Method for predicting state of object on basis of dynamic image data, and computing device performing same
The method processes PET scan data using prediction models to predict patient conditions more quickly, addressing the delay in PET image-based diagnosis by learning from initial and reduced-tracer dynamic image data, resulting in faster and more efficient diagnostic insights.
Patent Information
- Application Number
- PCT/KR2024/095818
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-05-17
- Filing Date
- 2024-05-17
- Publication Date
- 2024-11-21
AI Technical Summary
Current PET image-based diagnosis in medical settings is delayed due to the time required for tracer distribution and binding, necessitating a method to predict patient conditions more promptly based on dynamic image data.
A method utilizing a computing device to process initial dynamic image data from PET scans, learning prediction models for anatomical and disease-specific information, allowing for earlier prediction of patient conditions by reducing the amount of tracer injected and applying spatial normalization techniques.
This approach significantly reduces the waiting time for diagnosis, improving user convenience by generating initial blood flow and delayed intake image data simultaneously, thereby enhancing the efficiency of PET image analysis.
Smart Images

Figure KR2024095818_21112024_PF_FP_ABST
Abstract
Description
Method for predicting the state of an object based on dynamic image data and a computing device for performing the same
[0001] The present disclosure relates to medical technology. More specifically, the present disclosure relates to a method for predicting the condition of a subject for diagnosis based on dynamic image data, and a computing device for performing the method.
[0002] Positron Emission Tomography (PET) is a cutting-edge nuclear medicine imaging technique that administers a positron-emitting radioisotope to a subject and captures the radiation emitted from the body, providing useful diagnostic information about metabolic changes and receptor distribution in the body. Recently, PET has evolved beyond simple PET imaging to include hybrid scanners that integrate CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) technology.
[0003] Accordingly, recent PET examinations have used PET / CT scanners that combine PET devices and CT devices into one, which allows for obtaining anatomical information from CT images, thereby providing accurate location and depth information of lesions identified in PET images.
[0004] Typically, PET scans are performed to diagnose patients based on PET images acquired after a certain period of time (e.g., 1 hour 30 minutes to 3 hours) has elapsed after a radioactive isotope-labeled tracer is injected into the body, during which time the tracer-specific binding and non-specific binding reach a stable state or the difference between them becomes maximum.
[0005] However, since PET imaging-based diagnosis is performed relatively late after tracer injection, a method to improve this is needed.
[0006] The task to be solved in this disclosure is to provide a method for predicting the condition of a diagnosis subject based on dynamic image data.
[0007] The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.
[0008] A method performed by at least one processor for predicting a state of a subject based on dynamic image data according to one embodiment of the present invention comprises the steps of: acquiring initial dynamic image data corresponding to an initial period up to a preset time point after a drug is injected into a subject for learning; learning a first prediction model that predicts first image data representing anatomical information or blood flow information about the subject for learning corresponding to a time point before the first period by inputting the initial dynamic image data corresponding to a first period among the initial periods; and learning a second prediction model that predicts second image data representing disease-specific information about the subject for learning corresponding to a reference time point after the initial period by inputting the initial dynamic image data corresponding to a second period among the initial periods.
[0009] The method may further include a step of acquiring initial dynamic image data corresponding to the first section after the drug is injected into the diagnostic subject; and a step of inputting the initial dynamic image data corresponding to the first section of the diagnostic subject into the learned first prediction model to predict first image data representing anatomical information or blood flow information for the diagnostic subject corresponding to the preceding time point.
[0010] The method may further include a step of acquiring initial dynamic image data corresponding to the second section after the drug is injected into the diagnostic subject; and a step of inputting the initial dynamic image data corresponding to the second section of the diagnostic subject into the learned second prediction model to predict second image data representing disease-specific information about the diagnostic subject corresponding to the reference time point.
[0011] The processor may inject the drug, which has been reduced by a preset amount, into the diagnostic subject, and based on this, acquire and process initial dynamic image data corresponding to the first section and initial dynamic image data corresponding to the second section, acquire the processed initial dynamic image data corresponding to the first section and the processed initial dynamic image data corresponding to the second section, input the processed initial dynamic image data corresponding to the first section into the first prediction model, and input the processed initial dynamic image data corresponding to the second section into the second prediction model.
[0012] The amount of the drug injected into the learning subject is a reduced amount compared to the reference amount, the first prediction model is learned based on first label image data corresponding to the first image data, and the second prediction model is learned based on second label image data corresponding to the second image data, wherein the first label image data and the second label image data may be label image data processed according to the amount of drug injected into the learning subject.
[0013] The method may further include a step of performing spatial normalization before initial dynamic image data corresponding to a first section of the learning object and the diagnosis object are input to a first prediction model, so as to enable machine learning using a smaller number of learning object data; and a step of performing spatial normalization before initial dynamic image data corresponding to a second section of the learning object and the diagnosis object are input to a second prediction model.
[0014] The processor may set an acquisition time section of initial dynamic image data corresponding to a first section among the initial sections of the learning object and the diagnosis object, and may set an acquisition time section of initial dynamic image data corresponding to a second section among the initial sections of the learning object and the diagnosis object.
[0015] The processor can identify the initial section when a drug is injected into the diagnostic subject, and obtain initial dynamic image data corresponding to the first section and the second section based on the identified initial section.
[0016] The acquired initial dynamic image data, the first image data, and the second image data may be positron emission tomography (PET) image data.
[0017] A dynamic image data-based target state prediction device according to one embodiment of the present invention may include a memory; and at least one processor that communicates with the memory.
[0018] The processor may acquire initial dynamic image data corresponding to an initial section up to a preset time point after a drug is injected into a learning subject, learn a first prediction model that uses the initial dynamic image data corresponding to a first section of the initial section as input to predict first image data representing anatomical information or blood flow information about the learning subject corresponding to a time point before the first section, and learn a second prediction model that uses the initial dynamic image data corresponding to a second section of the initial section as input to predict second image data representing disease-specific information about the learning subject corresponding to a reference time point after the initial section.
[0019] The processor may obtain initial dynamic image data corresponding to the first section after the drug is injected into the diagnostic subject, and input the initial dynamic image data corresponding to the first section of the diagnostic subject into the learned first prediction model, thereby predicting first image data representing anatomical information or blood flow information about the diagnostic subject corresponding to the preceding time point.
[0020] The processor can obtain initial dynamic image data corresponding to the second section after the drug is injected into the diagnostic subject, and input the initial dynamic image data corresponding to the second section of the diagnostic subject into the learned second prediction model, thereby predicting second image data representing disease-specific information about the diagnostic subject corresponding to the reference time point.
