A method and system for knee joint MR image enhancement

Abstract

The present application relates to the technical field of image data processing, and especially relates to a knee joint MR image enhancement method and system. The method comprises the following steps: determining a noise map and separating it according to the gradient direction similarity between pixels in different scales in the knee joint MR image, and the difference between the amplitude value of the pixels and the maximum amplitude value in the knee joint MR image, to obtain a target knee joint MR image; obtaining a cartilage collagen fiber region according to the ratio of the area of a superpixel block to the area of the minimum circumscribed rectangle of the superpixel block, the difference between the average gray value of the superpixel block and the maximum gray value, and the average gray value difference between the superpixel block and its peripheral pixels; using different coefficients in the Laplace operator to sharpen and enhance the cartilage collagen fiber region and other regions in the target knee joint MR image, to obtain an enhanced knee joint MR image, which effectively improves the accuracy of knee joint MR image enhancement.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and in particular to a method and system for enhancing MR images of the knee joint. Background Technology

[0002] Knee osteoarthritis is a degenerative disease. Before using platelet-rich plasma to treat knee osteoarthritis, it is necessary to use knee magnetic resonance imaging (MR) to examine the knee cartilage. MR can clearly show the morphology, structure and lesions of soft tissues, providing efficient and objective diagnostic support for clinical practice.

[0003] During the capture of MR images of the knee joint, radio frequency pulses are used to excite hydrogen nuclei within the body, and the radio frequency signals emitted by these nuclei are received, generating a two-dimensional intensity matrix. The value of each pixel in the knee joint MR image represents the signal intensity at that location; the higher the signal intensity, the brighter the corresponding pixel; and the lower the signal intensity, the darker the corresponding pixel. To reduce the lack of detail in MR images caused by the complex structure of the knee joint, the Laplacian operator is used to sharpen and enhance the knee joint MR image, improving image contrast and clarity by highlighting high-frequency details (such as edges and textures).

[0004] However, the Laplacian operator is a global edge enhancement algorithm, which cannot accurately distinguish between useful high-frequency regions that need to be identified (such as the boundary between cartilage and bone) and useless low-frequency regions that do not need to be enhanced (such as randomly distributed colored noise points generated by random thermal noise of the device, and the microscopic arrangement of cartilage collagen fibers). This leads to the enhancement of useless low-frequency regions as well, resulting in misjudgment.

[0005] Therefore, how to enhance only the useful high-frequency regions on knee joint MR images when using the Laplacian operator to enhance images is a problem that urgently needs to be solved. Summary of the Invention

[0006] To address the technical problem of how to enhance only the useful high-frequency regions on knee joint MR images when using the Laplacian operator for image enhancement, this invention provides a method and system for enhancing knee joint MR images.

[0007] In a first aspect, the present invention provides a method for enhancing knee joint MR images, employing the following technical solution:

[0008] A method for enhancing MR images of the knee joint, comprising the following steps:

[0009] The gradient directions of each pixel in the knee joint MR image are obtained at different scales. Based on the similarity of the gradient directions between different scales and the difference between the pixel's amplitude value and the maximum amplitude value in the knee joint MR image, the degree to which a pixel belongs to noise is determined. A noise map is generated in response to the comparison between the degree to which a pixel belongs to noise and a noise threshold. This noise map is then separated from the knee joint MR image to obtain the target knee joint MR image. The target knee joint MR image is divided into multiple superpixel blocks, and the area of ​​each superpixel block is determined. The gradient direction of each superpixel block is compared with the maximum amplitude value in the knee joint MR image. The probability that a superpixel belongs to cartilage collagen fibers is obtained by comparing the ratio of the area of ​​the small bounding rectangle and the difference between the mean and maximum gray values ​​of the superpixel. In response to the comparison between the probability of a superpixel belonging to cartilage collagen fibers and a probability threshold, candidate cartilage collagen fiber regions are obtained. Based on the difference in the mean gray values ​​between the candidate cartilage collagen fiber regions and their surrounding pixels, the cartilage collagen fiber region is determined. In the Laplacian operator, different coefficients are used to sharpen and enhance the cartilage collagen fiber region and other regions in the target knee joint MR image, resulting in an enhanced knee joint MR image.

[0010] This invention uses the Laplacian operator to sharpen and enhance knee joint MR images, highlighting details and improving image clarity. During image enhancement, this invention considers that different regions in the knee joint MR image have different enhancement needs, such as cartilage collagen fibers. Therefore, by analyzing the grayscale features, brightness features, and local contrast features of superpixel blocks in the image, this invention can accurately calculate and identify cartilage collagen fiber regions and other regions in the image. This allows for precise enhancement based on the sharpening needs of different regions, effectively improving the accuracy of knee joint MR image enhancement. Before identifying cartilage collagen fiber regions, this invention also considers that noise in the knee joint MR image can affect the accuracy of cartilage collagen fiber region identification and may affect subsequent diagnosis. Therefore, this invention analyzes the degree to which the amplitude and gradient features of each pixel in the image match noise features, accurately removing noise from the knee joint MR image, thereby effectively improving the accuracy and efficiency of knee joint MR image enhancement.

