Knee joint ROI extraction method and system
By using bone segmentation mask, edge line extraction and correction techniques in the ROI extraction method for knee osteoarthritis, the problems of incomplete ROI selection and poor consistency in the existing technology are solved, and more accurate ROI area extraction and Consistency between samples provides a good foundation for machine learning and deep learning analysis.
Patent Information
- Application Number
- CN202410977971.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2025-01-03
AI Technical Summary
In the early prediction of knee osteoarthritis, the selection of ROI is mostly based on rectangular shapes, resulting in the loss of information on the lower bone edge, incomplete extraction and large differences in position between samples, and research deviations, making it difficult to effectively extract the complete edge on the tibia. And consistent ROI.
A knee joint ROI extraction method is adopted, including extraction of bone segmentation masks, edge line extraction, edge line stretching and correction of ROI areas. Through manual annotation, machine learning or deep learning methods, pixel translation, subtraction and threshold interception are used. , ensure the integrity and consistency of ROI.
More accurate ROI area extraction is achieved, and the information on the lower bone edge is fully obtained, ensuring the consistency of ROI positions between samples, providing a solid foundation for machine learning or deep learning analysis.
Smart Images
Figure CN119251249A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an extraction method, and in particular to a knee joint ROI extraction method and system. Background Art
[0002] Early prediction and treatment of knee osteoarthritis are very important. Currently, tibial plateau subchondral bone has been identified as an imaging biomarker that can provide information on subchondral bone changes due to knee osteoarthritis.
[0003] With the advancement of medical imaging technology, some studies have proposed that artificial intelligence and other methods can predict the onset of knee osteoarthritis and identify early bone changes associated with the onset of osteoarthritis. The success of the study depends on the selection of ROI location. Therefore, the selection of ROI is a key step in bone structure analysis.
[0004] In existing studies, most ROI selection shapes are based on rectangular selection, which seriously loses the edge information of the lower bone, resulting in incomplete extraction of the lower bone, and there are differences in ROI position information between samples, which leads to research bias. In addition, due to the differences in human bones, it is particularly important to effectively extract subchondral bone with complete edges and consistency between samples on the tibia as ROI. Summary of the invention
[0005] The present invention aims at the defects of the prior art and provides a knee joint ROI extraction method and system.
[0006] To achieve the above-mentioned purpose, the present invention adopts the following technical scheme, a knee joint ROI extraction method, comprising: extracting a bone segmentation mask for a knee joint image to be processed.
[0007] Extract the bone edge line and intercept the edge line according to the threshold.
[0008] Stretch the edge lines and correct the (redundant) ROI area.
[0009] Furthermore, the extracted bone segmentation mask is obtained by using any one of a manual labeling method, a machine learning method, and a deep learning method.
[0010] Furthermore, extracting the bone edge line includes:
[0011] The skeleton segmentation mask image M0 is shifted upward by one pixel as a whole to obtain the mask image M1; it can be expressed as:
[0012] M1 x，y =M0 x-1,y
[0013] The mask image M1 is subtracted from the bone segmentation mask image M0 to obtain the mask image M2; the formula is:
[0014] M2=M0-M1
[0015] The points of the mask image M2 whose values are greater than 1 are retained, and the pixel sizes of other points are changed to 0 to obtain the bone edge line.
[0016] Furthermore, the edge line interception according to the threshold includes: marking the lowest point of the skeleton edge line as x, and all points in the skeleton edge line whose horizontal coordinates are less than x minus the threshold (the threshold can be 30) become 0, thereby obtaining a mask image M3.
[0017] Furthermore, the stretching edge line includes: calculating the number of pixels n required to be translated according to the bone ring length l (in millimeters) to be taken and the spacing parameter (also known as spacing interval, in millimeters / pixel) of the medical image; as follows:
[0018]
[0019] The mask image M3 is translated from 1 to n pixels to obtain n images in total; the obtained mask image is added to the original image to obtain the mask image M4; the values of all points greater than 1 in the mask image M4 are changed to 1.
[0020] Furthermore, the modified ROI area includes: removing the protrusions on both sides of the mask image M4; finding a point at a certain pixel distance (for example, 30 pixels) from the leftmost side of the mask image M3, and calculating the coordinates (x l ,y l ); In the mask image M4, the coordinate (x l -n,y l ) are all changed to 0.
