Apparatus for detecting cerebral aneurysm based on multiple segmentation of cerebrovascular system

Abstract

The present invention relates to a cerebral aneurysm detection device based on multiple segmentation of the cerebral vascular system, comprising: an image interface unit for receiving a 3D cerebral vascular image; an image preprocessing unit for preprocessing the 3D cerebral vascular image received through the image interface unit; a cerebral artery loop segmentation model for extracting a 3D cerebral artery loop segmentation map divided into multiple artery segments by artery of the cerebral artery loop from the preprocessed 3D cerebral vascular image through a pre-set artificial intelligence-based segmentation model; a blood vessel map extraction module for extracting a 3D cerebral vascular map from the preprocessed 3D cerebral vascular image through a pre-set blood vessel skeleton extraction algorithm; and a final segmentation map generation module for generating a cerebral vascular system segmentation map for the entire cerebral vascular system together with the cerebral artery loop within the cerebral artery loop region, based on the cerebral artery loop segmentation map, by dividing the cerebral blood vessels within the cerebral vascular map that are excluding the cerebral artery loop region within the cerebral artery loop segmentation map into multiple artery segments by artery. It is characterized by including an artificial intelligence-based cerebral aneurysm extraction model that receives the cerebral vascular segmentation map generated by the final segmentation map generation module and extracts the cerebral aneurysm region. Through this, it becomes possible to detect cerebral aneurysms by segmenting the entire cerebral vascular system without requiring vast computational resources for training.

Description

Technology Field

[0001] The present invention relates to a cerebral aneurysm detection device, and more specifically, to a cerebral aneurysm detection device that detects cerebral aneurysms based on multiple segmentation of the cerebral vascular system. Background Technology

[0002] In modern medicine, medical imaging is a crucial tool for the effective diagnosis of diseases and the treatment of patients. Furthermore, advancements in imaging technology have made it possible to acquire increasingly sophisticated medical image data. As a trade-off for this sophistication, the volume of data is becoming increasingly massive, posing significant challenges to the analysis of medical image data relying solely on human vision. Consequently, over the past decade, clinical decision support systems and computer-assisted reading systems have played an essential role in the automated analysis of medical images.

[0003] Conventional clinical decision support systems or computer-assisted reading systems detect and display lesion areas or present reading information to medical staff or medical professionals (hereinafter referred to as users).

[0004] For example, the 'Method and Apparatus for Calculating Disease Diagnosis Information Based on Medical Images' disclosed in Korean Published Patent No. 10-2017-0017614 includes the steps of detecting a region of interest in which an object to be analyzed is captured, calculating a coefficient of variation, creating a coefficient of variation image, and comparing it with a reference sample, and describes the effect of diagnosing the degree of a patient's disease by utilizing medical images obtained through CT, MRI, MRA, and ultrasound imaging devices.

[0005] In particular, recently, artificial intelligence (AI) technology based on machine learning, such as deep learning, has been the foundation for rapid advancements in diagnosing patients' diseases using medical images.

[0006] Deep learning refers to a machine learning method based on artificial neural networks that enables machines to learn by mimicking human biological neurons. Recently, deep learning technology has been advancing rapidly in the field of image recognition and is also widely used in the field of medical image interpretation.

[0007] Deep learning technology in medical imaging involves machine learning being performed using multiple medical images containing diseases and the diseases as training data to generate a machine learning model (hereinafter referred to as the 'diagnosis model'), and when a medical image to be analyzed is input into the diagnosis model, the presence of a lesion is diagnosed.

[0008] As described above, deep learning-based machine learning methods collect training data used to generate a diagnosis model, such as multiple images of lesions and multiple images of normal people, and the diagnosis model generated by learning from the collected images of lesions and normal people determines whether a newly input medical image to be diagnosed has a lesion. Thus, the accuracy of the diagnosis result can be improved when diverse and large amounts of training data are used for training. Furthermore, the accuracy of the diagnosis result can be enhanced when the medical image to be diagnosed is similar to the training data.