[0011] According to the present invention, a method for enhancing a knee joint MR image includes, prior to obtaining the gradient direction of each pixel in the knee joint MR image at different scales, converting the knee joint MR image into a spectrum image by Fourier transform to obtain the amplitude value of each pixel and the maximum amplitude value in the knee joint MR image.

[0012] According to the present invention, a method for enhancing knee joint MR images includes obtaining the gradient direction of each pixel in the knee joint MR image at different scales, which includes obtaining the gradient direction of each pixel at different scales using the Sober operator.

[0013] According to a knee joint MR image enhancement method provided by the present invention, the step of obtaining the degree to which a pixel belongs to noise based on the gradient direction similarity of pixels at different scales and the difference between the amplitude value of the pixel and the maximum amplitude value in the knee joint MR image includes: recording the mean cosine similarity between the gradient direction of each pixel at one scale and the gradient direction at the next scale as the similarity index of the pixel; calculating the degree to which the i-th pixel belongs to noise. :

[0014] ;

[0015] , These are the amplitude value and similarity index of the i-th pixel, respectively. The maximum amplitude value in the knee joint MR image. It is an exponential function with base e.

[0016] This invention takes into account that noise in knee joint MR images manifests as high-frequency signals in the frequency domain and appears chaotic and inconsistent at different scales. Therefore, by analyzing the amplitude of each pixel and the similarity of gradient directions at different scales, the degree to which each pixel belongs to noise can be accurately calculated.

[0017] According to a method for enhancing a knee joint MR image provided by the present invention, generating a noise map in response to a comparison result of the degree to which a pixel belongs to noise and a noise threshold includes: taking pixels in the knee joint MR image whose degree to which they belong to noise is greater than the noise threshold as noise points; and generating a noise map based on the noise points in the knee joint MR image.

[0018] According to the present invention, a method for enhancing a knee joint MR image includes dividing the target knee joint MR image into multiple superpixel blocks and determining the area of ​​each superpixel block, comprising: determining seed points in the target knee joint MR image; performing region growing starting from the seed points to finally divide the target knee joint MR image into multiple superpixel blocks; and using the number of pixels in a superpixel block as the area of ​​the superpixel block.

[0019] This invention takes into account that the cartilage collagen fiber region has similar features, and superpixel blocks can group pixels with similar features together. Therefore, this invention divides the target knee joint MR image into multiple superpixel blocks and analyzes the features within each superpixel block, which effectively reduces the amount of data processing while completely identifying cartilage collagen fibers.

[0020] According to a method for enhancing MR images of a knee joint provided by the present invention, the step of obtaining the probability that a superpixel block belongs to cartilage collagen fibers based on the ratio of the area of ​​the superpixel block to the area of ​​its minimum bounding rectangle and the difference between the mean gray value and the maximum gray value of the superpixel block includes: filtering out superpixel blocks whose difference between the mean gray value of the superpixel block and the maximum gray value of all superpixel blocks is not greater than a difference threshold; calculating the probability that the superpixel block belongs to cartilage collagen fibers based on the ratio of the area of ​​the superpixel block to the area of ​​its minimum bounding rectangle and the difference between the mean gray value and the maximum gray value of all superpixel blocks; and calculating the probability that the superpixel block belongs to cartilage collagen fibers based on the ratio of the area of ​​the superpixel block to the area of ​​its minimum bounding rectangle and the difference between the mean gray value and the maximum gray value of all superpixel blocks. The probability that a superpixel belongs to cartilage collagen fiber :

[0021] ;

[0022] For the first The average grayscale value of each superpixel block The maximum grayscale value. , The first The area of ​​each superpixel block and the area of ​​the minimum bounding rectangle. It is an exponential function with base e.

[0023] This invention takes into account that the shape of the cartilage collagen fiber region is close to rectangular and its grayscale characteristics are darker than other regions. Therefore, by analyzing the grayscale characteristics and rectangularity of each superpixel block, the possibility of each superpixel block being a cartilage collagen fiber can be accurately assessed, thus preparing for the subsequent accurate screening of cartilage collagen fibers.

[0024] According to a method for enhancing MR images of the knee joint provided by the present invention, the step of obtaining the cartilage collagen fiber region based on the difference in gray-level mean between the candidate cartilage collagen fiber region and its surrounding pixels includes: recording the normalized value of the difference in gray-level mean between the candidate cartilage collagen fiber region and its surrounding pixels as the gray-level contrast of the candidate cartilage collagen fiber region; if the gray-level contrast of any candidate cartilage collagen fiber region is not less than 0, then after sorting the gray-level contrasts of the candidate cartilage collagen fiber regions with gray-level contrasts not less than 0, selecting the candidate cartilage collagen fiber regions with gray-level contrasts greater than or equal to the third quartile as the cartilage collagen fiber regions.

