Intelligent Recognition System for Gout Crystal Areas in the Knee Joint Based on CT Images

Abstract

This invention relates to the field of image recognition technology, specifically to an intelligent recognition system for gouty crystallization regions of the knee joint based on CT images. The system includes: identifying high-density regions in knee CT images by utilizing the absorption of X-rays by urate crystals produced by gout; identifying non-dense calcified regions within the high-density regions based on the grayscale values ​​of the calcified foci and high-density regions; determining a clouding coefficient based on the degree of tissue infiltration in the non-dense calcified regions, the standard deviation of adjacent edge pixels, and the range of pixel counts in the through-pixel segment; determining a crystallization coefficient based on the distance between edge pixels of non-skeletal connected regions and skeletal connected regions in the non-dense calcified regions and the clouding coefficient; labeling the non-dense calcified regions as crystalline regions using the crystallization coefficient; and training a pre-set neural network model using the labeled CT images to obtain a knee joint gouty crystallization region recognition model. This invention can effectively identify crystalline regions in CT images.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to an intelligent recognition system for gout crystal areas in the knee joint based on CT images. Background Technology

[0002] Currently, CT scans are widely used in the diagnosis of musculoskeletal diseases. Their high-resolution cross-sectional imaging capabilities can clearly display the bony structures, soft tissues, and subtle lesions within the knee joint cavity. In the diagnosis of gouty arthritis, urate crystals deposited on CT images typically appear as needle-like or granular shadows with a density significantly higher than that of surrounding tissues. This imaging characteristic provides important evidence for the early identification and quantitative assessment of gout.

[0003] However, traditional manual image interpretation mainly relies on the experience and subjective judgment of personnel, which is not only time-consuming and laborious, but also prone to failure to identify multi-layered, small-volume or low-density crystalline lesions, resulting in inaccurate identification and low identification efficiency of crystalline areas in CT images. Summary of the Invention

[0004] To address the technical problems of inaccurate and low-efficiency identification of crystalline regions in CT images, the present invention aims to provide an intelligent identification system for gouty crystalline regions of the knee joint based on CT images. The specific technical solution adopted is as follows:

[0005] The information processing module is used to identify the corresponding high-density areas in the acquired knee joint CT images of the target object by utilizing the absorption of X-rays by urate crystals produced by gout.

[0006] The filtering module is used to identify non-dense calcification areas in the high-density areas based on the gray values ​​of the calcification areas and the high-density areas in the knee joint CT images.

[0007] The cloud and fog coefficient determination module is used to determine the cloud and fog coefficient of the non-dense calcified region based on the degree of body tissue infiltration in the non-dense calcified region, the standard deviation of adjacent edge pixels in the non-dense calcified region, and the range of the number of pixels in the through-hole pixel segment in the non-dense calcified region.

[0008] The crystallization coefficient determination module is used to determine the crystallization coefficient based on the distance between some pixels in the non-skeletal connected domain of the non-dense calcified region and the edge pixels of the skeletal connected domain in the non-dense calcified region, as well as the cloud coefficient of the non-dense calcified region.

[0009] The annotation module is used to annotate non-dense calcification areas as crystalline areas in knee CT images when the crystallization coefficient is greater than a preset crystallization threshold.

[0010] The identification module is used to train a preset neural network model to identify crystal regions using knee CT images with marked crystal regions, thereby obtaining a trained knee joint gout crystal region identification model; wherein, the knee joint gout crystal region identification model is used to mark the crystal regions in knee CT images.

[0011] Preferably, the high-density region in the acquired knee joint CT image of the target object is determined by utilizing the absorption of X-rays by urate crystals produced by gout, including:

[0012] By using the grayscale values ​​and pixel gradient values ​​of the leg images in knee CT images, the urate crystal region in the leg image was identified.

[0013] Based on the urate crystal area in the leg and the urate crystal area in the knee joint in the leg images, the high-density area corresponding to the knee joint CT image is determined.

[0014] Preferably, the grayscale values ​​and pixel gradient values ​​of the leg image in a knee CT scan are used to determine the urate crystal region in the leg image, including:

[0015] Calculate the ratio of grayscale value to pixel gradient value of each pixel in the grayscale image of the leg image, and normalize the ratio to obtain the high density coefficient of each pixel.

[0016] The Otsu threshold segmentation of the leg image was performed using the high density coefficient of each pixel to obtain the urate crystallization region of the leg.

