A MRI image analysis method for lumbar lesion detection
By using local similarity screening and adaptive thresholds to obtain the nucleus pulposus region of the lumbar spine in the detection of lumbar spine, the problem of incomplete boundary segmentation of the nucleus pulposus region in the prior art is solved, and the accuracy of lumbar spine judgment and treatment plan are improved. Suitability.
Patent Information
- Application Number
- CN202411497382.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2024-12-31
- Estimated Expiration
- 2044-10-25
AI Technical Summary
The existing lumbar lesion detection methods are based on a simple regional growth method. The use of a single threshold leads to incomplete boundary segmentation of the nucleus pulposus region, affecting the accuracy of judgment.
By acquiring the initial growth points in the MRI image, the target growth points are screened according to the local similarity, and the nucleus pulposus area of the lumbar vertebra is obtained according to the adaptive threshold of the target growth points, reducing the incomplete boundary segmentation caused by the fixed threshold.
The accuracy of judging the nucleus pulposterior region of lumbar spondylosis is improved, so that the nucleus pulposterior region of lumbar spondylosis can be obtained completely, thereby improving the accuracy of doctors in judging the degree of lumbar spondylosis and the appropriateness of treatment plans by the doctor.
Smart Images
Figure CN119006479B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to an MRI image analysis method for lumbar lesion detection. Background Art
[0002] In MRI images, the shape of the nucleus pulposus of lumbar lesions depends on the nature and severity of the lesions. The nucleus pulposus of a herniated disc is usually irregular in shape, while degeneration is flat in shape. Protruding and ruptured nuclei pulposus show different morphological characteristics in the images, which help doctors accurately assess the severity of the lesions and develop appropriate treatment plans. Since the shape of the nucleus pulposus of lumbar lesions in MRI images may have unclear boundaries or complex shapes, and there may be significant differences in the brightness of the internal areas of the nucleus pulposus, there are many interferences in judging the nucleus pulposus area of lumbar lesions.
[0003] The existing lumbar lesion detection method is based on simple region growing, which determines the nucleus pulposus region of lumbar lesions by setting a growth threshold. However, the threshold used in the simple region growing method is a single threshold, which cannot represent the growth degree of the nucleus pulposus at each position. When performing nucleus pulposus segmentation, it will lead to incomplete boundary segmentation, and thus it is impossible to obtain the complete nucleus pulposus region of lumbar lesions, which reduces the accuracy of doctors' judgment on the degree of lumbar lesions.
[0004] Therefore, how to obtain accurate and complete nucleus pulposus area of the lumbar spine becomes an urgent problem to be solved. Summary of the invention
[0005] In view of this, an embodiment of the present invention provides an MRI image analysis method for lumbar lesion detection to solve the problem of how to obtain accurate and complete nucleus pulposus areas of the lumbar spine.
[0006] An embodiment of the present invention provides an MRI image analysis method for lumbar lesion detection, the method comprising the following steps:
[0007] Acquire an MRI image of the lumbar spine, and preprocess the MRI image to obtain a grayscale image;
[0008] Acquire an initial growth point in the grayscale image, take the initial growth point as the center, acquire a first local area including at least one suspected non-growth point according to a preset initial growth threshold, screen the suspected non-growth points in the first local area according to the local similarity between each suspected non-growth point in the first local area and the previous grown point in its growth direction to obtain a target growth point, and acquire an adaptive threshold of each target growth point according to the data difference between each target growth point and the grown point in the first local area;
[0009] For any target growth point, according to the adaptive threshold of the target growth point, a second local area including at least one suspected non-growth point is obtained, wherein the area of the second local area is smaller than the area of the first local area, a to-be-grown point in the second local area is obtained, the to-be-grown point is used as the next target growth point, and the method for obtaining the to-be-grown point is repeated until no to-be-grown point is obtained, thereby obtaining all to-be-grown points corresponding to the target growth point;
[0010] All the waiting growth points corresponding to each of the target growth points are obtained, and the initial growth points, the grown points and the target growth points in the first local area, and all the waiting growth points corresponding to each of the target growth points form a nucleus pulposus area of the lumbar vertebra in the grayscale image.
[0011] Preferably, taking the initial growth point as the center and obtaining the first local area including at least one suspected non-growth point according to a preset initial growth threshold value comprises:
[0012] Taking the initial growth point as the center and the neighborhood of the initial growth point as the growth direction, for any growth direction of the initial growth point, if there is any pixel point in the growth direction and the grayscale value difference between the pixel point and the previous pixel point is less than or equal to the preset initial growth threshold, then the pixel point is taken as a grown point; if there is any pixel point in the growth direction and the grayscale value difference between the pixel point and the previous pixel point is greater than the preset initial growth threshold, then the pixel point is taken as a suspected non-growth point; obtain the suspected non-growth points in each growth direction of the initial growth point, and establish a first local area with the initial growth point as the center and the suspected non-growth points in each growth direction of the initial growth point as the boundary.
