A rapid full-automatic intracranial aneurysm auxiliary detection post-processing system and method

Abstract

The application belongs to the field of biological medicine and medical imaging technology, and relates to a rapid full-automatic intracranial aneurysm auxiliary detection post-processing system and method. The system comprises an image preprocessing module, an image adaptive threshold classification module, an automatic blood vessel seed point searching module, a blood vessel segmentation module, a blood vessel center line extraction, segmentation and classification module, an aneurysm enhancement module, an aneurysm screening module and an aneurysm display module. The application can be used for full-automatic detection of three common medical image data (MRI, CT and 3D-DSA) intracranial aneurysms, and has high processing speed.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of biological medicine and medical imaging technology, and relates to a rapid full-automatic intracranial aneurysm auxiliary detection post-processing system and method. BACKGROUND

[0002] The prior art discloses that imaging equipment such as magnetic resonance (MRI), computed tomography (CT), three-dimensional digital subtraction angiography (3D-DSA) can realize the development of intracranial aneurysms. At present, the detection method of intracranial aneurysm mainly depends on the artificial recognition of professional physicians based on the above medical images, which has strong subjectivity, low repeatability and is easy to miss. Other methods such as: 1) medical image recognition scheme based on deep learning [1], on the one hand, a large amount of high-quality artificial annotation data is needed to achieve high accuracy; on the other hand, the generalization ability is poor, such as the model based on MRI training set cannot be used for CT image, the same model has large difference on the data of different medical institutions and different medical equipment manufacturers. 2) The existing non-deep learning scheme often takes a long time and needs manual interaction, which is not automatic [2]. The enhancement of aneurysm in three-dimensional image is the key step to realize the automatic detection. The Hessian filter analysis based on multi-scale is the method widely used at present [3]. This kind of method such as Frangi, Sato, Jerman and others is based on Hessian matrix, the Hessian eigenvalue is calculated, and different eigenvalue response functions are designed to obtain the enhancement value of aneurysm. Due to the relatively time-consuming eigenvalue calculation, the enhancement of the whole three-dimensional image needs several minutes or even longer time.

[0003] Based on the present situation of the prior art, the inventors of the present application provide a rapid full-automatic intracranial aneurysm auxiliary detection post-processing system and method.

[0004] The references related to the present application are:

[0005] [1] Ueda Daiju, Yamamoto Akira, Nishimori et Masataka, et al. (2018). Deep Learning for MR Angiography: Automated Detection of Cerebral Aneurysms. Radiology. 290. 180901.

[0006] [2] Lawonn Kai, Meuschke Monique, Ralph, et al. (2019). A Geometric Optimization Approach for the Detection and Segmentation of Multiple Aneurysms. Computer Graphics Forum. 38. 10.1111 / cgf.13699.

[0007] [3] Jerman Tim, Pernus Franjo, Likar Bostjan, et al. (2015). Blob Enhancement and Visualization for Improved Intracranial Aneurysm Detection. IEEE Transactions on Visualization and Computer Graphics. 22. 1-1. 10.1109 / TVCG.2015.2446493. SUMMARY

[0008] The present application is based on the current situation of the prior art, and provides a rapid full-automatic intracranial aneurysm auxiliary detection post-processing system and method. In particular, a system based on existing medical image data (MRI, CT, 3D-DSA) is used to realize full-automatic and rapid identification of intracranial aneurysms, so as to assist doctors in efficiently and accurately positioning the position of the aneurysm.

[0009] The present application provides an aneurysm auxiliary detection post-processing system, which realizes the identification of intracranial aneurysms of MRI, CT and 3D-DSA three kinds of medical images through the following functional modules;

[0010] The functional modules include but are not limited to:

[0011] The image preprocessing module unifies MRI, CT or 3D-DSA medical image;

[0012] The image adaptive threshold classification module analyzes the gray histogram of the input image to obtain the threshold, and classifies the image into background, low signal blood vessel region or high signal blood vessel region according to the signal value of the image;

[0013] The automatic blood vessel seed point searching module performs connected domain analysis on the high signal blood vessel region output by the second module, and finds the corresponding blood vessel;

