A method and system for rib positioning, segmentation and post-processing

Abstract

The present invention relates to medical image processing technology, and discloses a method and system for rib positioning, segmentation, and post-processing. The method comprises: storing image data, rib segmentation results, and rib post-processing results in a relational database; a rib segmentation module that processes and segments ribs from chest CT scans using a segmentation model; and a rib post-processing module that corrects mis-segments and segments each rib based on the rib segmentation results. This method eliminates the need to distinguish and label individual ribs. After deep learning-based rib segmentation, the method determines whether adjacent ribs are connected by the three-dimensional slope of their centerlines and then segments the connected areas.

Description

Technical Field

[0001] The present invention relates to medical image processing technology, and in particular to a method and system for rib positioning, segmentation and post-processing of chest CT plain scans. Background Art

[0002] Chest CT scans are the most widely used imaging examinations in medical settings, and rib fractures are a common screening test during chest CT scans. Rib fractures can have diverse morphologies, and because ribs bend downward, it's difficult to determine the fractured rib from a horizontal axial perspective. This requires simultaneous interpretation from both sagittal and coronal planes, making the image reading process laborious, especially for younger physicians. Therefore, an automated method that can automatically locate, segment, and identify ribs could greatly improve the efficiency of radiology departments in reading and reporting.

[0003] For example, the existing technology CN114049358A method first performs segmentation based on deep learning, and then obtains the outline of each rib through the connected area to achieve rib positioning. This method cannot remove non-rib defect points that appear in the segmentation results, and for ribs that are connected due to segmentation errors or patient defects, this method cannot separate the connected ribs.

[0004] Existing patent technology 2, CN111915620A, uses deep learning to segment ribs and then delete non-rib defect points, but does not distinguish between each rib;

[0005] Prior art three CN110992376A This patent distinguishes and labels each rib, and directly segments and distinguishes each rib through deep learning, thereby increasing the data cost required for deep learning. Summary of the Invention

[0006] The present invention addresses the problem in the prior art that connected ribs are difficult to cut and the cost of cutting is high, and provides a method and system for rib positioning, segmentation and post-processing.

[0007] In order to solve the above technical problems, the present invention is solved by the following technical solutions:

[0008] A method for rib positioning, segmentation and post-processing, the method comprising:

[0009] Step 1: The rib segmentation module reads the chest CT plain scan original image I from the image database, takes the HU threshold ε, obtains the binary segmentation result according to I ≥ ε, and fills it with the water filling method to obtain the bone segmentation result B;

[0010] Step 2: Extract the largest connected region from the bone segmentation result B, set the values ​​of other connected regions to 0, and crop the unsegmented slices along the positive and negative directions of the axial, coronal, and sagittal planes to obtain the rib segmentation candidate region C;

[0011] Step 3: Build a 3D Unet network to obtain rib segmentation results

[0012] Step 4: Use a spherical kernel with size r0 to expand the segmentation result At the same time, set the volume threshold v of the connected domain, determine the number n of connected domains whose volume is greater than the volume threshold ν, iterate k times until n = 1; set the values ​​of other connected regions to 0, that is, eliminate other connected regions, and obtain in, for The result after dilation and removal of incorrect segmentation;

[0013] Step 5, rib segmentation results and Obtain rib segmentation results after removing redundant connected areas in for The segmentation results of the incorrectly segmented areas were removed;

[0014] Step 6: Use spherical core erosion with size r1 to segment the result Get segmentation results in, for The segmentation result after re-erosion;

[0015] Step 7, Extract The center line of the rib is taken as the rib center line image C1, and a sliding window of m×m×m (m≥11) is taken to traverse all points on the center line to obtain the first vertical discrimination result center line image C2;

[0016] Step 8: Reduce the value of m to ma, and ma is greater than 3; repeat step 7 to obtain the second vertical discrimination result center line image C3;

[0017] Step 9: Calculate C2+C3 using the centerline image C2 of the first vertical discrimination result and the centerline image C3 of the second vertical discrimination result, and take the point with a point value of 4 as the vertical point in the rib centerline;

[0018] Step 10, Take the point set Q with C2+C3=4 from the equation i ,y i ,z i ),i∈(1,…,q)}; truncate the segmentation results of the upper and lower connected ribs;

[0019] Step 11, Extract Connected area To achieve the positioning of rib segmentation. As a preference, step 7, extract The center line of the rib is used as the rib center line image C1, and a sliding window of m×m×m (m≥11) is taken to traverse all points on the center line to obtain the vertical discrimination result C2; which includes:

[0020] Step 7.1: Take an m×m×m window M around the center point c = (x, y, z) and set the value of the central 3×3×3 area to 0;

[0021] Step 7.2, extract the connected area of ​​M {P i ,i∈{1,…,t}}, for each connected region point set {c ij =(x ij ,y ij ,z ij ), i∈{1,…,t}, j∈{1,…,p}}, calculate the average slope k of the connected area to point c i =(k ix ,k iy ,k iz ), calculated as follows,

[0022]

[0023] Set the vertical threshold k z and the horizontal threshold k xy , satisfying k ix ≤k xy ,k iy ≤k xy ,k iz ≥k z The value of all points in the connected area i is 2, and the value of all points in the unconnected area is 1;

[0024] Step 7.3, traverse all points of C1 and obtain the vertical discrimination result C2.

[0025] As a preference, in step 10, Take the point set Q with C2+C3=4 from the equation i ,y i ,z i ),i∈(1,…,q)}, which includes:

[0026] Step 10.1, for point ( x i, y i, z i), take z=z iThe fault is divided into two parts, with the point (x i ,y i ) as the center, take a square area with a side length of a and set it to 0;

[0027] Step 10.2, traverse each point in the point set Q to obtain the processed rib segmentation result As a result of segmentation by cutting off the upper and lower connected ribs.

[0028] As a preference, step 11, extract Connected area It includes:

[0029] Step 11.1: Along the central sagittal plane, It is divided into two parts, namely the right rib image and left rib

[0030] Step 11.2: Starting from the head, go down along the spine and take The i-th connected region is the right i rib, The i-th connected region is the left i-th rib, until i=12, achieving the segmentation and positioning of 12 pairs of ribs.

[0031] In order to solve the above technical problems, the present invention further provides a system for rib positioning, segmentation and post-processing, which is implemented by a method for rib positioning, segmentation and post-processing, comprising:

[0032] Storage of image database: image data, rib segmentation results, and rib post-processing results are stored in the form of a relational database;

[0033] The rib segmentation module processes and segments ribs from chest CT scans using a segmentation model;

[0034] The rib post-processing module corrects mis-segmentation and segments each rib based on the rib segmentation results.

[0035] In order to solve the above technical problems, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for rib positioning, segmentation and post-processing are implemented.

[0036] In order to solve the above technical problems, the present invention also provides an electronic device, which includes: a memory for storing a computer program; and a processor for implementing the steps of the method for rib positioning, segmentation and post-processing when executing the computer program.

[0037] The present invention has significant technical effects due to the adoption of the above technical solutions:

[0038] Based on deep learning-based rib segmentation, the present invention separates incorrectly segmented areas and areas where upper and lower adjacent ribs are connected to facilitate the positioning of each rib, thereby expanding the application scenarios of rib segmentation and sorting.

[0039] The present invention does not need to distinguish and mark each rib. After the rib segmentation based on deep learning, the three-dimensional slope of the center line is used to determine whether there are upper and lower adjacent ribs connected, and the connected area is cut. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the present invention;

[0041] Figure 2 This is a system block diagram of embodiment 2 of the present invention;

[0042] Figure 3 It is the vertical determination result of the centerline of some ribs of the present invention;

[0043] Figure 4 This is the rib segmentation result of the three-dimensional Unet model of the present invention;

[0044] Figure 5 This is the positioning result of the 12 pairs of left and right ribs of the present invention; DETAILED DESCRIPTION

[0045] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0046] Example 1

[0047] A method for rib positioning, segmentation and post-processing, through the process Figure 1 It can be seen that the methods include:

[0048] Step 1: The rib segmentation module reads the chest CT plain scan original image I from the image database, takes the HU threshold of 100, obtains the binary segmentation result according to I ≥ 100, and fills it with the water filling method to obtain the bone segmentation result B;

[0049] Step 2: Extract the largest connected region from the bone segmentation result B, set the values ​​of other connected regions to 0, and crop the unsegmented slices along the positive and negative directions of the axial, coronal, and sagittal planes to obtain the rib segmentation candidate region C;

[0050] Step 3: Build a 3D Unet network to obtain rib segmentation results

[0051] Step 3.1, build the Unet network, set the input data size to 1*128*128*128, and the output data size to 1*128*128*128;

[0052] Step 3.2: Construct a labeled dataset D and train the Unet model to obtain a trained Unet network model.