[0013] Preferably, the screening of the suspected non-growth points in the first local area to obtain the target growth points according to the local similarity between each suspected non-growth point in the first local area and the last grown point in the growth direction thereof comprises:
[0014] For any suspected non-growth point in the first local area, in the growth direction of the suspected non-growth point, obtain a grown point adjacent to the suspected non-growth point, record it as the previous grown point, and obtain the local similarity between the suspected non-growth point and the previous grown point;
[0015] When the local similarity between the suspected non-growth point and the previous grown point is greater than or equal to a preset local similarity threshold, all grown points in the opposite growth direction of the suspected non-growth point are obtained, and all grown points in the opposite growth direction of the suspected non-growth point and the suspected non-growth point are combined into a growth group, and a local similarity prediction value of the suspected non-growth point is obtained according to the data difference of each pixel point in the growth group, and when the local similarity prediction value of the suspected non-growth point is within a preset range, the suspected non-growth point is recorded as a target growth point.
[0016] Preferably, the obtaining of the local similarity between the suspected non-growing point and the last grown point includes:
[0017] Taking the suspected non-growth point and the previous grown point as the center, respectively, windows of preset sizes are established to obtain a suspected non-growth point window and a grown point window;
[0018] According to the data difference of the pixel points in the suspected non-growth point window and the grown point window, the similarity of the grayscale distribution of the suspected non-growth point window and the grown point window is obtained;
[0019] Acquiring the structural similarity between the suspected non-growth point window and the grown point window as the window structural similarity between the suspected non-growth point window and the grown point window;
[0020] According to the grayscale distribution similarity and the window structure similarity between the suspected non-growing point window and the grown point window, the local similarity between the suspected non-growing point and the last grown point is obtained.
[0021] Preferably, obtaining the grayscale distribution similarity between the suspected non-growth point window and the grown point window according to the data difference between the pixel points in the suspected non-growth point window and the grown point window includes:
[0022] The grayscale value difference between each pixel in the suspected non-growing point window and the pixel at the same position in the grown point window is obtained respectively to form a grayscale value difference sequence, the square value of each grayscale value difference in the grayscale value difference sequence is obtained, and the square value mean is obtained accordingly, and the reciprocal of the square value mean is normalized to obtain the grayscale distribution similarity between the suspected non-growing point window and the previous grown point window.
[0023] Preferably, obtaining the local similarity between the suspected non-growth point and the previous grown point according to the grayscale distribution similarity and the window structure similarity between the suspected non-growth point window and the grown point window includes:
[0024] The addition result between the grayscale distribution similarity and the window structure similarity of the suspected non-growth point window and the grown point window is obtained as the local similarity between the suspected non-growth point and the last grown point.
[0025] Preferably, obtaining the local similarity prediction value of the suspected non-growth point according to the data difference of each pixel point in the growth group includes:
[0026] According to the data difference of each pixel point in the growth group, the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is obtained, the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is fitted to obtain a local similarity fitting function, and the local similarity fitting function is used to predict the local similarity between the suspected non-growth point and the previous grown point to obtain a local similarity prediction value of the suspected non-growth point.
[0027] Preferably, obtaining the adaptive threshold of each target growth point according to the data difference between each target growth point and the already grown points in the first local area includes:
[0028] For any target growth point, obtain the local similarity between the target growth point and the previous grown point in its growth direction, obtain the absolute value of the difference between the local similarity prediction value of the target growth point and the local similarity between the target growth point and the previous grown point in its growth direction as the error coefficient of the target growth point, and obtain the adaptive threshold of the target growth point according to the local similarity between the target growth point and the previous grown point in its growth direction, the error coefficient of the target growth point and a preset initial growth threshold.
[0029] Preferably, the step of obtaining the adaptive threshold of the target growth point according to the local similarity between the target growth point and the previous growth point in its growth direction, the error coefficient of the target growth point and a preset initial growth threshold comprises:
[0030] Obtaining a first weight coefficient of the local similarity between the target growth point and the previous grown point in its growth direction, obtaining a first range difference of the local similarity between the target growth point and the previous grown point in its growth direction, calculating the product of the first weight coefficient, the inverse of the first range difference, and the local similarity between the target growth point and the previous grown point in its growth direction to obtain a first product, calculating the product of the first weight coefficient, the inverse of the first range difference, and a preset local similarity threshold to obtain a second product, and calculating the difference between the first product and the second product to obtain a first difference;
[0031] Obtaining a second weight coefficient of the error coefficient of the target growth point, obtaining a second range difference of the error coefficient, calculating a third product as a product of the second weight coefficient, the inverse of the second range difference of the error coefficient, and the error coefficient, and calculating a difference between the second weight coefficient and the third product to obtain a second difference;
[0032] A sum of the preset initial growth threshold, the first difference and the second difference is obtained as an adaptive threshold of the target growth point.
[0033] Preferably, the step of obtaining the to-be-grown point in the second local area includes:
[0034] The suspected non-growth points in the second local area are screened to obtain target growth points in the second local area, and the target growth points in the second local area are used as the to-be-grown points in the second local area.
