Apparatus and method for analyzing brain image

Abstract

Disclosed are an apparatus and method for analyzing a brain image. The present invention trains an artificial intelligence model with an MRA brain image that is easily captured and a DSA brain image effective for detecting a brain aneurysm to output information about the brain aneurysm, and when the training of the artificial intelligence model is completed, inputs the MRA brain image to the artificial intelligence model to output brain aneurysm information, thereby exhibiting an effect of detecting a brain aneurysm by using an MRA brain image with substantially the same accuracy as when detecting a brain aneurysm with a DSA brain image.

Description

Brain imaging analysis device and method

[0001] The present invention relates to a brain image analysis device and method, and more specifically, to a brain image analysis device and method for analyzing a brain aneurysm using a brain image.

[0002] Recently, vascular diseases are on the rise due to changes in dietary habits and the aging of the population. Among vascular diseases, cerebrovascular disease, in particular, develops suddenly, and if medical measures are not taken promptly, it can lead to death or leave severe aftereffects, highlighting its severity.

[0003] Cerebrovascular diseases, represented by stroke (ischemic and hemorrhagic), are diseases caused by pathological phenomena such as stenosis, occlusion, rupture, or degeneration of blood vessels. In terms of blood flow, they can be described as a sudden decrease in blood flow (cerebral infarction), persistently low blood flow (cerebrovascular disease, vascular dementia), and leakage of blood out of the blood vessels.

[0004] A representative case of blood leakage is cerebral hemorrhage, and one type of this is bleeding caused by the rupture of a cerebral aneurysm. A cerebral aneurysm is a condition in which a portion of a cerebral artery weakens and swells up like a balloon or sac, and it mainly occurs at large branching points of blood vessels.

[0005] The walls of cerebral arteries are very thin and structurally different from normal blood vessels, so they rupture easily, and it is known that if a cerebral aneurysm ruptures, there is a high possibility of subarachnoid hemorrhage.

[0006] Cerebral aneurysms are diagnosed through CT scans of the brain and magnetic resonance angiography (MRA), and if medical staff determine that it is necessary to examine the patient's aneurysm and surrounding blood vessels in more detail, they perform digital subtraction angiography (DSA) to make a diagnosis.

[0007] Cerebral angiography is a diagnostic method that involves inserting a tube less than 2 mm in diameter into a blood vessel to directly inject a contrast agent and then taking X-ray images to check for abnormalities in the cerebral blood vessels. Digital subtraction angiography (DSA) is a fluoroscopic technique used in radiology to visualize blood vessels in bone or high-density soft tissue environments. Fluoroscopy is a technology that allows physicians to obtain real-time moving images of a patient's internal structures, with X-rays serving as the simplest form of fluoroscopic imaging. DSA images are generated by subtracting the pre-contrast image from the image after the contrast agent has been injected.

[0008] Although cerebral angiography is invasive compared to CT or MRI scans, it is the most accurate method for identifying overall cerebrovascular abnormalities, such as cerebral aneurysms, cerebral vascular malformations, Moyamoya disease, arteriovenous fistulas, carotid artery stenosis, cerebral vascular occlusion, and vasospasm after cerebral aneurysm surgery. It is utilized not only as a diagnostic tool but also as a treatment method.

[0009] However, this type of cerebral angiography (DSA) has disadvantages such as being invasive, exposure to X-rays, and the possibility of side effects such as allergic reactions to the contrast agent, as a contrast agent must be administered.

[0010] The problem that the present invention aims to solve is to provide a brain imaging analysis device and method that can identify brain aneurysms with the same performance as DSA using MRA images, which have relatively few side effects and are easy to take.

[0011] A brain image analysis device according to a preferred embodiment of the present invention for solving the above-mentioned problem is a brain image analysis device comprising a processor and a memory storing predetermined instructions, wherein the processor that executes the instructions stored in the memory performs the steps of: (a) receiving an MRA image and a DSA image of the same person's brain from an image providing device, and training the artificial intelligence model so that the artificial intelligence model receives the MRA image and outputs brain aneurysm information; (b) applying the MRA image of the brain to be analyzed to the artificial intelligence model that has completed training; and (c) outputting the brain aneurysm image output by the artificial intelligence model.