[0025] This invention takes into account an important characteristic of the cartilage collagen fiber region: it has a strong grayscale contrast compared to the surrounding area and is relatively brighter than the surrounding cartilage matrix. Therefore, this invention accurately obtains the grayscale contrast of each candidate cartilage collagen fiber region by analyzing the grayscale difference between the candidate cartilage collagen fiber region and its surrounding pixels, so that the cartilage collagen fiber region in the target knee joint MR image can be accurately obtained based on this feature.

[0026] According to a method for enhancing knee joint MR images provided by the present invention, the method involves using different coefficients to sharpen and enhance the cartilage collagen fiber region and other regions in the target knee joint MR image using the Laplacian operator. This includes: using the cartilage collagen fiber region as a low-sharpening mask; denoteing the region in the edge image of the knee joint MR image after removing the low-sharpening mask as a high-sharpening mask; and using the remaining region other than the low-sharpening and high-sharpening masks as a medium-sharpening mask. The Laplacian operator then uses low-sharpening coefficients, medium-sharpening coefficients, and high-sharpening coefficients respectively to sharpen and enhance the low-sharpening mask, medium-sharpening mask, and high-sharpening mask. The low-sharpening coefficient is smaller than the medium-sharpening coefficient, and the medium-sharpening coefficient is smaller than the high-sharpening coefficient.

[0027] Secondly, the present invention provides a knee joint MR image enhancement system, which adopts the following technical solution:

[0028] A knee joint MR image enhancement system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned knee joint MR image enhancement method is implemented.

[0029] By adopting the above technical solution, a computer program is generated from the above-mentioned method for enhancing MR images of the knee joint and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0030] The present invention has the following technical effects:

[0031] Based on the above technical solutions, this invention provides a method and system for enhancing knee joint MR images. By using the Laplacian operator to sharpen and enhance knee joint MR images, it can highlight details and improve image clarity. During image enhancement, this invention analyzes the grayscale features, brightness features, and local contrast features of superpixel blocks in the image to accurately calculate and identify cartilage collagen fiber regions and other regions. This allows for precise enhancement tailored to the sharpening needs of different regions, effectively improving the accuracy of knee joint MR image enhancement. Before identifying the cartilage collagen fiber regions, this invention also analyzes the degree to which the amplitude and gradient features of each pixel in the image conform to noise characteristics, accurately removing noise from the knee joint MR image, thereby effectively improving the accuracy and efficiency of knee joint MR image enhancement. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating a method for enhancing MR images of the knee joint, as provided in an embodiment of the present invention. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0034] To enhance only the useful high-frequency regions in knee joint MR images when using the Laplacian operator, this invention discloses a knee joint MR image enhancement method. This method analyzes the probability that each superpixel block in the knee joint MR image belongs to the cartilage collagen fiber region, thereby obtaining the cartilage collagen fiber region and other regions in the knee joint MR image. This allows for the accurate setting of the corresponding sharpening coefficient, effectively improving the accuracy of knee joint MR image enhancement.

[0035] Please see details. Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for enhancing MR images of the knee joint according to an embodiment of the present invention. The method specifically includes the following steps:

[0036] S1: Obtain the grayscale value, amplitude value, and gradient direction of each pixel in the knee joint MR image at different scales.

[0037] It should be noted that during the acquisition of knee joint MR images, random thermal noise inside the acquisition device can cause noise interference to the knee joint MR images. This noise not only belongs to the useless low-frequency region in the knee joint MR images that does not need to be enhanced, but also affects the subsequent diagnostic results.

[0038] Based on this, in this embodiment of the invention, before identifying the cartilage collagen fiber region in the knee joint MR image, the knee joint MR image can be denoised first. While denoising, the details and clear edges of the image need to be preserved.

[0039] It is understandable that noise manifests as a high-frequency signal in the frequency domain, and the gradient direction of the knee joint edge in a knee joint MR image is usually relatively consistent at different scales, while noise is random and independent, and its gradient direction will be chaotic at different scales.

[0040] Therefore, embodiments of the present invention can accurately calculate the probability that each pixel is noise by combining the amplitude value of each pixel to identify high-frequency noise components in the image and the gradient direction consistency of pixels at different scales.

[0041] For example, in an embodiment of the present invention, obtaining the grayscale value and amplitude value of each pixel in the knee joint MR image includes: converting the knee joint MR image into a spectrum by Fourier transform to obtain the amplitude value of each pixel and the maximum amplitude value in the knee joint MR image.

[0042] Specifically, MR images of the knee joint can be acquired by MRI3d-Dess cartilage scanning technology and then processed into grayscale to obtain the knee joint MR image. The grayscale value range of the pixels in the knee joint MR image is [0, 255]. The parameters during the scanning process can be set as follows: field of view 160mm, slice thickness 3mm, repetition time 3900ms, echo time 43ms, matrix 256×320. The scanning parameters can be set according to actual needs, and the embodiments of the present invention do not impose too many restrictions here.