[0035] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:
[0036] The present invention obtains an MRI image of the lumbar spine, preprocesses the MRI image to obtain a grayscale image; obtains an initial growth point in the grayscale image, takes the initial growth point as the center, and obtains a first local area including at least one suspected non-growth point according to a preset initial growth threshold; based on the local similarity between each suspected non-growth point in the first local area and the previous grown point in its growth direction, the suspected non-growth points in the first local area are screened to obtain target growth points; based on the data difference between each target growth point and the grown point in the first local area, the adaptive threshold of each target growth point is respectively obtained; for any target growth point , according to the adaptive threshold of the target growth point, a second local area including at least one suspected non-growth point is obtained, wherein the second local area is smaller than the first local area, a to-be-grown point in the second local area is obtained, the to-be-grown point is used as the next target growth point, and the method for obtaining the to-be-grown point is repeated until no to-be-grown point is obtained, and all to-be-grown points corresponding to the target growth point are obtained; all to-be-grown points corresponding to each of the target growth points are obtained, and the initial growth point, the grown points and the target growth point in the first local area, and all to-be-grown points corresponding to each of the target growth points are combined to form a nucleus pulposus area of the lumbar vertebra in the grayscale image. The present invention determines suspected non-growth points by presetting an initial growth threshold, and then screens the suspected non-growth points to obtain target growth points, and further determines the points to be grown in the area where the target growth points are located by obtaining an adaptive threshold of the target growth points. The adaptive threshold of the target growth points can better represent the growth degree of the location where the target growth points are located, reduces the incomplete segmentation of the nucleus pulposus area boundary caused by a fixed growth threshold, improves the accuracy of judging the nucleus pulposus area of lumbar lesions, and enables the nucleus pulposus area of lumbar lesions to be completely obtained, thereby improving the accuracy of doctors' judgment on the degree of lumbar lesions in patients and formulating more appropriate treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0038] Figure 1 is a method flow chart of an MRI image analysis method for lumbar lesion detection provided by Embodiment 1 of the present invention;
[0039] Figure 2 This is a schematic diagram of a growth group provided in Example 1 of the present invention. DETAILED DESCRIPTION
[0040] Embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.
[0041] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.
[0042] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.
[0043] See also Figure 1 , is a method flow chart of an MRI image analysis method for lumbar lesion detection provided by Embodiment 1 of the present invention, such as Figure 1 As shown, the method may include:
[0044] Step S101, acquiring an MRI image of the lumbar spine, and preprocessing the MRI image to obtain a grayscale image.
[0045] MRI images of the lumbar spine were obtained. In order to reduce unnecessary calculations and interference caused by noise, bilateral filtering was used to denoise the MRI images. The denoised MRI images were then converted into grayscale images. The grayscale images were read using OpenCV, and the grayscale value of each pixel in the grayscale images was extracted.
[0046] Step S102, obtaining an initial growth point in the grayscale image, taking the initial growth point as the center, obtaining a first local area including at least one suspected non-growth point according to a preset initial growth threshold, screening the suspected non-growth points in the first local area according to the local similarity between each suspected non-growth point in the first local area and the previous grown point in its growth direction to obtain target growth points, and obtaining an adaptive threshold for each target growth point according to the data difference between each target growth point and the grown point in the first local area.
[0047] When performing regional growth, it is first necessary to obtain the initial growth point based on the grayscale image. To obtain an image of the nucleus pulposus region, the growth effect is best when the initial growth point is fixed at the approximate center of the nucleus pulposus, so it is necessary to set the initial growth point at the approximate center of the nucleus pulposus based on experience. In one embodiment, by observing the grayscale image of the lumbar region, it can be known that the approximate center of the nucleus pulposus is located at the intersection of one-third of the height (H) and one-half of the width (W) in the grayscale image, that is, the coordinates of the initial growth point are determined to be represented by (H / 3, W / 2), and the initial growth threshold k is set to 10 based on experience. There is no limitation here and it can be set according to the specific implementation scenario.
[0048] Because different forms of the nucleus pulposus region represent different degrees of pathological changes in the lumbar region, segmenting the nucleus pulposus region can more conveniently observe the degree of pathological changes in the lumbar region. The method of using a fixed growth threshold to determine the lumbar nucleus pulposus region is to select an initial growth point during regional growth, obtain a fixed growth threshold through experience, perform regional growth according to the fixed growth threshold, and finally determine the lumbar nucleus pulposus region. Since the lumbar nucleus pulposus region appears as a high signal in the MRI image, specifically as a clustered distribution with high brightness in the center gradually decreasing to the surrounding brightness, the initial growth point in the method of using a fixed growth threshold to determine the lumbar nucleus pulposus region has a general growth effect during regional growth, making it difficult to accurately segment the lumbar nucleus pulposus region.
[0049] Therefore, when performing regional growth at the initial growth point, it is necessary to first obtain the first local area through the preset initial growth threshold, then screen the suspected non-growth points in the first local area to obtain the target growth point, and finally obtain the adaptive threshold of the target growth point. The subsequent lumbar nucleus pulposus area is judged through the adaptive threshold of the target growth point.