[0053] In step 3.3, a 1*128*128*128 three-dimensional sliding window is used with a step size of 64 to slide and crop the candidate rib segmentation region C. The image is then input into the trained Unet model and the segmentation probability result is obtained by applying the sigmoid function. The results are then concatenated to form the segmentation result R′.

[0054] Step 3.4, R ′ Each pixel in the probability is set as 1, and the pixel points with probability ≥ 0.5 are set as 1, and the pixel points with probability < 0.5 are set as 0 to obtain the binary rib segmentation result. get Figure 4 The rib segmentation result of the three-dimensional Unet model is shown.

[0055] Step 4: Use a spherical kernel with size r0 to expand the segmentation result At the same time, the connected domain volume threshold is set to 2000, and the number of connected domains with a volume greater than 2000 is determined, and iterated 5 times until n = 1. Then the values ​​of other connected regions are set to 0, that is, other connected regions are eliminated, and the obtained

[0056] Step 5, Obtain rib segmentation results after removing redundant connected areas

[0057] Step 6: Use spherical core erosion with a size of 5 to segment the result Get segmentation results

[0058] Step 7, Extract The center line of the rib is used as the rib center line image C1. A 13×13×13 sliding window is taken to traverse all points on the center line, and the following steps are performed respectively:

[0059] Step 7.1: Take a 13×13×13 window M around the center point c = (x, y, z) and set the value of the central 3×3×3 area to 0;

[0060] Step 7.2, extract the connected area of ​​M {P i ,i∈{1,…,t}}, for each connected region point set {c ij =(x ij ,y ij ,z ij ), i∈{1,…,t}, j∈{1,…,p}}, calculate the average slope k of the connected area to point c i =(k ix ,kiy ,k ij ), calculated as follows

[0061]

[0062] Set the vertical threshold k z and the horizontal threshold k xy , satisfying k ix ≤k xy ,k iy ≤k xy ,k iz ≥k z The value of all points in the connected area i is 2, and the value of all points in the unconnected area is 1;

[0063] Step 7.3, traverse all points of C1 and obtain the vertical discrimination result C2.

[0064] In step 8, reduce the value of m to 10 and repeat step 7 to obtain the vertical discrimination result C3.

[0065] Step 9: Calculate C2+C3 and select the point with a value of 4 as the vertical point in the rib centerline. Since the rib has a low slope in the axial plane, the vertical point can be used as a mis-segmentation of the connection between different ribs, especially the connection between upper and lower adjacent ribs; Figure 3 The vertical determination results of some rib centerlines are obtained. After the centerline is extracted, vertical and horizontal are also distinguished.

[0066] Step 10, Take the point set Q with C2+C3=4 from the equation i ,y i ,z i ),i∈(1,…,q)};

[0067] Step 10.1, for the point (x i ,y i ,z i ),Pick z=z i The fault is divided into two parts, with the point (x i ,y i ) as the center, take a square area with a side length of 21 and set it to 0;

[0068] Step 10.2, traverse each point in the point set Q to obtain the processed rib segmentation result As a result of segmentation by cutting off the upper and lower connected ribs.

[0069] Step 11, Extract Connected area

[0070] Step 11.1: Along the central sagittal plane, It is divided into two parts, namely the right rib image and left rib

[0071] Step 11.2: Starting from the head, go down along the spine and take The i-th connected region is the right i rib, The i-th connected region is the left i-th rib, until i=12, achieving the segmentation and positioning of 12 pairs of ribs. Thus, the final effect is obtained. Figure 5 , Figure 5 This is the positioning result diagram of 12 pairs of left and right ribs.

[0072] Example 2

[0073] Based on Example 1, this example is a system for rib positioning, segmentation, and post-processing, which is implemented by a method for rib positioning, segmentation, and post-processing. Figure 2 Among them, they include:

[0074] Storage of image database: image data, rib segmentation results, and rib post-processing results are stored in the form of a relational database;

[0075] The rib segmentation module processes and segments ribs from chest CT scans using a segmentation model;

[0076] The rib post-processing module corrects mis-segmentation and segments each rib based on the rib segmentation results.

[0077] Example 3

[0078] Based on Example 1, this embodiment is a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for rib positioning, segmentation and post-processing are implemented.

[0079] Example 4

[0080] Based on Example 1, this embodiment is an electronic device, which includes: a memory for storing a computer program; a processor for implementing the steps of the method for rib positioning, segmentation and post-processing when executing the computer program.