[0009] Meanwhile, a cerebral aneurysm refers to the abnormal expansion of a portion of a vital blood vessel supplying blood to the brain due to weakening; if the condition worsens, it can rupture at the affected location, making it a dangerous disease that can lead to a cerebral hemorrhage (subarachnoid hemorrhage).

[0010] In particular, subarachnoid hemorrhage, which can occur from a cerebral aneurysm, is a significantly fatal disease, and since most ruptures occur without warning signs, about 30 to 60 percent of patients die or suffer severe disabilities.

[0011] Cerebral aneurysms and subarachnoid hemorrhage caused by cerebral aneurysms occur most frequently in the working-age population in their 50s and 60s, and because the incidence is also high in those in their 30s and 40s, it is of significant importance not only to the individual patient but also at the social and economic level.

[0012] Since the management and treatment of cerebral aneurysms vary depending on the location, size, and health condition of the patient, early detection and accurate identification of the location can contribute to preventing serious complications associated with subarachnoid hemorrhage.

[0013] As medical imaging technologies that can be used to detect cerebral aneurysms, Computed Tomography Angiography (CTA), Magnetic Resonance Angiography (MRA), and Digital Subtraction Angiography (DSA) are utilized as technologies that visualize three-dimensional cerebral vascular structures and perform image analysis.

[0014] Conventional vascular imaging-based diagnosis and analysis of cerebral aneurysms have relied primarily on qualitative interpretation based on visual inspection and the experiential knowledge of experts, which requires a relatively large amount of time and effort. Furthermore, results can vary depending on the subjective opinion of the interpreter, leading to issues that undermine diagnostic consistency and accuracy.

[0015] As previously explained, automated vessel segmentation models based on artificial intelligence for the detection of cerebral aneurysms have recently been developed. For example, the paper "eICAB: A novel deep learning pipeline for Circle of Willis multiclass segmentation and analysis (NeuroImage 260. 2022. 119425)" by Flix Dumais et al. proposes an automated method that can accurately segment the Circle of Willis (CW) in a given 3D MRA image and label it according to anatomical nomenclature.

[0016] However, the aforementioned AI models have technical limitations in effectively segmenting the complex overall structure of cerebral arteries. In particular, because technical constraints exist that limit the segmentation to the loop of cerebral arteries rather than the entire cerebrovascular system in vascular images, it poses significant difficulties in detecting cerebral aneurysms occurring outside the loop of cerebral arteries.

[0017] Furthermore, considering the technical aspects, there is a problem in that the development and training of a whole cerebrovascular segmentation model requires vast computational resources. The problem to be solved

[0018] Accordingly, the present invention has been devised to resolve the aforementioned problems and aims to provide a cerebral aneurysm detection device based on multiple cerebral vascular system segmentation that enables the detection of cerebral aneurysms through segmentation of the entire cerebral vascular system without requiring vast computational resources for learning.

[0019] In addition, another objective of the present invention is to provide a cerebral vascular system multi-segmentation-based cerebral aneurysm detection device capable of detecting cerebral aneurysms in the entire cerebral vascular system for various data types, not limited to 3D MRA images. means of solving the problem

[0020] The above objective is achieved, according to the present invention, by an image interface unit receiving a three-dimensional cerebral blood vessel image; an image preprocessing unit preprocessing the three-dimensional cerebral blood vessel image received through the image interface unit; a cerebral artery loop segmentation model that extracts a three-dimensional cerebral artery loop segmentation map divided into multiple artery segments by artery of the cerebral artery loop from the preprocessed three-dimensional cerebral blood vessel image using a preset artificial intelligence-based segmentation model; a blood vessel map extraction module that extracts a three-dimensional cerebral blood vessel map from the preprocessed three-dimensional cerebral blood vessel image using a preset blood vessel skeleton extraction algorithm; and a final segmentation map generation module that, based on the cerebral artery loop segmentation map, divides the cerebral blood vessels within the cerebral blood vessel map that are excluding the cerebral artery loop region within the cerebral artery loop segmentation map into multiple artery segments by artery, thereby generating a cerebral blood vessel system segmentation map for the entire cerebral vascular system together with the cerebral artery loop within the cerebral artery loop region. This is achieved by a cerebrovascular multi-segmentation-based cerebrovascular aneurysm detection device characterized by including an artificial intelligence-based cerebrovascular aneurysm extraction model that receives the cerebrovascular segmentation map generated by the final segmentation map generation module and extracts the cerebrovascular aneurysm region.