[0012] In addition, in step (a) above, the processor can register the DSA image by aligning it with the MRA image, receive brain aneurysm information read from the registered DSA image from the administrator terminal and input it into the artificial intelligence model as Ground Truth (GT) information, and input the MRA image as a training image to train the artificial intelligence model.

[0013] In addition, in step (a) above, the processor displays the registered DSA image through the administrator terminal and, upon receiving location information in which the brain aneurysm area is set from the administrator terminal, can generate the maximum length and neck length of the brain aneurysm in the DSA image and input them into the artificial intelligence model as Ground Truth (GT) information along with the location information.

[0014] In addition, in step (a) above, the artificial intelligence model can generate brain aneurysm prediction information including location information, maximum length, and neck length of the brain aneurysm from the MRA image, and can be trained to minimize the difference between the generated brain aneurysm prediction information and the location information, maximum length, and neck length of the brain aneurysm included in the GT information.

[0015] Additionally, in step (a) above, the processor may apply the cerebral aneurysm prediction information and the GT information to a loss function, and if the difference value output from the loss function is not smaller than a predefined threshold, adjust the weights of the artificial intelligence model and regenerate the prediction information and input it into the loss function, and repeat the adjustment of the weights of the artificial intelligence model and the generation of prediction information until the difference value output from the loss function becomes smaller than the threshold.

[0016] Additionally, the above step (a) comprises: (a1) receiving the MRA image and the DSA image, and registering the DSA image by aligning it with the MRA image; (a2) receiving brain aneurysm information including location information for a brain aneurysm read from the registered DSA image from an administrator terminal, generating the maximum length and neck length of the brain aneurysm, and inputting it into the artificial intelligence model as Ground Truth (GT) information along with the location information; (a3) ​​applying the MRA image to the artificial intelligence model to generate prediction information including the location information, maximum length, and neck length of the brain aneurysm; (a4) applying the GT information and the prediction information to a loss function to generate a difference value, and comparing the difference value with a predefined threshold; and (a5) if the difference value is not smaller than the threshold, adjusting the weights of the artificial intelligence model and repeating from step (a3).

[0017] Meanwhile, a brain image analysis method according to a preferred embodiment of the present invention for solving the above-mentioned problem is a brain image analysis method performed in a brain image analysis device comprising a memory storing instructions and a processor executing instructions stored in the memory, comprising: (a) a step in which a processor executing instructions stored in the memory receives an MRA image and a DSA image of the same person’s brain from an image providing device, and trains an artificial intelligence model such that the artificial intelligence model receives the MRA image and outputs brain aneurysm information; (b) a step in which the processor applies an analysis target MRA image of the brain to the artificial intelligence model that has completed training; and (c) a step in which the processor outputs a brain aneurysm image output by the artificial intelligence model.

[0018] In addition, in step (a) above, the processor can register the DSA image by aligning it with the MRA image, receive brain aneurysm information read from the registered DSA image from the administrator terminal and input it into the artificial intelligence model as Ground Truth (GT) information, and input the MRA image as a training image to train the artificial intelligence model.

[0019] In addition, in step (a) above, the processor displays the registered DSA image through the administrator terminal and, upon receiving location information in which the brain aneurysm area is set from the administrator terminal, can generate the maximum length and neck length of the brain aneurysm in the DSA image and input them into the artificial intelligence model as Ground Truth (GT) information along with the location information.

[0020] In addition, in step (a) above, the artificial intelligence model can generate brain aneurysm prediction information including location information, maximum length, and neck length of the brain aneurysm from the MRA image, and can be trained to minimize the difference between the generated brain aneurysm prediction information and the location information, maximum length, and neck length of the brain aneurysm included in the GT information.