[0021] The processor may inject the drug, which has been reduced by a preset amount, into the diagnostic subject, and based on this, acquire and process initial dynamic image data corresponding to the first section and initial dynamic image data corresponding to the second section, acquire the processed initial dynamic image data corresponding to the first section and the processed initial dynamic image data corresponding to the second section, input the processed initial dynamic image data corresponding to the first section into the first prediction model, and input the processed initial dynamic image data corresponding to the second section into the second prediction model.
[0022] The amount of the drug injected into the learning subject is a reduced amount compared to the reference amount, the first prediction model is learned based on first label image data corresponding to the first image data, and the second prediction model is learned based on second label image data corresponding to the second image data, wherein the first label image data and the second label image data may be label image data processed according to the amount of drug injected into the learning subject.
[0023] The processor may perform spatial normalization before initial dynamic image data corresponding to a first section of the learning object and diagnosis object are input to a first prediction model, and may perform spatial normalization before initial dynamic image data corresponding to a second section of the learning object and diagnosis object are input to a second prediction model.
[0024] The waiting time for the distribution of the image tracer in the body after drug injection, which is a characteristic of nuclear medicine molecular imaging, can take up to several hours. However, according to various embodiments of the present disclosure, initial blood flow image data and delayed uptake image data can be generated simultaneously using only the initial dynamic image data after drug injection, and the waiting time for diagnosis can be drastically reduced, thereby greatly improving user convenience.
[0025] The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.
[0026] FIG. 1 is an overall schematic diagram showing a method for predicting the state of a target object based on dynamic image data according to the present disclosure.
[0027] FIG. 2 is a block diagram showing the configuration of a dynamic image data-based target state prediction device according to the present disclosure.
[0028] FIG. 3 is a sequence diagram illustrating a method for predicting a target state based on dynamic image data according to the present disclosure.
[0029] FIG. 4 is a diagram for explaining a process of learning a first prediction model and a second prediction model based on initial dynamic image data according to the present disclosure.
[0030] FIG. 5 is a sequence diagram illustrating a method for predicting first image data representing anatomical information or blood flow information of a diagnosis subject according to the present disclosure and second image data representing disease-specific information.
[0031] Figure 6 is a sequence diagram showing a preprocessing process for performing spatial normalization according to the present disclosure.
[0032] Figure 7 is a diagram for explaining spatial normalization performed in a preprocessing step before learning an artificial neural network model according to the present disclosure.
[0033] Figure 8 is a diagram showing changes in radiation dose over time after drug injection according to the present disclosure.
[0034] FIG. 9a is a graph for comparing delayed image data and label image data generated with a reference dose according to the present disclosure, FIG. 9b is a graph for comparing delayed image data and label image data generated with a low dose according to the present disclosure, and FIG. 9c is a graph for comparing delayed image data generated with a reference dose according to the present disclosure and delayed image data generated with a low dose.
[0035] FIG. 10a shows analog PET data used in the learning and testing stages for each of the AP (Anterior Putamen) and PP (Posterior Putamen) according to the present disclosure, and FIG. 10b shows digital PET data used in the testing stage for each of the AP and PP according to the present disclosure.
[0036] FIG. 11 is a diagram illustrating a method for obtaining ultra-early image data according to the present disclosure.
[0037] FIG. 12 is a diagram for explaining a method for obtaining delay image data according to the present disclosure.
[0038] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. The advantages and features of the present invention, and methods for achieving them, will become clearer with reference to the embodiments described in detail below together with the attached drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals designate like elements throughout the specification.
[0039] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in their common sense to those of ordinary skill in the art to which the present invention pertains. Furthermore, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.
[0040] The terminology used herein is for the purpose of describing embodiments and is not intended to limit the present invention. In this specification, singular forms also include plural forms, unless specifically stated otherwise. As used herein, the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components in addition to the components mentioned.
[0041] In this specification, the term "computing device" includes various devices that perform computational processing. A "computing device" may include one or more computers. For example, a computer may include not only a desktop PC or notebook, but also a smart phone, a tablet PC, a cellular phone, a PCS phone (Personal Communication Service phone), a synchronous / asynchronous IMT-2000 (International Mobile Telecommunication-2000) mobile terminal, a Palm PC (Palm Personal Computer), a PDA (Personal Digital Assistant), etc. In addition, a computer may also include medical equipment that acquires or observes medical images. In addition, a computer may be a server computer that is connected to various client computers.
[0042] In this specification, 'image data' means an image acquired by a medical imaging device.
[0043] As used herein, "medical imaging device" refers to a device used to acquire medical images. For example, "medical imaging device" may include a positron emission tomography (PET) imaging device, a magnetic resonance imaging (MRI) imaging device, and the like.
[0044] In this specification, 'delay image data' refers to diagnostic images acquired after a reference time and used for patient diagnosis.
[0045] In this specification, the term 'reference time' means the time from the initial injection of a drug (e.g., contrast agent or tracer) into the human body to the time at which image data capable of diagnosing the patient's condition can be obtained (i.e., the reference time).
[0046] In this specification, "drug" refers to something injected into the body during medical imaging. For example, "drug" may include contrast agents used in magnetic resonance imaging (MRI) or computed tomography (CT), or tracers used in positron emission tomography (PET).
[0047] In this specification, 'initial dynamic image data' is image data including a plurality of consecutive image frames, and 'initial dynamic image data' is acquired before the reference time point at which delayed image data is acquired, and is acquired in an initial time range (e.g., a time range a short time after insertion of a contrast agent or tracer administered during imaging).
[0048] In this specification, dynamic image data may mean image data excluding initial dynamic image data and delayed image data from image data.
[0049] Specifically, dynamic image data may refer to a playback section image from the time at which the blood flow effect begins to decrease to a predetermined reference time point after the drug injection time of the learning subject included in each of the plurality of image data.
[0050] In this specification, anatomical information may be information about the structure and function of the subject, and may include image data including blood flow information showing the flow of blood in the subject.
[0051] In this specification, disease-specific information may mean image data that can be used to determine the condition of a subject by allowing a drug to remain in a specific target location (e.g., an organ or tissue).
[0052] Hereinafter, a detailed description will be given of a method and program for generating diagnostic images based on initial dynamic image data according to embodiments of the present invention with reference to the drawings.
[0053] FIG. 1 is an overall schematic diagram showing a method for predicting the state of an object based on dynamic image data according to the present disclosure, and can be performed by a dynamic image data-based object state prediction device (100, hereinafter referred to as 'object state prediction device').
