Diagnostic assistance device and method for predicting worsening of intracranial hemorrhage in head trauma patient on basis of artificial intelligence model

Abstract

The present disclosure relates to a diagnostic assistance device and method for predicting worsening of intracranial hemorrhage in a head trauma patient on the basis of an artificial intelligence model. The diagnostic assistance device for predicting worsening of intracranial hemorrhage in a head trauma patient according to the present disclosure may comprise: a communication unit configured to receive a head CT image of a patient; and a processor including an artificial intelligence model trained to convert the input CT image into a 3D image and predict whether intracranial hemorrhage of the patient will worsen in the future by using the 3D image, wherein the CT image is an initial CT image captured after the patient has suffered head trauma.

Description

Diagnostic aid device and method for predicting exacerbation of intracranial hemorrhage in patients with head trauma based on an artificial intelligence model

[0001] The present disclosure relates to a diagnostic aid device and method for predicting the worsening of intracranial hemorrhage in patients with head trauma based on an artificial intelligence model.

[0002] Initial CT scans are crucial for patients with head trauma. This allows for an objective assessment of the intracranial condition and, by reviewing neurological abnormalities identified during the physical examination, helps determine the necessity of emergency surgery and establish a surgical plan.

[0003] However, in clinical practice, there are often cases where a patient whose condition was not severe initially suddenly deteriorates within just a few hours, requiring belated emergency surgery. Although the likelihood of deterioration can be predicted based on the patient's underlying diseases and medication history, the predictive accuracy is not high, so clinical practice typically performs additional CT scans on most patients.

[0004] If initial CT scans can not only assess the current intracranial condition but also predict the worsening of bleeding, they can be of great assistance in the management of head trauma patients in conjunction with their existing medical records. Accordingly, there is a need for technology capable of predicting future bleeding deterioration in head trauma patients based solely on initial CT scan results.

[0005] The present disclosure aims to provide a diagnostic aid and method capable of predicting whether a patient's future bleeding will worsen based solely on the results of an initial CT scan of a patient with head trauma.

[0006] The problems that this disclosure aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by a person skilled in the art from the description below.

[0007] A diagnostic aid for predicting the worsening of intracranial hemorrhage in a patient with head trauma to achieve the aforementioned technical task comprises a communication unit configured to receive a CT image of the patient's head, and a processor comprising an artificial intelligence model trained to convert the input CT image into a first 3D image, convert the resolution of the first 3D image to generate a plurality of second 3D images, and predict whether the patient's future intracranial hemorrhage will worsen using the first 3D image and the plurality of second 3D images, wherein the CT image may be a CT image taken for the first time after the patient suffered head trauma.

[0008] In one embodiment, the processor can generate prescription data corresponding to the patient based on the result of predicting whether the bleeding will worsen.

[0009] In one embodiment, the artificial intelligence model outputs a probability value indicating whether the patient's bleeding has worsened, and the processor can generate prescription data recommending hospitalization when the probability value is higher than a reference value, and generate prescription data recommending discharge when the probability value is lower than a reference value.

[0010] In one embodiment, the training data used for training the artificial intelligence model may be data in which the actual worsening of the patient's bleeding is labeled on a 3D image generated based on the first CT image taken after the patient suffered a head injury.

[0011] In one embodiment, whether the actual bleeding worsens can be determined based on an additional CT scan image taken after a predetermined time has elapsed from the time when the patient first took a CT scan after suffering a head injury.

[0012] In one embodiment, the input data input to the artificial intelligence model may not include any data other than a 3D image generated based on the first CT image taken after the patient suffered a head injury.

[0013] In one embodiment, the CT image used for the 3D image conversion may be an image in the DICOM format.

[0014] In one embodiment, the processor generates a plurality of second 3D images having a lower resolution than the first 3D image, normalizes the brain tissue density measured in the first 3D image and the plurality of second 3D images, segments the hemorrhage area in the first 3D image and the plurality of second 3D images and inputs it to the artificial intelligence model, and can predict whether the hemorrhage will worsen as the output of the artificial intelligence model for the input.

[0015] In one embodiment, the processor can predict a hemorrhage spread rate indicating the hourly increase rate of the hemorrhage area from the input CT image and indicate a hemorrhage risk based on the volume and location of the hemorrhage area shown in the input CT image.

[0016] In one embodiment, the processor sets a patient-specific threshold according to the patient's age and comorbidities and applies it to the artificial intelligence model, and can adjust and apply the threshold over time.

[0017] Additionally, a diagnostic aid method for predicting the worsening of intracranial hemorrhage in a patient with head trauma according to the present disclosure comprises the steps of receiving a CT image of the patient's head, converting the input CT image into a first 3D image, converting the resolution of the first 3D image to generate a plurality of second 3D images, and predicting whether the patient's future intracranial hemorrhage will worsen by utilizing the first 3D image and the plurality of second 3D images, wherein the CT image may be a CT image taken for the first time after the patient suffered head trauma.

[0018] The diagnostic aid according to the present disclosure can predict whether future bleeding will worsen in a patient with head trauma based solely on the results of an initial CT scan. Through this, the present disclosure can minimize the need for additional CT scans for patients with head trauma and concentrate medical resources on patients at risk.

[0019] FIG. 1 is an overall system diagram of the present disclosure.

[0020] FIG. 2 is a block diagram of a server included in the diagnostic assistance device of the present disclosure.

[0021] FIG. 3 is a block diagram of a terminal included in the diagnostic assistance device of the present disclosure.

[0022] FIG. 4 is a flowchart illustrating a diagnostic assistance method according to the present disclosure.

[0023] FIG. 5 is a flowchart illustrating a learning method for an artificial intelligence model according to the present disclosure.

[0024] Figure 6 is a detailed flowchart of step S120 disclosed in Figure 4.

