Method and device for providing information for cancer diagnosis by using extracellular vesicles

Abstract

This method utilizes machine learning on the basis of extracellular vesicles obtained through a liquid biopsy, so as to provide information necessary for efficient cancer diagnosis, and comprises: collecting learning data including surface characteristic images of extracellular vesicles extracted from cancer cells; using the learning data so as to construct an artificial intelligence diagnosis model; and deriving cancer-related disease information through the constructed artificial intelligence diagnosis model on the basis of surface characteristic data of extracellular vesicles to be diagnosed.

Description

Method and device for providing information for cancer diagnosis using extracellular vesicles

[0001] One embodiment of the present invention relates to a method and device for providing information for efficiently diagnosing cancer by utilizing machine learning based on a sample containing extracellular vesicles (EVs) obtained through liquid biopsy.

[0002] Cancer is a disease where early diagnosis and prompt treatment based on that diagnosis have a significant impact on survival rates. However, early diagnosis is difficult because cancer often has no early symptoms, and diagnostic methods vary depending on the type of cancer.

[0003] In particular, lung cancer is the cancer with the highest mortality rate worldwide as of 2024. Since patients rarely experience early symptoms, it is often diagnosed after the disease has progressed to a certain extent, so the survival rate is lower than that of other cancers, making early detection very important.

[0004] Lung cancer occurs when abnormal cells in the lungs proliferate uncontrollably, forming a mass and eventually leading to lung cancer. Lung cancer is classified into small cell lung cancer and non-small cell lung cancer (NSCLC) based on histological characteristics. Generally, more than 85% of lung cancers are NSCLC. Compared to small cell lung cancer, NSCLC grows relatively slowly and spreads to surrounding tissues before metastasizing throughout the body.

[0005] Existing lung cancer diagnosis methods include image analysis methods such as chest X-rays and computed tomography (CT), sputum cytology, tissue biopsies, and blood tests using single markers. However, image analysis methods have issues with radiation exposure and side effects from contrast agents. In addition, sputum cytology tests can be difficult to detect in the early stages of cancer because cancer cells may not be mixed in with sputum if they exist in the early stages or peripheral lung cancer. Tissue biopsies are risky, such as being invasive, and may not be possible depending on the location or size of the tumor. Furthermore, blood tests using single markers such as CEA (carcinoembryonic antigen) and Cyfra 21-1 (cytokeratin-19 fragments) have a high false positive error rate, limiting their use to only a limited number of purposes.

[0006] Likewise, research is steadily underway to overcome the limitations of difficult cancer diagnosis by leveraging AI machine learning technology. For example, Patent Publication No. 10-2020-0082660 discloses a method for diagnosing lung cancer by combining image analysis with AI machine learning technology to analyze images obtained from chest X-rays and computed tomography.

[0007] To address the limitations and problems of existing lung cancer diagnosis methods, there is a growing need for a method that combines information obtained through liquid biopsy using body fluids with artificial intelligence machine learning technology to diagnose lung cancer more safely, quickly, easily, conveniently, and accurately.

[0008] One embodiment of the present invention is intended to solve the problems of the above-mentioned prior art, and provides a method and device for providing information necessary for cancer diagnosis using an artificial intelligence-based diagnostic model that outputs disease information related to cancer of a subject based on features of surface characteristic images of extracellular vesicles (EVs).

[0009] However, the technical tasks to be achieved by the embodiments of the present invention are not limited to the technical tasks described above, and other technical tasks may exist.

[0010] The present disclosure proposes various solutions to solve the above-described technical problem, and according to one aspect, in a method for providing information necessary for cancer diagnosis using extracellular vesicles, each step is performed by an information providing device, and includes: (a) a step of collecting learning data including a surface characteristic image of an extracellular vesicle extracted from a cancer cell; (b) a step of constructing an artificial intelligence diagnostic model based on the learning data, which outputs disease information related to cancer of a subject of the extracellular vesicle surface characteristic data by using a feature associated with the surface characteristic image data of the extracellular vesicle to be diagnosed when the surface characteristic data of the extracellular vesicle to be diagnosed is input; and (c) a step of acquiring the surface characteristic data of the extracellular vesicle to be diagnosed and deriving the disease information related to cancer through the artificial intelligence diagnostic model.