[0050] In a grayscale image, with an initial growth point as the center, regional growth is performed according to a preset initial growth threshold to obtain a first local area including at least one suspected non-growth point: with the initial growth point as the center and the neighborhood of the initial growth point as the growth direction, for any growth direction of the initial growth point, if there is any pixel point in the growth direction and the grayscale value difference between the pixel point and the previous pixel point is less than or equal to the preset initial growth threshold, the pixel point is regarded as a grown point; if there is any pixel point in the growth direction and the grayscale value difference between the pixel point and the previous pixel point is greater than the preset initial growth threshold, the pixel point is regarded as a suspected non-growth point; the suspected non-growth points in each growth direction of the initial growth point are obtained, and a first local area with the initial growth point as the center and the suspected non-growth points in each growth direction of the initial growth point as the boundary is established.
[0051] In one embodiment, with the initial growth point as the center and the four neighborhoods of the initial growth point as the growth directions, that is, each neighborhood of the initial growth point represents a growth direction, the suspected non-growth points on each growth direction of the initial growth point are obtained, and all the suspected non-growth points are connected one by one to form a boundary. The first local area is formed by connecting all the suspected non-growth points one by one with the initial growth point as the center.
[0052] After obtaining the first local area, it is necessary to screen the suspected non-growth points in the first local area to obtain target growth points according to the local similarity between each suspected non-growth point in the first local area and the last grown point in its growth direction.
[0053] The method of obtaining the target growth point is as follows:
[0054] (1) For any suspected non-growth point in the first local area, a grown point adjacent to the suspected non-growth point in the growth direction of the suspected non-growth point is obtained, recorded as the previous grown point, and the local similarity between the suspected non-growth point and the previous grown point is obtained.
[0055] Among them, obtaining the local similarity between the suspected non-growing point and the last grown point includes:
[0056] (a) Establishing windows of preset sizes with the suspected non-growth point and the previous grown point as centers respectively, to obtain a suspected non-growth point window and a grown point window.
[0057] In one implementation, a window with a size of 5×5 is established with the suspected non-growth point and the last grown point as the center, respectively. There is no limitation here, and the suspected non-growth point window and the grown point window can be obtained according to the specific implementation scenario.
[0058] (b) obtaining a grayscale distribution similarity between the suspected non-growth point window and the grown point window according to a data difference between pixels in the suspected non-growth point window and the grown point window.
[0059] Specifically, the grayscale value difference between each pixel point in the suspected non-growing point window and the pixel point at the same position in the grown point window is obtained respectively to form a grayscale value difference sequence, the square value of each grayscale value difference in the grayscale value difference sequence is obtained, and the corresponding square value mean is obtained, and the inverse of the square value mean is normalized to obtain the grayscale distribution similarity between the suspected non-growing point window and the previous grown point window.
[0060] In one embodiment, the grayscale distribution similarity between the suspected non-growth point window 1 and the grown point window 2 is calculated:
[0061]
[0062] in, is the grayscale distribution similarity between the suspected non-growth point window 1 and the grown point window 2; is the gray value of the i-th pixel in the suspected non-growth point window 1; is the gray value of the i-th pixel in the grown point window 2; i is the sequence number of the pixel in the suspected non-growing point window 1 and the grown point window 2; is the number of pixels in the suspected non-growing point window 1 and the grown point window 2; norm() is the normalization function.
[0063] It should be noted that, in the suspected non-growth point window 1 and the grown point window 2, the smaller the difference in grayscale values of the pixels at the same position is, the more similar the grayscale distributions in the suspected non-growth point window 1 and the grown point window 2 are, and the greater the similarity of the grayscale distributions is; the greater the difference in grayscale values of the pixels at the same position is, the more dissimilar the grayscale distributions in the suspected non-growth point window 1 and the grown point window 2 are, and the smaller the similarity of the grayscale distributions is.
[0064] (c) Obtaining structural similarity between the suspected non-growth point window and the grown point window as the window structural similarity between the suspected non-growth point window and the grown point window.
[0065] In one embodiment, the structural similarity (SSIM) of the suspected non-growing point window 1 and the grown point window 2 is obtained, and the calculation formula of the structural similarity is:
[0066]
[0067] in, is the window structure similarity between the suspected non-growing point window 1 and the grown point window 2; is the mean gray value of the pixels in the suspected non-growth point window 1; is the mean gray value of the pixels in the grown point window 2; is the variance of the grayscale value of the pixel in the suspected non-growth point window 1; is the variance of the grayscale values of the pixels in the grown point window 2; is the covariance of the grayscale values of suspected non-growth point window 1 and grown point window 2; is a constant 1, It is a constant 3, which is not limited here and can be set according to the specific implementation scenario.
[0068] Since structural similarity belongs to the prior art, it will not be described here in detail, and the structural similarity (ie, SSIM) of the suspected non-growth point window 1 and the grown point window 2 is taken as the window structural similarity of the suspected non-growth point window 1 and the grown point window 2.
[0069] (d) obtaining a local similarity between the suspected non-growth point and the previous grown point according to a grayscale distribution similarity and a window structure similarity between the suspected non-growth point window and the grown point window.