[0025] FIG. 7 is a flowchart of a diagnostic assistance method according to another embodiment of the present disclosure.

[0026] Throughout this disclosure, the same reference numerals denote the same components. This disclosure does not describe all elements of the embodiments, and general content in the art to which this disclosure pertains or content that overlaps between embodiments is omitted. The terms 'part, module, component, block' as used in the specification may be implemented in software or hardware, and depending on the embodiments, a plurality of 'parts, modules, components, blocks' may be implemented as a single component, or a single 'part, module, component, block' may include a plurality of components.

[0027] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are directly connected but also cases where they are indirectly connected, and indirect connections include connections made via a wireless communication network.

[0028] Furthermore, when it is stated that a part "includes" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0029] Throughout the specification, when it is stated that a component is located "on" another component, this includes not only cases where a component is in contact with another component, but also cases where another component exists between the two components.

[0030] The terms first, second, etc. are used to distinguish one component from another, and the components are not limited by the aforementioned terms.

[0031] Singular expressions include plural expressions unless there is an obvious exception in the context.

[0032] In each step, identification codes are used for convenience of explanation and do not describe the order of the steps; the steps may be performed differently from the specified order unless a specific order is clearly indicated in the context.

[0033] The operating principles and embodiments of the present disclosure will be described below with reference to the attached drawings.

[0034] In this specification, the term "device according to the present disclosure" includes all various devices capable of performing computational processing and providing results to a user. For example, the device according to the present disclosure may include all of a computer, a server device, and a portable terminal, or may be in the form of any one of these.

[0035] Here, the computer may include, for example, a notebook, desktop, laptop, tablet PC, slate PC, etc. equipped with a web browser.

[0036] The above server device is a server that processes information by communicating with an external device, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, and a web server.

[0037] The above portable terminal may include, for example, all types of handheld-based wireless communication devices such as PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), WiBro (Wireless Broadband Internet) terminals, smartphones, etc., as well as wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head-mounted devices (HMDs).

[0038] A diagnostic assistance device according to the present disclosure may be implemented by at least one of a server and a terminal. Specifically, the device according to the present disclosure may be implemented by either a server or a terminal, or may be implemented as a system through data transmission and reception between a server and a terminal.

[0039] Hereinafter, a diagnostic assistance device according to the present disclosure will be described.

[0040] Referring to FIG. 1, a diagnostic assistance device according to the present disclosure may include a server (10) and a terminal (20).

[0041] The server (10) is connected to the terminal (20) via a network, and after receiving data necessary for diagnosis from the terminal (20), can transmit the diagnosis result to the terminal (20).

[0042] Meanwhile, it is obvious to a person skilled in the art that the above-described terminal (20) is not limited to the portable terminal described above and may include a laptop, desktop, laptop, tablet PC, slate PC, etc. equipped with a processor.

[0043] As described above, the diagnostic assistance device according to the present disclosure can be implemented through data transmission and reception between a server (10) and a terminal (20).

[0044] Hereinafter, each of the server (10) and terminal (20) for implementing the diagnostic assistance device according to the present disclosure will be described.

[0045] FIG. 2 is a block diagram of a server included in the diagnostic assistance device of the present disclosure.

[0046] A server (100) according to the present disclosure may include at least one of a communication unit (110), a storage unit (120), and a processor (130).

[0047] The communication unit (110) can communicate with at least one of a terminal, an external storage (e.g., a database (140)), an external server, and a cloud server.

[0048] Meanwhile, an external server or cloud server may be configured to perform at least a part of the role of the processor (130). That is, the performance of data processing or data operations, etc., can be performed on an external server or cloud server, and the present invention does not impose any special restrictions on such a method.

[0049] Meanwhile, the communication unit (110) can support various communication methods according to the communication standards of the target being communicated (e.g., electronic device, external server, device, etc.).

[0050] For example, the communication unit (110) may be configured to communicate with a communication target using at least one of the following technologies: WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, and Wireless USB (Wireless Universal Serial Bus).

[0051] Next, the storage unit (120) may be configured to store various information related to the present invention. In the present invention, the storage unit (120) may be provided in the device itself according to the present invention. Alternatively, at least part of the storage unit (120) may refer to at least one of a database (DB, 140) or a cloud storage (or cloud server). That is, the storage unit (120) is sufficient as a space where information necessary for the device and method according to the present invention is stored, and it can be understood that there are no restrictions on physical space. Accordingly, below, the storage unit (120), database (140), external storage, and cloud storage (or cloud server) will not be distinguished separately, and will all be referred to as the storage unit (120).

[0052] Next, the processor (130) may be configured to control the overall operation of the device related to the present invention. The processor (130) may process signals, data, information, etc. that are input or output through the components described above, or provide or process appropriate information or functions to the user.

[0053] The processor (130) includes at least one CPU (Central Processing Unit) and can perform the functions according to the present invention.

[0054] At least one component may be added or removed in response to the performance of the components illustrated in FIG. 2. Additionally, it will be readily understood by those skilled in the art that the relative positions of the components may be changed in response to the performance or structure of the device.

[0055] Hereinafter, a terminal included in the diagnostic assistance device of the present disclosure will be described in detail.

[0056] FIG. 3 is a block diagram of a terminal included in the diagnostic assistance device of the present disclosure.

[0057] Referring to FIG. 3, the terminal (200) according to the present disclosure may include a communication unit (210), an input unit (220), a display unit (230), and a processor (240), etc. Since the components illustrated in FIG. 3 are not essential for implementing the diagnostic assistance device according to the present disclosure, the terminal described in this specification may have more or fewer components than those listed above.

[0058] Among the above components, the communication unit (210) may include one or more components that enable communication with an external device, for example, at least one of a broadcast receiving module, a wired communication module, a wireless communication module, a short-range communication module, and a location information module.