[0017] Preferably, the high-density region corresponding to the knee joint CT image is determined based on the urate crystal region in the leg and the urate crystal region in the knee joint in the leg image, including:

[0018] Based on the urate crystal area in the leg and the urate crystal area in the knee joint in the leg images, the high-density candidate area corresponding to the knee joint CT image is determined.

[0019] If the difference between the gray values ​​of edge pixels and the gray values ​​of surrounding pixels in a high-density candidate region meets the preset growth conditions, the surrounding pixels are added to the high-density candidate region to obtain the high-density region.

[0020] Preferably, based on the urate crystal area in the leg and the urate crystal area in the knee joint in the leg image, a high-density candidate region corresponding to the knee joint CT image is determined, including:

[0021] By merging the connected components of the urate crystal region in the leg and the urate crystal region in the knee joint in the leg images, high-density candidate regions corresponding to the knee joint CT images are obtained.

[0022] Preferably, based on the grayscale values ​​of the calcified area and the high-density area in the knee joint CT image, a non-dense calcified area is identified within the high-density area, including:

[0023] The threshold for dense calcification is determined based on the mean and standard deviation of the gray values ​​in the calcified area.

[0024] If the average gray value of a high-density region is not greater than the dense calcification threshold, the high-density region is defined as a non-dense calcification region.

[0025] Preferably, the method for determining the degree of body tissue infiltration in non-dense calcified areas includes:

[0026] The degree of body tissue infiltration in the non-dense calcified region is determined based on the difference in the mean grayscale values ​​between the internal and external pixels of the non-dense calcified region, and the difference in the number of pixels between the non-dense calcified region and its corresponding high-density candidate region.

[0027] Preferably, the method for determining the mean grayscale difference between internal and external pixels in the non-dense calcified region includes:

[0028] On the straight line connecting the edge point of the non-dense calcified region and the centroid, select m pixel segments, and define the pixel segments located within the non-dense calcified region as internal pixel segments, and define the pixel segments located outside the non-dense calcified region as external pixel segments; where m is a positive integer.

[0029] The grayscale difference is obtained based on the average grayscale value of the outer pixel segment and the average grayscale value of the inner pixel segment in the non-dense calcified region.

[0030] Preferably, the cloud and fog coefficient of the non-dense calcified region is determined based on the degree of body tissue infiltration in the non-dense calcified region, the standard deviation of adjacent edge pixels in the non-dense calcified region, and the range of the number of pixels in the through-hole pixel segment in the non-dense calcified region, including:

[0031] Calculate the distance between the edge pixels of each non-dense calcified region and the centroid to obtain the first distance corresponding to the edge pixels of each non-dense calcified region.

[0032] The standard deviation of the first distance between the edge pixels of each non-dense calcified region and the adjacent edge pixels is calculated to obtain the first standard deviation of the edge pixels of each non-dense calcified region. The mean of the first standard deviation of the edge pixels of each non-dense calcified region is then calculated to obtain the mean standard deviation.

[0033] The maximum and minimum number of pixels are determined on the centroid-crossing pixel segment passing through the centroid of the non-dense calcified region, and the difference between the maximum and minimum number of pixels is calculated to obtain the range.

[0034] Calculate the product of the standard deviation mean and the range to obtain the adjustment factor;

[0035] The cloud and fog coefficient is obtained by adjusting the degree of infiltration into body tissues using an adjustment coefficient.

[0036] Preferably, the crystallization coefficient is determined based on the distance between some pixels in the non-skeletal connected domain of the non-dense calcified region and the edge pixels of the skeletal connected domain in the non-dense calcified region, as well as the cloudiness coefficient of the non-dense calcified region, including:

[0037] Calculate the minimum distance between pixels in the non-skeletal connected region of the non-dense calcified region and the edge pixels of the skeletal connected region, and calculate the mean of the minimum distance;

[0038] Calculate the ratio of the cloud and fog coefficient to the mean minimum distance, and normalize the ratio to obtain the crystallization coefficient.