[0070] Specifically, the addition result between the grayscale distribution similarity and the window structure similarity of the suspected non-growing point window and the grown point window is obtained as the local similarity between the suspected non-growing point and the previous grown point.
[0071] In one embodiment, the local similarity between the suspected non-growth point and the last grown point is calculated:
[0072]
[0073] in, is the local similarity between the suspected non-growth point and the last grown point; is the grayscale distribution similarity between the suspected non-growth point window 1 and the grown point window 2; It is the window structure similarity between the suspected non-growing point window 1 and the grown point window 2.
[0074] It should be noted that the local similarity is composed of the grayscale distribution similarity and the window structure similarity. The greater the grayscale distribution similarity, the more similar the grayscale distribution of pixels in the suspected non-growth point window 1 and the grown point window 2, and the greater the local similarity; the greater the window structure similarity, the more similar the window structures in the suspected non-growth point window 1 and the grown point window 2, and the greater the local similarity.
[0075] (2) When the local similarity between the suspected non-growth point and the previous grown point is greater than or equal to a preset local similarity threshold, all grown points in the opposite growth direction of the suspected non-growth point are obtained, and all grown points in the opposite growth direction of the suspected non-growth point and the suspected non-growth point are combined into a growth group. According to the data difference of each pixel point in the growth group, a local similarity prediction value of the suspected non-growth point is obtained. When the local similarity prediction value of the suspected non-growth point is within a preset range, the suspected non-growth point is recorded as a target growth point.
[0076] After obtaining the local similarity, the local similarity threshold f is set to 1.5. There is no restriction here and it can be set according to the specific implementation scenario. When the local similarity is less than 1.5, the suspected non-growth point is considered to be a non-growth point (that is, it does not belong to the lumbar nucleus pulposus area); when the local similarity is greater than or equal to 1.5, in order to reduce accidental similarity, a secondary judgment is required for the suspected non-growth point to confirm whether the suspected non-growth point is the target growth point for subsequent regional growth. Specifically, all grown points in the opposite direction of the growth of the suspected non-growth point are obtained, and all grown points in the opposite direction of the growth of the suspected non-growth point and the suspected non-growth point are combined into a growth group. According to the data difference of each pixel point in the growth group, a local similarity prediction value of the suspected non-growth point is obtained, and a preset range is set to [1.5, 2]. There is no limitation here, and it can be set according to the specific implementation scenario. When the local similarity prediction value is greater than the maximum value 2 of the local similarity, or the local similarity prediction value is less than the preset local similarity threshold 1.5, the suspected non-growth point is considered to be a non-growth point; when the local similarity prediction value is in the preset range [1.5, 2], the suspected non-growth point is recorded as a target growth point.
[0077] Among them, according to the data difference of each pixel in the growth group, the specific method of obtaining the local similarity prediction value of the suspected non-growth point is:
[0078] According to the data difference of each pixel point in the growth group, the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is obtained, the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is fitted to obtain a local similarity fitting function, and the local similarity fitting function is used to predict the local similarity between the suspected non-growth point and the previous grown point to obtain a local similarity prediction value of the suspected non-growth point.
[0079] In one embodiment, if Figure 2 Shown is a schematic diagram of a growth group. It is a suspected non-growth point, and the direction of the arrow is the growth direction of the suspected non-growth point. are all the growing points in the opposite direction of the growth of the suspected non-growing point. A growth group is formed, and the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is obtained according to the above-mentioned method for obtaining the local similarity between the suspected non-growth point and the previous grown point. The least squares method is used to perform data fitting on all local similarities of the growth group to obtain a local similarity fitting function, wherein the least squares method belongs to the prior art and will not be described in detail here.
[0080] The local similarity fitting function is used to predict the local similarity between the suspected non-growth point and the last grown point to obtain the local similarity prediction value.
[0081] According to the above-mentioned method for judging suspected non-growth points, all target growth points in the first local area are obtained. After the target growth points are obtained, it is necessary to obtain the adaptive threshold of each target growth point, and the subsequent lumbar nucleus pulposus area is judged by the adaptive threshold of the target growth point.
[0082] The method for obtaining the adaptive threshold of each target growth point in the first local area is as follows:
[0083] (1) For any target growth point, obtain the local similarity between the target growth point and the previous grown point in its growth direction, and obtain the absolute value of the difference between the local similarity prediction value of the target growth point and the local similarity between the target growth point and the previous grown point in its growth direction as the error coefficient of the target growth point.
[0084] In one embodiment, the error coefficient of the target growth point is calculated:
[0085]
[0086] Among them, r is the error coefficient of the target growth point; is the local similarity prediction value of the target growth point; It is the local similarity between the target growth point and the previous growth point in its growth direction.
[0087] It should be noted that the smaller the difference between the local similarity prediction value of the target growth point and the local similarity between the target growth point and the previous grown point in its growth direction, the more the target growth point conforms to the growth law of the growth group, and the smaller the error coefficient of the target growth point; the greater the difference between the local similarity prediction value of the target growth point and the local similarity between the target growth point and the previous grown point in its growth direction, the less the target growth point conforms to the growth law of the growth group, and the larger the error coefficient of the target growth point.