[0059] The wired communication module may include various wired communication modules such as a Local Area Network (LAN) module, a Wide Area Network (WAN) module, or a Value Added Network (VAN) module, as well as various cable communication modules such as USB (Universal Serial Bus), HDMI (High Definition Multimedia Interface), DVI (Digital Visual Interface), RS-1302 (recommended standard 1302), power line communication, or POTS (plain old telephone service).

[0060] In addition to Wi-Fi modules and WiBro (Wireless broadband) modules, the wireless communication module may include wireless communication modules that support various wireless communication methods such as GSM (global System for Mobile Communication), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access), UMTS (universal mobile telecommunications system), TDMA (Time Division Multiple Access), LTE (Long Term Evolution), 4G, 5G, and 6G.

[0061] The input unit (220) is for inputting video information (or signal), audio information (or signal), data, or information input from a user, and may include at least one of at least one camera, at least one microphone, and a user input unit. Voice data or image data collected from the input unit may be analyzed and processed into a user control command.

[0062] The display unit (230) is intended to generate outputs related to sight, hearing, or touch, and may include at least one of a display unit, an audio output unit, a haptic module, and an optical output unit. The display unit may form a layered structure with the touch sensor or be formed integrally to implement a touch screen. Such a touch screen functions as a user input unit that provides an input interface between the device and the user, and at the same time can provide an output interface between the device and the user.

[0063] The display unit displays (outputs) information processed by the device. For example, the display unit may display execution screen information of an application program (e.g., an application) running on the device, or UI (User Interface) and GUI (Graphic User Interface) information based on such execution screen information.

[0064] In addition to the components described above, the terminal described above may further include an interface unit and memory.

[0065] The interface section serves as a passage for various types of external devices connected to the present device. This interface section may include at least one of a wired / wireless headset port, an external charger port, a wired / wireless data port, a memory card port, a port for connecting a device equipped with a SIM card, an audio I / O (Input / Output) port, a video I / O (Input / Output) port, and an earphone port. The present device can perform appropriate control related to the external device connected to the interface section.

[0066] The memory can store data supporting various functions of the device and programs for the operation of the processor, and can store input / output data (e.g., music files, still images, videos, etc.), and can store multiple application programs (or applications) running on the device, data for the operation of the device, and instructions. At least some of these application programs may be downloaded from an external server via wireless communication.

[0067] Such memory may include at least one type of storage medium among flash memory type, hard disk type, SSD type (Solid State Disk type), SSD type (Silicon Disk Drive type), multimedia card micro type, card type memory (e.g., SD or XD memory, etc.), RAM (random access memory; RAM), SRAM (static random access memory), ROM (read-only memory; ROM), EEPROM (electrically erasable programmable read-only memory), PROM (programmable read-only memory), magnetic memory, magnetic disk, and optical disk. Additionally, the memory may be a database that is separate from the device but connected via wired or wireless connection.

[0068] Meanwhile, the above-described terminal includes a processor (240). The processor may be implemented as a memory that stores data for an algorithm or a program that reproduces the algorithm for controlling the operation of components within the device, and at least one processor (not shown) that performs the above-described operation using the data stored in the memory. In this case, the memory and the processor may each be implemented as separate chips. Alternatively, the memory and the processor may be implemented as a single chip.

[0069] Meanwhile, a processor included in at least one of the server and the terminal may include an artificial intelligence module for implementing the diagnostic assistance device described below. The artificial intelligence model learning method described below is described as being implemented by the operation of the said module, but the execution of the operation of each step described below does not necessarily have to be performed by the said module.

[0070] In addition, the processor may control one or a combination of the components described above in order to implement various embodiments according to the present disclosure described in the drawings below on the device.

[0071] Meanwhile, at least one component may be added or removed in response to the performance of the components illustrated in FIGS. 1 to 3. In addition, it will be readily understood by those skilled in the art that the relative positions of the components may be changed in response to the performance or structure of the device.

[0072] Below, the artificial intelligence described in the present invention will be explained in detail.

[0073] The artificial intelligence-related functions according to the present disclosure are operated through the processor and memory installed in the aforementioned server and terminal. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0074] The predefined operating rules or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a predefined operating rules or artificial intelligence models configured to perform desired characteristics (or objectives) are created by a basic artificial intelligence model being trained using multiple learning data by a learning algorithm. Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0075] An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values ​​and performs neural network operations through operations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights may be updated so that the loss value or cost value obtained from the artificial intelligence model during the learning process is reduced or minimized. Artificial neural networks may include deep neural networks (DNNs), such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Restricted Boltzmann Machines (RBMs), Deep Belief Networks (DBNs), Bidirectional Recurrent Deep Neural Networks (BRDNNs), or Deep Q-Networks, but are not limited to the examples mentioned above.

[0076] According to an exemplary embodiment of the present disclosure, a processor can implement artificial intelligence. Artificial intelligence refers to a machine learning method based on an artificial neural network that enables a machine to learn by mimicking human biological neurons. Methodologies of artificial intelligence can be classified according to the learning method into supervised learning, where input and output data are provided together as learning data and the solution (output data) to the problem (input data) is predetermined; unsupervised learning, where only input data is provided without output data and the solution (output data) to the problem (input data) is not predetermined; and reinforcement learning, where a reward is given from an external environment whenever an action is taken from the current state, and learning proceeds in a direction that maximizes such reward. In addition, artificial intelligence methodologies can be classified according to the architecture, which is the structure of the learning model. The architectures of widely used deep learning technologies can be classified into Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Transformers, and Generative Adversarial Networks (GAN).

[0077] The device and system may include an artificial intelligence model. The artificial intelligence model may be a single model or may be implemented as multiple models. The artificial intelligence model may be composed of a neural network (or artificial neural network) and may include statistical learning algorithms that mimic biological neurons in machine learning and cognitive science. A neural network may refer to a model that possesses problem-solving capabilities by having artificial neurons (nodes) that form a network through synaptic connections and change the strength of synaptic connections through learning. The neurons of a neural network may include combinations of weights or biases. A neural network may include one or more layers composed of one or more neurons or nodes. For example, the device may include an input layer, a hidden layer, and an output layer. The neural network constituting the device can infer a result (output) to be predicted from an arbitrary input by changing the weights of the neurons through learning.

