A radioactive drug distribution image generation system and method using deep learning

Abstract

The present invention relates to a radioactive drug distribution image generation system and method using deep learning, and more particularly, to a radioactive drug distribution image generation system and method using deep learning, in which dynamic medical images collected from multiple patients and time-radiation dose distribution curves of each organ can be learned through a deep learning network, and a spatial distribution image of a radioactive drug can be generated from static medical images obtained from a specific patient. According to the present invention, even if only medical images at a specific time are acquired after injecting a radioactive drug into a patient, a spatial distribution image of the radioactive drug over the entire time can be obtained by using a deep learning network, and based on this, by calculating the time-radiation dose distribution curve, quantitative analysis of the radioactive drug can be performed.

Description

Technical Field

[0001] The present disclosure relates to a system and method for generating a distribution image of a radiopharmaceutical based on deep learning, and more particularly, to a system and method for generating a distribution image of a radiopharmaceutical based on deep learning, wherein a deep learning network learns dynamic medical images collected from multiple patients and time-radiation dose distribution curves of organs, and generates a spatial distribution image of the radiopharmaceutical based on static medical images obtained from a specific patient. Background Art

[0002] Positron emission tomography (PET) and single photon emission computed tomography (SPECT) images are medical imaging devices that detect the energy emitted when labeled radionuclides decay, and they can be used as medical imaging tools for diagnosing patients and evaluating treatment responses. In particular, the advantage of PET and SPECT is that quantitative analysis can be performed based on the intensity of the images.

[0003] Research on quantitative analysis of PET- and SPECT-based images is divided into static medical image analysis in which a radiopharmaceutical for diagnosis or treatment is injected and the absorption at a specific time is analyzed, and dynamic medical image analysis in which the absorption rate of the radiopharmaceutical over time after injection of the radiopharmaceutical is quantitatively analyzed.

[0004] Dynamic medical image analysis includes a process of administering a radiopharmaceutical to a patient's body before acquiring a medical image, then continuously acquiring images while waiting until the radiopharmaceutical is sufficiently absorbed in the target area, and performing quantitative analysis based on a single acquired medical image. Representatively, the analysis using 18 F-FDG belongs to dynamic medical image analysis. 18 F-FDG PET imaging is performed by 18 intravenously injecting F-FDG into a patient's body and acquiring images for about 20 minutes after 1 hour.

[0005] Quantitative analysis based on dynamic medical images provides information about the change in the absorption rate of a radiopharmaceutical over time, which cannot be provided by static medical images because image acquisition starts simultaneously with the injection of the radiopharmaceutical and images are continuously acquired within a predetermined time period. However, unlike the acquisition of static medical images, the acquisition of dynamic medical images is very inconvenient for patients because the patients have to lie on the imaging device for a long time.

[0006] Therefore, analysis based on static medical images is less inconvenient for patients because medical images are acquired in a short time. However, the disadvantage of analysis based on static medical images is that it cannot reflect information about the change in the absorption rate of a radiopharmaceutical over time.

[0007] On the other hand, the analysis based on dynamic medical images can accurately evaluate information on the absorption rate of radiopharmaceuticals in the target area over time. However, the problem is that it becomes a heavy burden on patients because all medical images over time need to be acquired. Summary of the Invention

[0008] Technical Problem

[0009] One aspect of the present disclosure is to provide a system and method for generating a radiopharmaceutical distribution image based on deep learning, wherein a deep learning network learns dynamic medical images collected from multiple patients and time-radiation dose distribution curves according to organs, and generates a spatial distribution image of radiopharmaceuticals based on static medical images obtained from a specific patient.

[0010] Technical Solution

[0011] According to an embodiment of the present disclosure, a radiopharmaceutical distribution image generation system using deep learning can be provided. The system includes: a dynamic medical image acquisition unit configured to acquire dynamic medical images by continuously collecting medical images of a patient while injecting a radiopharmaceutical; a distribution curve acquisition unit configured to acquire a time-radiation dose distribution curve from the dynamic medical images, the time-radiation dose distribution curve representing the radiation dose of each organ of the human body over time; a static medical image acquisition unit configured to acquire static medical images for a specific period after injecting the radiopharmaceutical; a deep learning image generation network configured to predict and generate medical images corresponding to times before and after the static medical image by collecting and learning dynamic medical images of multiple patients and the corresponding time-radiation dose distribution curves; and a spatial distribution image acquisition unit configured to acquire a spatial distribution image of radiopharmaceuticals from the static medical images and the generated medical images.