[0054] The target state prediction device (100) can acquire initial dynamic image data (10). The initial dynamic image data (10) can correspond to an initial period from when a drug is injected into the target through a tracer to a preset time point. The initial dynamic image data (10) can include a plurality of image frames, and each of the plurality of image frames can be acquired at a 1-minute or 2-minute cycle, but the present disclosure is not limited thereto.
[0055] Here, the preset time point may be a predetermined range of time points based on the peak radiation dose contained in the subject's bloodstream. Furthermore, the initial interval may vary depending on the type of tracer and the underlying factors of the subject being diagnosed.
[0056] The target state prediction device (100) can input the acquired initial dynamic image data (10) into the first prediction model (EM1) and the second prediction model (EM2) (S11, S12).
[0057] In this case, initial dynamic image data corresponding to the first section of the initial dynamic image data (10) may be input into the first prediction model (EM1), and initial dynamic image data corresponding to the second section of the initial dynamic image data (10) may be input into the second prediction model (EM2). However, according to an embodiment, the object state prediction device (100) may input the same initial dynamic image data (10) into the first prediction model (EM1) and the second prediction model (EM2).
[0058] The first prediction model (EM1) can predict true early image data (20) based on the input initial dynamic image data (S21). The true early image data may be dynamic image data corresponding to a preset short period after drug injection or image data at a specific point in time.
[0059] Additionally, the second prediction model (EM2) can predict delayed image data (30) based on the input initial dynamic image data (S21). The delayed image data may be dynamic image data corresponding to a reference point in time after the initial period or image data at a specific point in time.
[0060] In addition, the initial dynamic image data corresponding to the first section, the initial dynamic image data corresponding to the second section, the ultra-early image data, and the delayed image data may all be positron emission tomography (PET) image data, but the present disclosure is not limited thereto.
[0061] FIG. 2 is a block diagram showing the configuration of a target state prediction device (100) according to the present disclosure.
[0062] Referring to FIG. 2, the object state prediction device (100) may include a medical imaging device or may use image data captured from a medical imaging device, and may include a communication unit (110), an input unit (120), a display (130), a memory (150), and at least one processor (190). The components of the object state prediction device (100) illustrated in FIG. 2 are not essential for implementing the object state prediction device (100) according to the present disclosure, and thus the object state prediction device (100) described in this specification may have more or fewer components than the components listed above.
[0063] Among the above components, the communication unit (110) may include one or more components that enable communication with various devices equipped with communication devices, and may include, for example, at least one of a satellite device, a wired communication device, a cellular-based wireless communication device, an IEEE 802.11-based (e.g., may be named Wifi), a short-range communication-based (e.g., may be, but is not limited to, Bluetooth, Bluetooth low energy, UWB, Zigbee) communication device, and a location information module.
[0064] The input unit (120) is for inputting image information (or signal), audio information (or signal), data, or information input from a user, and may include at least one camera, a touch input device equipped on a touch screen, and / or at least one microphone, but is not limited thereto. Touch input, voice data, and / or image data for the touch screen collected by the input unit (120) may be analyzed and processed as a user's control command. The input unit (120) may include various input devices (inputters).
[0065] The camera processes image frames, such as still images or moving images, obtained by the image sensor in shooting mode. The processed image frames may be displayed on a display (130, or the screen of the object state prediction device (100) of the present disclosure) or stored in a memory (150).
[0066] A microphone processes external acoustic signals into electrical voice data. The processed voice data can be utilized in various ways depending on the function (or application) being performed by the device. Meanwhile, the microphone can implement various noise-reduction algorithms to eliminate noise generated during the process of receiving external acoustic signals.
[0067] The output unit is for generating output related to visual, auditory, or tactile sensations, and may include at least one of a display (130), at least one speaker, a haptic module, and an optical output device. The display (130) may be formed as a layer structure with a touch input device or formed as an integral part, thereby implementing a touch screen. Such a touch screen may perform an output function and / or an input function. The output device may include a device (outputter) for various outputs.
[0068] The memory (150) can store at least one instruction that causes the execution of various functions of the object state prediction device (100). The memory (150) can store data for content expression (e.g., music files, still images, moving images, etc.). The memory (150) can store at least one application program (or application) that causes the execution of operations performed by various embodiments of the present disclosure driven by the object state prediction device (100), data for the operation of the object state prediction device (100), and commands. At least some of these application programs can be downloaded from an external server via wireless communication. The object state prediction device (100) can, for example, download the application and store it in the memory (150). The object state prediction device (100) can perform the operations performed by various embodiments of the present disclosure by executing the application. Alternatively, the target state prediction device (100) may temporarily download data causing the operations performed by various embodiments of the present disclosure from a server (for example) and store the data in the memory (150).
[0069] The memory (150) may include a storage medium corresponding to at least one type of a flash memory type, a hard disk type, an SSD (Solid State Disk type), an SDD (Silicon Disk Drive type), a multimedia card micro type, a card type memory (e.g., SD or XD memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk.
[0070] Those skilled in the art will understand that the memory (150) may mean, for example, a cache memory for linking with the processor (190) and / or a cache memory and / or register included in the processor (190). In addition, the memory (150) may be a database that is separate from the target state prediction device (100) but connected by wire or wirelessly, and may be implemented as a database system.
[0071] The processor (190) may include one or more processors and may include at least one core. The processor (190) may execute instructions stored in the memory (150). The processor (190) may be implemented with a memory that stores data regarding an algorithm for controlling the operation of components within the target state prediction device (100) or a program that reproduces the algorithm, and at least one processor (not shown) that performs the aforementioned operation using the data stored in the memory. In this case, the memory and the processor may be implemented as separate chips. Alternatively, the memory and the processor may be implemented as a single chip.
[0072] In an embodiment, the object state prediction device (100) may provide various UIs in the form of web services based on a platform, such as, but not limited to, a website or web application. Furthermore, the platform may be provided in the form of a PC application, a mobile application, or the like, but the embodiment is not limited thereto. In this case, various user terminals may utilize the various UIs provided by the object state prediction device (100) based on a platform.
[0073] The predefined operation rules or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that the basic artificial intelligence model is learned by a learning algorithm using a plurality of learning data, thereby creating a predefined operation rules or artificial intelligence model set to perform a desired characteristic (or purpose). This learning may be performed on the device itself on which the artificial intelligence according to the present disclosure is performed, or may be performed through a separate server and / or system. Examples of the learning algorithm include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.
[0074] 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 operation results of the previous layer and the multiple weights. The multiple weights of the multiple neural network layers may 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 is reduced or minimized during the learning process. The artificial neural network may include a deep neural network (DNN), and examples thereof include, but are not limited to, a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), or deep Q-networks.