[0039] The present invention has the following beneficial effects:

[0040] The information processing module is used to identify corresponding high-density areas in the acquired knee CT images of the target object by utilizing the absorption of X-rays by urate crystals produced by gout; the filtering module is used to identify non-dense calcified areas in the high-density areas based on the gray values ​​of calcified lesions and high-density areas in the knee CT images; the cloudiness coefficient determination module is used to determine the cloudiness coefficient of non-dense calcified areas based on the degree of body tissue infiltration in the non-dense calcified areas, the standard deviation of adjacent edge pixels in the non-dense calcified areas, and the range of the number of pixels in the through-hole pixel segment in the non-dense calcified areas; the crystallization coefficient determination module is used to... The crystallization coefficient is determined based on the distance between pixels in the non-skeletal connected domain of the non-dense calcified region and the edge pixels of the skeletal connected domain in the non-dense calcified region, as well as the cloudiness coefficient of the non-dense calcified region. A labeling module is used to label the non-dense calcified region as a crystalline region in the knee CT image when the crystallization coefficient is greater than a preset crystallization threshold. A recognition module is used to train a preset neural network model for crystallization region recognition using the knee CT images with labeled crystallized regions, resulting in a trained knee gout crystallization region recognition model. This model is used to label crystalline regions in the knee CT images. In this way, high-density regions in the image are extracted using pixel grayscale and gradient changes, and their crystallization coefficient is determined based on the morphology and infiltration degree of these high-density regions. Crystallized regions in the knee CT images are then selected using the crystallization coefficient. By using the knee CT images with labeled crystalline regions as training data to train the knee gout crystallization region recognition model, the influence of calcifications can be eliminated, thereby improving the accuracy of crystalline region recognition. Furthermore, model-based recognition of crystalline regions effectively improves recognition efficiency. Attached Figure Description

[0041] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of an intelligent identification system for gouty crystallization areas of the knee joint based on CT images, provided in one embodiment of the present invention.

[0043] Figure 2 This is a schematic flowchart of an information processing module provided in one embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram illustrating the specific process of the cloud and fog coefficient determination module provided in one embodiment of the present invention. Detailed Implementation

[0045] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an intelligent identification system for gouty crystallization areas of the knee joint based on CT images proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0047] The specific solution of the intelligent identification system for gout crystallization areas of the knee joint based on CT images provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Please see Figure 1 This illustration shows a schematic diagram of an intelligent identification system for gouty crystal regions of the knee joint based on CT images, according to an embodiment of the present invention. In an exemplary embodiment, an intelligent identification system for gouty crystal regions of the knee joint based on CT images is provided, comprising:

[0049] The information processing module 110 is used to determine the corresponding high-density area in the acquired knee joint CT image of the target object by utilizing the absorption of X-rays by urate crystals produced by gout.

[0050] The filtering module 120 is used to identify non-dense calcification areas in the high-density areas based on the gray values ​​of the calcification areas and the high-density areas in the knee joint CT images.

[0051] The cloud and fog coefficient determination module 130 is used to determine the cloud and fog coefficient of the non-dense calcified region based on the degree of body tissue infiltration in the non-dense calcified region, the standard deviation of adjacent edge pixels in the non-dense calcified region, and the range of the number of pixels in the through-pixel segment in the non-dense calcified region.

[0052] The crystallization coefficient determination module 140 is used to determine the crystallization coefficient based on the distance between some region pixels of the non-skeletal connected domain in the non-dense calcified region and the edge pixels of the skeletal connected domain in the non-dense calcified region, as well as the cloud coefficient of the non-dense calcified region.

[0053] The annotation module 150 is used to annotate non-dense calcification areas as crystalline areas in knee CT images when the crystallization coefficient is greater than a preset crystallization threshold.

[0054] The recognition module 160 is used to train a preset neural network model to identify crystal regions using knee CT images with marked crystal regions, thereby obtaining a trained knee joint gout crystal region recognition model; wherein, the knee joint gout crystal region recognition model is used to mark the crystal regions in the knee joint CT images.

[0055] For example, the target subject can be any patient who has taken CT images, or a designated patient; there is no limitation here. It is understood that all knee CT images were obtained with the patient's consent using a multi-slice spiral CT scanner for axial scanning. It should be noted that the scanning range extends from the superior border of the femoral condyle to the inferior border of the tibial plateau. A thin-slice axial image with a slice thickness of 0.625 mm is reconstructed using a standard bone algorithm, and coronal and sagittal multidimensional reconstructed images are generated simultaneously. All raw data and images are exported losslessly in DICOM format. In this embodiment, the knee CT image includes the knee joint portion and the leg region connected to the knee joint portion.

[0056] Preferably, such as Figure 2 As shown, by utilizing the absorption of X-rays by urate crystals produced by gout, the corresponding high-density regions in the acquired knee joint CT images of the target subject are identified, including:

[0057] S210. Using the grayscale values ​​and pixel gradient values ​​of the leg images in knee CT images, the urate crystal region in the leg images is determined.

[0058] S220. Based on the urate crystal area in the leg and the urate crystal area in the knee joint in the leg image, determine the high-density area corresponding to the knee joint CT image.