[0012] Preferably, the deep learning network may include: a deep learning unit configured to collect and learn dynamic medical images of multiple patients and the corresponding time-radiation dose distribution curves; and an image generation unit configured to predict and generate medical images corresponding to times before and after taking the static medical image.

[0013] Preferably, the deep learning network may use a generative adversarial network (GAN) to generate medical images based on the dynamic medical images and the corresponding time-radiation dose distribution curves.

[0014] Preferably, the system may further include a distribution curve predictor configured to acquire a time-radiation dose distribution curve by obtaining the radiation dose and position of each region over time from the spatial distribution image.

[0015] According to an embodiment of the present disclosure, a method performed by a radioactive drug distribution image generation system using deep learning can be provided. The method includes: (a) acquiring dynamic medical images by continuously acquiring medical images of a patient while injecting a radioactive drug; (b) obtaining a time-radiation dose distribution curve from the dynamic medical images, the distribution curve representing the radiation dose of each organ of the human body over time; (c) acquiring dynamic medical images and corresponding time-radiation dose distribution curves of multiple patients and performing learning through a deep learning network; (d) obtaining static medical images for a specific period after injecting the radioactive drug; (e) predicting and generating medical images corresponding to the times before and after the static medical images through the deep learning network; and (f) obtaining a spatial distribution image of the radioactive drug from the static medical images and the generated medical images.

[0016] Advantageous Effects

[0017] According to the present disclosure, even if only medical images for a specific period are acquired after injecting a radioactive drug into a patient, a spatial distribution image of the radioactive drug over the entire time can be obtained through a deep learning network, thereby calculating a time-radiation dose distribution curve and enabling quantitative analysis of the radioactive drug. Brief Description of the Drawings

[0018] Figure 1 is a block diagram of a radioactive drug distribution image generation system using deep learning according to the present disclosure.

[0019] Figure 2 is a flowchart of a radioactive drug distribution image generation method using deep learning according to the present disclosure.

[0020] Figure 3 is a diagram showing a time-radiation dose distribution curve.

[0021] Figure 4 is a diagram showing generating an image from a static medical image each time and extracting a time-radiation dose distribution curve. Detailed Description of the Embodiments

[0022] The present disclosure can be modified in various ways and includes various embodiments. Therefore, specific embodiments will be shown in the drawings and described in detail below. However, it should be understood that the present disclosure is not limited to specific embodiments, but includes all variations, equivalents, and replacements within the technical spirit and scope of the present disclosure.

[0023] The terms "first", "second", etc. may be used to describe different elements, but these elements are not limited by these terms. These terms are only used for the purpose of distinguishing one element from another. For example, without departing from the scope of the present disclosure, a first element that is referred to as the first element in one embodiment may be referred to as the second element in another embodiment.

[0024] The terms used in the present disclosure are intended to describe only specific embodiments and not to limit the present disclosure. Unless the context clearly dictates otherwise, the singular forms are also intended to include the plural forms. It should be further understood that the terms "comprising", "having", etc. used in the present disclosure are intended to specify the presence of the stated features, integers, steps, operations, elements, components, or combinations thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

[0025] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0026] Terms defined in commonly used dictionaries should be interpreted as having a meaning that matches the meaning in the context of the relevant art, and should not be interpreted as ideal or overly formal unless otherwise clearly defined.

[0027] Hereinafter, embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. The same numbers always refer to the same elements.

[0028] Figure 1 is a block diagram of a radiopharmaceutical distribution image generation system using deep learning according to the present disclosure, Figure 2 is a flowchart of a radiopharmaceutical distribution image generation method using deep learning according to the present disclosure.

[0029] The radiopharmaceutical distribution image generation method using deep learning according to the present disclosure refers to and is substantially the same as the method performed by the radiopharmaceutical distribution image generation system using deep learning according to the present disclosure, and thus they will be described together hereinafter.

[0030] Referring to Figure 1 , the radiopharmaceutical distribution image generation system 100 using deep learning according to the present disclosure may include a dynamic medical image acquisition unit 110, a distribution curve acquisition unit 120, a deep learning network 130, a static medical image acquisition unit 140, and a spatial distribution image acquisition unit 150, and may additionally include a distribution curve predictor 160.

[0031] First, the dynamic medical image acquisition unit 110 acquires a dynamic medical image by continuously acquiring a patient's medical images while injecting a radioactive drug (S100). To perform spatial analysis on diagnostic medical images, a positron emission tomography-computed tomography (PET-CT) scanner can be used. The diagnostic medical images obtained by the PET-CT scanner can include PET images and CT images, and both images can be used in the analysis.