[0075] 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 imitates human neurons (biological neurons) to enable machines to learn. Artificial intelligence methodologies can be categorized into supervised learning, in which input data and output data are provided together as training data depending on the learning method, so that the solution (output data) to the problem (input data) is determined; unsupervised learning, in which only input data is provided without output data, so that the solution (output data) to the problem (input data) is not determined; and reinforcement learning, in which a reward (Reward) is provided from an external environment whenever an action (Action) is taken in the current state (State), and learning is performed in a direction to maximize this reward. In addition, artificial intelligence methodologies can be categorized according to the architecture of the learning model. The architectures of widely used deep learning technologies can be categorized into convolutional neural networks (CNNs), recurrent neural networks (RNNs), transformers, and generative adversarial networks (GANs).
[0076] The present device and system may include an artificial intelligence model. The artificial intelligence model may be a single artificial intelligence model or may be implemented as multiple artificial intelligence 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 in general that has problem-solving capabilities by changing the binding strength of synapses through learning, formed by artificial neurons (nodes) that form a network by combining synapses. The neurons of the neural network may include a combination of weights or biases. The 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 desired result (output) from an arbitrary input (input) by changing the weights of neurons through learning.
[0077] The processor can create a neural network, train (or learn) a neural network, perform a calculation based on received input data, generate an information signal based on the calculation result, or retrain the neural network. The models of the neural network can include various types of models such as CNN (Convolution Neural Network) such as GoogleNet, AlexNet, VGG Network, R-CNN (Region with Convolution Neural Network), RPN (Region Proposal Network), RNN (Recurrent Neural Network), S-DNN (Stacking-based deep Neural Network), S-SDNN (State-Space Dynamic Neural Network), Deconvolution Network, DBN (Deep Belief Network), RBM (Restrcted Boltzman Machine), Fully Convolutional Network, LSTM (Long Short-Term Memory) Network, Classification Network, etc., but are not limited thereto. The processor can include one or more processors for performing calculations according to the models of the neural network. For example, the neural network can be a deep neural network. It may include a deep neural network.
[0078] 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), Generative Adversarial Network (GAN), Liquid State Machine (LSM), Extreme Learning Machine (ELM), It will be understood by those skilled in the art that any neural network may be included, including but not limited to ESN (Echo State Network), DRN (Deep Residual Network), DNC (Differentiable Neural Computer), NTM (Neural Turning Machine), CN (Capsule Network), KN (Kohonen Network), and AN (Attention Network).
[0079] At least one component may be added or deleted to correspond to the performance of the components illustrated in Figure 2. Furthermore, it will be readily apparent to those skilled in the art that the relative positions of the components may be altered to correspond to the performance or structure of the system.
[0080] Meanwhile, each component illustrated in FIG. 2 refers to software and / or hardware components such as a Field Programmable Gate Array (FPGA) and an Application Specific Integrated Circuit (ASIC).
[0081] FIG. 3 is a sequence diagram illustrating a method for predicting a state of an object based on dynamic image data according to the present disclosure (including steps S300, S310 to S330), and FIG. 4 is a diagram for explaining a process of learning a first prediction model (EM1) and a second prediction model (EM2) based on initial dynamic image data according to the present disclosure. When describing FIG. 3, FIG. 4 will be referred to together where necessary.
[0082] In step S310, the processor (190) can acquire initial dynamic image data corresponding to an initial section up to a preset point in time after the drug is injected into the learning subject.
[0083] Here, the drug may be a tracer. Tracers can come in various types depending on the target tissue or organ. For example, different types of tracers can be used to determine the location or size of a tumor, investigate brain functional activity, or diagnose cardiac blood flow and metabolic activity.
[0084] Additionally, the preset point in time may be a point within a preset range at which the radiation dose to a site other than the target tissue or organ reaches a peak, and may be used as a default value depending on the embodiment.
[0085] In step S320, the processor (190) may learn a first prediction model (EM1) that predicts first image data representing anatomical information (or blood flow information) for a learning subject corresponding to a time point before the first section by inputting initial dynamic image data corresponding to a first section among the initial sections. Here, the first image data may be extremely initial image data.
[0086] Referring to FIG. 4, the first prediction model (EM1) may include a generative model (EM1G) and a discriminative model (EM1D).
[0087] When the processor (190) inputs the initial dynamic image data (710a) corresponding to the first section into the generation model (EM1G) of the first prediction model (EM1), it can predict (generate) intermediate ultra-early image data (710b).
[0088] The processor (190) inputs the generated intermediate ultra-early image data (710b) and the label image data (710c, first label) of the ultra-early image data into the discrimination model (EM1D), and compares and verifies the generated intermediate ultra-early image data (710b) and the first label (710c) through the generation model (EM1G). The processor (190) can repeatedly perform generation and discrimination until the prediction model (EM1) satisfies a preset condition (e.g., a similarity of 95% or more).
[0089] The first prediction model (EM1) can be generated based on a GAN model, and a Conditional GAN model that additionally utilizes condition information (e.g., a label) can be applied, but the present disclosure is not limited thereto.
[0090] At step S330, the processor (190) can learn a second prediction model (EM2) that predicts second image data representing disease-specific information for a learning subject corresponding to a reference time point after the initial period by inputting initial dynamic image data corresponding to a second period among the initial periods. Here, the second image data may be delayed image data.
[0091] Referring to FIG. 4, the processor (190) can predict intermediate delay image data (720b) when inputting initial dynamic image data (720a) corresponding to the second section into the generation model (EM2G) of the second prediction model (EM2).
[0092] The processor (190) inputs the generated intermediate delay image data (720b) and the label image data (720c, second label) of the delayed image data into the discrimination model (EM2D), and can compare and verify the intermediate delay image data (720b) and the second label (730c) generated through the generation model (EM2G). The processor (190) can repeatedly perform generation and discrimination until the prediction model (EM2) satisfies a preset condition (e.g., a similarity of 95% or more).
[0093] The second prediction model (EM2) can be generated based on a GAN model, and a Conditional GAN model that additionally utilizes condition information (e.g., a label) can be applied, but the present disclosure is not limited thereto.
[0094] In an embodiment, the amount of drug injected into the learning subject may be reduced from the reference amount. For example, the reduced amount may be half of the reference amount, but the present disclosure is not limited thereto.
[0095] The first prediction model (EM1) may be trained based on first label image data corresponding to the very early image data, and the second prediction model (EM2) may be trained based on second label image data corresponding to the delayed image data. At this time, the first label image data and the second label image data may be label image data processed according to the amount of drug injected into the learning subject. When the amount of drug injected into the learning subject is half of the reference amount, the processor (190) needs to use the first label image data and the second label image data as label image data corresponding to half of the reference amount.