[0021] Similarly, find a point that is a certain distance away from the rightmost pixel of the mask image M3 (for example, 30 pixels), and calculate the coordinates of the point (x r ,y r ); In the mask image M4, the coordinate (x r -n,y r )The upper right points are all changed to 0; the final ROI mask image M5 is obtained.
[0022] The mask image M5 is multiplied by the knee joint image to be processed to obtain the ROI image.
[0023] A knee joint ROI extraction system comprises: a segmentation image extraction module, which is used to extract a bone segmentation mask for a knee joint image to be processed.
[0024] The edge line extraction module is used to extract the bone edge line; it includes: translating the bone segmentation mask image upward by one pixel as a whole to obtain the mask image M1; subtracting the mask image M1 from the bone segmentation mask image to obtain the mask image M2; retaining the points of the mask image M2 that are greater than 1, and changing the pixel size of other points to 0.
[0025] The interception module is used to intercept the edge line according to the threshold value; it includes: marking the lowest point of the skeleton edge line as x, and all points in the skeleton edge line whose horizontal coordinates are less than x minus the threshold value become 0, so as to obtain a mask image M3.
[0026] The stretching module is used for the edge line; it includes: according to the length l of the bone ring to be taken (in millimeters) and the spacing parameter of the medical image (also known as the spacing interval, in millimeters / pixel), the number of pixels n that need to be translated is calculated; as follows:
[0027]
[0028] The mask image M3 is translated from 1 to n pixels to obtain n images in total; the obtained mask image is added to the original image to obtain the mask image M4; the values of all points greater than 1 in the mask image M4 are changed to 1.
[0029] The correction module is used to correct the ROI area; that is, to remove the protrusions on both sides of the mask image M4.
[0030] Compared with the prior art, the present invention has beneficial effects.
[0031] The knee joint ROI extraction method and system of the present invention can obtain a more accurate ROI area, completely extract the edge information of the lower bone, and ensure the consistency of ROI positions between samples, providing a solid foundation for machine learning or deep learning analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. The protection scope of the present invention is not limited to the following description.
[0033] Figure 1 It is the overall framework diagram of the knee joint ROI extraction method and system.
[0034] Figure 2 It is a schematic diagram of extracting bone segmentation mask.
[0035] Figure 3 This is a schematic diagram of the bone edge line.
[0036] Figure 4 is the mask image M3.
[0037] Figure 5This is the image after the edge line is stretched.
[0038] Figure 6 is the final ROI mask image. DETAILED DESCRIPTION
[0039] Specific embodiment: Figure 1 As shown, the knee joint ROI extraction method and system include the following steps:
[0040] 1. Extract the bone segmentation mask.
[0041] Extracting the bone segmentation mask is the first step to extract the bone ROI. The bone segmentation mask can be obtained by manual annotation, machine learning, deep learning and other methods. Taking the femur as an example, the bone segmentation mask is as follows: Figure 2 shown.
[0042] Second, extract the bone edge line and intercept the edge line according to the threshold.
[0043] The skeleton segmentation mask (denoted as M0) is shifted upward by one pixel as a whole to obtain a new mask image (denoted as M1).
[0044] M1 x，y =M0 x-1，y .
[0045] Subtract the mask image M1 from the skeleton segmentation mask image M0 to obtain the mask image M2.
[0046] M2=M0-M1.
[0047] Keep the points on M2 that are greater than 1 and change the pixel size of other points to 0 to get the bone edge line, such as Figure 3 shown.
[0048] The lowest point of the skeleton edge line is marked as x. The points in the skeleton edge line whose horizontal coordinates are less than x minus the threshold (the threshold can be 30) are all changed to 0, and the mask image M3 is obtained, as shown in Figure 4 shown.
[0049] 3. Stretch the edge line and correct the redundant ROI area.
[0050] According to the bone ring length l (in millimeters) to be obtained and the spacing parameter of the medical image (in millimeters / pixel), the number of pixels n that need to be translated is calculated.
[0051]
[0052] Shift the mask image M3 by 1 to n pixels, and get a total of n images. Add the obtained mask image and the original image together to get the mask image M4. Change the values of all points greater than 1 in the mask image M4 to 1, such as Figure 5 shown.