[0088] Since the maximum value of local similarity is 2, the range of the local similarity prediction value of the target growth point and the local similarity between the target growth point and the previous grown point in its growth direction belongs to [f, 2], and the range of the error coefficient of the target growth point is [0, 2-f].
[0089] (2) Obtaining an adaptive threshold of the target growth point according to a local similarity between the target growth point and a previous growth point in its growth direction, an error coefficient of the target growth point, and a preset initial growth threshold.
[0090] Specifically, a first weight coefficient of the local similarity between the target growth point and the previous grown point in its growth direction is obtained, a first range difference of the local similarity between the target growth point and the previous grown point in its growth direction is obtained, the product of the first weight coefficient, the inverse of the first range difference, and the local similarity between the target growth point and the previous grown point in its growth direction is calculated to obtain a first product, the product of the first weight coefficient, the inverse of the first range difference, and a preset local similarity threshold is calculated to obtain a second product, and the difference between the first product and the second product is calculated to obtain a first difference;
[0091] Obtaining a second weight coefficient of the error coefficient of the target growth point, obtaining a second range difference of the error coefficient, calculating a third product as a product of the second weight coefficient, the inverse of the second range difference of the error coefficient, and the error coefficient, and calculating a difference between the second weight coefficient and the third product to obtain a second difference;
[0092] A sum of the preset initial growth threshold, the first difference and the second difference is obtained as an adaptive threshold of the target growth point.
[0093] In one embodiment, since the adaptive threshold needs to be set in a suitable range to ensure that the region growth is carried out normally, it is necessary to set an adaptive threshold upper and lower limits a, which are set according to experience as , which is not limited here and can be set according to the specific implementation scenario; since the local similarity has a greater impact on the target growth point, the first weight coefficient assigned to the local similarity according to the experimental data is , the second weight coefficient of the error coefficient is ; The range of local similarity between the target growth point and the previous growth point in its growth direction belongs to [f, 2], then the first range difference of local similarity between the target growth point and the previous growth point in its growth direction is ; The range of the error coefficient of the target growth point is [0, 2-f], then the second range difference of the error coefficient is , calculate the adaptive threshold of the target growth point:
[0094]
[0095] in, is the adaptive threshold of the target growth point; k is the preset initial growth threshold; f is the preset local similarity threshold; r is the error coefficient of the target growth point; is the local similarity between the target growth point and the previous growth point in its growth direction; a is the upper and lower limits of the adaptive threshold of the target growth point.
[0096] It should be noted that a ensures that the adaptive threshold is within a suitable range to ensure that the regional growth proceeds normally; the greater the local similarity between the target growth point and the previous growth point in its growth direction, the more the target growth point conforms to the growth law of the growth group, and the first difference The larger the error coefficient is, the larger the adaptive growth threshold of the target growth point is; the larger the error coefficient is, the less the target growth point conforms to the growth law of the growth group, and the second difference The smaller it is, the smaller the adaptive growth threshold of the target growth point is.
[0097] At this point, the grown points and the target growing points in the first local area can be obtained, and the adaptive threshold of the target growing point can be obtained.
[0098] Step S103, for any target growth point, according to the adaptive threshold of the target growth point, obtain a second local area including at least one suspected non-growth point, wherein the second local area is smaller than the first local area, obtain the to-be-grown point in the second local area, take the to-be-grown point as the next target growth point, repeat the method of obtaining the to-be-grown point until no to-be-grown point is obtained, and obtain all the to-be-grown points corresponding to the target growth point.
[0099] After obtaining the adaptive threshold of the target growth point, for any target growth point in the first local area, with the target growth point as the center, according to the adaptive threshold of the target growth point, obtain a second local area including at least one suspected non-growth point, wherein the area of the second local area is smaller than the area of the first local area, and according to the method for obtaining the target growth point in the first local area, the suspected non-growth points in the second local area are screened to obtain the target growth point in the second local area, and the target growth point in the second local area is used as the to-be-grown point in the second local area, and the to-be-grown point in the second local area is used as the next target growth point, and according to the method for obtaining all target growth points in the first local area, the to-be-grown point corresponding to the next target growth point is continuously obtained, and so on, until no to-be-grown point is obtained, and all to-be-grown points corresponding to the target growth point in the first local area are obtained. Similarly, all to-be-grown points corresponding to each target growth point in the first local area are obtained respectively, that is, all target growth points.
[0100] At this point, all the to-be-grown points corresponding to each target growing point in the first local area are obtained.
[0101] Step S104, obtaining all the to-be-grown points corresponding to each of the target growth points, and combining the initial growth points, the grown points and the target growth points in the first local area, and all the to-be-grown points corresponding to each of the target growth points to form a nucleus pulposus area of the lumbar vertebra in the grayscale image.