[0096] In an embodiment, the processor (190) may directly obtain first label image data and second label image data for cases where the drug injection amount is less than the reference amount through a clinical trial and use the same for learning.
[0097] However, since even when injecting a small amount of medicine compared to the standard amount, permission from a government agency is required and may be difficult to secure, the processor (190) can generate first label image data and second label image data corresponding to a case where the amount of medicine injected is less than the standard amount by shortening the frame generation time when generating the first label image data and second label image data in which the amount of medicine injected is the standard amount.
[0098] For example, when the processor (190) generates 5 image data in 2-minute units, it can generate image data in 1-minute units, then acquire frames with 2-minute intervals (not using the image data originally generated every 2 minutes), and use the acquired image data as the first label image data and the second image data. That is, the processor (190) can generate the first label image data and the second label image data corresponding to an injection amount equivalent to half of the reference amount. That is, the processor (190) can process the image to generate an image with a reduced drug injection amount using List mode data or apply an approximate method to generate image frames in 2-minute units by interleaving them in 1-minute units.
[0099] Depending on the implementation example, this can be performed by changing the order between steps S320 and S330.
[0100] FIG. 5 is a sequence diagram showing a method (S400) for predicting first image data representing anatomical information of a diagnosis subject according to the present disclosure and second image data representing disease-specific information.
[0101] At step S410, the processor (190) can acquire initial dynamic image data corresponding to the first section after the drug is injected into the diagnostic subject.
[0102] At step S420, the processor (190) inputs initial dynamic image data corresponding to a first section of the diagnostic subject into the learned first prediction model (EM1), thereby predicting first image data representing anatomical information (or blood flow information) for the diagnostic subject corresponding to the preceding time point.
[0103] Here, the first image data may be ultra-early image data, and the first period may be 10 to 30 minutes after drug injection, but the embodiment is not limited thereto. The preceding time point corresponds to the ultra-early image data, and may be within a preset range (e.g., 1 minute) after drug injection, but the embodiment is not limited thereto.
[0104] At step S430, the processor (190) can acquire initial dynamic image data corresponding to the second section after the drug is injected into the diagnostic subject.
[0105] The second period may be 10 to 20 minutes after drug injection, but the present disclosure is not limited thereto.
[0106] At step S440, the processor (190) inputs initial dynamic image data corresponding to the second section of the diagnostic subject into the learned second prediction model (EM2), thereby predicting second image data representing disease-specific information about the diagnostic subject corresponding to the reference time point.
[0107] In an embodiment, the processor (190) may inject a drug reduced by a preset amount into the diagnostic subject, and based on this, acquire and process initial dynamic image data corresponding to the first section and initial dynamic image data corresponding to the second section.
[0108] Here, the drug can be injected in a pre-set amount less than the reference amount, in order to reduce the injected radiation dose.
[0109] The processor (190) can process the initial dynamic image data corresponding to the first section and the initial dynamic image data corresponding to the second section before inputting them into the prediction model (EM1, EM2). This is to process the initial dynamic image data corresponding to the case where the injected radiation dose is less than the reference dose.
[0110] The processor (190) can process the movement path of the drug within the human body more vividly and clearly from the initial dynamic image data corresponding to the first section, and can process each pixel value into each pixel value when the reference amount is injected using a smoothing filter (e.g., a Gaussian filter). That is, the processor (190) can process each pixel value using a smoothing technique (e.g., a spatial smooting method).
[0111] In an embodiment, when the processor (190) inputs initial dynamic image data corresponding to a first section in which a small amount of a drug is injected into a processing model and initial dynamic image data corresponding to a second section, the processor (190) may output and use the initial dynamic image data corresponding to the first section in which a reference amount is injected and the initial dynamic image data corresponding to the second section through the processing model.
[0112] The processor (190) obtains initial dynamic image data corresponding to the processed first section and initial dynamic image data corresponding to the processed second section, and then inputs the initial dynamic image data corresponding to the processed first section into a first prediction model (EM1) to predict first image data representing corresponding anatomical information, and inputs the initial dynamic image data corresponding to the processed second section into a second prediction model (EM2) to predict second image data representing corresponding disease-specific information.
[0113] Fig. 6 is a sequence diagram illustrating a preprocessing process (including steps S500, S510, and S520) for performing spatial normalization according to the present disclosure. Fig. 7 is a diagram for explaining spatial normalization performed in a preprocessing step prior to learning an artificial neural network model according to the present disclosure.
[0114] At step S510, the processor (190) may perform spatial normalization before the initial dynamic image data corresponding to the first section of the learning object and the diagnosis object are input to the first prediction model (EM1) to enable machine learning using a smaller number of learning object data.
[0115] At step S520, the processor (190) can perform spatial normalization before the initial dynamic image data corresponding to the second section of the learning object and the diagnosis object is input to the second prediction model (EM2).
[0116] Referring to FIG. 7, the processor (190) may perform spatial normalization on the initial dynamic image data before inputting the initial dynamic image data corresponding to the first section of the learning subject and the diagnosis subject to the first prediction model (EM1) or before inputting the initial dynamic image data corresponding to the second section of the learning subject and the diagnosis subject to the second prediction model (EM2), so as to perform spatial normalization on the first image (1210), the second image (1220), and the third image (1230) based on the reference region.
[0117] For example, the processor (190) can spatially arrange the first image (1210) based on the reference center using a horizontal line (D1a) and a vertical line (D2a), spatially arrange the second image (1220) based on the reference center using a horizontal line (D1b) and a vertical line (D2a), and spatially arrange the third image (1230) based on the reference center using a horizontal line (D1a) and a vertical line (D2b). Accordingly, learning efficiency can be improved and diagnosis can be effective.
[0118] The processor (190) may set an acquisition time section of initial dynamic image data corresponding to a first section and / or a second section among the initial sections of the learning subject and the diagnosis subject in order to take into account the operability of the equipment and the predictability of the diagnosis subjects.
[0119] The processor (190) can identify an initial section when a drug is injected into a diagnostic subject, and acquire initial dynamic image data corresponding to the first section and the second section based on the identified initial section.
[0120] The processor (190) can set the time for acquiring initial dynamic image data corresponding to the first and second sections according to the characteristics of each individual, but the present disclosure is not limited thereto.
[0121] Figure 8 is a diagram showing changes in radiation dose over time after drug injection according to the present disclosure.