Claims

1. A method for rib location, segmentation, and post-processing, comprising: Step 1: The rib segmentation module reads the chest CT plain scan original image I from the image database, takes the HU threshold ε, obtains the binary segmentation result according to I ≥ ε, and fills it with the water filling method to obtain the bone segmentation result B; Step 2: Extract the largest connected region from the bone segmentation result B, set the values ​​of other connected regions to 0, and crop the unsegmented slices along the positive and negative directions of the axial, coronal, and sagittal planes to obtain the rib segmentation candidate region C; Step 3: Build a 3D Unet network to obtain rib segmentation results Step 4: Use a spherical kernel with size r0 to expand the segmentation result At the same time, the volume threshold ν of the connected domain is set, and the number n of connected domains whose volume is greater than the volume threshold ν is determined, and iterated k times until n = 1; the values ​​of other connected regions are set to 0, that is, other connected regions are eliminated, and the obtained in, for The result after dilation and removal of incorrect segmentation; Step 5, rib segmentation results and Obtain rib segmentation results after removing redundant connected areas in for The segmentation results of the incorrectly segmented areas are removed; Step 6: Use spherical core erosion with size r1 to segment the result Get segmentation results in, for The segmentation result after re-erosion; Step 7, Extract The center line of the rib is taken as the rib center line image C1, and a sliding window of m×m×m, m≥11 is taken to traverse all points on the center line to obtain the first vertical discrimination result center line image C2; Step 8: Reduce the value of m to ma, and ma is greater than 3; repeat step 7 to obtain the second vertical discrimination result center line image C3; Step 9: Calculate C2+C3 using the centerline image C2 of the first vertical discrimination result and the centerline image C3 of the second vertical discrimination result, and take the point with a point value of 4 as the vertical point in the rib centerline; Step 10, Take the point set Q with C2+C3=4 from the equation i ,y i ,z i ),i∈(1,…,q)}; truncate the segmentation results of the upper and lower connected ribs; Step 11, Extract Connected area Realize the positioning of rib segmentation.

2. The method for rib positioning, segmentation and post-processing according to claim 1, characterized in that: Step 7, Extract The center line of the rib is taken as the rib center line image C1, and a sliding window of m×m×m, m≥11 is taken to traverse all points on the center line to obtain the vertical discrimination result C2; It includes: Step 7.1: Take an m×m×m window M around the center point c = (x, y, z) and set the value of the central 3×3×3 area to 0; Step 7.2, extract the connected regions of M {P i ,i∈{1,…,t}}, for each connected region point set {c ij =(x ij ,y ij ,z ij ), i∈{1,…,t}, j∈{1,…,p}}, calculate the average slope k of the connected area to point c i =(k ix ,k iy ,k iz ), calculated as follows, Set the vertical threshold k z and the horizontal threshold k xy , satisfying k ix ≤k xy ,k iy ≤k xy ,k iz ≥k z The value of all points in the connected area i is 2, and the value of all points in the unconnected area is 1; Step 7.3, traverse all points of C1 and obtain the vertical discrimination result C2.

3. The method for rib positioning, segmentation and post-processing according to claim 1, characterized in that: Step 10, Take the point set Q with C2+C3=4 from the equation i ,y i ,z i ),i∈(1,…,q)}, which includes: Step 10.1, for the point (x i ,y i ,z i ),Pick z=z i The fault is divided into two parts, with the point (x i ,y i ) as the center, take a square area with a side length of a and set it to 0; Step 10.2, traverse each point in the point set Q to obtain the processed rib segmentation result As a result of segmentation by cutting off the upper and lower connected ribs.

4. The method for rib positioning, segmentation and post-processing according to claim 1, characterized in that: Step 11, Extract Connected area It includes: Step 11.1: Along the central sagittal plane, It is divided into two parts, namely the right rib image and left rib Step 11.2: Starting from the head, go down along the spine and take The i-th connected region is the right i rib, The i-th connected region is the left i-th rib, until i=12, achieving the segmentation and positioning of 12 pairs of ribs.

5. A system for rib positioning, segmentation and post-processing, characterized in that: A system implemented by the method for rib positioning, segmentation, and post-processing according to any one of claims 1 to 4, comprising: Storage of image database: image data, rib segmentation results, and rib post-processing results are stored in the form of a relational database; The rib segmentation module processes and segments ribs from chest CT scans using a segmentation model; The rib post-processing module corrects mis-segmentation and segments each rib based on the rib segmentation results.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for rib positioning, segmentation, and post-processing according to any one of claims 1 to 4.

7. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is configured to implement the steps of a rib positioning, segmentation and post-processing method as claimed in any one of claims 1 to 4 when executing the computer program.

Images