[0021] Additionally, in step (a) above, the processor may apply the cerebral aneurysm prediction information and the GT information to a loss function, and if the difference value output from the loss function is not smaller than a predefined threshold, adjust the weights of the artificial intelligence model and regenerate the prediction information and input it into the loss function, and repeat the adjustment of the weights of the artificial intelligence model and the generation of prediction information until the difference value output from the loss function becomes smaller than the threshold.

[0022] Additionally, the above step (a) may include: (a1) the processor receiving the MRA image and the DSA image, and registering the DSA image by aligning it with the MRA image; (a2) the processor receiving brain aneurysm information including location information for a brain aneurysm read from the registered DSA image from an administrator terminal, generating the maximum length and neck length of the brain aneurysm, and inputting it into the artificial intelligence model as Ground Truth (GT) information along with the location information; (a3) ​​the processor applying the MRA image to the artificial intelligence model to generate prediction information including the location information, maximum length, and neck length of the brain aneurysm; (a4) the processor applying the GT information and the prediction information to a loss function to generate a difference value, and comparing the difference value with a predefined threshold; and (a5) if the difference value is not smaller than the threshold, the processor adjusting the weights of the artificial intelligence model and performing the process again starting from step (a3).

[0023] Meanwhile, a computer program according to a preferred embodiment of the present invention for solving the above-described problem is stored in a non-transient storage medium and executed on a computer including a processor to perform the brain image analysis method described above.

[0024] The present invention enables the detection of brain aneurysms using MRA brain images with substantially the same accuracy as when detecting brain aneurysms using DSA brain images by training an artificial intelligence model with MRA brain images, which are simple to capture, and DSA brain images, which are effective for detecting brain aneurysms, to output information regarding brain aneurysms, and by inputting MRA brain images into the artificial intelligence model when the training of the artificial intelligence model is completed to output information regarding brain aneurysms.

[0025] FIG. 1 is a drawing illustrating the configuration of an overall system to which a brain image analysis device is connected according to a preferred embodiment of the present invention and the detailed configuration of the brain image analysis device.

[0026] FIG. 2 is a flowchart illustrating a brain image analysis method according to a preferred embodiment of the present invention.

[0027] Figure 3 is a diagram illustrating the process of performing the brain image analysis method of the present invention.

[0028] FIG. 4 is a flowchart illustrating the learning process of an artificial intelligence model according to a preferred embodiment of the present invention.

[0029] Fig. 5a is a diagram explaining how a doctor inputs brain aneurysm information using DSA images, Fig. 5b is a diagram explaining an example of displaying brain aneurysm information on a DSA image, and Fig. 5c is a diagram explaining the maximum length and neck length of a brain aneurysm.

[0030] Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

[0031] Hereinafter, the aforementioned objects, features, and advantages of the present invention will become more apparent from the following detailed description in conjunction with the accompanying drawings. However, as the present invention is subject to various modifications and may have various embodiments, specific embodiments are illustrated in the drawings and described in detail below.

[0032] Throughout the specification, identical reference numbers indicate identical components in principle. Additionally, components with identical functions within the scope of the same concept appearing in the drawings of each embodiment are described using the same reference numeral.

[0033] When a part of a specification is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part" or "module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software.

[0034] If it is determined that a detailed description of known functions or configurations related to the present invention may unnecessarily obscure the essence of the present invention, such detailed description is omitted. Additionally, numbers used in the description of this specification (e.g., 1st, 2nd, etc.) are merely identification symbols to distinguish one component from another.

[0035]

[0036] FIG. 1 is a drawing illustrating the configuration of an overall system to which a brain image analysis device is connected according to a preferred embodiment of the present invention and the detailed configuration of the brain image analysis device.

[0037] Referring to FIG. 1, a brain image analysis device (100) according to a preferred embodiment of the present invention is connected to an image providing device (300) and an administrator terminal (200) through a wired or wireless communication network.