[0011] Here, the step (a) may collect training data including surface characteristic images of extracellular vesicles extracted from non-small cell lung cancer cells having normal epidermal growth factor receptor (EGFR) but including KRAS mutation; surface characteristic images of extracellular vesicles extracted from non-small cell lung cancer cells having one genetically mutant epidermal growth factor receptor (EGFR); and surface characteristic images of extracellular vesicles extracted from non-small cell lung cancer cells having a mutant epidermal growth factor receptor (EGFR) with drug resistance.

[0012] Here, in the step (a), the surface characteristic image may include a topography image, a stiffness image, a Young's modulus image obtained from an atomic force microscope, and an image of the binding force between an antibody introduced into a probe of the atomic force microscope and a specific protein on the surface of an extracellular vesicle.

[0013] Here, the step (c) may output disease information related to the cancer, including at least one of disease information on non-small cell lung cancer of the subject having the normal epidermal growth factor receptor (EGFR) but including a KRAS mutation, disease information on non-small cell lung cancer of the subject having the one genetic mutation epidermal growth factor receptor (EGFR), and disease information on non-small cell lung cancer of the subject having the mutant epidermal growth factor receptor (EGFR) with drug resistance, through the artificial intelligence diagnostic model.

[0014] Here, the artificial intelligence diagnostic model may be an algorithm based on a convolutional neural network (CNN).

[0015] Here, the step of providing a class activate map indicating the basis for deriving disease information related to the cancer may be further included.

[0016] Here, an image processing process may further be included to further generate a processed surface characteristic image obtained by processing the surface characteristic image.

[0017] Here, the image processing process may be to obtain a first processed image with an extracellular vesicle boundary line added to the topographic image and a second processed image with an indication of a binding force greater than a predetermined standard added to the binding force image; and overlay the first processed image and the second processed image on the stiffness image or the Young's modulus image to generate the processed surface characteristic image.

[0018] Here, the step of constructing the artificial intelligence diagnostic model based on the learning data may be to input a 3D quantitative image, which is a 3D stack of the surface characteristic image and the processed surface characteristic image, into the artificial intelligence diagnostic model and learn it.

[0019] Here, the step of constructing the artificial intelligence diagnostic model based on the learning data may be such that the artificial intelligence diagnostic model is an algorithm based on a parallel convolutional neural network (CNN), and the surface characteristic image and the processed surface characteristic image are inputted into the artificial intelligence diagnostic model in parallel to learn.

[0020] In another aspect, the present disclosure provides a device that provides information necessary for cancer diagnosis using extracellular vesicles, the device including: a data collection unit that collects learning data including surface characteristic images of extracellular vesicles extracted from cancer cells; a machine learning unit that, when surface characteristic data of an extracellular vesicle to be diagnosed is input, constructs an artificial intelligence diagnostic model based on the learning data that outputs disease information related to cancer of a subject of the extracellular vesicle surface characteristic data by using features associated with the surface characteristic image data of the extracellular vesicle to be diagnosed; and a determination unit that obtains the surface characteristic data of the extracellular vesicle to be diagnosed and derives disease information related to cancer through the artificial intelligence diagnostic model.

[0021] In another aspect, the present disclosure may provide a computer program product including one or more computer-readable recording media storing a program that performs the steps of: collecting learning data including surface characteristic images of extracellular vesicles extracted from cancer cells; constructing an artificial intelligence diagnostic model based on the learning data, which outputs cancer-related disease information of a subject of the surface characteristic image data of the extracellular vesicles to be diagnosed by using features associated with the surface characteristic image data of the extracellular vesicles to be diagnosed when the surface characteristic data of the extracellular vesicles to be diagnosed is input; and obtaining the surface characteristic data of the extracellular vesicles to be diagnosed and deriving the cancer-related disease information through the artificial intelligence diagnostic model.

[0022] According to the aforementioned means for solving the problem of the present invention, information for diagnosing cancer is provided based on a sample obtained by separating extracellular vesicles (EVs) from a body fluid of a subject, thereby enabling non-invasive testing and requiring only a minimal amount of sample, thereby minimizing the burden on the subject.