[0043] For example, when converting a knee joint MR image into a spectrogram using Fourier transform to obtain the amplitude values ​​of each pixel and the maximum amplitude value in the knee joint MR image, a spectrogram can be constructed with frequency as the horizontal axis and the amplitude value corresponding to each frequency as the vertical axis to obtain the amplitude value of each pixel; the maximum amplitude value in the spectrogram is then obtained as the maximum amplitude value in the knee joint MR image.

[0044] The step of converting the knee joint MR image into a spectrogram through Fourier transform can be obtained by existing technology, and will not be described in detail here in the embodiments of the present invention.

[0045] For example, in an embodiment of the present invention, obtaining the gradient direction of each pixel in a knee joint MR image at different scales includes: using different scales in the Sober operator to obtain the gradient direction of each pixel at different scales.

[0046] In the Sober operator, the scale is determined by the size of the Gaussian kernel. The size of each scale can be obtained through the formula of the scale sequence, which usually increases in the order of 2.

[0047] In this embodiment of the invention, the number of scales can be set to 3, and the initial value of the scales can be 0, resulting in sizes of 1, 2, and 4 for each scale; the specific settings can be adjusted according to actual needs. The formula for the scale sequence and the specific steps for obtaining the gradient direction of the pixel at each scale can be obtained through existing technologies, and will not be elaborated upon here.

[0048] After obtaining the amplitude value and gradient direction of each pixel based on the above steps, the probability that each pixel is noise data can be accurately obtained by combining frequency features and gradient features, thereby accurately identifying noise points in the knee joint MR image, i.e., performing the following steps.

[0049] S2: The degree to which a pixel belongs to noise is obtained based on the gradient direction similarity of pixels at different scales and the difference between the amplitude value of a pixel and the maximum amplitude value in the knee joint MR image.

[0050] It should be noted that noise in images typically manifests as random, irregular fluctuations. Therefore, in knee joint MR images, the maximum amplitude value usually represents extreme noise rather than a true representation of the tissue signal. The smaller the difference between each pixel in a knee joint MR image and the maximum amplitude value, and the more chaotic the gradient directions of the pixels at different scales, indicating poor similarity, the greater the likelihood that the pixel is noise.

[0051] Based on this, embodiments of the present invention can accurately determine the degree to which a pixel belongs to noise by considering the gradient direction similarity of pixels at different scales and the difference between the amplitude value of a pixel and the maximum amplitude value in the knee joint MR image.

[0052] For example, in an embodiment of the present invention, the degree to which a pixel belongs to noise is obtained based on the gradient direction similarity of the pixel between different scales and the difference between the amplitude value of the pixel and the maximum amplitude value in the knee joint MR image. The method further includes: recording the mean cosine similarity between the gradient direction of each pixel at one scale and the gradient direction at the next scale as the similarity index of the pixel.

[0053] The similarity index is used to characterize the consistency of gradient direction of pixels at different scales.

[0054] Here's an example of how to obtain the similarity index for a pixel: There are 3 scales, numbered 0, 1, and 2. Calculate the cosine similarity between the gradient directions of scale 0 and scale 1, scale 1 and scale 2, and scale 2 and scale 3 respectively. Use the average of the three cosine similarities as the similarity index for that pixel.

[0055] For example, to calculate the degree to which a pixel belongs to noise, please refer to the following formula:

[0056] ;

[0057] Let represent the degree to which the i-th pixel belongs to noise. Let be the amplitude value of the i-th pixel. Let be the similarity index for the i-th pixel. The maximum amplitude value in the knee joint MR image. It is an exponential function with base e, where e is the natural constant.

[0058] In the above formula, Let be the normalized representation of the frequency domain features of the i-th pixel. If the difference between the amplitude value of the i-th pixel and the maximum amplitude value... The smaller the value, the greater the probability that the i-th pixel is noise in the frequency domain.

[0059] This indicates the consistency of the gradient direction of the i-th pixel at different scales. The smaller this value, the greater the disorder of the gradient direction of the i-th pixel at different scales in the spatial domain, the smaller the similarity, and the greater the possibility that it belongs to noise.

[0060] In summary, the smaller the gradient direction similarity of a pixel across different scales and the smaller the difference between the pixel's amplitude value and the maximum amplitude value in the knee joint MR image, the greater the likelihood that the pixel is noise.

[0061] Thus, by combining frequency and gradient features, this embodiment of the invention can more comprehensively identify different types of noise, including high-frequency noise and random noise caused by devices. It avoids misclassifying normal texture details as noise, thereby preserving useful information in the image.

[0062] After obtaining the degree of noise for each pixel based on the above steps, the noise in the knee joint MR image can be accurately determined according to the degree of noise for each pixel.

[0063] S3: Generate a noise map in response to the comparison between the degree to which a pixel belongs to noise and the noise threshold. Separate the noise map from the knee joint MR image to obtain the target knee joint MR image. Divide the target knee joint MR image into multiple superpixel blocks and determine the area of ​​each superpixel block.

[0064] The noise threshold can be set to 0.6; the specific noise threshold can be set according to actual needs.