[0059] For example, because some gout crystals can be confused with calcifications, and gout crystals can be indistinct from body tissues, it is necessary to first pre-identify the leg region in the knee CT images to better identify the gout crystal areas later. Specifically, the knee CT images are input into a pre-trained leg region recognition model, which outputs the leg image from the knee CT images. The leg region recognition model is obtained by training a neural network model using a large number of knee CT training images with labeled leg regions as training data.

[0060] Because urate crystals produced by gout absorb X-rays much more readily than soft tissues in the body, they appear as high-grayscale crystals in images. Therefore, by analyzing the grayscale values ​​and gradient values ​​in a grayscale image, regions of urate crystals in the leg can be identified in leg images.

[0061] In this embodiment, after determining the leg image in the knee joint CT image, a grayscale image corresponding to the leg image is obtained. After obtaining the grayscale image, the leg region needs to be further segmented. First, the maximum value point in the grayscale image is obtained, and then the point with the smallest grayscale value adjacent to this maximum value point is obtained as the segmentation point. It should be noted that this segmentation point can segment out the cortical bone with a maximum grayscale value in the image.

[0062] If the grayscale value of a pixel in a grayscale image is less than the grayscale value of the segmentation point, that pixel is identified as part of the leg region. This process is repeated, comparing the grayscale values ​​of all pixels in the grayscale image with the grayscale values ​​of the segmentation points to obtain the final leg region, resulting in more accurate leg segmentation. The gradient values ​​of each pixel in the leg region are then obtained using Canny edge detection, and the grayscale values ​​of each pixel in the leg region are acquired.

[0063] Optionally, step S210 includes: calculating the ratio of gray value to pixel gradient value of each pixel in the grayscale image of the leg image, and normalizing the ratio to obtain the high density coefficient of each pixel; using the high density coefficient of each pixel to perform Otsu threshold segmentation on the leg image to obtain the urate crystallization region of the leg.

[0064] Specifically, because uric acid crystals have a high absorption rate for X-rays, they appear as high-level grayscale with small gradient changes in grayscale images. Therefore, the high-density coefficient is determined by calculating the ratio of the grayscale value to the pixel gradient value. Taking the i-th pixel as an example, the formula for calculating the high-density coefficient of the i-th pixel is as follows:

[0065]

[0066] in, This represents the high density coefficient of the i-th pixel. This represents the grayscale value of the i-th pixel. This represents the gradient value of the i-th pixel. It should be noted that a larger high-density coefficient indicates a larger grayscale value and a lower gradient for the pixel.

[0067] It should be understood that in the various calculation formulas described in this embodiment, when the denominator in the calculation formula is zero in extreme cases, the system will automatically introduce a positive number with the same dimension and non-zero (e.g., 10^-6) to add to it, in order to prevent the calculation from being meaningless due to division by zero, and to ensure the continuity and feasibility of the data processing flow.

[0068] After determining the high density coefficient of each pixel, use the high density coefficient The patient's leg tissue was segmented using the Otsu threshold method, and the high-threshold portion was selected as the urate crystal crystallization region in the leg.

[0069] Optionally, based on the urate crystal area in the leg and the urate crystal area in the knee joint in the leg images, the corresponding high-density area in the knee joint CT image is determined, including:

[0070] Based on the urate crystal area in the leg and the urate crystal area in the knee joint in the leg images, the high-density candidate area corresponding to the knee joint CT image is determined.

[0071] If the difference between the gray values ​​of edge pixels and the gray values ​​of surrounding pixels in a high-density candidate region meets the preset growth conditions, the surrounding pixels are added to the high-density candidate region to obtain the high-density region.

[0072] For example, a leg image and the high-density coefficients of each pixel are input into a trained model for recognizing urate crystal regions in the knee joint, and the model outputs the urate crystal regions of the knee joint in the leg image. The knee joint urate crystal region recognition model is obtained by training a neural network model using leg images labeled with high-density coefficients as training data.

[0073] Further, step S220 includes: merging the connected components of the urate crystal region in the leg and the urate crystal region in the knee joint in the leg image to obtain the high-density candidate region corresponding to the knee joint CT image.

[0074] Specifically, after identifying the urate crystal region in the leg and the urate crystal region in the knee joint of the leg image, connected component analysis is performed to merge adjacent high-threshold pixels into independent high-density candidate regions. Objects that are too small are then removed, considered noise, thus obtaining the high-density candidate regions corresponding to the knee joint CT image.