[0078] The processor can create a neural network, train a neural network, perform computations based on received input data, generate an information signal based on the results of the computation, or retrain the neural network. The neural network models may include, but are not limited to, various types of models such as Convolutional Neural Networks (CNN), Region with Convolutional Neural Networks (R-CNN), Region Proposal Networks (RPN), Recurrent Neural Networks (RNN), Stacking-based Deep Neural Networks (S-DNN), State-Space Dynamic Neural Networks (S-SDNN), Deconvolution Networks, Deep Belief Networks (DBN), Restructured Boltzmann Machines (RBM), Fully Convolutional Networks, Long Short-Term Memory Networks (LSTM), and Classification Networks, such as GoogleNet, AlexNet, and VGG Network. The processor may include one or more processors to perform computations according to the neural network models. For example, the neural network is a deep neural network It may include a (Deep Neural Network).

[0079] Neural networks include CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), perceptron, multilayer perceptron, FF (Feed Forward), RBF (Radial Basis Network), DFF (Deep Feed Forward), LSTM (Long Short Term Memory), GRU (Gated Recurrent Unit), AE (Auto Encoder), VAE (Variational Auto) Encoder), DAE (Denoising Auto Encoder), SAE (Sparse Auto Encoder), MC (Markov Chain), HN (Hopfield Network), BM (Boltzmann Machine), RBM (Restricted Boltzmann Machine), DBN (Depp Belief Network), DCN (Deep Convolutional Network), DN (Deconvolutional Network), DCIGN (Deep Convolutional Inverse Graphics Network), GAN (Generative Adversarial Network), LSM (Liquid State Machine), ELM (Extreme Learning Machine), ESN (Echo It will be understood by a person skilled in the art that any neural network may be included, but is not limited to, State Network, Deep Residual Network, Differential Neural Computer, Neural Turning Machine, Capsule Network, Kohonen Network, and Attention Network.

[0080] According to an exemplary embodiment of the present disclosure, the processor comprises a Convolutional Neural Network (CNN) such as GoogleNet, AlexNet, VGG Network, Region with Convolutional Neural Network (R-CNN), Region Proposal Network (RPN), Recurrent Neural Network (RNN), Stacking-based Deep Neural Network (S-DNN), State-Space Dynamic Neural Network (S-SDNN), Deconvolution Network, Deep Belief Network (DBN), Restructured Boltzmann Machine (RBM), Fully Convolutional Network, Long Short-Term Memory (LSTM) Network, Classification Network, Generative Modeling, eXplainable AI, Continual AI, Representation Learning, AI for Material Design, BERT, SP-BERT, MRC / QA, Text Analysis, Dialog System, GPT-3, GPT-4 for Natural Language Processing, Visual Analytics, Visual Understanding, Video Synthesis for Vision Processing, Anomaly Detection, Prediction, Time-Series Forecasting, Optimization for ResNet Data Intelligence, Various artificial intelligence structures and algorithms, such as recommendation and data creation, may be used, but are not limited thereto. Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings.

[0081] Hereinafter, a diagnostic assistance method utilizing the aforementioned components is described in detail. The diagnostic assistance method described below is implemented through the transmission and reception of data between the server and the terminal described above, and some steps of the method may be performed on either the server or the terminal. However, it is obvious to a person skilled in the art that the method described below may be performed independently on either the server or the terminal, without being limited thereto.

[0082] Hereinafter, a diagnostic assistance method utilizing the components described in FIGS. 1 to 3 will be described.

[0083] FIG. 4 is a flowchart of a diagnostic assistance method according to the present disclosure, and FIG. 5 is a flowchart showing a method for training an artificial intelligence model according to the present disclosure.

[0084] Referring to Fig. 4, the step of receiving the patient's head CT image is performed (S110).

[0085] Here, the patient's head CT image may be the first CT image taken after the patient suffered a head injury. The artificial intelligence model described below predicts whether the patient's future bleeding will worsen based solely on the first CT image taken after the patient suffered a head injury.

[0086] Meanwhile, the CT image may be an image in DICOM format. In one embodiment, the CT image may include a plurality of photos extracted from a video format file.

[0087] DICOM (Digiral Imaging and Communications in Medicine) is the standard format for medical imaging information. Generally, when performing analysis, it is not used as is but is converted into a JPG or PNG file for analysis; at this time, the window can be adjusted so that specific body parts are clearly visible.

[0088] For example, an image file in DICOM format can be converted so that the bones or the brain are clearly visible through window adjustment.

[0089] In one embodiment, the CT image may be a 124×124 image, and images of different sizes may be converted to the said size. However, the size of the CT image is not limited to the said size.

[0090] In one embodiment, the preprocessing process of the CT image may include 1) Color-to-grayscale, 2) Zero-pedding, and Resize steps. In the Color-to-grayscale step, the computational load is reduced by converting 3 channels to 1 channel. In the Zero-pedding step, the image is adjusted to a 1:1 ratio. Finally, in the Resize step, the parameter (volume 33, height 128, width 128) is adjusted so that the pixel value closest to a point on the original image is assigned as the pixel value of the corresponding point in the output image.

[0091] Next, a step of converting the input CT image into a 3D image is performed (S120).

[0092] The above 3D converted image may be a 3D image of the patient's head.

[0093] In one embodiment, the 3D image may be an image converted using 20 photos.

[0094] The above 3D image can be used as input data for an artificial intelligence model.

[0095] Finally, a step is performed in which an artificial intelligence model predicts whether the patient's future bleeding will worsen using the converted 3D image as input data (S130).