[0065] For example, in an embodiment of the present invention, generating a noise map in response to a comparison result of the degree to which a pixel belongs to noise and a noise threshold includes: taking pixels in the knee joint MR image that belong to noise at a degree greater than the noise threshold as noise points; and generating a noise map based on the noise points in the knee joint MR image.

[0066] Understandably, if the degree to which a pixel in a knee joint MR image belongs to noise is not greater than the noise threshold, it means that such pixels do not conform to the frequency domain characteristics and gradient characteristics of noise and do not need to be removed. The value of each pixel in the noise image is the degree to which each pixel belongs to noise.

[0067] After separating the noise map from the knee joint MR image based on the above steps, a noise-free target knee joint MR image can be obtained. In the target knee joint MR image, the cartilage collagen fiber region has specific morphological and texture features. Superpixel blocks can group pixels with similar features together to form more meaningful regional units. This allows the superpixel blocks to accurately capture the boundaries and internal features of these regions when analyzing the cartilage collagen fiber region, avoiding the information fragmentation problem that may occur when analyzing based on a single pixel point, and helping to more completely identify the structure of cartilage collagen fibers. Therefore, this embodiment of the invention can divide the target knee joint MR image into multiple superpixel blocks, and analyze each superpixel block to accurately select superpixel blocks that conform to the characteristics of cartilage collagen fibers.

[0068] For example, in an embodiment of the present invention, the target knee joint MR image is divided into multiple superpixel blocks, and the area of ​​each superpixel block is determined, including: determining seed points in the target knee joint MR image; performing region growing with the seed points as the starting points, and finally dividing the target knee joint MR image into multiple superpixel blocks; and using the number of pixels in the superpixel block as the area of ​​the superpixel block.

[0069] The size of the seed point can be set to 3×3; the specific size of the seed point can be set according to actual needs, and this embodiment of the invention does not impose too many restrictions here.

[0070] Specifically, every four pixels, a 3×3 pixel block is selected on the target knee joint MR image as a seed point. For each 3×3 seed point, its gray value variance is calculated. Seed points with gray value variance greater than a preset threshold are selected as starting points for region growth, and finally multiple superpixel blocks are obtained.

[0071] The preset threshold can be set according to actual needs.

[0072] After dividing the target knee joint MR image into multiple superpixel blocks based on the above steps, each superpixel block can be precisely analyzed to accurately determine the probability that it belongs to cartilage collagen fibers, i.e., the following steps are performed.

[0073] S4: Based on the ratio of the area of ​​the superpixel block to the area of ​​the minimum bounding rectangle of the superpixel block, and the difference between the average gray value and the maximum gray value of the superpixel block, the probability that the superpixel block belongs to cartilage collagen fiber is obtained.

[0074] It should be noted that cartilage collagen fibers are linear or ribbon-like structures. These structures appear as long, continuous, rectangular lines in MR images. Furthermore, collagen fibers typically exhibit high grayscale contrast with the surrounding cartilage matrix, and their boundaries are usually well-defined. Therefore, collagen fibers appear as relatively bright areas in images, while the surrounding cartilage matrix appears darker, with a clear transition in grayscale values ​​at the boundaries, resulting in distinct line edges in MR images.

[0075] Based on this, embodiments of the present invention can accurately determine the probability that a superpixel block belongs to cartilage collagen fibers by using the ratio of the area of ​​the superpixel block to the area of ​​the minimum bounding rectangle of the superpixel block and the difference between the average gray value and the maximum gray value of the superpixel block.

[0076] It should be further noted that the superpixel blocks obtained based on the above steps include both cartilage collagen fiber regions and brighter remaining regions. The brighter remaining regions are clearly non-cartilage collagen fiber regions, and the superpixel blocks obtained based on region growth are not regularly shaped, with low rectangularity, which is significantly inconsistent with the brightness and rectangular characteristics of cartilage collagen fiber regions. Therefore, before calculating the probability that each superpixel block belongs to cartilage collagen fibers based on the following steps, non-cartilage collagen fiber regions that clearly do not conform to the characteristics of cartilage collagen fibers can be screened out first, thereby effectively improving data processing efficiency and reducing the amount of data processed.

[0077] For example, in an embodiment of the present invention, the probability that a superpixel block belongs to cartilage collagen fiber is obtained based on the ratio of the area of ​​the superpixel block to the area of ​​the minimum bounding rectangle of the superpixel block and the difference between the mean gray value and the maximum gray value of the superpixel block. The method further includes filtering out superpixel blocks whose difference between the mean gray value of the superpixel block and the maximum mean gray value of all superpixel blocks is not greater than a difference threshold.

[0078] The difference threshold can be set to 0.5; the specific difference threshold can be set according to actual needs, and this embodiment of the invention does not impose too many restrictions here.