[0075] Since the area outside the high-density candidate region is body tissue, there may be areas with small grayscale differences when there are incompletely segmented lesions outside the high-density candidate region. Therefore, further constraints are needed on the high-density candidate region. Taking high-density candidate region A as an example, the edge pixels of the obtained high-density candidate region A are set as seed points for growth, and the standard deviation of the grayscale values ​​of the pixels in high-density candidate region A is calculated. Then, the minimum circumcircle of high-density candidate region A is determined, and the grayscale values ​​of the pixels in the minimum circumcircle that do not belong to high-density candidate region A are determined. Finally, the mean of the grayscale values ​​of the pixels that do not belong to high-density candidate region A is calculated.

[0076] For a high-density candidate region A, starting from its seed point, the surrounding pixels are determined within a 3x3 grid. Pixels within high-density candidate region A are removed from the surrounding pixels, and the remaining pixels are used as growth points. The preset growth conditions are that the grayscale value of the growth point is less than the standard deviation of the grayscale values ​​of the pixels in high-density candidate region A, and the grayscale value of the growth point is greater than the mean grayscale value of the pixels in high-density candidate region A. If the preset growth conditions are met, growth is performed (i.e., the surrounding pixels are added to high-density candidate region A), and the growth point is then used as a seed point to continue growth until growth stops (i.e., the surrounding pixels no longer meet the preset growth conditions). In this way, high-density candidate regions are grown based on the extraction defects of urate crystals in the high-density candidate regions to obtain high-density regions.

[0077] For example, in the high-density region obtained above, since the X-ray absorption rate of some calcifications is similar to that of urate crystals, the high-density region may contain calcifications, so it is necessary to remove the calcified areas. Preferably, based on the gray values ​​of the calcified areas and the high-density region in the knee joint CT image, non-dense calcification areas are identified in the high-density region, including:

[0078] The threshold for dense calcification is determined based on the mean and standard deviation of the gray values ​​in the calcified area.

[0079] If the average gray value of a high-density region is not greater than the dense calcification threshold, the high-density region is defined as a non-dense calcification region.

[0080] For example, blood vessel tracking is performed on knee CT images to identify vascular regions, and areas with CT values ​​≥ 130 HU are identified as calcification lesions. Then, the mean and standard deviation of gray values ​​of the calcification lesion regions are calculated based on the gray values ​​of the pixels in these regions, and the mean gray value of the high-density regions is calculated based on their gray values.

[0081] The formula for calculating the dense calcification threshold is:

[0082]

[0083] in, Indicates the dense calcification threshold. This represents the average gray value of the calcified area. This represents the standard deviation of the grayscale values ​​in the calcified area.

[0084] If the average gray value of a high-density region is greater than the dense calcification threshold, the high-density region is determined to be a dense calcification region; if the average gray value of a high-density region is not greater than the dense calcification threshold, the high-density region is determined to be a non-dense calcification region.

[0085] For example, since non-dense calcification areas contain both ordinary calcification foci and urate crystals, the precipitation characteristics of urate crystals cause some edges of the crystals to infiltrate into body tissue, resulting in a less obvious difference between the urate crystals and the body tissue. Therefore, the degree of infiltration into body tissue can be used to further delineate non-dense calcification areas.

[0086] Preferably, the method for determining the degree of body tissue infiltration in the non-dense calcified region includes: determining the degree of body tissue infiltration in the non-dense calcified region based on the difference in the average grayscale values ​​between the internal and external pixels of the non-dense calcified region, and the difference in the number of pixels in the non-dense calcified region and its corresponding high-density candidate region.

[0087] Optionally, the method for determining the mean grayscale difference between internal and external pixels in the non-dense calcified region includes:

[0088] On the straight line connecting the edge point of the non-dense calcified region and the centroid, select m pixel segments, and define the pixel segments located within the non-dense calcified region as internal pixel segments, and define the pixel segments located outside the non-dense calcified region as external pixel segments.

[0089] The grayscale difference is obtained by considering the average grayscale value of the outer pixel segment and the average grayscale value of the inner pixel segment in the non-dense calcified region. Here, m can be set according to actual needs and is not limited here. For example, m can be 10.

[0090] Specifically, a line is obtained connecting each edge point and the center point in the non-dense calcified region. Ten pixels are then taken from each side of the endpoint of this line, dividing the region into multiple pixel segments. Based on the position of these segments, they are distinguished into internal pixel segments located within the non-dense calcified region and external pixel segments located outside the region. The average grayscale value of each internal and external pixel segment is calculated, and then the difference between the average grayscale value of the internal and external pixel segments is calculated.