[0021] Here, the three-dimensional cerebrovascular image received through the image interface unit may include any one of a CTA image, an MRA image, and a DSA image in DICOM format.

[0022] In addition, the image preprocessing unit converts the 3D cerebral blood vessel image in the DICOM format into a 3D cerebral blood vessel image in the Nifti format; and can preprocess the 3D cerebral blood vessel image in the Nifti format so that it can be input into the cerebral aneurysm extraction model.

[0023] In addition, the above-mentioned vascular skeleton extraction algorithm may include the Gumbel algorithm or the unilateral normal distributions algorithm.

[0024] And, the final segmentation map generation module comprises: a linear sum module that generates a cerebral blood vessel multi-map by linearly summing the cerebral artery loop segmentation map and the cerebral blood vessel map; a clustering module that sets a 3D cube box region including the cerebral artery loop region in the cerebral blood vessel multi-map and clusters the remaining cerebral blood vessels existing within the region excluding the 3D cube box region into multiple blood vessel clusters using a pre-set clustering algorithm; and a residual segmentation model that divides the remaining cerebral blood vessels excluding the cerebral artery loop within the cerebral blood vessel multi-map into multiple artery segments by artery through a pre-set segmentation algorithm; It may include a resegmentation module that performs a Hadamard product on the blood vessel clusters clustered by the clustering module and the artery segments divided by the residual division model, and for one blood vessel cluster clustered by the clustering module, resegments target segments divided into two or more artery segments by the residual division model into one artery segment based on the Euclidean distance from the center of the cube box area; and a segment integration module that generates the final cerebrovascular system segmentation map by reflecting the artery segments resegments divided into one segment by the resegmentation module into the artery segments divided by the residual division model.

[0025] In addition, the clustering algorithm may include a connected component algorithm.

[0026] In addition, the above partitioning algorithm may include the KNN (K-Nearest Neighbors) algorithm.

[0027] In addition, the above KNN (K-Nearest Neighbors) algorithm can learn the cerebral artery loop segmentation map and perform resegmentation on the remaining cerebral blood vessels.

[0028] In addition, the re-segmentation module can extract multiple voxel coordinates from each of two or more arterial segments constituting the re-segmentation target segment, calculate the Euclidean distance between the center coordinates of the cube box area and each of the voxel coordinates, and calculate the average value of the Euclidean distance for each arterial segment, thereby re-segmenting the remaining arterial segments into the arterial segment having the smallest average value.

[0029] In addition, the above-described cerebral aneurysm detection model can count the number of voxels of the cerebral aneurysm region included in each artery when the cerebral aneurysm region is detected at a branching point of an artery constituting the cerebral vascular system, and provide cerebral aneurysm information based on the counted number of voxels. Effects of the invention

[0030] According to the above configuration, the present invention provides a cerebral vascular system multi-segmentation-based cerebral aneurysm detection device capable of detecting cerebral aneurysms through segmentation of the entire cerebral vascular system without requiring vast computational resources for learning.

[0031] In addition, according to the present invention, a cerebral vascular system multi-segmentation-based cerebral aneurysm detection device is provided that is not limited to 3D MRA images and can detect cerebral aneurysms in the entire cerebral vascular system for various data types. Brief explanation of the drawing

[0032] FIG. 1 is a diagram showing an example of the configuration of a multi-segmentation-based cerebral aneurysm detection system according to an embodiment of the present invention, and FIG. 2 is a diagram showing an example of the configuration of a multi-segmentation-based cerebral aneurysm detection device according to an embodiment of the present invention, and FIG. 3 is a diagram showing an example of the configuration of a final segmentation map generation module according to an embodiment of the present invention, and FIGS. 4 to 7 are drawings illustrating the operation process of a multi-segmentation-based cerebral aneurysm detection device according to an embodiment of the present invention. Specific details for implementing the invention

[0033] The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail.