[0102] Since all the to-be-growing points corresponding to each target growth point in the first local area are target growth points, all the to-be-growing points corresponding to each target growth point in the first local area are acquired, and the initial growth points, the grown points and the target growth points in the first local area, and all the to-be-growing points corresponding to each target growth point constitute the nucleus pulposus area of the lumbar vertebra in the grayscale image.
[0103] At this point, the nucleus pulposus region of the lumbar spine in the grayscale image is obtained.
[0104] Since the normal lumbar nucleus pulposus is a regular ellipse, the shape of the nucleus pulposus area of the diseased lumbar spine is complex and has obvious differences compared to the normal nucleus pulposus. Through the above operation, a segmented image of the nucleus pulposus area is obtained, and the edge shape of the nucleus pulposus area is obtained by morphological processing (opening and closing operations). The segmented nucleus pulposus edge shape is handed over to the doctor, and the doctor is assisted in judging the lesions of the lumbar MRI image by comparing the edge morphology. Among them, morphological processing belongs to the existing technology and will not be repeated here.
[0105] In summary, the present embodiment acquires an MRI image of the lumbar spine, preprocesses the MRI image to obtain a grayscale image; acquires an initial growth point in the grayscale image, takes the initial growth point as the center, acquires a first local area including at least one suspected non-growth point according to a preset initial growth threshold, screens the suspected non-growth points in the first local area according to the local similarity between each suspected non-growth point in the first local area and the previous grown point in its growth direction to obtain a target growth point, and acquires an adaptive threshold value of each target growth point according to the data difference between each target growth point and the grown point in the first local area; for any target Growth point, according to the adaptive threshold of the target growth point, obtain a second local area including at least one suspected non-growth point, wherein the second local area is smaller than the first local area, obtain the to-be-grown point in the second local area, take the to-be-grown point as the next target growth point, repeat the method for obtaining the to-be-grown point until no to-be-grown point is obtained, and obtain all the to-be-grown points corresponding to the target growth point; obtain all the to-be-grown points corresponding to each of the target growth points, and combine the initial growth point, the grown points and the target growth point in the first local area, and all the to-be-grown points corresponding to each of the target growth points to form the nucleus pulposus area of the lumbar vertebra in the grayscale image. This embodiment preliminarily determines the suspected non-growth point through the initial growth point and the initial growth threshold, and screens the suspected non-growth point to obtain the target growth point by obtaining the local similarity between the suspected non-growth point and the previous grown point in its growth direction, and obtains the target growth point by obtaining the adaptive threshold of the target growth point, and obtains the to-be-grown point in the local area where the target growth point is located, and then obtains the lumbar nucleus pulposus area in the grayscale image, and judges by the adaptive threshold, which improves the accuracy of judging the nucleus pulposus area of lumbar lesions, so that the nucleus pulposus area of lumbar lesions can be completely obtained, thereby improving the accuracy of doctors' judgment on the degree of lumbar lesions in patients and formulating more appropriate treatment plans.
[0106] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. An MRI image analysis method for lumbar lesion detection, characterized in that: The method comprises: Acquire an MRI image of the lumbar spine, and preprocess the MRI image to obtain a grayscale image; Acquire an initial growth point in the grayscale image, take the initial growth point as the center, acquire a first local area including at least one suspected non-growth point according to a preset initial growth threshold, screen the suspected non-growth points in the first local area according to the local similarity between each suspected non-growth point in the first local area and the previous grown point in its growth direction to obtain a target growth point, and acquire an adaptive threshold of each target growth point according to the data difference between each target growth point and the grown point in the first local area; For any target growth point, according to the adaptive threshold of the target growth point, a second local area including at least one suspected non-growth point is obtained, wherein the area of the second local area is smaller than the area of the first local area, a to-be-grown point in the second local area is obtained, the to-be-grown point is used as the next target growth point, and the method for obtaining the to-be-grown point is repeated until no to-be-grown point is obtained, thereby obtaining all to-be-grown points corresponding to the target growth point; Acquire all the to-be-grown points corresponding to each of the target growth points, and form the nucleus pulposus region of the lumbar vertebra in the grayscale image with the initial growth points, the grown points and the target growth points in the first local area, and all the to-be-grown points corresponding to each of the target growth points; The step of taking the initial growth point as the center and acquiring a first local area including at least one suspected non-growth point according to a preset initial growth threshold comprises: Taking the initial growth point as the center and the neighborhood of the initial growth point as the growth direction, for any growth direction of the initial growth point, if there is any pixel point in the growth direction and the grayscale value difference between the pixel point and the previous pixel point is less than or equal to the preset initial growth threshold, then the pixel point is taken as a grown point; if there is any pixel point in the growth direction and the grayscale value difference between the pixel point and the previous pixel point is greater than the preset initial growth threshold, then the pixel point is taken as a suspected non-growth point; obtain the suspected non-growth points in each growth direction of the initial growth point, and establish a first local area with the initial growth point as the center and the suspected non-growth points in each growth direction of the initial growth point as the boundary; The method of screening the suspected non-growth points in the first local area to obtain target growth points according to the local similarity between each suspected non-growth point in the first local area and the last grown point in the growth direction thereof includes: For any suspected non-growth point in the first local area, in the growth direction of the suspected non-growth point, obtain a grown point adjacent to the suspected non-growth point, record it as the previous grown