[0091] For non-dense calcified regions, first determine the number of internal pixels, then determine the number of internal pixels in the high-density candidate region before growth, and calculate the difference between the number of internal pixels in the non-dense calcified region and the number of internal pixels in the high-density candidate region before growth.

[0092] The formula for calculating the wetting coefficient is as follows:

[0093]

[0094] in, Indicates the wetting coefficient. This represents the difference between the number of internal pixels in a non-dense calcified region and the number of internal pixels in a high-density candidate region before growth. This represents the difference between the average gray value of the inner pixel segment and the average gray value of the outer pixel segment.

[0095] The degree of infiltration of body tissue in the non-dense calcified area is the result obtained after normalizing the infiltration coefficient. The larger the value, the smaller the gray difference between the body tissue area surrounding the non-dense calcified area and the edge of the non-dense calcified area, and the more it grows during the above-mentioned growth process. Therefore, it indicates that the infiltration of the non-dense calcified area is more obvious.

[0096] For example, since urate crystals are formed through continuous precipitation, while calcifications are formed through continuous accumulation, urate crystals in the knee joint will appear as a cloud-like dispersion. However, the edges of these cloud-like urate crystals are relatively irregular and have low sphericity. Therefore, a cloudiness coefficient is used to further screen for crystalline regions. Preferably, as... Figure 3 As shown, the cloud and fog coefficient of the non-dense calcified region is determined based on the degree of body tissue infiltration in the non-dense calcified region, the standard deviation of adjacent edge pixels in the non-dense calcified region, and the range of the number of pixels in the through-hole pixel segment in the non-dense calcified region. This includes:

[0097] S310. Calculate the distance between the edge pixels of each non-dense calcified region and the centroid to obtain the first distance corresponding to the edge pixels of each non-dense calcified region.

[0098] S320. Calculate the standard deviation of the first distance between the edge pixels of each non-dense calcified region and the adjacent edge pixels to obtain the first standard deviation of the edge pixels of each non-dense calcified region, and calculate the mean of the first standard deviation of the edge pixels of each non-dense calcified region to obtain the mean standard deviation.

[0099] S330. Determine the maximum and minimum number of pixels on the centroid-through pixel segment passing through the centroid of the non-dense calcified region, and calculate the difference between the maximum and minimum number of pixels to obtain the range.

[0100] S340. Calculate the product of the mean and range of the standard deviation to obtain the adjustment factor;

[0101] S350. The cloud and fog coefficient is obtained by adjusting the degree of infiltration into body tissues using an adjustment coefficient.

[0102] Specifically, the centroid of each non-dense calcified region is first obtained, and then the distance between the edge pixels of the non-dense calcified region and the centroid is determined, thus obtaining the first distance corresponding to the edge pixels of each non-dense calcified region. Simultaneously, based on the first distance of edge pixel B, the first distance of adjacent edge pixel C, and the first distance of adjacent edge pixel D of the non-dense calcified region, the first standard deviation corresponding to edge pixel B is calculated. This method is then used to calculate the first standard deviation of each edge pixel in each non-dense calcified region. Finally, the mean of the first standard deviations of each edge pixel is calculated to obtain the mean standard deviation of each non-dense calcified region.

[0103] For each centroid passing through the non-dense calcified region, obtain the passing-through line, taking the portion within the non-dense calcified region as the passing-through pixel segment, and determine the number of pixels in each passing-through pixel segment. Obtain the maximum and minimum number of pixels, and calculate the difference between the maximum and minimum number of pixels to obtain the range. It should be noted that the non-dense calcified region is irregular, therefore the range will never be 0.

[0104] The formula for calculating the adjustment factor is as follows:

[0105]

[0106] in, Indicates the adjustment factor. This represents the standard deviation and mean. This indicates the range. It should be noted that... A larger value indicates a larger standard deviation between the distance between the edge pixels and the centroid of the non-dense calcified region and a more disordered growth of the non-dense calcified region, thus indicating a larger cloud and fog coefficient in the non-dense calcified region.

[0107] The formula for calculating the cloud and fog coefficient is as follows:

[0108]

[0109] in, Indicates the cloud and fog coefficient. This represents the wetting coefficient.