[0079] Specifically, when obtaining the difference between the average gray value in a superpixel block and the maximum average gray value of all superpixel blocks, the absolute value of the difference between the average gray value in a superpixel block and the maximum average gray value of all superpixel blocks can be normalized to obtain the difference between the average gray value in a superpixel block and the maximum average gray value of all superpixel blocks. After filtering out non-cartilage collagen fiber superpixel blocks that obviously do not conform to the characteristics of cartilage collagen fibers based on the difference, the following steps are continued to calculate the probability that each superpixel block belongs to cartilage collagen fibers.

[0080] For example, in an embodiment of the present invention, the probability of a superpixel block belonging to cartilage collagen fibers is calculated, as shown in the following formula:

[0081] ;

[0082] For the first The probability that a superpixel belongs to cartilage collagen fiber. For the first The average grayscale value of each superpixel block The maximum grayscale value. For the first The area of ​​each superpixel block For the first The area of ​​the minimum bounding rectangle of a superpixel block. It is an exponential function with base e.

[0083] In the above formula, Indicates the first The rectangularity of the nth superpixel block; the closer this value is to 1, the more rectangular the nth superpixel block. The closer the minimum bounding rectangle of a superpixel block is to a rectangle, the more it conforms to the shape characteristics of cartilage collagen fibers.

[0084] Indicates the first The normalized approximation value between the mean and maximum gray values ​​of the nth superpixel block. Since the gray values ​​of cartilage collagen fibers belong to a region with relatively low overall gray values ​​in the image, a smaller this value indicates a higher gray level. The closer the average grayscale value of each superpixel block is to the maximum grayscale value, the less it conforms to the brightness characteristics of cartilage collagen fibers, even at this point. It is close to 1 and does not conform to the characteristics of cartilage collagen fibers, therefore the corresponding number is... The lower the probability that a superpixel belongs to cartilage collagen fibers, the lower the probability. Conversely, the higher the probability, the lower the probability that the superpixel belongs to cartilage collagen fibers. The smaller the average gray value of each superpixel block is compared to the maximum gray value, the more it matches the brightness characteristics of cartilage collagen fibers, and the greater the probability that it belongs to cartilage collagen fibers.

[0085] Based on the above steps, the probability of each superpixel block belonging to cartilage collagen fibers can be obtained. Based on the probability of a superpixel block belonging to cartilage collagen fibers, the superpixel blocks that belong to cartilage collagen fibers can be accurately identified.

[0086] S5: In response to the comparison result of the probability of a superpixel block belonging to cartilage collagen fiber and the probability threshold, a candidate cartilage collagen fiber region is obtained, the peripheral pixels of the candidate cartilage collagen fiber region are determined, and the cartilage collagen fiber region is obtained based on the difference in grayscale mean between the candidate cartilage collagen fiber region and its peripheral pixels.

[0087] For example, when setting the probability threshold, the probability of all superpixel blocks belonging to cartilage collagen fibers can be sorted, and the third quartile of the sorted value can be used as the probability threshold.

[0088] The third quartile of the sorted data can be obtained by linear interpolation. The specific steps can be obtained by existing technology, and will not be described in detail in this embodiment of the invention.

[0089] Understandably, because cartilage collagen fibers exhibit relatively high local contrast, if a superpixel block corresponds to a complete cartilage collagen fiber region, the brightness inside the superpixel block will be higher than the brightness of its surrounding pixels. Therefore, if the brightness inside a superpixel block is not less than the brightness of its surrounding pixels, it indicates that the characteristics of the candidate cartilage collagen fiber region conform to the local characteristics of cartilage collagen fibers; conversely, it indicates that the characteristics of the candidate cartilage collagen fiber region do not conform to the local characteristics of cartilage collagen fibers.

[0090] Based on this, the true cartilage collagen fiber region can be obtained from all candidate cartilage collagen fiber regions.

[0091] For example, in an embodiment of the present invention, obtaining a cartilage collagen fiber region based on the difference in the mean grayscale value between a candidate cartilage collagen fiber region and its surrounding pixels includes: recording the normalized value of the difference in the mean grayscale value between the candidate cartilage collagen fiber region and its surrounding pixels as the grayscale contrast of the candidate cartilage collagen fiber region; if the grayscale contrast of any candidate cartilage collagen fiber region is not less than 0, then after sorting the grayscale contrasts of the candidate cartilage collagen fiber regions with grayscale contrasts not less than 0, selecting the candidate cartilage collagen fiber regions with grayscale contrasts greater than or equal to the third quartile as the cartilage collagen fiber regions.

[0092] Among them, the outermost pixels of the candidate cartilage collagen fiber region are the pixels immediately adjacent to the outer ring of the candidate cartilage collagen fiber region.

[0093] Understandably, if the grayscale contrast of any candidate cartilage collagen fiber region is not less than 0, it means that the average grayscale value of the candidate cartilage collagen fiber region is not less than the average grayscale value among its surrounding pixels, and the features of the candidate cartilage collagen fiber region conform to the local features of cartilage collagen fibers. If the grayscale contrast of any candidate cartilage collagen fiber region is not less than 0, it means that at least one candidate cartilage collagen fiber region conforms to the local features of cartilage collagen fibers. In this case, the grayscale contrast of all candidate cartilage collagen fiber regions with a grayscale contrast not less than 0 can be sorted, and then the candidate cartilage collagen fiber regions can be screened. Conversely, if the grayscale contrast of a candidate cartilage collagen fiber region is less than 0, it means that it does not conform to the local features of cartilage collagen fibers.