[0038] The image providing device (300) stores training MRA images and DSA images for training an artificial intelligence model and provides them to the brain image analysis device (100). In addition, it acquires an MRA image that is the actual subject of analysis and provides it to the brain image analysis device (100).

[0039] The image providing device (300) may be implemented as an image capturing device that captures human brain regions and outputs MRA brain images and DSA brain images in real time, or it may be implemented as a database server that stores previously captured MRA brain images and DSA brain images and provides them upon request from the brain image analysis device (100).

[0040] The administrator terminal (200) is connected to the brain image analysis device (100) and can input various setting information into the brain image analysis device (100). It should be noted that the administrator terminal (200) is a concept that includes not only an administrator operating the brain image analysis device (100) of the present invention, but also a terminal device used by a doctor who sets the brain aneurysm area in the DSA image during the learning process of the artificial intelligence model.

[0041] The brain image analysis device (100) receives an MRA brain image and a DSA brain image from an image providing device (300) and learns an artificial intelligence model stored internally, and after the learning is completed, receives an MRA image of the brain and applies it to the learned artificial intelligence model to output an analysis result of analyzing brain aneurysms in the MRA image.

[0042] A brain image analysis device (100) according to a preferred embodiment of the present invention includes a processor (110), a memory (120), an input interface (130), an output interface (140), and a communication interface (150).

[0043] A memory (120) according to a preferred embodiment of the present invention can store instructions executable by a processor (110) and programs executed by a processor (110), and can also store input / output data.

[0044] Memory (120) may include a kernel, middleware, an application programming interface (API), and / or an application program. At least some of the kernel, middleware, or API may be referred to as an operating system. The kernel may control or manage system resources (e.g., a bus, a processor, or memory, etc.) used to execute operations or functions implemented in other programs (e.g., middleware, API, or application program). Additionally, the kernel may provide an interface that allows the control or management of system resources by accessing individual components of the server through the middleware, API, or application program.

[0045] Examples of memory (120) may include SSD (Solid State Drive), flash memory, ROM (Read-Only Memory), RAM (Random Access Memory), etc. Memory (120) may be replaced by web storage or a cloud server that performs the function of a storage medium on the internet, or may be implemented by replacing it with a hard disk drive (HDD).

[0046] A processor (110) according to a preferred embodiment of the present invention may be implemented as a CPU (Central Processing Unit) or a similar device (e.g., MPU (Micro Processing Unit), MCU (Micro Control Unit), etc.) and performs each step of the brain image analysis method described below with reference to FIGS. 2 to 5c by executing instructions stored in memory (120). In addition, the processor (110) can control the overall function of the brain image analysis device (100).

[0047] The input interface (130) is connected to a typical input means such as a mouse and keyboard, and can receive setting information and selection information from the user through the input means and output it to the processor (110).

[0048] The output interface (140) is connected to typical output means such as a monitor and a printer to output data generated by the processor (110) to the user.

[0049] The communication interface (150) communicates with external servers, video providing devices (300), administrator terminals (200), and user terminals via wired or wireless communication methods. Representative examples of wired communication methods used by the communication interface (150) include LAN (Local Area Network) or USB (Universal Serial Bus) communication, and other methods are also possible.

[0050] In addition, for wireless types, communication methods of the WPAN (Wireless Personal Area Network) family, such as Bluetooth or Zigbee, can be primarily used, and it is also possible to use communication methods of the WLAN (Wireless Local Area Network) family, such as Wi-Fi, or other known communication methods.

[0051] Hereinafter, with further reference to FIGS. 2 to 5c, the function of a brain image analysis device (100) and a brain image analysis method according to a preferred embodiment of the present invention will be described in more detail.

[0052]

[0053] FIG. 2 is a flowchart illustrating a brain image analysis method according to a preferred embodiment of the present invention.