[0032] The distribution curve acquisition unit 120 acquires a time-radiation dose distribution curve from the dynamic medical image, and this distribution curve represents the radiation dose of each organ of the human body over time (S200).

[0033] The distribution curve acquisition unit 120 divides the diagnostic medical image based on the CT image in a predetermined manner, sets the boundary between the organ and the target area based on the anatomical information obtained through the CT image, and acquires the radiation dose and position of the radioactive isotope over time for each divided area based on the PET image.

[0034] Figure 3 is a graph showing the time-radiation dose distribution curve.

[0035] Reference Figure 3 to the graph, the time-radiation dose distribution curve shows the radiation dose over time measured in a specific organ or a specific area. In the graph of the time-radiation dose distribution curve, the x-axis represents time and the y-axis represents the measured radiation dose.

[0036] Return to reference Figure 1 and Figure 2 will continue to be described.

[0037] Finally, the distribution curve acquisition unit 120 acquires a diagnostic distribution curve based on the time-radiation dose distribution for each organ region of the human body from the dynamic medical image obtained after injecting the radioactive drug into the patient.

[0038] The dynamic medical image acquisition unit 110 and the distribution curve acquisition unit 120 acquire dynamic medical images of different patients and time-radiation dose distribution curves according to organs, and transmit the acquired dynamic medical images and the acquired time-radiation dose distribution curves according to organs to the deep learning network 130 (described later) for deep learning.

[0039] The deep learning network 130 includes a deep learning unit 131 and an image generation unit 132.

[0040] Here, deep learning is defined as a set of machine learning algorithms that attempt high-level abstractions through a combination of several non-linear transformation techniques and, in a broad sense, refers to the field of machine learning that trains computers in a human-like thinking manner. Machine learning enables computers to learn large amounts of data on their own like humans based on artificial neural networks (ANNs). Deep learning trains computers to distinguish and classify objects by mimicking the information processing method of the human brain, that is, by discovering patterns in large amounts of data and differentiating objects. Through deep learning, a computer can perform recognition, reasoning, and decision-making on its own without any criteria set by humans. Deep learning can be widely used, for example, for image recognition, analysis, classification, etc.

[0041] According to the present disclosure, after learning dynamic medical images collected from different patients, a generative adversarial network (GAN) algorithm can be used to generate medical images corresponding to times other than the time when static medical images are taken. Based on the competition between a model that generates seemingly real pseudo-images and a model that recognizes their authenticity, the GAN algorithm can generate the best medical images corresponding to times other than the time when static medical images are taken. In the process of deep learning based on the GAN algorithm, by learning dynamic medical images and simultaneously learning the time-radiation dose distribution curve according to the organ related to the dynamic medical images, the best medical images can be generated when considering the time-radiation dose distribution curve.

[0042] Finally, the GAN algorithm can obtain medical images corresponding to the entire time from static medical images just like taking dynamic medical images.

[0043] For this purpose, the deep learning unit 131 of the deep learning network 130 applies deep learning to the dynamic medical images collected from multiple patients and the corresponding time-radiation dose distribution curves through the deep learning network (S300). In this case, as described above, medical images can be generated through the GAN algorithm based on the dynamic medical images and the corresponding time-radiation dose distribution curves.

[0044] Meanwhile, the static medical image acquisition unit 140 acquires static medical images for a specific period after injecting a radioactive drug into a specific patient (S400). The acquired static medical images are sent to the deep learning network 130.

[0045] The image generation unit 132 of the deep learning network 130 predicts and generates medical images corresponding to the times before and after the static medical image is taken (S500). The image generation unit 132 of the deep learning network 130 is capable of predicting and generating medical images corresponding to times other than the time when the static medical image is taken, so as to generate medical images corresponding to the entire time just like taking dynamic medical images.

[0046] The spatial distribution image acquisition unit 150 acquires a spatial distribution image of a radiopharmaceutical from a static medical image and a medical image generated by the deep learning network 130 (S600).

[0047] Finally, the distribution curve predictor 160 acquires the radiation dose and position of each region over time from the acquired spatial distribution image, thereby obtaining a time-radiation dose distribution curve (S700). The difference is that the time-radiation dose distribution curve obtained in operation S200 is a distribution curve obtained from a dynamic medical image, but the time-radiation dose distribution curve obtained in operation S700 is a distribution curve obtained from a medical image generated at a time corresponding to a time other than the time when the static medical image was taken.

[0048] Figure 4 It is a diagram showing that an image is generated from a static medical image each time and a time-radiation dose distribution curve is extracted.

[0049] Reference Figure 4 , the first row shows the dynamic medical images taken at all times of interest, and the other three rows below show the static medical images taken only at specific times.