[0014] Vessel segmentation module, region growing with the seed points obtained in module three, the growing range is limited to the low signal and high signal vessel regions in module two, and the result is refined by level set to obtain the vessel;

[0015] Vessel centerline extraction, segmentation and classification module, based on the vessel segmentation result of module four, the centerline is extracted, and the centerline is divided into different segments according to the bifurcation point;

[0016] Aneurysm enhancement module, which skips the Hessian eigenvalue calculation and only relies on the analysis of Hessian Matrix to determine the sign of the corresponding eigenvalue and selects all regions with negative eigenvalue signs as potential regions of aneurysm, thereby greatly improving the speed of the algorithm;

[0017] Aneurysm screening module, which needs to screen the suspected aneurysm to eliminate false positive results;

[0018] Aneurysm display module, which directly overlays the obtained aneurysm with the original image or overlays it with the maximum intensity projection image, volume rendering image, and vessel segmentation image of module four.

[0019] The image preprocessing refers to removing the scalp from the MRI or CT image to obtain the brain and cerebellum regions, and marking other regions as background; or performing an inverse operation on the 3D-DSA image to achieve uniformity of the input image: blood vessels as high signal, other tissues as low signal.

[0020] The image adaptive threshold classification is to analyze the gray level histogram of the input image to obtain threshold T1 and T2, and divide the image into three categories: signal value < T1 as background, T1 ≤ signal value < T2 as low signal vessel region, and signal value ≥ T2 as high signal vessel region.

[0021] The automatic search for vessel seed points is to perform connected component analysis on the high signal vessel region output by module two, and sort the three largest connected regions M1, M2, and M3 in descending order of volume, and satisfy the conditions: M2 ≥ 0.1 * M1, M3 ≥ 0.1 * M1;

[0022] If the M2 or M3 region does not meet the above conditions, it is excluded;

[0023] Finally, a pixel point is randomly selected from each region that meets the conditions as a seed point;

[0024] For MRI and CT images, up to three seed points can be found, corresponding to the left common carotid artery, right common carotid artery, and vertebral artery, respectively; for 3D-DSA images, one seed point can be found, corresponding to a certain branch of the developed blood vessels.

[0025] The aneurysm refers to a saccular aneurysm or an inter-wall aneurysm or a fusiform aneurysm.

[0026] The saccular aneurysm satisfies all the following conditions:

[0027] V>4 / 3*3.14*R0 3 ; wherein V is the volume of the aneurysm, and R0 is a threshold value;

[0028] contains an endpoint segment;

[0029] K / N>Th0. wherein N is the number of center points of the endpoint segment, K is the number of center points of the endpoint segment located in the aneurysm, and Th0 is a threshold value.

[0030] The inter-wall aneurysm or the fusiform aneurysm satisfies all the following conditions:

[0031] V>4 / 3*3.14*R0 3 , V is the volume of the aneurysm, and R0 is a threshold value;

[0032] contains a main segment;

[0033] R I / R O >Th1; wherein R I is the average radius of the center points of Ps located in the aneurysm, R O is the average radius of the center points of Ps located outside the aneurysm, and Th1 is a threshold value (here, Th1 can be set as 2.0).

[0034] Correspondingly, the application provides a fast and full-automatic intracranial aneurysm auxiliary detection post-processing method, i.e. using the above system to realize the recognition of intracranial aneurysms in MRI, CT or 3D-DSA medical images. The method comprises the following steps:

[0035] Image pre-processing, to unify MRI, CT or 3D-DSA medical images;

[0036] Image adaptive threshold classification, to obtain a threshold value by analyzing the gray histogram of the input image, and to divide the image into a background, a low-signal blood vessel region or a high-signal blood vessel region according to the signal value of the image;

[0037] Automatic searching for a blood vessel seed point, to perform connected domain analysis on the high-signal blood vessel region output by the image adaptive threshold classification, and to find the corresponding blood vessel;

[0038] Blood vessel segmentation, to perform region growing on the seed point obtained above, to limit the growing range to the low-signal and high-signal blood vessel regions in the image adaptive threshold classification, and to perform level set fine segmentation on the result to obtain the blood vessel;