[0096] The above artificial intelligence model may be ResNET or CNN, but is not limited thereto.

[0097] The above artificial intelligence model can be trained to output whether the patient's future bleeding will worsen by using a 3D image generated based on the initial CT image of a patient with head trauma as input data.

[0098] Referring to FIG. 5, a step of taking a first CT image of the patient's head for artificial intelligence model training is performed (S210).

[0099] It is obvious to a person skilled in the art that the patient described in Fig. 5 is a patient involved in the formation of training data for an artificial intelligence model, and is a different patient from the patient described in Fig. 4.

[0100] The training data for the above artificial intelligence model may be data in which the actual worsening of the patient's bleeding is labeled on a 3D image generated based on the first CT image (first CT image) taken after the patient suffered a head injury.

[0101] In order to check whether the bleeding has worsened, a step of taking a second CT image of the patient's head is performed after a predetermined time has elapsed since the time of the first CT image (S220).

[0102] Whether the actual bleeding has worsened can be determined by an additional CT image (second CT image) taken after a predetermined period has elapsed from the time the patient first underwent a head trauma. In other words, the above-mentioned worsening of bleeding can be determined not by the initial CT image after the head trauma, but by a CT image taken after a sufficient amount of time has passed since the head trauma.

[0103] Next, a step of labeling the first CT image is performed based on whether the hemorrhage has worsened in the second CT image (S230).

[0104] The above labeling can be matched to the generated 3D image when a 3D image is generated based on the first CT image.

[0105] In one embodiment, the labeling may be performed as either aggravated bleeding or not aggravated bleeding.

[0106] Finally, a step is performed to train an artificial intelligence model using the labeled first CT image as training data (S240).

[0107] The training data of the artificial intelligence model may be the labeled first CT image or a 3D image generated based on the labeled first CT image.

[0108] In one embodiment, the training data of the artificial intelligence model may not include any data other than a 3D image generated based on the first CT image taken after the patient suffered a head injury and whether the actual bleeding worsens labeled on the 3D image. That is, the artificial intelligence model may not use any other information about the patient other than the first CT image taken after the patient's head injury to predict whether the bleeding worsens.

[0109] In one embodiment, CT images of patients who visited the emergency room at the Gyeonggi Northern Regional Trauma Center of Uijeongbu St. Mary's Hospital for head trauma and underwent head CT scans were utilized to form training data. Specifically, patients with head trauma who visited the said emergency room from 2020 to 2021 were included, while pediatric patients under the age of 19 were excluded. CT images of the patients were collected in DICOM format. Subsequently, the patients were classified into two groups by labeling them based on whether the condition worsened during additional CT scans. The CT images and labels of the two groups were used as training data for an artificial intelligence model.

[0110] The ResNET model was used as the artificial intelligence model, and the AI ​​model was trained using CT images of 303 patients as training data, resulting in a prediction accuracy of 70.1%.

[0111] As described above, the trained artificial intelligence model can predict with high accuracy whether a patient's future bleeding will worsen.

[0112] Meanwhile, the processor can generate prescription data corresponding to the patient based on the prediction results of the artificial intelligence model. Specifically, the artificial intelligence model outputs a probability value regarding whether the patient's bleeding worsens, and the processor can generate prescription data recommending hospitalization if the probability value is higher than a reference value, and generate prescription data recommending discharge if the probability value is lower than the reference value.

[0113] For example, the processor may output prescription data recommending admission to the intensive care unit and additional CT scans to the patient when the probability of the patient's bleeding worsening exceeds 70%, and may output prescription data recommending discharge to the patient when the probability of the patient's bleeding worsening is less than 70%.

[0114] As described above, the diagnostic aid according to the present disclosure can predict whether future bleeding will worsen in a patient with head trauma based solely on the results of an initial CT scan. Through this, the present disclosure can minimize the need for additional CT scans for patients with head trauma and concentrate medical resources on patients at risk.

[0115] Figure 6 is a detailed flowchart of step S120 disclosed in Figure 4.

[0116] Referring to FIG. 6, a step of converting an input CT image into a first 3D image is performed (S121).

[0117] The first 3D image may be an image converted through an input CT image.

[0118] Next, a step of generating a plurality of second 3D images by converting the resolution of the first 3D image is performed (S122).

[0119] A plurality of second 3D images each have a resolution different from the resolution (V_original) of the first 3D image, and for example, three second 3D images can be generated having resolutions (V_1 / 2, V_1 / 4 and V_1 / 8) of 1 / 2, 1 / 4, and 1 / 8 of the resolution (V_original) of the first 3D image.

[0120] When multiple 3D images with different resolutions are applied as input data for an artificial intelligence model, bleeding patterns of different sizes can be effectively detected.

[0121] Next, a step of normalizing the brain tissue density (HU_raw) measured in the first 3D image and a plurality of second 3D images is performed (S123).

[0122] For example, the brain tissue density (HU_raw) measured in the first 3D image and a plurality of second 3D images, respectively, can be normalized using a brain tissue reference parameter standardized according to Equation 1 below.

[0123] [Mathematical Formula 1]

[0124] HU_normalized = (HU_raw - μ_brain) / σ_brain

[0125] In Equation 1, HU_normalized represents the normalized brain tissue density, HU_raw represents the brain tissue density measured in the 3D image, μ_brain represents the standardized average brain tissue density, which is applied as 30HU, for example, and σ_brain represents the standard deviation of the standardized brain tissue density, which is applied as 15HU, for example.

[0126] In this way, when normalizing brain tissue density in 3D images, consistency of input data for artificial intelligence models can be ensured, and the effect of improving analysis accuracy can be expected.

[0127] Next, a step of segmenting the hemorrhage area in the first 3D image and a plurality of second 3D images is performed (S124).