[0054] Referring to FIG. 2, when the brain image analysis method of the present invention is executed, the processor (110) of the present invention executes instructions stored in memory (120), and the processor (110) that executed the instructions receives a pair of MRA images and DSA images of the same person's brain from an image providing device (300), and trains an artificial intelligence model so that when the artificial intelligence model receives an MRA image of the brain, it generates and outputs brain aneurysm information (S100).

[0055] When the training of the artificial intelligence model is completed, the processor (110) receives the MRA image to be analyzed from the image providing device (300) and applies it to the artificial intelligence model (S200).

[0056] After that, the processor (110) outputs brain aneurysm information generated from the artificial intelligence model to the user (S300). As described below, the brain aneurysm information includes location information, maximum length, neck length, etc. of the brain aneurysm.

[0057]

[0058] FIG. 3 is a diagram illustrating the process of performing the brain image analysis method of the present invention, and FIG. 4 is a flowchart illustrating the learning process of an artificial intelligence model according to a preferred embodiment of the present invention.

[0059] With further reference to FIGS. 3 and FIGS. 4, the step S100 of training an artificial intelligence model according to a preferred embodiment of the present invention will be described in more detail.

[0060] First, it should be noted that the registration module, artificial intelligence model, loss function, information extraction module, and optimization module illustrated in FIG. 3 are implemented in software by the processor (110) executing instructions.

[0061] First, in the process of learning an artificial intelligence model, the processor (110) receives an MRA image and a DSA image of the same person's brain, and the input MRA image is input to the registration module and the artificial intelligence model, respectively, and the DSA image is input to the registration module (S110).

[0062] After that, the registration module aligns the input DSA image with the MRA image and registers the DSA image (S120).

[0063] Medical image registration is a process that aligns images within the same coordinate system, so that brain regions displayed in the aligned MRA and DSA images are located at the same coordinates. This registration process is generally performed by calculating the similarity between the two images through mutual information and maximizing it. Representative registration methods include Rigid Registration using rotation and translation, Affine Registration allowing size and tilt, and Non-Rigid Registration applying complex deformations, and these can be applied to the registration module of the present invention.

[0064] After that, the processor (110) displays the aligned DSA image to the doctor through the administrator terminal (200) and receives information on the cerebral aneurysm on the DSA image from the doctor through the administrator terminal (200) (S130).

[0065] In step S130, the processor (110) receives input from the administrator terminal (200) regarding the brain aneurysm area designated by the doctor in the DSA image and displays the area.

[0066] In a preferred embodiment of the present invention, as described in FIG. 5a, a physician draws the region corresponding to the cerebral aneurysm on a DSA image using an open source labeling tool such as ITK-snap. That is, the physician opens the DSA image using ITK-snap and draws a segmentation map by designating the region corresponding to the cerebral aneurysm through a segmentation process in which the region is defined by coloring the location where the cerebral aneurysm exists. In addition, the vessel neck of the cerebral aneurysm, which is the contact part between the cerebral aneurysm and the blood vessel, is drawn in the same way.

[0067] The segmentation map generated in this way is the same size as the original DSA image and indicates only the location where the cerebral aneurysm is located. When this is overlaid on the original DSA image, an image is obtained in which the original DSA image is colored with the area marked by the doctor (500), as shown in FIG. 5b. Then, the processor (110) receives the name of the blood vessel to which the cerebral aneurysm is attached as a blood vessel number from the doctor through the administrator terminal (200).

[0068] In conclusion, the processor (110) receives the region (location) and blood vessel name of the brain aneurysm in the DSA image from the doctor through the administrator terminal (200) as brain aneurysm information.

[0069] After that, the information extraction module of the processor (110) calculates the maximum length and neck length of the cerebral aneurysm from the DSA image showing the cerebral aneurysm region (S140).

[0070] Referring to Fig. 5c, the information extraction module finds the two points furthest apart among the pixels corresponding to the cerebral aneurysm in the segmented map, and calculates and sets the distance (A) between these two points as the largest length.

