the model and monitor the performance on the validation set;

[0152] Use the Mean Squared Error (MSE) loss function for performance evaluation.

[0153] Next, describe the specific steps for outputting the evaluation results of the degree of calcification of the aortic valve:

[0154] S1101: Model inference;

[0155] Use the trained model to perform inference on the aortic valve region in the test set or new data,

[0156] Input the segmented aortic valve region image,

[0157] Output the regression result of the degree of calcification (such as a score).

[0158] S1102: Visualize the results

[0159] Quantification results of visualizing the degree of calcification: Mark the calcified area on the original image and use a heatmap to display the distribution of the degree of calcification.

[0160] In the actual operation process, when this device is used, it is first necessary to collect medical image data containing the aortic valve. This step usually involves using imaging techniques such as CT scans or echocardiograms to obtain high-quality images that can clearly show the heart structure, especially the position and morphology of the aortic valve. Subsequently, based on machine learning algorithms, the aortic valve area in the obtained medical image data is accurately segmented and identified. This step is crucial for subsequent assessment of the degree of calcification because it ensures that the algorithm can accurately focus on the aortic valve area and exclude the influence of other irrelevant tissues. To achieve this, a convolutional neural network (CNN) or other advanced machine learning models trained with a large amount of labeled data are usually used for automatic segmentation. Next, a pre-trained deep learning model is used to perform a quantitative analysis of the degree of calcification on the already segmented aortic valve area. This process involves further processing of the segmented image, such as evaluating the severity of calcification by calculating the distribution of specific pixel values. Finally, the system will output an assessment report on the degree of calcification of the aortic valve based on the above analysis results, providing valuable diagnostic information for clinicians to help them formulate more reasonable treatment plans. In the entire process, from data collection to the output of the final assessment results, each component collaborates closely to jointly complete an efficient and accurate assessment of the degree of aortic valve calcification.

[0161] The methods, programs, systems, devices, etc. of the embodiments of the present invention can be executed or implemented in a single or multiple networked computers, and can also be practiced in a distributed computing environment. In the embodiments of this specification, in these distributed computing environments, tasks can be executed by remote processing devices connected through a communication network.

[0162] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, system, or computer program product. Therefore, those skilled in the art can think that the implementation of the functional modules / units or controllers and related method steps clarified in the above embodiments can be achieved in a software, hardware, or a combination of software and hardware manner.

[0163] Unless explicitly stated, the actions or steps of the methods and programs recorded according to the embodiments of the present invention do not necessarily have to be executed in a specific order and can still achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be beneficial.

[0164] In this document, multiple embodiments of the present invention are described. For the sake of brevity, the description of each embodiment is not exhaustive, and features or parts that are the same or similar among the various embodiments may be omitted. In this document, "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" mean applicable to at least one embodiment or example according to the present invention, rather than all embodiments. The above terms do not necessarily refer to the same embodiment or example. Without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples.

[0165] Exemplary systems and methods of the present invention have been specifically shown and described with reference to the above embodiments, which are only examples of the best mode for implementing the systems and methods. Those skilled in the art can understand that various changes can be made to the embodiments of the systems and methods described herein when implementing the systems and / or methods without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. An aortic valve calcification degree evaluation method based on autonomous learning, characterized in that, Including: Collecting medical image data including the aortic valve; Segmenting and identifying the aortic valve region in the medical image data based on a feature extraction algorithm of adaptive pixel distance; Quantitatively analyzing the calcification degree of the segmented aortic valve region using a pre-trained deep learning model; Outputting the evaluation result of the calcification degree of the aortic valve.

2. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 1, wherein The collecting medical image data including the aortic valve includes: Obtaining original image data through a medical imaging device; Preprocessing the original image data to remove noise; Enhancing the contrast of the aortic valve region based on image enhancement technology; Aligning the enhanced image with a standard template using image registration technology.

3. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 2, wherein The enhancing the contrast of the aortic valve region based on image enhancement technology includes: Applying histogram equalization technology to adjust the pixel intensity distribution; Smoothing the image based on a Gaussian filter to reduce high-frequency noise, where the standard deviation σ of the Gaussian filter needs to satisfy the condition σ>0; Adopting an adaptive threshold segmentation method to highlight the aortic valve structure; Adjusting the contrast according to the contrast enhancement coefficient CEC, where the calculation formula of CEC is CEC=(MaxI-MinI) / AvgI, and MaxI, MinI, and AvgI represent the maximum pixel value, the minimum pixel value, and the average pixel value respectively.

4. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 3, wherein The adopting an adaptive threshold segmentation method to highlight the aortic valve structure includes: Determining the local threshold T, where T=Tavg+K*Tstd, Tavg is the local average gray level, Tstd is the local standard deviation, and K is a constant; Performing binarization processing on the image based on the local threshold T; Using morphological operations to remove small non-target regions; Performing post-processing according to the shape characteristics of the aortic valve, where the shape characteristics include the area A and the perimeter P, ensuring that A and P satisfy the conditions A>A_min and P>P_min, and A_min and P_min are the preset minimum area threshold and minimum perimeter threshold respectively.

5. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 4, wherein The performing post-processing according to the shape characteristics of the aortic valve includes: Performing connected component analysis on the binarized image; Screening out possible aortic valve regions based on shape characteristics, where the shape characteristics include the roundness R, and the calculation formula is R=4*π*A / P^2, A is the area, and P is the perimeter; Retaining the connected components that satisfy the condition R>R_min, and R_min is the preset minimum roundness threshold; Performing thinning processing on the retained connected components to further optimize the boundary.

6. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 5, wherein, The retaining the connected components that satisfy the condition R>R_min further includes: Calculating the roundness R of each connected component; Judging whether R is greater than R_min, if not, excluding the connected component; If so, further checking whether the area A of the connected component is greater than A_min, if so, retaining the connected component; If not, excluding the connected component.

7. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 1, wherein The segmenting and identifying the aortic valve region in the medical image data based on a feature extraction algorithm of adaptive pixel distance includes: Using image preprocessing technology to remove noise and enhance the contrast; Extracting key features of the aortic valve region based on a feature extraction algorithm of adaptive pixel distance; Applying a machine learning model to classify the extracted features to identify the aortic valve region; Post-process the recognition results to optimize the segmentation boundary.

8. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 7, characterized in that, The key features of the aortic valve region extracted by the feature extraction algorithm based on the adaptive pixel distance include: Select different directions; For each pixel (x, y), calculate the gray-level co-occurrence matrix P(i, j) according to the selected direction and the adaptive pixel distance; Extract the key texture features through the gray-level co-occurrence matrix P(i, j); Among them, d0 is the basic pixel distance, |κ| is the normalized local curvature, and G is the local gradient amplitude.

9. The method for evaluating the degree of aortic valve calcification based on autonomous learning according to claim 8, wherein The extraction of the key texture features through the gray-level co-occurrence matrix P(i, j) includes: Calculate the contrast C = ∑ based on the gray-level co-occurrence matrix i,j P(i,j)*(i - j) 2 ; If the contrast C is greater than the threshold T, the texture features are considered obvious.

Images