[0094] Thus, the embodiments of the present invention reflect the distribution of gray-scale contrast data of all candidate regions through the third quartile, thereby accurately selecting candidate cartilage collagen fiber regions with higher gray-scale contrast from all candidate cartilage collagen fiber regions with gray-scale contrast not less than 0 as cartilage collagen fiber regions, reducing the influence of other factors on the recognition results.

[0095] After obtaining the cartilage collagen fiber region in the target knee joint MR image based on the above steps, different degrees of sharpening and enhancement can be performed on different regions, i.e., the following steps are performed.

[0096] S6: In the Laplacian operator, different coefficients are used to sharpen and enhance the cartilage collagen fiber region and other regions in the target knee joint MR image to obtain the enhanced knee joint MR image.

[0097] It should be noted that in the target knee joint MR image, the cartilage collagen fiber area is a low-intensity enhancement low-sharpening area to avoid affecting subsequent diagnosis; the edges in the target knee joint MR image are the true edges of the anatomical structure and belong to the high-sharpening areas that require high enhancement to significantly improve the edge clarity of the knee joint; the remaining areas other than the low-sharpening and high-sharpening areas are medium-sharpening areas, and a medium sharpening coefficient needs to be set to balance the detail preservation effect.

[0098] For example, in an embodiment of the present invention, different coefficients are used to sharpen and enhance the cartilage collagen fiber region and other regions in the target knee joint MR image in the Laplacian operator, including: using the cartilage collagen fiber region as a low-sharpening mask, denoteing the region after removing the low-sharpening mask in the edge image of the knee joint MR image as a high-sharpening mask, and using the remaining region other than the low-sharpening mask and the high-sharpening mask as a medium-sharpening mask; and using low-sharpening coefficients, medium-sharpening coefficients, and high-sharpening coefficients respectively in the Laplacian operator to sharpen and enhance the low-sharpening mask, medium-sharpening mask, and high-sharpening mask.

[0099] The low sharpening factor is smaller than the medium sharpening factor, and the medium sharpening factor is smaller than the high sharpening factor. The high-sharpening mask and the medium-sharpening mask together form the other regions.

[0100] Specifically, edge maps of knee joint MR images can be obtained using Canny edge detection.

[0101] For example, the range of low sharpening factor can be 0.1 to 0.4; the range of medium sharpening factor can be 0.5 to 0.6; and the range of high sharpening factor can be 0.7 to 1.0.

[0102] For example, in an embodiment of the present invention, a standard Laplacian convolution kernel can be used to convolve the noise-removed target knee joint MR image to generate a Laplacian response map.

[0103] For example, an embodiment of the present invention provides a formula for calculating the sharpening result, which can be seen in the following relationship:

[0104] ;

[0105] This indicates the sharpening result of the target knee joint MR image. This represents an MR image of the target knee joint. Represents the Laplace response diagram. This represents the sharpening factor.

[0106] The sharpening factor can be low, medium, or high.

[0107] Thus, this embodiment of the invention has accurately sharpened and enhanced the denoised target knee joint MR image.

[0108] As can be seen, in this embodiment of the invention, when enhancing a knee joint MR image, the gradient direction of each pixel in the knee joint MR image at different scales can be obtained; based on the similarity of the gradient directions of pixels at different scales and the difference between the amplitude value of the pixel and the maximum amplitude value in the knee joint MR image, the degree to which a pixel belongs to noise is obtained; in response to the comparison result of the degree to which a pixel belongs to noise and a noise threshold, a noise map is generated, and the noise map is separated from the knee joint MR image to obtain the target knee joint MR image; the target knee joint MR image is divided into multiple superpixel blocks, and the area of ​​each superpixel block is determined; based on the area of ​​the superpixel block and the... The probability that a superpixel block belongs to cartilage collagen fibers is obtained by comparing the ratio of the area of ​​the minimum bounding rectangle of the superpixel block and the difference between the mean and maximum gray values ​​of the superpixel block. In response to the comparison between the probability of a superpixel block belonging to cartilage collagen fibers and a probability threshold, candidate cartilage collagen fiber regions are obtained. Based on the difference in the mean gray values ​​between the candidate cartilage collagen fiber regions and their surrounding pixels, the cartilage collagen fiber regions are then identified. In the Laplacian operator, different coefficients are used to sharpen and enhance the cartilage collagen fiber regions and other regions in the target knee joint MR image, resulting in an enhanced knee joint MR image, effectively improving the accuracy of knee joint MR image enhancement.

[0109] This invention also discloses a knee joint MR image enhancement system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a knee joint MR image enhancement method provided by this invention.