[0122] Referring to Figure 8, when FP-CIT is used as a drug (i.e., tracer), in the initial period of drug administration, the radiation dose may be high throughout the brain due to the influence of blood flow, making diagnosis difficult. However, as time passes, the drug is excreted in the urine or metabolized by the liver, and the concentration in the plasma decreases, and the drug in the blood can be confirmed to decrease. In addition, the drug flows out from the brain tissue into the blood to balance the decrease in the drug in the blood. The processor (190) can acquire initial dynamic image data in the initial period of change in the radiation dose in the brain tissue.
[0123] After that, since FP-CIT has a strong binding affinity with dopamine and is effective in differentiating Parkinson's disease, when FP-CIT binds to the dopamine neurotransmitter, which is the target area, the target area (640, 650) maintains a high concentration even if the concentration in other areas (610 to 630) decreases due to reduced blood flow. The processor (190) can determine the initial section of the initial dynamic image data based on the point at which the dose concentration in other areas (e.g., the reference area (610 to 630)) peaks.
[0124] The processor (190) can acquire delayed image data at a point in time when a reference time has elapsed after drug administration and the effect of blood flow has decreased. The delayed image data may be image data that is currently used by medical staff to diagnose the size, shape, etc. of a target area.
[0125] In an embodiment, if the drug used for imaging data acquisition is a tracer that binds to a specific target region, the initial interval for acquiring initial dynamic image data may be determined based on various criteria. For example, the initial interval may be set to the time range after the point in time when the influence of blood flow in the reference region begins to decrease. The reference region may be a specific region of the brain where there is generally no or little substance to which the tracer binds, and thus no difference between diseases. The reference region varies depending on the type and characteristics of the tracer. For example, if the tracer is FP-CIT, the reference region may be a region with low dopaminergic neurotransmitters, such as the cerebellum or the occipital cortex (OC). In another example, the initial interval may be set around the point in time when the dose ratio difference between the target region and the reference region exceeds a specific value or when the ratio difference reaches a maximum.
[0126] When the drug used for capturing image data is a tracer (e.g., Fluorodeoxyglucose (hereinafter, FDG) PET) whose amount continuously increases over time in the target area, the dose to each area of the brain increases without a peak point, so the processor (190) acquires initial dynamic image data in an initial section where the difference in radiation dose is not large, and acquires delayed image data at a reference point where the ratio of the difference in radiation dose between areas increases.
[0127] FIG. 9a is a graph for comparing delayed image data and label image data generated with a reference dose according to the present disclosure, FIG. 9b is a graph for comparing delayed image data and label image data generated with a low dose according to the present disclosure, and FIG. 9c is a graph for comparing delayed image data generated with a reference dose according to the present disclosure and delayed image data generated with a low dose.
[0128] Referring to Figure 9a, for the AP (Anterior Putamen, the front part of the basal ganglia), a correlation of 0.95 can be obtained by comparing the DAT (Dopamine Transporter) uptake between the delayed image data generated with the reference dose and the labeled image data, and for the PP (posterior putamen, the back part of the basal ganglia), a correlation of 0.96 can be obtained.
[0129] Referring to Figure 9b, for AP, a correlation of 0.93 can be obtained by comparing DAT uptake between delayed image data generated at low dose (half of the reference dose) and labeled image data, and for PP (posterior putamen, the posterior part of the basal ganglia), a correlation of 0.95 can be obtained.
[0130] Referring to Fig. 9c, for AP, a correlation of 0.99 can be calculated by comparing the DAT uptake between the delayed image data generated with the reference dose and the delayed image data generated with the low dose, and for PP, a correlation of 1.00 can be calculated between the delayed image data generated with the reference dose and the delayed image data generated with the low dose.
[0131] Referring to Figures 9a to 9c, since images are obtained in a state where attenuation due to half-life, which is a characteristic of nuclear medicine molecular imaging, is less, delayed uptake images (delay images) of the same quality can be predicted even through low-dose images with a reduced isotope injection amount.
[0132] FIG. 10a shows analog PET data used in the learning and testing stages for each of the AP (Anterior Putamen) and PP (Posterior Putamen) according to the present disclosure, and FIG. 10b shows digital PET data used in the testing stage for each of the AP and PP according to the present disclosure.
[0133] Referring to Fig. 10a, for analog PET data, the test set results can be confirmed to have a correlation of 0.95 for AP and a correlation of 0.96 for PP.
[0134] Referring to Fig. 10b, for the digital PET data, a correlation of 0.93 can be confirmed for the independent test set, for AP, and a correlation of 0.97 can be confirmed for PP.
[0135] FIG. 11 is a diagram illustrating a method for obtaining ultra-early image data according to the present disclosure.
[0136] Within the initial period, a correlation of 0.94 can be confirmed between the dynamic image data 10 minutes after drug injection and the very early image data (1010), a correlation of 0.77 can be confirmed between the dynamic image data 20 minutes after drug injection and the very early image data, and a correlation of 0.53 can be confirmed between the dynamic image data 30 minutes after drug injection and the very early image data. In this way, the highest correlation can be confirmed for the dynamic image data 10 minutes after drug injection, which is closest in time to the very early image data.
[0137] The processor (190) can predict the very early image data (1040) based on the dynamic image data of 10 to 30 minutes after the drug injection. That is, the prediction of the very early image data (1040) can be effective using the initial dynamic image data (e.g., the first section of the initial section).
[0138] FIG. 12 is a diagram for explaining a method for obtaining delay image data according to the present disclosure.
[0139] Within the initial period, a correlation of 0.80 can be confirmed between the dynamic image data 10 minutes after drug injection and the delayed image data (1110), and a correlation of 0.94 can be confirmed between the dynamic image data 20 minutes after drug injection and the delayed image data. In this way, the correlation can be confirmed to be the highest for the dynamic image data 20 minutes after drug injection, which is closest in time to the delayed image data.
[0140] The processor (190) can predict the delayed image data (1130) based on the dynamic image data of 10 to 20 minutes after the drug injection. That is, the prediction of the delayed image data (1130) can be effective using the initial dynamic image data (e.g., the second section of the initial section).
[0141] Meanwhile, for dynamic image data, the image quality of dynamic image frames can be set based on the number of frames required. That is, since the maximum amount of radiation emitted externally when using the same amount of drug is limited, thereby limiting the amount of signal used to generate image frames, a computing device can set the image quality of each image frame differently based on the number of image frames based on the same amount of signal.
[0142] For example, increasing the number of video frames reduces the amount of time required to generate each video frame, reducing the amount of signal available to generate one video frame, and causing the computing device to generate each video frame with lower quality.
[0143] In one embodiment, the computing device can learn by matching changes in each pixel in the dynamic image data over time with each pixel in the initial image data and the delayed image data.