[0110] Furthermore, based on the distance between some pixels in the non-skeletal connected domain of the non-dense calcified region and the edge pixels of the skeletal connected domain in the non-dense calcified region, and the cloudiness coefficient of the non-dense calcified region, the crystallization coefficient is determined, including:

[0111] Calculate the minimum distance between pixels in the non-skeletal connected region of the non-dense calcified region and the edge pixels of the skeletal connected region, and calculate the mean of the minimum distance;

[0112] Calculate the ratio of the cloud and fog coefficient to the mean minimum distance, and normalize the ratio to obtain the crystallization coefficient.

[0113] Specifically, cortical bone with extremely high grayscale values ​​is segmented from the previous image, and its edges are extracted using a bone edge extraction model. This bone edge extraction model is obtained by training a neural network model using cortical bone images labeled with bone edges as training data.

[0114] Because urate crystals, compared to bone and joints, appear as a coating, tightly covering the surface of the articular cartilage, non-skeletal connected regions are extracted from the non-dense calcified areas. Then, the minimum distance between pixels in the non-skeletal connected regions and edge pixels in the skeletal connected regions is obtained, and the mean of the minimum distances is calculated.

[0115] The formula for calculating the crystallinity coefficient is as follows:

[0116]

[0117] in, Indicates the crystallinity coefficient. Indicates the cloud and fog coefficient. This represents the minimum mean distance. express The function is used for normalization. It should be noted that... A higher value indicates a greater cloudiness coefficient in the non-dense calcified area and a smaller distance between the non-dense calcified area and the bone joint, thus indicating that it belongs to crystals covering the bone, and therefore its crystallization coefficient is greater.

[0118] Furthermore, the preset crystallization threshold can be set according to actual needs. When the crystallization coefficient is greater than the preset crystallization threshold, the non-dense calcification area is marked as a crystallization area in the knee CT image; when the crystallization coefficient is not greater than the preset crystallization threshold, the non-dense calcification area is not marked in the knee CT image, and it is considered an interference area.

[0119] Furthermore, by processing multiple knee CT images in the above manner, the knee CT images with annotated crystalline areas are used as training data and input into a preset neural network model (such as the U-Net network) for recognition training. In this way, the model learns a large amount of such annotated data and autonomously masters the characteristics of gout crystal areas in CT images, thereby achieving automatic recognition of crystal areas, eliminating the influence of calcifications, and improving the accuracy and efficiency of recognition.

[0120] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0121] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A CT image-based intelligent identification system for gout crystal area of knee joint, characterized in that, The method comprises the following steps: An information processing module is used to determine a corresponding high-density region in a knee joint CT image of a target object by using the absorption of X-rays by urate crystals generated by gout; A screening module is used to determine a non-dense calcification region in the high-density region according to the gray value of the calcification region and the gray value of the high-density region in the knee joint CT image; A cloud and mist coefficient determination module is used to determine the cloud and mist coefficient of the non-dense calcification region according to the body tissue infiltration degree of the non-dense calcification region, the standard deviation of adjacent edge pixel points in the non-dense calcification region, and the range of the number of pixel points of the through-pixel segment in the non-dense calcification region; A crystal coefficient determination module is used to determine the crystal coefficient according to the distance between the partial region pixel points of the non-skeletal connected domain in the non-dense calcification region and the edge pixel points of the skeletal connected domain in the non-dense calcification region, and the cloud and mist coefficient of the non-dense calcification region; A labeling module is used to label the non-dense calcification region as a crystal region in the knee joint CT image when the crystal coefficient is greater than a preset crystal threshold value; An identification module is used to perform crystal region identification training on a preset neural network model by using the knee joint CT image in which the crystal region is labeled, to obtain a trained knee joint gout crystal region identification model; wherein the knee joint gout crystal region identification model is used to label the crystal region in the knee joint CT image.