[0128] In one embodiment, segmentation of the hemorrhage region can utilize an Attention U-Net-based deep learning model. The hemorrhage region can be segmented into corresponding voxels using the output of an Attention U-Net-based deep learning model for a 3D image.

[0129] In one embodiment, the total hemorrhage volume in a 3D image can be quantified by summing the volumes of voxels divided into hemorrhage regions. Such a parameter for quantifying the hemorrhage volume in a 3D image can be applied as an input to an artificial intelligence model.

[0130] [Mathematical Formula 2]

[0131] V_hemorrhage = Σ(segmented_voxels×voxel_size³)

[0132] In mathematical formula 2, V_hemorrhage represents the hemorrhage volume in the 3D image, segmented_voxels represents the number of segmented voxels, and voxel_size represents the volume of the segmented voxels.

[0133] FIG. 7 is a flowchart of a diagnostic assistance method according to another embodiment of the present disclosure.

[0134] Referring to FIG. 7, after the step illustrated in FIG. 4, a step of predicting the bleeding spread rate is performed (S140).

[0135] In one embodiment, the bleeding spread rate, which is the rate of increase per hour of the bleeding area from the first CT image and the second CT image, can be modeled.

[0136] [Mathematical Formula 3]

[0137] Rate_expansion = α × V_current^β× P_intracranial^γ

[0138] In mathematical formula 3, Rate_expansion represents the rate of hemorrhage diffusion, α is a hemorrhage characteristic coefficient determined in the system, V_current is the volume of the hemorrhage area appearing in the first CT image, β is a volume dependency index, for example, set between 0.6 and 0.8, P_intracranial is intracranial pressure, and γ is an intracranial pressure influence index, for example, set between 1.2 and 1.5.

[0139] The rate of hemorrhage spread can be predicted from the input CT image using mathematical formula 3 as described above.

[0140] Next, a step to calculate a risk score that evaluates the risk of bleeding is performed (S150).

[0141] In one embodiment, the risk score can be calculated according to the following mathematical formula 4.

[0142] [Mathematical Formula 4]

[0143] risk_Score = w1×V_ratio + w2×Shape_complexity + w3×Location_factor + w4×Density_gradient

[0144] In mathematical formula 4, V_ratio is the ratio of the volume of the hemorrhage area (V_hemorrhage) and the total brain volume (V_brain_total) shown in the CT image, Shape_complexity is an indicator of the complexity of the hemorrhage area, calculated as, for example, Perimeter (total length of the boundary line of the hemorrhage area)² / (4π × Area (area occupied by the hemorrhage area)), Location_factor is a factor indicating how close the hemorrhage area is to major structures, calculated as higher the closer the distance is, for example, through Euclidean distance measurement with any of the structures among the brainstem, internal capsule, and thalamus, Density_gradient represents the gradient of the maximum HU (Hounsfield Unit) density change within the hemorrhage area, and w1, w2, w3, and w4 each represent weights set in the system.

[0145] In this way, it is possible to provide not only a prediction of bleeding but also a comprehensive prediction of the rate of bleeding spread and the risk of bleeding.

[0146] Next, a step to provide prediction reliability is performed (S160).

[0147] In one embodiment, the processor can quantify the prediction uncertainty of the artificial intelligence model by applying a Bayesian deep learning technique as shown in Equation 5 below.

[0148] [Mathematical Formula 5]

[0149] P(y|x) = ∫ P(y|x,θ)P(θ|D)dθ

[0150] In mathematical equation 5, P(y|x) represents the predicted probability distribution of output data y for input data x, P(y|x,θ) represents the predicted probability distribution of the parameter θ of the artificial intelligence model, and P(θ|D) represents the posterior probability distribution of the parameter θ for training data D.

[0151] The processor can provide a prediction confidence interval ([P_lower, P_upper]) from Equation 5.

[0152] In one embodiment, the processor may calculate and provide an uncertainty index that estimates the uncertainty of a prediction by applying Monte Carlo Dropout as shown in Equation 6 below. The lower the prediction uncertainty variance according to Equation 6, the more consistent and reliable the prediction can be evaluated.

[0153] [Mathematical Formula 6]

[0154] Prediction_final = (1 / T) × Σ(f(x, θt))

[0155] Uncertainty = Var(f(x, θt))

[0156] In mathematical equation 6, Prediction_final represents the average predicted value, T represents the number of samples, f(x, θt) represents the model predicted value for the parameter θt, and Uncertainty represents the prediction uncertainty.

[0157] Meanwhile, in one embodiment, the artificial intelligence model may have a multimodal fusion feature that learns spatial features and temporal patterns simultaneously. For example, the artificial intelligence model may learn spatial features and temporal patterns simultaneously by applying a hybrid architecture that combines a 3D CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory).

[0158] 3D CNN can extract spatial features from training data. Spatial features can be represented as shown in Equation 7 below.

[0159] [Mathematical Formula 7]

[0160] F_spatial = CNN_3D(V_input)

[0161] In mathematical equation 7, F_spatial is a parameter representing spatial features, and CNN_3D(V_input) represents the output of the 3D CNN for the volume (V_input) of the hemorrhage area appearing in the input data.

[0162] LSTM can learn temporal patterns from sequences of spatial feature parameters, which can be expressed as Equation 8 below.

[0163] [Mathematical Formula 8]

[0164] F_temporal = LSTM(F_spatial_sequence)

[0165] In mathematical equation 8, F_temporal is a parameter representing a temporal pattern feature, and LSTM(F_spatial_sequence) represents the output of the LSTM for the spatial feature parameter (F_spatial_sequence) per sequence.

[0166] In one embodiment, the artificial intelligence model may derive a final prediction probability by integrating the results of several individual prediction models in an ensemble manner.

[0167] For example, the artificial intelligence model may include a first prediction model (P1) that outputs volume-based prediction results, a second prediction model (P2) that outputs morphological feature-based prediction results, and a third prediction model that outputs location-based prediction results.