[0071] In addition, the information extraction module obtains a plane that is the neck of the cerebral aneurysm, which is the part where the area segmented as a cerebral aneurysm is in contact with the blood vessel, finds the two points with the shortest distance from this plane, and calculates and sets the distance (B) between these two points as the neck length.

[0072] The maximum length and neck length of the cerebral aneurysm calculated in the information extraction module are input into the loss function as Ground Truth (GT) information along with the location information of the cerebral aneurysm (S150).

[0073] Meanwhile, the processor (110) applies the training MRA image to an artificial intelligence model to analyze the MRA image, generates the location of the cerebral aneurysm, the maximum length of the cerebral aneurysm, and the neck length as prediction information, outputs it as a loss function, and the loss function receives this as input (S160).

[0074] In a preferred embodiment of the present invention, various previously known artificial intelligence models may be used as the artificial intelligence model. A preferred embodiment of the present invention utilizes a deep learning-based Convolutional Neural Network (CNN) model, which is a neural network structure specialized for image processing that automatically extracts important features by hierarchically processing input images.

[0075] CNNs use convolution layers, activation functions, and pooling layers to learn increasingly abstract information while preserving the spatial structure of images; when trained with a sufficient amount of data, they can accurately recognize various patterns, which enables them to produce stable and consistent results even with new data.

[0076] Representative artificial intelligence models applicable to the present invention include U-Net, Fully Convolutional Network (FCN), and DeepLab, and the characteristics of each model are briefly summarized as follows.

[0077] U-Net: A model frequently used for medical image segmentation, featuring a symmetrical U-shaped structure. It consists of an encoder (reduction) and a decoder (expansion), and reuses features obtained from each encoder stage in the decoder stage to reconstruct accurate boundaries. It is widely used in the medical field because it enables effective training even with small amounts of data.

[0078] Fully Convolutional Network (FCN): A model that eliminates the fully connected layers from the existing CNN structure and replaces all layers with convolutional layers. It outputs feature maps regardless of the size of the input image and performs segmentation through pixel-level prediction. FCN has the advantage of being able to operate flexibly even at various image resolutions.

[0079] DeepLab: DeepLab uses atrous convolution and Conditional Random Fields (CRF) to effectively reflect image contextual information while maintaining spatial resolution. Capable of recognizing objects at various scales, it delivers excellent performance in complex scenes or object segmentation.

[0080] Referring again to FIG. 4, in the above-mentioned step S160, when the artificial intelligence model generates prediction information and inputs it into the loss function, the processor (110) applies the input GT information and prediction information to the loss function to calculate the difference between the GT information and the prediction information, and inputs the difference value into the optimization module (S170).

[0081] The loss function evaluates how well an AI model is learning by calculating the difference between the predicted information and the actual ground truth (GT), and it adjusts the model's weights in a direction that minimizes the loss to perform predictions more accurately as the model learns.

[0082] When the processor (110) calculates the difference between the location prediction information of the location of the cerebral aneurysm, which is a categorical prediction, and the location information of the GT, it uses Cross Entropy loss as the loss function (predicting the probability of belonging to a specific class and calculating the loss by comparing it with the actual correct class).

[0083] Additionally, when the processor (110) calculates the difference between the maximum length and neck length of the brain aneurysm and the maximum length and neck length of the GT, which is a problem of predicting numbers, it uses MSE loss (calculating the average after squaring the difference between the predicted value and the actual value) as the loss function.

[0084] The processor (110) compares the difference value calculated in the loss function with a predefined threshold value, terminates the learning process if the difference value is less than the threshold value, and outputs the difference value to the optimization module to continue the learning process (S180).

[0085] The optimization module adjusts the weights of the artificial intelligence model so that the difference value output from the loss function is minimized (S190).

[0086] In other words, the optimization module determines the direction and magnitude of the AI ​​model's weight updates based on the measured loss (difference value). Through this, the AI ​​model learns to make better predictions and ultimately minimizes the loss.

[0087] In a preferred embodiment of the present invention, various optimization modules may be applied, and representative examples are as follows.