[0034] However, this is not intended to limit the invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention.

[0035] The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0036] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0037] Hereinafter, embodiments according to the present invention will be described in detail with reference to the attached drawings.

[0038] FIG. 1 is a diagram showing an example of the configuration of a multi-segmentation-based cerebral aneurysm detection system according to an embodiment of the present invention.

[0039] Referring to FIG. 1, a cerebral aneurysm detection system according to an embodiment of the present invention may be configured to include a plurality of 3D imaging devices (50), a DICOM server (30), and a cerebral aneurysm detection device (10).

[0040] Each 3D imaging device (50) can image the cerebral blood vessels of a patient, etc. to generate a three-dimensional image of the cerebral blood vessels. In one embodiment, the 3D imaging device (50) may include at least one of a CTA imaging device (50), an MRA imaging device (50), and a DSA imaging device (50), and each may generate a three-dimensional CTA image, an MRA image, and a DSA image.

[0041] CTA is a technique that uses X-rays to image blood vessels. It is a non-invasive imaging technology that visualizes vascular structures in three dimensions by injecting an iodine contrast agent and then scanning with a CT scanner. While it is relatively inexpensive and allows for rapid results due to the short examination time, there is a possibility of patient exposure to radiation because it uses X-rays.

[0042] MRA is a technique that uses a strong magnetic field to generate images of blood vessels and is a non-invasive imaging technology that visualizes vascular structures without the use of contrast agents. While it involves no radiation exposure and has fewer side effects because it does not require contrast agents, the examination time is relatively long and the cost is also relatively high.

[0043] DSA is an imaging technique that visualizes blood vessels through X-ray imaging after injecting a contrast agent. While it can detect vascular abnormalities with very high resolution and accuracy, it is an invasive method that requires the insertion of a contrast agent and a catheter, and because it uses X-rays, there is a risk of radiation exposure.

[0044] The three-dimensional brain blood vessel image generated by being captured by the three-dimensional imaging equipment (50) is transmitted to the DICOM server (30) via the network (70) and stored in the DICOM server (30). Here, the DICOM server (30) converts the three-dimensional brain blood vessel image transmitted from each three-dimensional imaging equipment (50) into a DICOM format and stores it.

[0045] A cerebral aneurysm detection device (10) according to an embodiment of the present invention receives a 3D cerebral blood vessel image of a target patient from a 3D cerebral blood vessel image stored in a DICOM server (30) via a network (70), and can detect a cerebral aneurysm in the 3D cerebral blood vessel image.

[0046] FIG. 2 is a diagram showing an example of the configuration of a multi-segment-based cerebral aneurysm detection device (10) according to an embodiment of the present invention.

[0047] Referring to FIG. 2, the cerebral aneurysm detection device (10) according to an embodiment of the present invention may be configured to include a device control unit (700). The device control unit (700) may include a hardware configuration and a software configuration that controls the overall function of the cerebral aneurysm detection device (10) according to an embodiment of the present invention.

[0048] A cerebral aneurysm detection device (10) according to an embodiment of the present invention may be configured to include an image interface unit (100). The image interface unit (100) connects to a DICOM server (30) via a network (70) and receives a 3D cerebral blood vessel image that is a target for cerebral aneurysm detection among the 3D cerebral blood vessel images stored in the DICOM server (30).

[0049] In one embodiment, the video interface unit (100) may be a TCP-IP-based wired or wireless internet and may be a mobile communication network such as 5G. Here, it goes without saying that the communication protocol used by the video interface unit (100) may take various forms.

[0050] A cerebral aneurysm detection device (10) according to an embodiment of the present invention may be configured to include an image preprocessing unit (200). Here, the image preprocessing unit (200) preprocesses a three-dimensional cerebral blood vessel image received through an image interface unit (100), and can preprocess the three-dimensional cerebral blood vessel image to match the format of the input data of a cerebral aneurysm extraction model (600) to be described later.