[0168] The artificial intelligence model can provide a final prediction result by integrating the prediction results of the first prediction model (P1) to the third prediction model (P3) according to the mathematical formula 9 below.

[0169] [Mathematical Formula 9]

[0170] P_final = Σ(wi Х Pi) / Σ(wi)

[0171] In mathematical formula 9, P_final is the final prediction probability of the artificial intelligence model, wi is set by the system as the confidence weight of prediction model i, and pi represents the prediction probability of prediction model i.

[0172] In this way, the above artificial intelligence model may maximize prediction performance by simultaneously learning spatial-temporal features or by fusing the prediction probabilities of multiple prediction models to present a final prediction probability.

[0173] Meanwhile, in one embodiment, the artificial intelligence model can be dynamically applied by setting a threshold value for each patient.

[0174] For example, the processor can set patient-specific thresholds based on the patient's age and comorbidities as shown in Equation 10 below, and dynamically adjust the set thresholds over time as shown in Equation 11 below. These thresholds can be applied as thresholds used for various decision-making in learning or prediction in the artificial intelligence model.

[0175] [Mathematical Formula 10]

[0176] Threshold_personalized = Threshold_base

[0177] In mathematical formula 10, Threshold_personalized means the threshold for a specific patient, Threshold_base means the reference threshold, Age_factor means the age factor set in the system based on the patient's age, Comorbidity_score means the comorbidity score set in the system based on the patient's comorbidity, and α and β mean the weights set in the system.

[0178] [Mathematical Formula 11]

[0179] Threshold_t = Threshold_0 Х exp(-λt) + Threshold_min

[0180] In Equation 11, Threshold_t means the adjusted threshold after time t has elapsed, Threshold_0 means the initial threshold, λ means the speed control parameter set on the system, and Threshold_min means the minimum threshold.

[0181] In this way, the above artificial intelligence model can optimize the sensitivity and specificity of prediction by applying customized thresholds that consider patient-specific characteristics and dynamically changing the thresholds to detect sensitive changes over time.

[0182] Meanwhile, the processor can perform system processing optimization to support rapid decision-making in emergency situations.

[0183] In one embodiment, the processor can reduce the number of parameters of the artificial intelligence model to less than 1,000 (< 10M) and shorten the inference time to less than 30 seconds (< 30 seconds) by applying a lightweight network structure based on MobileNet-3D.

[0184] In addition, the processor can calculate the reliability of the artificial intelligence model according to the processing time using the following mathematical formula 12, and perform incremental prediction by immediately returning an intermediate prediction result when the reliability reaches a threshold (e.g., 0.8).

[0185] [Mathematical Formula 12]

[0186] Confidence_level = min(Processing_time Х Speed_factor, 1.0)

[0187] In mathematical formula 12, Confidence_level represents confidence, Processing_time represents the processing time of the artificial intelligence model, and Speed_factor represents the speed factor, which can be set on the system, for example, between 0.1 and 1.0.

[0188] Meanwhile, the processor may also provide the basis for the prediction results by the above artificial intelligence model.

[0189] In one embodiment, the processor can visualize and provide the area of ​​the 3D image that has the greatest influence on the prediction of the artificial intelligence model using a Grad-CAM (Gradient-weighted Class Activation Mapping) based heatmap.

[0190] In one embodiment, the processor can utilize SHAP (SHapley Additive exPlanations) values ​​to calculate and present the degree to which at least one feature element included in the input data (e.g., volume, location, etc. of a bleeding area) contributed to the prediction result of the artificial intelligence model.

[0191] [Mathematical Formula 13]

[0192] ϕi = Σ [|S|!(n-|S|-1)! / n!] × [f(S∪{i}) - f(S)]

[0193] In Equation 13, φi represents the contribution of feature i included in the input data, n represents the total number of feature elements included in the input data, S represents a subset of feature elements excluding i among the feature elements included in the input data, f(S) represents the predicted value of the artificial intelligence model when the subset S is applied, and f(S∪{i}) represents the predicted value of the artificial intelligence model when the subset S is applied including feature i.

[0194] Meanwhile, the processor can verify the artificial intelligence model by comparing the prediction results of the above artificial intelligence model with the patient's actual progress.

[0195] In one embodiment, the processor may obtain the actual course of the patient's bleeding and perform cross-validation of the artificial intelligence model by comparing it with the prediction result of the artificial intelligence model.

[0196] For example, the processor can calculate performance indicators such as AUC-ROC, Sensitivity (True Positive Rate), Specificity (True Negative Rate), and PPV (Positive Predictive Value) for the artificial intelligence model, and perform cross-validation based on threshold values ​​for each performance indicator (e.g., AUC-ROC > 0.85, Sensitivity > 0.90, Specificity > 0.80, PPV > 0.75).

[0197] Meanwhile, the processor can update the above artificial intelligence model based on the collection of real-time learning data.

[0198] In one embodiment, the artificial intelligence model can update the training data in real time by performing a progressive learning function as shown in Equation 14 below.

[0199] [Mathematical Formula 14]

[0200] Model_t+1 = Model_t + η × ∇L(New_data)

[0201] In Equation 14, Model_t+1 represents the parameters of the updated model, Model_t represents the model parameters before the update, η represents the learning rate set in the system, and ∇L(New_data) represents the gradient of the loss function for the newly added training data (New_data).

[0202] In this way, the processor aims to maintain and improve the system's continuous performance through cross-validation and real-time model updates.

[0203] Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium that stores instructions executable by a computer. The instructions may be stored in the form of program code and, when executed by a processor, may generate a program module to perform the operation of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

[0204] Computer-readable recording media include all types of recording media that store instructions that can be decoded by a computer. Examples include ROM (Read Only Memory), RAM (Random Access Memory), magnetic tape, magnetic disk, flash memory, optical data storage devices, etc.