[0088] SGD (Stochastic Gradient Descent): A basic form of gradient descent that uses a randomly selected batch of data instead of the entire dataset to calculate weight updates each time. This speeds up computation but can result in slow convergence due to frequent oscillations.

[0089] Adam (Adaptive Moment Estimation): The most widely used optimizer, it tracks first and second moments (gradient and the square of the gradient) to maintain a consistent learning rate for each weight. It is fast and stable in training speed and works well even in complex problems.

[0090] RMSprop: An algorithm that dynamically adjusts the learning rate, using smaller learning rates for gradients with large changes and larger learning rates for gradients with small changes. This improves convergence speed and reduces oscillations, enabling stable training.

[0091] Until the difference value is determined to be smaller than the threshold value in the aforementioned step S180, the processor (110) continuously repeats steps S160 through S190 to improve the accuracy of the artificial intelligence model, and when the difference value becomes smaller than the threshold value, receives other DSA images and MRA images as input and repeats the aforementioned learning process.

[0092] Meanwhile, when the artificial intelligence model is trained through the process described above, the processor (110) receives the MRA image to be analyzed and applies it to the artificial intelligence model (S200), and the artificial intelligence model generates and outputs brain aneurysm information (location information of the brain aneurysm, maximum length, neck length) (S300).

[0093]

[0094] The brain image analysis method according to the preferred embodiment of the present invention described so far can be implemented as a computer program that is implemented as computer-executable instructions and stored on a non-transient storage medium.

[0095] Storage media include all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable storage media include ROM, RAM, CD-ROM, and optical data storage devices. Additionally, computer-readable storage media are distributed across networked computer systems, allowing computer-readable code to be stored and executed in a distributed manner.

[0096] The present invention has been described above with reference to its preferred embodiments. Those skilled in the art will understand that the present invention may be embodied in modified forms without departing from the essential characteristics of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

Claims

1. A brain imaging analysis device comprising a processor and a memory for storing predetermined instructions, The processor that executed the instructions stored in the memory above (a) receiving MRA images and DSA images of the same person's brain from an image providing device, and training the artificial intelligence model so that the artificial intelligence model receives the MRA images and outputs brain aneurysm information; (b) a step of applying the MRA image of the brain to be analyzed to a trained artificial intelligence model; and (c) A brain image analysis device characterized by performing the step of outputting a brain aneurysm image output from the above artificial intelligence model.

2. In claim 1, in step (a), the processor A brain image analysis device characterized by aligning the above DSA image with the above MRA image and registering it, receiving brain aneurysm information read from the registered DSA image from an administrator terminal and inputting it into the artificial intelligence model as Ground Truth (GT) information, and inputting the above MRA image as a training image to train the artificial intelligence model.

3. In claim 2, in step (a), the processor A brain image analysis device characterized by displaying a registered DSA image through the above-mentioned administrator terminal, and receiving location information in which a brain aneurysm area is set from the above-mentioned administrator terminal, generating the maximum length and neck length of the brain aneurysm in the above-mentioned DSA image, and inputting them into the above-mentioned artificial intelligence model as Ground Truth (GT) information along with the location information.

4. In Clause 2, at step (a) above A brain image analysis device characterized by the artificial intelligence model generating brain aneurysm prediction information including location information, maximum length, and neck length of the brain aneurysm from the MRA image, and being trained such that the difference between the generated brain aneurysm prediction information and the location information, maximum length, and neck length of the brain aneurysm included in the GT information is minimized.

5. In claim 4, in step (a), the processor A brain image analysis device characterized by applying the above-mentioned cerebral aneurysm prediction information and the above-mentioned GT information to a loss function, and if the difference value output from the loss function is not smaller than a predefined threshold, adjusting the weights of the artificial intelligence model and regenerating the prediction information and inputting it into the loss function, and repeating the adjustment of the weights of the artificial intelligence model and the generation of prediction information until the difference value output from the loss function becomes smaller than the threshold.