point, and obtain the local similarity between the suspected non-growth point and the previous grown point; When the local similarity between the suspected non-growth point and the last grown point is greater than or equal to a preset local similarity threshold, all grown points in the opposite direction of the growth of the suspected non-growth point are obtained, and all grown points in the opposite direction of the growth of the suspected non-growth point and the suspected non-growth point are combined into a growth group, and a local similarity prediction value of the suspected non-growth point is obtained according to the data difference of each pixel point in the growth group, and when the local similarity prediction value of the suspected non-growth point is within a preset range, the suspected non-growth point is recorded as a target growth point; The step of respectively obtaining the adaptive threshold of each target growth point according to the data difference between each target growth point and the already grown point in the first local area comprises: For any target growth point, obtain the local similarity between the target growth point and the previous grown point in its growth direction, obtain the absolute value of the difference between the local similarity prediction value of the target growth point and the local similarity between the target growth point and the previous grown point in its growth direction as the error coefficient of the target growth point, and obtain the adaptive threshold of the target growth point according to the local similarity between the target growth point and the previous grown point in its growth direction, the error coefficient of the target growth point and a preset initial growth threshold; The step of obtaining the adaptive threshold of the target growth point according to the local similarity between the target growth point and the previous growth point in its growth direction, the error coefficient of the target growth point and a preset initial growth threshold comprises: Obtaining a first weight coefficient of the local similarity between the target growth point and the previous grown point in its growth direction, obtaining a first range difference of the local similarity between the target growth point and the previous grown point in its growth direction, calculating the product of the first weight coefficient, the inverse of the first range difference, and the local similarity between the target growth point and the previous grown point in its growth direction to obtain a first product, calculating the product of the first weight coefficient, the inverse of the first range difference, and a preset local similarity threshold to obtain a second product, and calculating the difference between the first product and the second product to obtain a first difference; Obtaining a second weight coefficient of the error coefficient of the target growth point, obtaining a second range difference of the error coefficient, calculating a third product as a product of the second weight coefficient, the inverse of the second range difference of the error coefficient, and the error coefficient, and calculating a difference between the second weight coefficient and the third product to obtain a second difference; A sum of the preset initial growth threshold, the first difference and the second difference is obtained as an adaptive threshold of the target growth point.
2. The MRI image analysis method for lumbar lesion detection according to claim 1, characterized in that: The obtaining of the local similarity between the suspected non-growing point and the last grown point includes: Taking the suspected non-growth point and the previous grown point as the center, respectively, windows of preset sizes are established to obtain a suspected non-growth point window and a grown point window; According to the data difference of the pixel points in the suspected non-growth point window and the grown point window, the similarity of the grayscale distribution of the suspected non-growth point window and the grown point window is obtained; Acquiring the structural similarity between the suspected non-growth point window and the grown point window as the window structural similarity between the suspected non-growth point window and the grown point window; According to the grayscale distribution similarity and the window structure similarity between the suspected non-growing point window and the grown point window, the local similarity between the suspected non-growing point and the last grown point is obtained.
3. The MRI image analysis method for lumbar lesion detection according to claim 2, characterized in that: The obtaining, according to the data difference of the pixel points in the suspected non-growing point window and the grown point window, the grayscale distribution similarity between the suspected non-growing point window and the grown point window comprises: The grayscale value difference between each pixel in the suspected non-growing point window and the pixel at the same position in the grown point window is obtained respectively to form a grayscale value difference sequence, the square value of each grayscale value difference in the grayscale value difference sequence is obtained, and the square value mean is obtained accordingly, and the inverse of the square value mean is normalized to obtain the grayscale distribution similarity between the suspected non-growing point window and the grown point window.
4. The MRI image analysis method for lumbar lesion detection according to claim 2, characterized in that: The obtaining, according to the grayscale distribution similarity and the window structure similarity between the suspected non-growing point window and the grown point window, the local similarity between the suspected non-growing point and the last grown point comprises: The addition result between the grayscale distribution similarity and the window structure similarity of the suspected non-growth point window and the grown point window is obtained as the local similarity between the suspected non-growth point and the last grown point.
5. The MRI image analysis method for lumbar lesion detection according to claim 1, characterized in that: The obtaining, according to the data difference of each pixel point in the growth group, the local similarity prediction value of the suspected non-growth point comprises: According to the data difference of each pixel point in the growth group, the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is obtained, the local similarity between each pixel point in the growth group and the previous grown point in its growth direction is fitted to obtain a local similarity fitting function, and the local similarity fitting function is used to predict the local similarity between the suspected non-growth point and the previous grown point to obtain a local similarity prediction value of the suspected non-growth point.
6. The MRI image analysis method for lumbar lesion detection according to claim 1, characterized in that: The obtaining of the to-be-grown point in the second local area includes: The suspected non-growth points in the second local area are screened to obtain target growth points in the second local area, and the target growth points in the second local area are used as the to-be-grown points in the second local area.