[0205] As described above, the disclosed embodiments have been explained with reference to the attached drawings. Those skilled in the art will understand that the present disclosure may be practiced in forms different from the disclosed embodiments without changing the technical spirit or essential features of the present disclosure. The disclosed embodiments are illustrative and should not be interpreted restrictively.

Claims

1. A diagnostic aid for predicting the worsening of intracranial hemorrhage in patients with head trauma, A communication unit configured to receive a patient's head CT image; and The above input CT image is converted into a first 3D image, and A plurality of second 3D images are generated by converting the resolution of the first 3D image, and A processor comprising an artificial intelligence model trained to predict whether the patient's future intracranial hemorrhage will worsen by utilizing the first 3D image and the plurality of second 3D images, and The above CT image is, A diagnostic aid characterized by being the first CT image taken after the patient suffered a head injury.

2. In Paragraph 1, The above processor is, A diagnostic aid characterized by generating prescription data corresponding to the patient based on the result of predicting whether bleeding will worsen.

3. In Paragraph 2, The above artificial intelligence model is, Output the probability value of whether the above patient's bleeding worsens, and The above processor is, If the above probability value is higher than the reference value, prescription data recommending hospitalization is generated, and A diagnostic aid characterized by generating prescription data recommending discharge when the above probability value is lower than a reference value.

4. In Paragraph 3, The training data used for training the above artificial intelligence model is, A diagnostic aid characterized by data labeled with whether the patient's actual bleeding has worsened, based on a 3D image generated from the first CT image taken after the patient suffered a head injury.

5. In Paragraph 4, Whether the above actual bleeding worsens is, A diagnostic aid characterized by being determined based on additional CT images taken after a predetermined time has elapsed from the point in time when the patient first took a CT image after suffering a head injury.

6. In Paragraph 5, The input data input to the above artificial intelligence model is, A diagnostic aid characterized by not including any data other than a 3D image generated based on the first CT image taken after the patient suffered head trauma.

7. In Paragraph 6, A diagnostic aid device characterized in that the CT image used for the above 3D image conversion is an image in DICOM format.

8. In Paragraph 1, The above processor is, Generating the plurality of second 3D images having a lower resolution than the first 3D image, and Normalizing the brain tissue density measured in the first 3D image and the plurality of second 3D images, and The bleeding area in the first 3D image and the plurality of second 3D images is segmented and input into the artificial intelligence model, and A diagnostic aid that predicts whether the bleeding will worsen based on the output of the artificial intelligence model for the above input.

9. In Paragraph 1, The above processor is, Predicting the hemorrhage spread rate, which represents the hourly increase rate of the hemorrhage area from the above input CT image, and A diagnostic aid that calculates and provides a risk score indicating the risk of bleeding based on the volume and location of the bleeding area shown in the above-mentioned input CT image.

10. In Paragraph 1, The above processor is, A diagnostic aid device that sets a patient-specific threshold based on the patient's age and comorbidities and applies it to the artificial intelligence model, while adjusting and applying the threshold over time.

11. A diagnostic aid method for predicting the worsening of intracranial hemorrhage in patients with head trauma, Step of receiving the patient's head CT image; A step of converting the above-mentioned input CT image into a first 3D image; A step of generating a plurality of second 3D images by converting the resolution of the first 3D image; and The artificial intelligence model includes a step of predicting whether the patient's future intracranial hemorrhage will worsen by utilizing the first 3D image and the plurality of second 3D images, and The above CT image is, A diagnostic assistance method characterized by the fact that the CT image is the first image taken after the patient suffered a head injury.

12. In Paragraph 11, A diagnostic assistance method characterized by further including the step of generating prescription data corresponding to the patient based on the result of predicting whether bleeding will worsen.

13. In Paragraph 12, The above artificial intelligence model is, Output the probability value of whether the above patient's bleeding worsens, and The step of generating the above prescription data is, If the above probability value is higher than the reference value, prescription data recommending hospitalization is generated, and A diagnostic assistance method characterized by generating prescription data recommending discharge when the above probability value is lower than a reference value.

14. In Paragraph 13, The training data used for training the above artificial intelligence model is, A diagnostic assistance method characterized by data in which the actual worsening of bleeding in the patient is labeled in a 3D image generated based on the first CT image taken after the patient suffered head trauma.

15. In Paragraph 14, Whether the above actual bleeding worsens is, A diagnostic assistance method characterized by being determined based on additional CT images taken after a predetermined time has elapsed from the point in time when the patient first took a CT image after suffering head trauma.

16. In Paragraph 15, The input data input to the above artificial intelligence model is, A diagnostic assistance method characterized by not including any data other than a 3D image generated based on the first CT image taken after the patient suffered head trauma.

17. In Paragraph 16, A diagnostic assistance method characterized in that the CT image used for the above 3D image conversion is an image in the DICOM format.

18. In Paragraph 11, A step of generating a plurality of second 3D images having a lower resolution than the first 3D image; A step of normalizing the brain tissue density measured in the first 3D image and the plurality of second 3D images; A step of segmenting hemorrhage regions in the first 3D image and the plurality of second 3D images and inputting them into the artificial intelligence model; and A diagnostic assistance method further comprising the step of predicting whether the bleeding will worsen based on the output of the artificial intelligence model for the above input.

19. In Paragraph 11, A step of predicting a hemorrhage spread rate indicating the hourly increase rate of the hemorrhage area from the above-mentioned input CT image; and A diagnostic assistance method further comprising the step of calculating and providing a risk score indicating the risk of bleeding based on the volume and location of the bleeding area shown in the input CT image.

20. In Paragraph 11, A diagnostic assistance method further comprising the step of setting a patient-specific threshold according to the patient's age and comorbidities and applying it to the artificial intelligence model, wherein the threshold is adjusted and applied over time.

Images