6. In claim 1, the above step (a) (a1) A step of receiving the MRA image and the DSA image, and registering the DSA image by aligning it with the MRA image; (a2) receiving brain aneurysm information including location information for a brain aneurysm read from a registered DSA image from an administrator terminal, generating the maximum length and neck length of the brain aneurysm, and inputting it into the artificial intelligence model as Ground Truth (GT) information along with the location information; (a3) A step of applying the MRA image to the artificial intelligence model to generate predictive information including location information, maximum length, and neck length of the cerebral aneurysm; (a4) a step of generating a difference value by applying the above GT information and the above prediction information to a loss function, and comparing the difference value with a predefined threshold; and (a5) A brain image analysis device characterized by including the step of adjusting the weights of the artificial intelligence model and repeating from step (a3) ​​if the difference value is not smaller than the threshold value.

7. A brain image analysis method performed in a brain image analysis device comprising a memory storing instructions and a processor executing instructions stored in said memory, wherein (a) A step in which a processor that has executed the instruction stored in the memory receives an MRA image and a DSA image of the same person's brain from an image providing device, and trains the artificial intelligence model so that the artificial intelligence model receives the MRA image and outputs brain aneurysm information; (b) a step of applying the MRA image of the brain captured by the processor to an artificial intelligence model that has completed training; and (c) A brain image analysis method characterized by including the step of the processor outputting a brain aneurysm image output from the artificial intelligence model.

8. In claim 7, in step (a), the processor A brain image analysis method characterized by aligning the above DSA image with the above MRA image and registering it, receiving brain aneurysm information read from the registered DSA image from an administrator terminal and inputting it into the above artificial intelligence model as Ground Truth (GT) information, and inputting the above MRA image as a training image to train the above artificial intelligence model.

9. In claim 8, in step (a), the processor A brain image analysis method characterized by displaying a registered DSA image through the above-mentioned administrator terminal, receiving location information in which a brain aneurysm area is set from the above-mentioned administrator terminal, generating the maximum length and neck length of the brain aneurysm in the above-mentioned DSA image, and inputting them into the above-mentioned artificial intelligence model as Ground Truth (GT) information along with the location information.

10. In claim 8, at step (a) above A brain image analysis method characterized in that the artificial intelligence model generates brain aneurysm prediction information including location information, maximum length, and neck length of the brain aneurysm from the MRA image, and is trained such that the difference between the generated brain aneurysm prediction information and the location information, maximum length, and neck length of the brain aneurysm included in the GT information is minimized.

11. In claim 10, in step (a), the processor A brain image analysis method characterized by applying the above-mentioned cerebral aneurysm prediction information and the above-mentioned GT information to a loss function, and if the difference value output from the loss function is not smaller than a predefined threshold, adjusting the weights of the artificial intelligence model and regenerating the prediction information and inputting it into the loss function, and repeating the adjustment of the weights of the artificial intelligence model and the generation of prediction information until the difference value output from the loss function becomes smaller than the threshold.

12. In claim 7, the above step (a) (a1) A step in which the processor receives the MRA image and the DSA image, and registers the DSA image by aligning it with the MRA image; (a2) The processor receives brain aneurysm information including location information for a brain aneurysm read from a registered DSA image from an administrator terminal, generates the maximum length and neck length of the brain aneurysm, and inputs it into the artificial intelligence model as Ground Truth (GT) information along with the location information; (a3) A step in which the processor applies the MRA image to the artificial intelligence model to generate predictive information including location information, maximum length, and neck length of the cerebral aneurysm; (a4) A step in which the processor applies the GT information and the prediction information to a loss function to generate a difference value, and compares the difference value with a predefined threshold; and (a5) A brain image analysis method characterized by including the step of, if the difference value is not smaller than the threshold, adjusting the weights of the artificial intelligence model and performing the step (a3) ​​again.

13. A computer program stored in a non-transient storage medium and executed on a computer including a processor to perform the brain image analysis method of any one of claims 7 to 12.

Images