[0023] According to the aforementioned means for solving the problem of the present invention, by applying an analysis method using an atomic force microscope image of a sample, analysis is possible at the single molecule level without optical or chemical molecular marking of the sample, thereby simplifying the inspection process and enabling rapid data processing.

[0024] According to the aforementioned means for solving the problem of the present invention, by applying an analysis method using an atomic force microscope image of a sample, quantitative analysis and location analysis of a specific protein that causes cancer on the surface of a sample are possible, thereby providing additional information necessary for cancer diagnosis.

[0025] According to the aforementioned means for solving the problem of this invention, by subdividing the diagnostic criteria for lung cancer into non-small cell lung cancer having a normal epidermal growth factor receptor (EGFR) but including a KRAS mutation, non-small cell lung cancer having a single genetically mutated epidermal growth factor receptor (EGFR), and non-small cell lung cancer having a mutated epidermal growth factor receptor (EGFR) with drug resistance, it is possible to assist in the diagnosis of lung cancer with high accuracy.

[0026] According to the above-described means for solving the problem of the present invention, when using an atomic force microscope image of a sample as learning data and test data, a machine learning model is constructed and applied to the processed surface characteristic image that indicates the boundary of the extracellular vesicle and the location, size, number, etc. of a specific protein related to lung cancer expression on the extracellular vesicle through an image processing process, thereby assisting in more accurate lung cancer diagnosis.

[0027] The aforementioned solution to this problem provides a class activation map (CAM) that illustrates the basis for deriving lung cancer-related diagnostic information, making it useful for lung cancer diagnostic research. However, the benefits achieved through this method are not limited to the aforementioned benefits, and other benefits may also exist.

[0028] Figure 1 is a schematic diagram illustrating a system for providing cancer diagnosis information using surface characteristic images of extracellular vesicles according to one embodiment of the present invention.

[0029] FIG. 2a and FIG. 2b are conceptual diagrams for explaining an artificial intelligence-based diagnostic model of a cancer diagnosis information providing device utilizing surface characteristic images of extracellular vesicles according to one embodiment.

[0030] Figure 3 is a drawing for explaining an image processing process for processing an image of the surface characteristics of extracellular vesicles.

[0031] FIGS. 4 and 5 are conceptual diagrams for explaining a process in which an information providing device according to one embodiment provides result data by an artificial intelligence-based diagnostic model.

[0032] FIGS. 6A to 6C are tables and graphs showing evaluation results of the discrimination performance of an artificial intelligence-based diagnostic model for lung cancer, as verification experiment examples related to an information providing device according to one embodiment of the present invention.

[0033] Figure 7 is a schematic diagram of a device for providing cancer diagnosis information using extracellular vesicle surface characteristic data according to one embodiment of the present invention.

[0034] Figure 8 is a schematic flowchart of a method for providing cancer diagnosis information using the surface characteristics of extracellular vesicles according to one embodiment.

[0035] Figure 9 is a schematic diagram illustrating extracellular vesicles extracted from normal lung cells and lung cancer cells.

[0036] Figure 10 is a conceptual diagram illustrating a probe of an image generating device.

[0037] Figure 11 is a diagram illustrating an image processing process for generating processed surface characteristic image data.

[0038] The present disclosure proposes various solutions to solve the above-described technical problem, and according to one aspect, in a method for providing information necessary for cancer diagnosis using extracellular vesicles, each step is performed by an information providing device, and includes: (a) a step of collecting learning data including a surface characteristic image of an extracellular vesicle extracted from a cancer cell; (b) a step of constructing an artificial intelligence diagnostic model based on the learning data, which outputs disease information related to cancer of a subject of the extracellular vesicle surface characteristic data by using a feature associated with the surface characteristic image data of the extracellular vesicle to be diagnosed when the surface characteristic data of the extracellular vesicle to be diagnosed is input; and (c) a step of acquiring the surface characteristic data of the extracellular vesicle to be diagnosed and deriving the disease information related to cancer through the artificial intelligence diagnostic model.

[0039] The advantages and features of the present invention, and the methods for achieving them, will become clearer with reference to the embodiments described in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments presented below, but may be implemented in various different forms, and it should be understood that it includes all transformations, equivalents, and substitutes included in the spirit and technical scope of the present invention. The embodiments presented below are provided 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. In describing the present invention, if a detailed description of a related known technology is judged to obscure the gist of the present invention, the detailed description thereof will be omitted.