[0110] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0111] In this invention, the aforementioned memory can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0112] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for enhancing MR images of the knee joint, characterized in that, include: Obtain the gradient direction of each pixel in the knee joint MR image at different scales; The degree to which a pixel belongs to noise is determined based on the gradient direction similarity of pixels across different scales and the difference between the pixel's amplitude value and the maximum amplitude value in the knee joint MR image. This includes: recording the mean cosine similarity between the gradient directions of each pixel at one scale and the gradient directions at the next scale as the similarity index of that pixel; and calculating the degree to which the i-th pixel belongs to noise. : ; , These are the amplitude value and similarity index of the i-th pixel, respectively. The maximum amplitude value in the knee joint MR image. It is an exponential function with base e; A noise map is generated in response to the comparison between the degree to which a pixel belongs to noise and a noise threshold. The noise map is then separated from the knee joint MR image to obtain the target knee joint MR image. The target knee joint MR image is then divided into multiple superpixel blocks, and the area of ​​each superpixel block is determined. Based on the ratio of the area of ​​a superpixel block to the area of ​​its minimum bounding rectangle, and the difference between the average grayscale value and the maximum grayscale value of the superpixel block, the probability that the superpixel block belongs to cartilage collagen fiber is obtained, including: filtering out superpixel blocks whose average grayscale value differs from the maximum average grayscale value of all superpixel blocks by no more than a difference threshold; calculating the... The probability that a superpixel belongs to cartilage collagen fiber : ; For the first The average grayscale value of each superpixel block The maximum grayscale value. , The first The area of ​​each superpixel block and the area of ​​the minimum bounding rectangle; In response to the comparison between the probability of a superpixel block belonging to cartilage collagen fibers and the probability threshold, candidate cartilage collagen fiber regions are obtained. Based on the difference in grayscale mean between the candidate cartilage collagen fiber regions and their surrounding pixels, the cartilage collagen fiber regions are obtained. In the Laplacian operator, different coefficients are used to sharpen and enhance the cartilage collagen fiber regions and other regions in the target knee joint MR image, resulting in the enhanced knee joint MR image.

2. The method for enhancing MR images of the knee joint according to claim 1, characterized in that, The step of obtaining the gradient direction of each pixel in the knee joint MR image at different scales includes, prior to: The knee joint MR image is converted into a spectrogram by Fourier transform to obtain the amplitude value of each pixel and the maximum amplitude value in the knee joint MR image.

3. The method for enhancing MR images of the knee joint according to claim 1, characterized in that, The acquisition of the gradient direction of each pixel in the knee joint MR image at different scales includes: The gradient direction of each pixel at different scales is obtained by using different scales in the Sober operator.

4. The method for enhancing MR images of the knee joint according to claim 1, characterized in that, The generation of a noise map in response to a comparison between the degree to which a pixel belongs to noise and a noise threshold includes: Pixels in the knee joint MR image that are considered noise with a noise level greater than the noise threshold are identified as noise points; a noise map is generated based on the noise points in the knee joint MR image.

5. The method for enhancing MR images of the knee joint according to claim 1, characterized in that, The step of dividing the target knee joint MR image into multiple superpixel blocks and determining the area of ​​each superpixel block includes: Seed points are identified in the target knee joint MR image; region growing is performed starting from the seed points to divide the target knee joint MR image into multiple superpixel blocks; the number of pixels in a superpixel block is used as the area of ​​that superpixel block.

6. The method for enhancing MR images of a knee joint according to claim 1, characterized in that, The step of obtaining the cartilage collagen fiber region based on the difference in average grayscale values ​​between the candidate cartilage collagen fiber region and its surrounding pixels includes: The normalized value of the difference between the mean gray level of the candidate cartilage collagen fiber region and its surrounding pixels is recorded as the gray level contrast of the candidate cartilage collagen fiber region. If the gray level contrast of any candidate cartilage collagen fiber region is not less than 0, the gray level contrasts of the candidate cartilage collagen fiber regions with gray level contrasts not less than 0 are sorted, and the candidate cartilage collagen fiber regions with gray level contrasts greater than or equal to the third quartile are selected as the cartilage collagen fiber regions.

7. The method for enhancing MR images of the knee joint according to claim 1, characterized in that, The method of sharpening and enhancing the cartilage collagen fiber region and other regions in the target knee joint MR image using different coefficients in the Laplacian operator includes: The cartilage collagen fiber region was used as a low-sharpening mask, the region in the edge image of the knee joint after removing the low-sharpening mask was designated as a high-sharpening mask, and the remaining region excluding the low-sharpening and high-sharpening masks was designated as a medium-sharpening mask. The Laplacian operator was used to sharpen and enhance the low-sharpening, medium-sharpening, and high-sharpening masks respectively. The low-sharpening coefficient was smaller than the medium-sharpening coefficient, and the medium-sharpening coefficient was smaller than the high-sharpening coefficient.

8. A knee joint MR image enhancement system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a knee joint MR image enhancement method according to any one of claims 1-7.

Images