[0144] That is, in a combination of image data for a specific patient (i.e., a combination of initial dynamic image data, dynamic image data, and delayed image data), the computing device predicts the initial dynamic image data and the delayed image data from the dynamic image data for each point (i.e., pixel) of a body tissue (e.g., brain tissue), builds a dataset for each image data combination by matching anatomical information in the initial dynamic image data with disease-specific information in the delayed image data, and learns datasets for multiple patients for each point (i.e., pixel) to build a diagnostic image prediction model.
[0145] In this embodiment, the diagnostic image prediction model is built using a deep neural network (DNN). That is, the diagnostic image prediction model applies a deep learning algorithm to learn dynamic image data for one or more patients.
[0146] The processor can be generated by normalizing the brightness of dynamic image data based on the maximum or average value.
[0147] In an embodiment, when a drug used for capturing image data is a tracer (e.g., Fluorodeoxyglucose (hereinafter, FDG) PET) whose amount continuously increases over time when bound to a target area, the brightness ratio of the reference area and the target area does not change depending on the drug dosage due to the characteristic of the PET tracer being ingested in proportion to the drug dosage, and therefore, the computing device generates image data by normalizing based on the maximum or average value of the brightness of the dynamic image data.
[0148] Meanwhile, although the present disclosure describes the operation using PET, the above-described operation can be applied not only to PET but also to MR and can be applied to medical imaging in general.
[0149] However, in the case of MR, disease-specific information can be derived from initial dynamic image data, and anatomical information can be derived from delayed image data.
[0150] When the drug used for capturing image data is a tracer that binds to a specific target area (when FP-CIT is used as a tracer), since the parameters of the tracer model are the same even if the input and output are linearly increased and decreased simultaneously in a linear tracer kinetic model that inputs the reference area and outputs the target area, the computing device can generate new delayed image data by normalizing the brightness of the initial dynamic image data for learning and the initial dynamic image data for diagnosis based on the maximum or average value (count normalization or intensity normalization).
[0151] That is, when the drug used for capturing image data is a tracer that binds to a specific target area, the dynamic image data for learning can be formed from image data acquired from the time when the dose ratio difference between the target area and the reference area exceeds a specific value or the ratio difference value becomes maximum until the time when the influence of blood flow is excluded.
[0152] Additionally, the learning dynamic image data can be formed as image data acquired from the point at which the difference in dose ratio between the target area and the reference area exceeds a specific value or the amount of dose change in the reference area due to blood flow decreases to the point at which the difference in dose ratio between the target area and the reference area reaches a maximum value.
[0153] The dose ratio value can mean the ratio of the radiation dose at the initial point in time to the radiation dose at a specific measurement point in time.
[0154] Therefore, the dose ratio difference value can mean the difference value between the dose ratio of the target area and the dose ratio of the reference area.
[0155] Specifically, the processor can obtain information about the dose of the target image and the reference image.
[0156] Additionally, the processor can obtain a dose corresponding to the target image and a dose corresponding to the reference, calculate a ratio value of each dose, and calculate a difference value of each dose ratio.
[0157] Additionally, the processor can determine when the dose variation of the reference decreases and when the dose ratio difference becomes maximum.
[0158] Typically, when medical professionals diagnose based on delayed imaging data, they administer a large dose of medication early on to ensure that the delayed imaging data captures sufficient radiation dose to perform a diagnosis even after the radioisotope's physical half-life and physiological attenuation (i.e., physiological attenuation) due to urination and defecation can be used. This approach poses the problem of increasing the radiation dose delivered to the patient.
[0159] Accordingly, by using one embodiment of the present invention, when acquiring diagnostic dynamic image data, only an amount of tracer capable of acquiring a sufficient radiation dose is inserted, and then the diagnostic dynamic image data is acquired, and the diagnostic dynamic image data is inserted into a prediction model to acquire initial dynamic image data and final diagnostic delayed image data, thereby reducing the amount of radioactive material injected into the patient, thereby reducing the radiation dose provided to the patient's body.
[0160] The predetermined reference time can be determined based on the type of tracer.
[0161] Meanwhile, the processor can form second learning data for predicting anatomical information and disease-specific information from the initial dynamic image data and the delayed image data based on a second diagnostic image prediction model using the initial dynamic image data and the delayed image data.
[0162] In summary, the first diagnosis prediction model predicts initial dynamic image data and delayed image data from dynamic image data and learns and predicts corresponding anatomical information and disease-specific information, while the second diagnosis prediction model learns based on initial dynamic image data and delayed image data corresponding to dynamic image data, and when initial dynamic image data or delayed image data for diagnosis are input later, it can predict anatomical information and disease-specific information in an integrated manner.
[0163] Meanwhile, the processor can determine the condition of the subject based on anatomical information and disease-specific information.
[0164] Specifically, the processor can derive information on whether the subject has Parkinson's disease or dementia by using the disease-specific information and anatomical information of the subject derived by the above-described method.
[0165] Specifically, if the tracer remains in the dopamine neurotransmitter of the subject, the subject can be judged to have Parkinson's disease.
[0166] Additionally, if tracers of amyloid remain, the subject can be judged to have dementia.
[0167] The processor can determine a diagnostic image prediction model having a higher accuracy between the accuracy of the first diagnostic image prediction model and the accuracy of the second diagnostic image prediction model.
[0168] For example, if a diagnostician performs a diagnosis using initial dynamic imaging data, the accuracy of the Parkinson's disease assessment for the first diagnostic prediction model and the accuracy of the second diagnostic prediction model can be calculated. In this case, the processor can select the diagnostic prediction model with the highest accuracy among the diagnostic prediction models to determine the subject's condition.
[0169] The method for generating a diagnostic image based on initial dynamic image data according to an embodiment of the present invention, which is described above in detail, such as the timing of acquiring an appropriate dynamic image, can be implemented as a program (or application) to be executed in combination with a computer as hardware and stored in a medium.
[0170] The method for generating a diagnostic image based on initial dynamic image data according to one embodiment of the present invention described above can be implemented as a program (or application) to be executed in conjunction with a computer as hardware and stored in a medium.
[0171] The above-described program may include codes coded in a computer language, such as C, C++, JAVA, or machine language, that can be read by the processor (CPU) of the computer through the device interface of the computer, so that the computer reads the program and executes the methods implemented as a program. Such codes may include functional codes related to functions that define functions necessary for executing the methods, and may include control codes related to execution procedures necessary for the processor of the computer to execute the functions according to a predetermined procedure. In addition, such codes may further include memory reference-related codes regarding which location (address address) of the internal or external memory of the computer should reference additional information or media necessary for the processor of the computer to execute the functions. In addition, if the processor of the computer needs to communicate with any other computer or server located remotely in order to execute the functions, the code may further include communication-related code regarding how to communicate with any other computer or server located remotely using the communication module of the computer, and what information or media to send and receive during communication.