[0050] In the case of the dynamic medical image in the first row, the boundary between the organ and the target region is set based on the anatomical information obtained from the CT image, and the radiation dose and position of the radioactive isotope in each divided region over time are obtained based on the PET image, thereby finally obtaining the time-radiation dose distribution curve without difficulty.

[0051] On the other hand, in the case of the static medical images in the second to fourth rows, since the medical image only exists at a certain time, a time-radiation dose distribution curve corresponding to the entire time cannot be obtained.

[0052] In this case, the deep learning network 130 is used to generate medical images corresponding to all time points before and after the acquired image, and the generated medical images and the previously acquired static medical images are analyzed to predict the time-radiation dose distribution curve.

[0053] The present disclosure can be implemented by methods, apparatuses, systems, etc. The method of the present disclosure can also be implemented as a computer-readable program, code, application, software, etc. in a computer-readable recording medium. When implemented and executed by software or an application, the elements of the present disclosure are program or code segments for performing necessary tasks. The program or code segments can be stored in a computer-readable recording medium, or can be transmitted by a computer data signal combined with a carrier through a transmission medium or a communication network.

[0054] A computer-readable recording medium may include any type of recording device in which data readable by a computer system is stored. Examples of the computer-readable recording medium may include a hard disk, a floppy disk, a magnetic medium such as magnetic tape, an optical recording medium such as a compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a magneto-optical medium (such as a floppy disk), a random access memory (RAM), a read-only memory (ROM), a flash memory, or a similar hardware device specifically configured to store and execute program instructions. In addition, the computer-readable recording medium may also be distributed over network-coupled computer systems so that the computer-readable code is stored and executed in a distributed manner.

[0055] Although the present disclosure has been described above with reference to the embodiments depicted in the drawings, this description is for illustrative purposes only, and thus various changes and modifications can be made by those of ordinary skill in the art to which the present disclosure pertains. Therefore, the scope of the present disclosure is defined by the appended claims, and it should be understood that all technical ideas within the same and equivalent scope also fall within the scope of the present disclosure.

Claims

1. A radioactive drug distribution image generation system using deep learning, the system comprising: A dynamic medical image acquisition unit configured to acquire dynamic medical images by continuously collecting medical images of a patient while injecting a radioactive drug; A distribution curve acquisition unit configured to obtain a time-radiation dose distribution curve from the dynamic medical images, the time-radiation dose distribution curve representing the radiation dose of each organ of the human body over time; A static medical image acquisition unit configured to acquire static medical images for a specific period of time after injecting the radioactive drug; A deep learning network configured to predict and generate medical images corresponding to the times before and after the static medical image by collecting and learning the dynamic medical images of multiple patients and the corresponding time-radiation dose distribution curves; And A spatial distribution image acquisition unit configured to obtain a spatial distribution image of the radioactive drug from the static medical images and the generated medical images, wherein the deep learning network comprises: A deep learning unit configured to collect and learn the dynamic medical images of the multiple patients and the corresponding time-radiation dose distribution curves; and An image generation unit configured to predict and generate medical images corresponding to the times before and after the static medical image is taken.

2. The system according to claim 1, wherein, The deep learning network generates the medical images using a generative adversarial network (GAN) based on the dynamic medical images and the corresponding time-radiation dose distribution curves.

3. The system according to claim 1, further comprising a distribution curve predictor configured to obtain a time-radiation dose distribution curve by obtaining the radiation dose and position of each region over time from the spatial distribution image.

4. A method performed by a radioactive drug distribution image generation system using deep learning, the method comprising: (a) Acquiring dynamic medical images by continuously collecting medical images of a patient while injecting a radioactive drug; (b) Obtaining a time-radiation dose distribution curve from the dynamic medical images, the distribution curve representing the radiation dose of each organ of the human body over time; (c) Collecting the dynamic medical images of multiple patients and the corresponding time-radiation dose distribution curves and performing learning through a deep learning network; (d) Acquiring static medical images for a specific period of time after injecting the radioactive drug; (e) Predicting and generating medical images corresponding to the times before and after the static medical image through a deep learning network; And (f) Obtaining a spatial distribution image of the radioactive drug from the static medical images and the generated medical images, wherein the deep learning network comprises: A deep learning unit configured to collect and learn the dynamic medical images of the multiple patients and the corresponding time-radiation dose distribution curves; and An image generation unit configured to predict and generate medical images corresponding to the times before and after the static medical image is taken.

5. The method according to claim 4, wherein The deep learning network generates medical images using a generative adversarial network (GAN) based on the dynamic medical images and the corresponding time-radiation dose distribution curves.

Images