[0053] Finally, the bulges on both sides need to be removed. Find a point that is a certain distance away from the leftmost side of the mask image M3 (for example, 30 pixels), and calculate the coordinates of the point (x l ,y l ). In the mask image M4, the coordinate (x l -n,y l ) are all changed to 0. Similarly, find a point that is a certain distance away from the rightmost side of the mask image M3 (for example, 30 pixels), and calculate the coordinates of the point (x r ,y r ). In the mask image M4, the coordinate (x r -n,y r )The points on the upper right are all changed to 0. The final ROI mask image M5 is obtained, as shown in Figure 6 shown.
[0054] Use M5 to multiply the original image to get the ROI image.
[0055] It can be understood that the above specific description of the present invention is only used to illustrate the present invention and is not limited to the technical solutions described in the embodiments of the present invention. Those skilled in the art should understand that the present invention can still be modified or replaced by equivalents to achieve the same technical effects; as long as the use requirements are met, they are within the protection scope of the present invention.
Claims
1. A knee joint ROI extraction method, characterized by: include: For the knee joint image to be processed, extract the bone segmentation mask; Extract the bone edge line and intercept the edge line according to the threshold; Stretch the edge line and correct the ROI area.
2. The knee joint ROI extraction method according to claim 1, characterized in that: The extracted bone segmentation mask is obtained by using any one of a manual labeling method, a machine learning method, and a deep learning method.
3. The knee joint ROI extraction method according to claim 1, characterized in that: The bone edge line extraction comprises: The skeleton segmentation mask image MO is shifted upward by one pixel as a whole to obtain the mask image M1; it can be expressed as: M1 x，y =M0 x-1，y The mask image M1 is subtracted from the bone segmentation mask image M0 to obtain the mask image M2; the formula is: M2=M0-M1 The points of the mask image M2 whose values are greater than 1 are retained, and the pixel sizes of other points are changed to 0 to obtain the bone edge line.
4. The knee joint ROI extraction method according to claim 1, characterized in that: The method of intercepting the edge line according to the threshold value comprises: The lowest point of the skeleton edge line is marked as x, and the points in the skeleton edge line whose horizontal coordinates are less than x minus the threshold value become 0, thereby obtaining a mask image M3.
5. The knee joint ROI extraction method according to claim 1, characterized in that: The stretching edge line comprises: According to the bone ring length / and the spacing parameter of the medical image, the number of pixels n required to be translated is calculated as follows: The mask image M3 is translated from 1 to n pixels to obtain n images in total; the obtained mask image is added to the original image to obtain the mask image M4; the values of all points greater than 1 in the mask image M4 are changed to 1.
6. The knee joint ROI extraction method according to claim 5, characterized in that: The modified ROI area includes: To remove the protrusions on both sides of the mask image M4, find a point that is a certain pixel distance from the leftmost side of the mask image M3 and calculate the coordinates of the point (x l ,y l ); In the mask image M4, the coordinate (x l -n,y l )The upper left points are all changed to 0; Similarly, find a point that is a certain pixel distance from the rightmost side of the mask image M3 and calculate the coordinates of the point (x r ,y r ); In the mask image M4, the coordinate (x r -n,y r )The upper right points are all changed to 0; the final ROI mask image M5 is obtained; The mask image M5 is multiplied by the knee joint image to be processed to obtain the ROI image.
7. Knee joint ROI extraction system, characterized by: include: A segmentation image extraction module, which is used to extract a bone segmentation mask for a knee joint image to be processed; The edge line extraction module is used to extract the bone edge line; it includes: Shift the skeleton segmentation mask image upward by one pixel as a whole to obtain a mask image M1; subtract the mask image M1 from the skeleton segmentation mask image to obtain a mask image M2; retain the points of the mask image M2 that are greater than 1, and change the pixel size of other points to 0; The interception module is used to intercept the edge line according to the threshold value; comprising: The lowest point of the skeleton edge line is marked as x, and the points in the skeleton edge line whose horizontal coordinates are less than x minus the threshold value become 0, thereby obtaining a mask image M3. A stretching module for edge lines; comprising: According to the bone ring length / and the spacing parameter of the medical image, the number of pixels n required to be translated is calculated as follows: Shift the mask image M3 by 1 to n pixels to obtain n images in total; add the obtained mask image to the original image to obtain the mask image M4; change the values of all points greater than 1 in the mask image M4 to 1; The correction module is used to correct the ROI area; that is, to remove the protrusions on both sides of the mask image M4.