[0051] In one embodiment, the image preprocessing unit (200) can convert a 3D brain blood vessel image in DICOM format into a 3D brain blood vessel image in Nifti format. In addition, the image preprocessing unit (200) can perform at least one of re-orientation, skull-stripping, resampling, and image normalization in accordance with the format of the input data of the brain aneurysm extraction model (600).

[0052] A cerebral aneurysm detection device (10) according to an embodiment of the present invention may be configured to include a cerebral artery loop segmentation model (300). The cerebral artery loop segmentation model (300) receives a 3D cerebral blood vessel image preprocessed by an image preprocessing unit (200) and can extract a 3D cerebral artery loop segmentation map divided into multiple artery segments for each artery of the cerebral artery loop through a pre-set artificial intelligence-based segmentation model.

[0053] In one embodiment, the cerebral artery loop segmentation model (300) can extract a three-dimensional cerebral artery loop segmentation map by dividing the cerebral artery loop into multiple artery segments for each artery using the vascular segmentation model disclosed in the paper by Flix Dumais et al. described above.

[0054] FIG. 4(a) is a diagram showing an example of a cerebral artery loop segmentation map extracted by the cerebral artery loop segmentation model (300). FIG. 4(b) is a diagram showing an example of dividing the entire cerebral vascular system by artery, and is a cerebral vascular system segmentation map generated by the final segmentation map generation module (500) to be described later.

[0055] In FIG. 4, the Intracranial Carotid Artery (ICA) is the internal carotid artery, the Basilar Artery (BA) is the basilar artery, the Anterior Communicating Artery (ACA) is the anterior communicating artery, the Posterior Cerebral Artery (PCA) is the posterior cerebral artery, the Middle Cerebral Artery (MCA) is the middle cerebral artery, and the Superior cerebellar artery (SCA) is the superior cerebellar artery. In an embodiment of the present invention, the artery segments divided by artery are shown as being distinguished by color.

[0056] Here, as shown in FIG. 4, the cerebral artery loop segmentation model (300) segments and extracts only the cerebral artery loop from the entire cerebral vascular system.

[0057] Meanwhile, a cerebral aneurysm extraction device according to an embodiment of the present invention may be configured to include a blood vessel map extraction module (400). Here, the blood vessel map extraction module (400) receives a 3D cerebral blood vessel image preprocessed by an image preprocessing unit (200) and can extract a 3D cerebral blood vessel map using a pre-set blood vessel skeleton extraction algorithm. In one embodiment, the blood vessel skeleton extraction algorithm may include a Gumbel algorithm or a unilateral normal distributions algorithm.

[0058] FIG. 5(a) is a diagram showing an example of a cerebral blood vessel map extracted by a blood vessel map extraction module (400) according to an embodiment of the present invention. As shown in FIG. 5(a), the blood vessel map extracted by the blood vessel map extraction module (400) extracts only the skeleton of blood vessels constituting the cerebral vascular system, thereby reflecting the entire cerebral vascular system, but the skeleton of the entire blood vessel is extracted without being divided by artery.

[0059] Generally, when extracting the cerebral vascular skeleton using the Gumbel algorithm or unilateral normal distributions algorithm, the vessels constituting the cerebral arterial loop tend to be extracted relatively thinly; therefore, using this to detect cerebral aneurysms is undesirable as it often results in a significant decrease in detection accuracy.

[0060] Accordingly, the cerebral aneurysm detection device (10) according to an embodiment of the present invention may be configured to include a final segmentation map generation module (500).

[0061] Here, the final segmentation map generation module (500) generates a cerebral vascular system segmentation map for the entire cerebral vascular system using the cerebral artery loop segmentation map extracted by the cerebral artery loop segmentation model (300) and the cerebral blood vessel map extracted by the blood vessel map extraction module (400).

[0062] More specifically, the final segmentation map generation module (500) can generate a cerebrovascular system segmentation map for the entire cerebrovascular system, as shown in FIG. 4 (b), by dividing the cerebrovascular vessels within the cerebrovascular map, excluding the cerebral artery loop region within the cerebral artery loop segmentation map, into multiple artery segments by artery based on the cerebral artery loop segmentation map, together with the cerebral artery loop within the cerebral artery loop region.