[0039] Vessel centerline extraction, segmentation and classification, based on the results of vessel segmentation, centerline extraction is carried out, and the centerline is divided into different segments according to the bifurcation point;

[0040] Aneurysm enhancement, skip Hessian eigenvalue calculation, only rely on the analysis of Hessian Matrix, judge the sign of its corresponding eigenvalue, and select all the regions with negative eigenvalue sign as the potential region of aneurysm, so as to greatly improve the speed of the algorithm;

[0041] Aneurysm screening, this module needs to screen the above-mentioned suspected aneurysm to eliminate false positive results;

[0042] Aneurysm display, the above-mentioned aneurysm is directly superimposed and displayed with the original image, or superimposed and displayed with the maximum density projection image, volume rendering image and vessel segmentation image of the original image.

[0043] The system of the application can be used for aneurysm auxiliary detection post-processing.

[0044] Preferably, the above-mentioned system can automatically complete the intracranial aneurysm recognition or display.

[0045] The advantages of the application are:

[0046] 1) The three types of medical image data (MRI, CT, 3D-DSA) can realize the detection of intracranial aneurysm.

[0047] 2) The detection process is fully automated.

[0048] 3) Fast processing speed. For example, for 500*500*250 three-dimensional data, it can be completed within 10s on a computer with ordinary configuration (cpu: i5-7400, memory 16GB). BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0050] Figure 1 . Aneurysm automatic recognition process.

[0051] Figure 2 . Segmentation and classification of vessel centerline. The solid arrow is the end point, and the dashed arrow corresponds to the bifurcation point. S1-S5 correspond to different vessel segments.

[0052] Figure 3Different types of aneurysms. a) saccular aneurysm and its corresponding centerline; b) fusiform aneurysm and its corresponding centerline; c) dissecting aneurysm and its corresponding centerline. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the protection scope of the present application.

[0054] As shown in the accompanying drawings, Figure 1 The present application realizes the rapid and automatic recognition of intracranial aneurysms in MRI, CT and 3D-DSA medical images through the following functional modules.

[0055] Embodiment 1: Image preprocessing

[0056] For MRI and CT images, the main function of this module is to remove the scalp to obtain the region where the brain and cerebellum are located, and mark other regions as background (assign the signal value to 0). For 3D-DSA images, since the blood vessel signal is low signal and the background is high signal after subtraction, this module performs an inversion operation to realize the unification of the input image: the blood vessel is high signal and other tissues are low signal.

[0057] Embodiment 2: Image adaptive threshold classification

[0058] Through the analysis of the gray level histogram of the input image, the threshold values T1 and T2 are obtained, and the image is divided into three categories: background (signal value < T1), low signal blood vessel region (T1≤ signal value < T2), and high signal blood vessel region (signal value ≥ T2).

[0059] Embodiment 3: Automatic search for blood vessel seed points

[0060] The high signal blood vessel region output by module two is subjected to connected domain analysis, and the three connected regions M1, M2 and M3 with the largest volume are obtained in descending order of region volume, and the conditions are met: M2 ≥ 0.1 * M1, M3 ≥ 0.1 * M1. If the M2 or M3 region does not meet the above conditions, it is excluded. Finally, a pixel point is randomly selected as a seed point from each region that meets the conditions, and generally the pixel point at the bottom layer can be selected. For MRI and CT images, up to three seed points can be found, corresponding to the left common carotid artery, the right common carotid artery and the vertebral artery respectively. For 3D-DSA images, one seed point can be found, corresponding to a certain branch of the blood vessel.

[0061] Vessel segmentation

[0062] Region growing is performed on the seed points obtained in module three, with the growing range limited to the low signal and high signal vessel regions in module two, and level set fine segmentation is performed on the results to obtain the blood vessels.

[0063] Vessel centerline extraction, segmentation and classification

[0064] Based on the vessel segmentation results of module four, the centerline is extracted, and the centerline is divided into different segments according to the bifurcation points. Finally, each segment is classified as a terminal segment or an intermediate segment. As shown in Figure 2 the center point marked by the black dashed arrow is a bifurcation point, and the centerline is divided into a total of five segments S1-S5. The center point marked by the black solid arrow is a termination point, and S1-S5 is divided into two categories: terminal segments (S1, S2, S3, S5) and intermediate segments (S3).