2. The intelligent identification system for gouty crystallization areas of the knee joint based on CT images according to claim 1, characterized in that, Determine the corresponding high-density region in the knee joint CT image of the target object by using the absorption of X-rays by urate crystals generated by gout, comprising: Determine the leg urate crystal region in the leg image by using the gray value and pixel gradient value of the gray image of the leg image in the knee joint CT image; Determine the corresponding high-density region of the knee joint CT image according to the leg urate crystal region and the urate crystal region of the knee joint in the leg image. 3.The CT image-based gouty tophi region intelligent identification system of the knee joint according to claim 2, characterized in that, Determine the leg urate crystal region in the leg image by using the gray value and pixel gradient value of the gray image of the leg image in the knee joint CT image, comprising: Calculate the ratio of the gray value and the pixel gradient value of each pixel point in the gray image of the leg image, and normalize the ratio to obtain the high-density coefficient of each pixel point; Perform Otsu threshold segmentation on the leg image by using the high-density coefficient of each pixel point to obtain the leg urate crystal region. 4.The CT image-based gouty tophi region intelligent identification system of the knee joint according to claim 2, characterized in that, Determine the corresponding high-density region of the knee joint CT image according to the leg urate crystal region and the urate crystal region of the knee joint in the leg image, comprising: Determine the corresponding high-density candidate region of the knee joint CT image according to the leg urate crystal region and the urate crystal region of the knee joint in the leg image; In the case that the difference between the gray value of the edge pixel point of the high-density candidate region and the gray value of the surrounding pixel point meets the preset growth condition, the surrounding pixel point is added to the high-density candidate region to obtain the high-density region. 5.The CT image-based gouty tophi region intelligent identification system of the knee joint according to claim 4, characterized in that, Determine the corresponding high-density candidate region of the knee joint CT image according to the leg urate crystal region and the urate crystal region of the knee joint in the leg image, comprising: The high-density candidate region corresponding to the knee joint CT image is obtained by merging the connected components of the urate crystal region in the leg and the urate crystal region of the knee joint in the leg image. 6.The CT image-based gouty tophi region intelligent identification system of the knee joint according to claim 1, characterized in that, According to the gray value of the calcification region and the gray value of the high-density region in the knee joint CT image, the non-dense calcification region is determined in the high-density region, including: According to the mean value and standard deviation of the gray value of the calcification region, a dense calcification threshold is determined; In the case where the mean value of the gray value of the high-density region is not greater than the dense calcification threshold, the high-density region is determined as a non-dense calcification region. 7.The CT image-based gouty tophi region intelligent identification system of the knee joint according to claim 4, characterized in that, The method for determining the body tissue infiltration degree of the non-dense calcification region includes: According to the gray mean value difference between the internal and external pixel points of the non-dense calcification region, and the number difference of the pixel points of the non-dense calcification region and its corresponding high-density candidate region, the body tissue infiltration degree of the non-dense calcification region is determined.

8. The intelligent identification system for gouty crystallization areas of the knee joint based on CT images according to claim 7, characterized in that, The method for determining the gray mean value difference between the internal and external pixel points of the non-dense calcification region includes: On the straight line where the edge point of the non-dense calcification region and the center of mass are connected, m pixel point segments are selected, and the pixel segments located in the non-dense calcification region are determined as internal pixel segments, and the pixel segments located outside the non-dense calcification region are determined as external pixel segments; wherein m is a positive integer; According to the gray mean value of the external pixel segment and the gray value mean value of the internal pixel segment of the non-dense calcification region, the gray mean value difference is obtained. 9.The CT image-based gouty tophi region intelligent identification system of the knee joint according to claim 1, characterized in that, According to the body tissue infiltration degree of the non-dense calcification region, the standard deviation of the adjacent edge pixel points in the non-dense calcification region, and the range of the number of pixel points of the through-pixel segment in the non-dense calcification region, the cloud and mist coefficient of the non-dense calcification region is determined, including: The distances between the edge pixel points of each non-dense calcification region and the center of mass are calculated to obtain the first distances corresponding to the edge pixel points of each non-dense calcification region. The first standard deviation of the edge pixel points of each non-dense calcification region is calculated by calculating the first distances corresponding to the edge pixel points and the adjacent edge pixel points of each non-dense calcification region, and the mean value of the first standard deviation of the edge pixel points of each non-dense calcification region is calculated to obtain the standard deviation mean value. The maximum pixel point number and the minimum pixel point number are determined on the through-pixel segment passing through the center of mass of the non-dense calcification region, and the difference between the maximum pixel point number and the minimum pixel point number is calculated to obtain the range. The product of the standard deviation mean value and the range is calculated to obtain the adjustment coefficient. The cloud and mist coefficient is obtained by adjusting the body tissue infiltration degree using the adjustment coefficient. 10.The CT image-based intelligent identification system for gouty tophi region of knee joint according to claim 1, wherein, According to the distance between the part region pixel points of the non-skeletal connected domain in the non-dense calcification region and the edge pixel points of the skeletal connected domain in the non-dense calcification region, and the cloud and mist coefficient of the non-dense calcification region, the crystallization coefficient is determined, including: The minimum distance between the part region pixel points of the non-skeletal connected domain in the non-dense calcification region and the edge pixel points of the skeletal connected domain is calculated, and the minimum distance mean value is calculated. The ratio of the cloud and mist coefficient to the minimum distance mean value is calculated, and the ratio is normalized to obtain the crystallization coefficient.

Images