[0063] Hereinafter, a final segmentation map generation module (500) according to an embodiment of the present invention will be described in detail with reference to FIGS. 3 to 7.

[0064] As illustrated in FIG. 3, the final partition map generation module (500) according to an embodiment of the present invention may be configured to include a linear sum module (510), a clustering module, a residual partition model (530), a re-partition module (540), and a segment integration module (550).

[0065] A linear sum module (510) according to an embodiment of the present invention can generate a cerebral blood vessel multi-map by linearly summing a cerebral artery loop segmentation map and a cerebral blood vessel map. FIG. 5(b) is a diagram showing an example of a cerebral blood vessel multi-map generated by the linear sum module (510).

[0066] In one embodiment, the linear sum module (510) can binaryize the cerebral artery loop segmentation map. That is, among the voxels of the cerebral artery loop segmentation map, the voxels corresponding to the cerebral blood vessels can be binaryized to '1' and the rest to '0'. Here, the cerebral blood vessel map is generally binaryized such that the blood vessel region is '1' and the remaining region is '0'.

[0067] In this way, the linear sum module (510) binary-encodes the cerebral artery loop segmentation map and then performs a linear sum between the corresponding voxels of the cerebral artery loop segmentation map and the cerebral blood vessel map. Voxels with overlapping blood vessels are calculated as '2', voxels with blood vessels on only one side are calculated as '1', and voxels with no blood vessels on both sides are calculated as '0'. At this time, if the voxels with a value of '2' are converted to '1', a cerebral blood vessel multi-map having a mutually complementary framework of the entire cerebral blood vessel system can be generated.

[0068] Meanwhile, the clustering module (520) according to an embodiment of the present invention can set a three-dimensional cube box area including a cerebral artery loop area in a cerebral blood vessel multi-map. In addition, the clustering module (520) can cluster the remaining cerebral blood vessels existing within the area excluding the three-dimensional cube box area in the cerebral blood vessel multi-map into a plurality of clusters using a pre-set clustering algorithm.

[0069] In one embodiment, the clustering module (520) can set a 3D cube box based on the position of a voxel located at the outermost edge in each direction in 3D among the blood vessels corresponding to the cerebral artery loop within the cerebral blood vessel multimap. FIG. 6 (a) is a drawing showing an example of a 3D cube box, FIG. 6 (b) is a 2D drawing showing an example of a cube box set in a cerebral blood vessel multimap, and FIG. 6 (c) is a 2D drawing showing an example of a remaining cerebral blood vessel existing within an area excluding the 3D cube box area.

[0070] Here, the clustering module (520) clusters the remaining cerebral blood vessels into multiple clusters using a clustering algorithm for the remaining cerebral blood vessels, as shown in (c) of FIG. 6.

[0071] In one embodiment, the clustering algorithm applied to the clustering module (520) may include a connected component algorithm.

[0072] For example, the residual cerebral blood vessels shown in Fig. 6 (c) can be clustered into eight blood vessel clusters as shown in Fig. 7 (a).

[0073] Meanwhile, the residual division model (530) according to an embodiment of the present invention can divide the residual cerebral blood vessels, excluding the cerebral artery loop within the cerebral blood vessel multi-map, into a plurality of artery segments by artery through a pre-set division algorithm.

[0074] In one embodiment, the segmentation algorithm applied to the residual segmentation model (530) may include a KNN (K-Nearest Neighbors) algorithm. Here, the KNN (K-Nearest Neighbors) algorithm can perform re-segmentation on residual cerebral blood vessels by learning a cerebral artery loop segmentation map.

[0075] Figure 7(b) shows an example of an image re-segmented by applying the KNN (K-Nearest Neighbors) algorithm to the cerebral blood vessel multi-map shown in Figure 5(b), and it can be seen that the unsegmented blood vessels corresponding to the cerebral blood vessel map have been re-segmented by referring to the segmentation of the cerebral artery loop segmentation map.

[0076] The re-segmentation module (540) according to an embodiment of the present invention can perform a Hadamard product on the blood vessel clusters clustered by the clustering module (520) and the artery segments segmented by the residual segmentation model (530).