[0040] The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present invention. The appearance of phrases such as "in some embodiments" or "in one embodiment" in various places in this specification does not necessarily all refer to the same embodiment. The singular expression includes plural expressions unless the context clearly indicates otherwise. In this application, it should be understood that the terms "comprises" or "has" specify the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but do not exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0041] Additionally, the connecting lines or connecting members between components depicted in the drawings are merely exemplary representations of functional connections and / or physical or circuit connections. In an actual device, connections between components may be represented by various functional connections, physical connections, or circuit connections that may be replaced or added.

[0042] These embodiments may be modified in various ways and may take on various forms. Therefore, some embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit these embodiments to a specific disclosed form, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and technical scope of these embodiments. The terminology used herein is solely for the purpose of describing the embodiments and is not intended to limit these embodiments.

[0043] Unless otherwise defined, the terms used in these examples have the same meaning as commonly understood by those of ordinary skill in the technical field to which these examples pertain. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense unless explicitly defined in these examples.

[0044] The following detailed description of the present invention refers to the accompanying drawings, which illustrate specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention, while different from each other, are not necessarily mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be modified and implemented from one embodiment to another without departing from the spirit and scope of the present invention. Furthermore, it should be understood that the positions or arrangements of individual components within each embodiment may also be modified without departing from the spirit and scope of the present invention. Accordingly, the following detailed description is not to be taken in a limiting sense, and the scope of the present invention is to be construed to encompass the scope of the claims and all equivalents thereof. Like reference numerals in the drawings represent the same or similar elements throughout the several aspects.

[0045] Hereinafter, various embodiments of the present invention will be described in detail with reference to the attached drawings so that a person having ordinary skill in the art to which the present invention pertains can easily practice the present invention.

[0046] The terms used in this specification are defined as follows.

[0047] In this specification, the subject may include a human or an animal.

[0048] In this specification, body fluids may be cell culture fluid, blood, urine, saliva, tears, semen, breast milk, ascites, etc.

[0049] In this specification, organs may include lungs, thyroid glands, stomachs, colons, breasts, prostate glands, livers, pancreas, gallbladder, kidneys, etc., and tissues may include nerves, muscles, blood vessels, etc.

[0050] In this specification, cancer is described based on lung cancer, but is not limited thereto and may encompass all types of cancer that may occur in the organs or tissues of the subject.

[0051] In this specification, a diagnostic model means a diagnostic assistance tool that can be used to assist a specialist in diagnosing a disease, and may be, for example, a diagnostic assistance model.

[0052] Figure 1 is a schematic diagram illustrating an information provision system according to one embodiment of the present invention.

[0053] Referring to FIG. 1, the information providing system may include an image generating device (100), a cancer diagnosis information providing device (200) (hereinafter, “information providing device”) utilizing surface characteristic images of extracellular vesicles (EVs), and a user terminal (300). In FIG. 1, the image generating device (100) and the information providing device (200) are illustrated as separate devices, but the present invention is not limited thereto, and the image generating device (100) and the information providing device (200) may be implemented as a single device. In addition, in FIG. 1, only components related to one embodiment of the present invention are illustrated, but it will be apparent to those skilled in the art that other general components may be further included in addition to the illustrated components.

[0054] An image generating device (100), an information providing device (200), and a user terminal (300) can communicate with each other through a network (10). The network (10) refers to a connection structure that enables information exchange between each node, such as terminals and servers, and examples of such a network (10) include, but are not limited to, a 3GPP (3rd Generation Partnership Project) network, an LTE (Long Term Evolution) network, a 5G network, a WIMAX (World Interoperability for Microwave Access) network, the Internet, a LAN (Local Area Network), a Wireless LAN (Wireless Local Area Network), a WAN (Wide Area Network), a PAN (Personal Area Network), a Wi-Fi network, a Bluetooth network, a satellite broadcasting network, an analog broadcasting network, a DMB (Digital Multimedia Broadcasting) network, etc.