[0065] Aneurysm enhancement

[0066] This module skips the Hessian eigenvalue calculation and only relies on the analysis of the Hessian Matrix to determine the sign (positive or negative) of the corresponding eigenvalue, and selects all regions with negative eigenvalue signs as potential regions of aneurysm, thereby greatly improving the speed of the algorithm. Limiting the calculation range to the region of the blood vessels in module four can further improve the calculation speed. The results obtained by this module are referred to as suspected aneurysms.

[0067] Aneurysm screening

[0068] This module needs to screen the above-mentioned suspected aneurysms to eliminate false positive results. For Figure 3 the type differences of aneurysms, the present application uses the following two methods to screen aneurysms:

[0069] 1) Saccular aneurysm. If a suspected aneurysm meets all the following conditions, it is classified as a saccular aneurysm:

[0070] a) V > 4 / 3 * 3.14 * R0 3 . Where V is the volume of the aneurysm, and R0 is a threshold value (which can be set to 1.0 mm).

[0071] b) Contains a terminal segment;

[0072] c) K / N > Th0. Where N is the number of center points of the terminal segment, K is the number of center points of the terminal segment located within the aneurysm, and Th0 is a threshold value (which can be set to Th0 = 0.3).

[0073] 2) Dissecting aneurysm or fusiform aneurysm. If a suspected aneurysm meets all the following conditions, it will be classified into this category:

[0074] a) V > 4 / 3 * 3.14 * Ro 3 where V is the volume of the aneurysm and Ro is a threshold value (which can be set as 1.0 mm here).

[0075] b) Contains a main segment, hereinafter referred to as Ps. The aneurysm can contain multiple segments, and Ps is defined as the segment containing the largest number of center points within the aneurysm.

[0076] c) R I / R O > Th1, where R I is the average radius of the center points of Ps located within the aneurysm, R O is the average radius of the center points of Ps located outside the aneurysm (if the number of center points outside the aneurysm is more than the number of center points within the aneurysm, only the center points outside the aneurysm equal to the number of center points within the aneurysm are taken), and Th1 is a threshold value (which can be set as Th1 = 2.0 here).

[0077] Example 8: Aneurysm display

[0078] The aneurysm obtained by module seven can be directly displayed with the original image superimposed, or can be displayed with the maximum density projection image, volume rendering image, and blood vessel segmentation image of module four superimposed.

[0079] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or replacements easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A rapid full-automatic intracranial aneurysm auxiliary detection post-processing system, characterized in that, The function modules include but are not limited to: The image preprocessing module makes MRI, CT or 3D-DSA medical image uniform; The image adaptive threshold classification module analyzes the gray histogram of the input image to obtain a threshold, and classifies the image into a background, a low-signal blood vessel region or a high-signal blood vessel region according to the signal value of the image; The automatic blood vessel seed point searching module performs connected domain analysis on the output high-signal blood vessel region and finds corresponding blood vessels; The blood vessel centerline extraction, segmentation and classification module extracts the centerline based on the blood vessel segmentation result of the fourth module, and classifies the centerline into different segments according to the bifurcation points; The aneurysm enhancement module skips the Hessian eigenvalue calculation, only relies on the analysis of the Hessian Matrix, judges the sign of the corresponding eigenvalue, and selects all regions with negative eigenvalue signs as potential regions of aneurysms to improve the speed of the algorithm; The aneurysm screening module needs to screen suspected aneurysms to eliminate false positive results; The aneurysm display module directly superimposes the screened aneurysms on the original image or superimposes them on the maximum intensity projection image, volume rendering image and blood vessel segmentation image of the original image. The image preprocessing is to remove the scalp from the MRI or CT image to obtain the brain and cerebellum regions, and mark other regions as background; or perform an inverse operation on the 3D-DSA image to unify the input image: the blood vessels are high-signal and the other tissues are low-signal.