[0077] And, the resegmentation module (540) can resegment a segment to be resegmented, which is divided into two or more arterial segments by the residual segmentation model (530) for a single blood vessel clustered by the clustering module (520), into one arterial segment based on the Euclidean distance from the center of the cube box area.

[0078] Referring to Figures 7 (a) and (b), it can be seen that a cluster of blood vessels within the dotted circle in Figure 7 (a) is divided into two arterial segments within the dotted circle in Figure 7 (b), that is, at the same coordinates.

[0079] That is, through the clustering algorithm, it was clustered into a single blood vessel cluster, but the KNN (K-Nearest Neighbors) algorithm divides it into two different artery segments.

[0080] Accordingly, the re-segmentation module (540) integrates into one of the two artery segments based on the Euclidean distance from the center coordinates of the cube box to each artery segment.

[0081] More specifically, the re-segmentation module (540) extracts multiple voxel coordinates from each of the two arterial segments, for example, as shown in FIG. 7 (b). Here, the extracted voxel coordinates are exemplified by the extraction of the five closest to the center of the cube box.

[0082] And, the re-segmentation module (540) can calculate the Euclidean distance between the center coordinates of the cube box area and each voxel coordinate, and calculate the average value of the Euclidean distance for each artery segment. And, the re-segmentation module (540) can re-segment the remaining artery segments into the artery segment having the smallest average value and integrate them.

[0083] As described above, when the re-segmentation process by the re-segmentation module (540) is completed, the segment integration module (550) reflects the arterial segment re-segmented into one segment by the re-segmentation module (540) to the arterial segment divided by the residual segmentation model (530), thereby generating a cerebrovascular segmentation map as shown in FIG. 4 (b).

[0084] Referring again to FIG. 2, the cerebral aneurysm detection device (10) according to an embodiment of the present invention may be configured to include an artificial intelligence-based cerebral aneurysm extraction model (600).

[0085] A cerebral aneurysm extraction model (600) according to an embodiment of the present invention can extract a cerebral aneurysm region by receiving a cerebral vascular segmentation map generated by a final segmentation map generation module (500). Here, various known algorithms may be applied to the artificial intelligence model applied to the cerebral aneurysm extraction model (600).

[0086] Here, the cerebral aneurysm detection model can count the number of voxels of the cerebral aneurysm region contained in each artery when the cerebral aneurysm region is detected at a branching point of an artery constituting the cerebral vascular system, and provide cerebral aneurysm information based on the counted number of voxels.

[0087] In one embodiment, the location of the cerebral aneurysm and information about the cerebral aneurysm can be provided together on the image display unit (800) by listing the order from 1 based on the largest number of voxels or listing the number of voxels together.

[0088] Although some embodiments of the present invention have been illustrated and described, those skilled in the art will understand that modifications can be made to these embodiments without departing from the principles or spirit of the invention. The scope of the invention will be defined by the appended claims and their equivalents. Explanation of the symbols

[0089] 10: Cerebral Aneurysm Detection Device 30: DICOM Server 50 : 3D shooting equipment 70 : Network 100: Image Interface Unit 200: Image Preprocessing Unit 300: Cerebral artery loop segmentation model 400: Vascular map extraction module 500: Final Partition Map Generation Module 510: Linear Sum Module 520: Clustering Module 530: Residual Partition Model 540: Re-partitioning Module 550: Segment Integration Module 600: Cerebral Aneurysm Extraction Model 700: Device Control Unit 800 : Image display unit