[0172] The above storage medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, examples of the storage medium include, but are not limited to, ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. That is, the program can be stored in various recording media on various servers that the computer can access or in various recording media on the user's computer. In addition, the medium can be distributed across network-connected computer systems, so that computer-readable code can be stored in a distributed manner.
[0173] While the embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without altering the technical concept or essential features thereof. Therefore, the embodiments described above should be understood to be illustrative in all respects and not restrictive.
Claims
1. A method performed by at least one processor, A step of acquiring initial dynamic image data corresponding to an initial section up to a preset time point after a drug is injected into a learning subject; A step of learning a first prediction model that predicts first image data representing anatomical information or blood flow information for the learning subject corresponding to a time point before the first section by inputting initial dynamic image data corresponding to the first section among the initial sections; and A method comprising a step of learning a second prediction model that predicts second image data representing disease-specific information for the learning subject corresponding to a reference point after the initial period by inputting initial dynamic image data corresponding to the second period among the initial periods.
2. In paragraph 1, A step of acquiring initial dynamic image data corresponding to the first section after the drug is injected into the diagnostic subject; and A method further comprising a step of inputting initial dynamic image data corresponding to the first section of the diagnostic subject into the learned first prediction model to predict first image data representing anatomical information or blood flow information for the diagnostic subject corresponding to the previous time point.
3. In paragraph 2, A step of acquiring initial dynamic image data corresponding to the second section after the drug is injected into the diagnostic subject; and A method further comprising the step of inputting initial dynamic image data corresponding to the second section of the diagnostic subject into the learned second prediction model to predict second image data representing disease-specific information about the diagnostic subject corresponding to the reference time point.
4. In paragraph 3, The above processor, The drug, which has been reduced by a preset amount, is injected into the diagnostic subject, and based on this, initial dynamic image data corresponding to the first section and initial dynamic image data corresponding to the second section are acquired and processed, thereby acquiring initial dynamic image data corresponding to the processed first section and initial dynamic image data corresponding to the processed second section. A method configured to input initial dynamic image data corresponding to the processed first section into the first prediction model, and input initial dynamic image data corresponding to the processed second section into the second prediction model.
5. In paragraph 3, The amount of drug injected into the above learning subject is less than the standard amount, The first prediction model is learned based on first label image data corresponding to the first image data, and the second prediction model is learned based on second label image data corresponding to the second image data, A method wherein the first label image data and the second label image data are label image data processed according to the amount of drug injected into the learning subject.
6. In paragraph 3, A step of performing spatial normalization on the initial dynamic image data corresponding to the first section of the learning object and the diagnosis object before being input to the first prediction model; and A method further comprising the step of performing spatial normalization before the initial dynamic image data corresponding to the second section of the learning object and the diagnosis object is input to the second prediction model.
7. In paragraph 3, The above processor, Setting an acquisition time section of initial dynamic image data corresponding to a first section among the initial sections of the learning object and the diagnosis object, A method configured to set an acquisition time section of initial dynamic image data corresponding to a second section among the initial sections of the learning object and the diagnosis object.
8. In paragraph 3, The above processor, A method configured to identify the initial section when a drug is injected into the diagnostic subject, and acquire initial dynamic image data corresponding to the first section and the second section based on the identified initial section.
9. In paragraph 1, A method wherein the acquired initial dynamic image data, the first image data, and the second image data are positron emission tomography (PET) image data.
10. As a device for predicting the state of an object based on dynamic image data, memory; and at least one processor for performing communication with the memory, The above processor, A device configured to acquire initial dynamic image data corresponding to an initial section up to a preset time point after a drug is injected into a learning subject, learn a first prediction model that predicts first image data representing anatomical information or blood flow information about the learning subject corresponding to a time point before the first section by inputting the initial dynamic image data corresponding to a first section among the initial sections, and learn a second prediction model that predicts second image data representing disease-specific information about the learning subject corresponding to a reference time point after the initial section by inputting the initial dynamic image data corresponding to a second section among the initial sections.
11. In paragraph 10, The above processor, After the drug is injected into the diagnostic subject, initial dynamic image data corresponding to the first section is acquired, A device configured to input initial dynamic image data corresponding to the first section of the diagnostic subject into the learned first prediction model, thereby predicting first image data representing anatomical information or blood flow information for the diagnostic subject corresponding to the preceding time point.
12. In paragraph 11, The above processor, After the drug is injected into the diagnostic subject, initial dynamic image data corresponding to the second section is acquired, A device configured to input initial dynamic image data corresponding to the second section of the diagnostic subject into the learned second prediction model, thereby predicting second image data representing disease-specific information about the diagnostic subject corresponding to the reference time point.
13. In paragraph 12, The above processor, The drug, which has been reduced by a preset amount, is injected into the diagnostic subject, and based on this, initial dynamic image data corresponding to the first section and initial dynamic image data corresponding to the second section are acquired and processed, thereby acquiring initial dynamic image data corresponding to the processed first section and initial dynamic image data corresponding to the processed second section. A device configured to input initial dynamic image data corresponding to the processed first section into the first prediction model, and input initial dynamic image data corresponding to the processed second section into the second prediction model.
14. In paragraph 12, The amount of drug injected into the above learning subject is less than the standard amount, The first prediction model is learned based on first label image data corresponding to the first image data, The second prediction model is learned based on second label image data corresponding to the second image data, A device wherein the first label image data and the second label image data are label image data processed according to the amount of drug injected into the learning subject.
15. In paragraph 12, The above processor, Spatial normalization is performed before the initial dynamic image data corresponding to the first section of the above learning object and diagnosis object are input to the first prediction model, A device configured to perform spatial normalization before initial dynamic image data corresponding to the second section of the learning object and the diagnosis object is input to the second prediction model.
Citation Information
Patent Citations
Filter apparatus and odor reduction apparatus comprising the same
KR1020240080597A
Manufacturing Method of Aluminum Pouch Film For Secondary Battery
KR1020240139853A
Electronic device and method for wire and wireless charging in electronic device
KR102719820B1
Dynamic dual-tracer pet reconstruction method based on hybrid-loss 3D convolutional neural networks
US20200167974A1
KR20200055425A
Cited By
Joint positioning method and device for lesion boundary, medium and equipment
CN120918551A