2. The rapid fully automatic intracranial aneurysm auxiliary detection post-processing system according to claim 1, characterized in that, The image adaptive threshold classification is to analyze the gray histogram of the input image to obtain thresholds T1 and T2, and classify the image into three categories: signal value < T1 is background, T1 ≤ signal value < T2 is a low-signal blood vessel region, and signal value ≥ T2 is a high-signal blood vessel region.

3. The rapid fully automatic intracranial aneurysm auxiliary detection post-processing system according to claim 1, characterized in that, The automatic blood vessel seed point searching is to perform connected domain analysis on the high-signal blood vessel region, sort the regions by volume from large to small, obtain the three largest connected regions M1, M2 and M3, and satisfy the conditions: M2 ≥ 0.1*M1 and M3 ≥ 0.1*M1; if M2 or M3 does not meet the above conditions, it is excluded; 4. The rapid fully automatic intracranial aneurysm auxiliary detection post-processing system according to claim 1, characterized in that, Finally, a pixel point is randomly selected as a seed point from each region that meets the conditions; For MRI and CT images, the above analysis and sorting can find up to three seed points corresponding to the left common carotid artery, the right common carotid artery and the vertebral artery; for 3D-DSA images, one seed point can be found corresponding to a certain branch of the blood vessel. The aneurysm refers to a saccular aneurysm or an inter-mural aneurysm, and a fusiform aneurysm.

5. The rapid fully automatic intracranial aneurysm auxiliary detection post-processing system according to claim 1, characterized in that, ​ 6. The rapid fully automatic intracranial aneurysm auxiliary detection post-processing system according to claim 5, characterized in that, The saccular aneurysm satisfies all the following conditions: a) V>4 / 3*3.14*R03; wherein, V is the volume of the aneurysm, R0 is a threshold value; b) contains a terminal segment; c) K / N>Th0; wherein N is the number of center points of the terminal segment, K is the number of center points of the terminal segment located in the aneurysm, Th0 is a threshold value.

7. The rapid fully automatic intracranial aneurysm auxiliary detection post-processing system according to claim 5, characterized in that, The inter-wall aneurysm or fusiform aneurysm satisfies all the following conditions: a) V>4 / 3*3.14*R03; wherein, V is the volume of the aneurysm, R0 is a threshold value; b) contains a main segment; c) RI / RO>Th1; wherein, RI is the average radius of the center points of Ps located in the aneurysm, RO is the average radius of the center points of Ps located outside the aneurysm, Th1 is a threshold value, which can be set as Th1=2.

0.

8. A method for rapid fully automated post-processing of an intracranial aneurysm assist detection, characterized in that The method realizes the identification of intracranial aneurysms in MRI, CT and 3D-DSA medical images through the following steps: The steps include: Image preprocessing, unifying MRI, CT or 3D-DSA medical images; Image adaptive threshold classification, obtaining the threshold value by analyzing the gray histogram of the input image, and classifying it into background, low signal blood vessel region or high signal blood vessel region according to the signal value of the image; Automatic search for blood vessel seed points, performing connected domain analysis on the high signal blood vessel region output by the image adaptive threshold classification, and finding the corresponding blood vessels; Vessel segmentation, region growing from the obtained seed points, the growth range being limited to the low signal and high signal blood vessel regions in the image adaptive threshold classification, and performing level set fine segmentation on the results to obtain blood vessels; Vessel centerline extraction, segmentation and classification, based on the results of vessel segmentation, performing centerline extraction, and dividing the centerline into different segments according to the bifurcation points; Aneurysm enhancement, skipping Hessian eigenvalue calculation, only relying on the analysis of Hessian Matrix to judge the sign of the corresponding eigenvalue, and selecting all regions with negative eigenvalue signs as the potential region of the aneurysm, thereby improving the speed of the algorithm; Aneurysm screening, screening the suspected aneurysms detected by the above detection method to eliminate false positive results; Aneurysm display, directly superimposing the screened aneurysms on the original image, or superimposing them on the maximum density projection image, volume rendering image and blood vessel segmentation image of the original image.

9. The system according to any one of claims 1 to 7 for use in a post-processing method for assisting detection of an aneurysm.

10. The use according to claim 9, characterized in that, The system automatically completes the identification or display of intracranial aneurysms.

Images