Claims

Claim 1 An image interface unit for receiving a 3D cerebral vascular image; an image preprocessing unit for preprocessing the 3D cerebral vascular image received through the image interface unit; a cerebral artery loop segmentation model for extracting a 3D cerebral artery loop segmentation map divided into multiple artery segments by artery of the cerebral artery loop from the preprocessed 3D cerebral vascular image using a preset AI-based segmentation model; a blood vessel map extraction module for extracting a 3D cerebral vascular map from the preprocessed 3D cerebral vascular image using a preset blood vessel skeleton extraction algorithm; a final segmentation map generation module for generating a cerebral vascular system segmentation map for the entire cerebral vascular system together with the cerebral artery loop within the cerebral artery loop region, based on the cerebral artery loop segmentation map, by dividing the cerebral blood vessels within the cerebral vascular map that are excluding the cerebral artery loop region within the cerebral artery loop segmentation map into multiple artery segments by artery; and an AI-based A cerebrovascular system multi-segmentation-based cerebral aneurysm detection device characterized by including a cerebral aneurysm extraction model. Claim 2 A cerebral aneurysm detection device based on multiple segmentation of the cerebral vascular system according to claim 1, characterized in that the three-dimensional cerebral vascular image received through the image interface unit includes any one of a CTA image, an MRA image, and a DSA image in DICOM format. Claim 3 A cerebrovascular system multi-segmentation-based cerebrovascular aneurysm detection device according to claim 2, wherein the image preprocessing unit converts the 3D cerebrovascular image in the DICOM format into a 3D cerebrovascular image in the Nifti format and preprocesses the 3D cerebrovascular image in the Nifti format so that it can be input into the cerebrovascular aneurysm extraction model. Claim 4 A cerebral aneurysm detection device based on multiple vascular segmentation according to claim 1, characterized in that the vascular skeleton extraction algorithm includes a Gumbel algorithm or a unilateral normal distributions algorithm. Claim 5 In claim 1, the final segmentation map generation module comprises: a linear sum module that generates a cerebral blood vessel multi-map by linearly summing the cerebral artery loop segmentation map and the cerebral blood vessel map; a clustering module that sets a 3D cube box area including the cerebral artery loop region in the cerebral blood vessel multi-map and clusters the remaining cerebral blood vessels existing within the area excluding the 3D cube box area into multiple blood vessel clusters using a pre-set clustering algorithm; a residual segmentation model that divides the remaining cerebral blood vessels excluding the cerebral artery loop within the cerebral blood vessel multi-map into multiple artery segments by artery using a pre-set segmentation algorithm; and a Hadamard product of the blood vessel clusters clustered by the clustering module and the artery segments divided by the residual segmentation model, and for one blood vessel clustered by the clustering module, a re-segmentation target segment divided into two or more artery segments by the residual segmentation model based on the Euclidean distance from the center of the cube box area A cerebrovascular multi-segmentation-based cerebral aneurysm detection device characterized by comprising: a resegmentation module that resegments into arterial segments; and a segment integration module that generates a cerebrovascular segmentation map by reflecting the arterial segment resegmented into one segment by the resegmentation module into the arterial segment divided by the residual segmentation model. Claim 6 A cerebral aneurysm detection device based on multiple segmentation of the cerebrovascular system, characterized in that, in claim 5, the clustering algorithm includes a connected component algorithm. Claim 7 A cerebrovascular multi-segmentation-based cerebral aneurysm detection device according to claim 5, characterized in that the segmentation algorithm includes a KNN (K-Nearest Neighbors) algorithm. Claim 8 A cerebral aneurysm detection device based on multiple segmentation of the cerebral vascular system, characterized in that, in claim 7, the KNN (K-Nearest Neighbors) algorithm learns the cerebral artery loop segmentation map and performs re-segmentation on the remaining cerebral blood vessels. Claim 9 A cerebral aneurysm detection device based on multiple segmentation of a cerebrovascular system, wherein, in claim 5, the re-segmentation module extracts a plurality of voxel coordinates from each of two or more arterial segments constituting the re-segmentation target segment, calculates the Euclidean distance between the center coordinates of the cube box area and each of the voxel coordinates, calculates the average value of the Euclidean distance for each arterial segment, and re-segments the remaining arterial segments into the arterial segment having the smallest average value. Claim 10 A cerebral vascular system multi-segmentation-based cerebral aneurysm detection device according to claim 1, wherein the cerebral aneurysm detection model counts the number of voxels of the cerebral aneurysm region included in each artery when the cerebral aneurysm region is detected at a branching point of an artery constituting the cerebral vascular system, and provides cerebral aneurysm information based on the counted number of voxels.

Images