[0055] The user terminal (300) may be any type of wireless communication device, such as, for example, a smartphone, a smart pad, a tablet PC, a PCS (Personal Communication System), a GSM (Global System for Mobile communication), a PDC (Personal Digital Cellular), a PHS (Personal Handyphone System), a PDA (Personal Digital Assistant), an IMT (International Mobile Telecommunication)-2000, a CDMA (Code Division Multiple Access)-2000, a W-CDMA (W-Code Division Multiple Access), or a Wibro (Wireless Broadband Internet) terminal.

[0056] In the description of the embodiments of the present invention, the surface characteristic data of the extracellular vesicles may be image data. In this regard, the image generation device (100) may be, but is not limited to, an atomic force microscope (AFM) that detects changes in the bending or vibration of a cantilever due to the interaction force acting between the probe and the surface when the probe approaches or contacts the sample surface, through a photodetector that detects a laser beam reflected from the cantilever, and generates an image related to the detected sample surface characteristic by a computing device.

[0057] Meanwhile, the image data may be multi-channel image data, and in one embodiment, includes a morphology image, a stiffness image, a Young's modulus image, and an adhesion image. The morphology image represents information measuring the surface height of the extracellular vesicle, and the unit is nm. The stiffness image represents stiffness, which is a change in bending that occurs when the probe presses the surface of the extracellular vesicle, and the unit is N / m. The Young's modulus image represents the elasticity of the surface of the extracellular vesicle, and the unit is Pa. The adhesion image represents the binding force between the antibody introduced into the probe and the epidermal growth factor receptor (EGFR) distributed on the surface of the extracellular vesicle, and the unit is pN. In one embodiment, the probe may be attached to an antibody for acquiring an adhesion image, and for example, the antibody may be an epidermal growth factor receptor (EGFR) antibody.

[0058] Hereinafter, with reference to FIGS. 2a and 2b, the specific functions and operations of the information providing device (200) will be described.

[0059] FIG. 2a and FIG. 2b are conceptual diagrams for explaining an artificial intelligence-based diagnostic model of a cancer diagnosis information providing device (200) utilizing surface characteristic images of extracellular vesicles according to one embodiment.

[0060] Referring to FIGS. 2A and 2B, when surface characteristic data of an extracellular vesicle to be diagnosed is input as input data, the information providing device (200) can operate to derive disease information related to cancer of the subject to be diagnosed as output data through an artificial intelligence-based diagnostic model constructed using learning data including an extracellular vesicle surface characteristic image of a normal group and an extracellular vesicle surface characteristic image of a lung cancer group.

[0061] In addition, the surface characteristic data of the extracellular vesicle to be diagnosed, which is authorized by the information providing device (200), includes a topographic image, a stiffness image, a Young's modulus image, and a bonding force image obtained from the image generating device (100) for the extracellular vesicle to be diagnosed, and may further include a processed surface characteristic image obtained by additionally processing at least two or more of the surface characteristic images. That is, the information providing device (200) uses a multi-channel image representing various surface characteristics of a single object and a processed image processed therefrom as input data.

[0062] In addition, the input data applied to the information providing device (200) may be input as a 3D quantitative image in which a topographic image, a stiffness image, a Young's modulus image, a bonding force image, and a processed surface characteristic image for a single object obtained from the image generating device (100) are stacked in 3D, as illustrated in FIG. 2A, but is not limited thereto. In this case, the artificial intelligence diagnostic model may include an algorithm based on a convolutional neural network (CNN). In another embodiment, the input data applied to the information providing device (200) may be input in parallel to each algorithm each image (topographic image, stiffness image, Young's modulus image, bonding force image, and processed surface characteristic image) for a single object, as illustrated in FIG. 2B. In this case, the artificial intelligence diagnostic model may include an algorithm based on a parallel convolutional neural network (Parallel-CNN). Here, the algorithm based on a parallel convolutional neural network means an algorithm in which multiple algorithms are arranged in parallel, each receiving multiple different images of the same object as input data, but the output data is a single judgment result for the object.

[0063] In addition, the output data of the information providing device (200) may be output as binary type data corresponding to information on the presence or absence of lung cancer (for example, output value 1 indicates a result in which the subject is determined to be a lung cancer patient, and output value 0 indicates a result in which the subject is determined to be a lung cancer patient), but is not limited thereto. In another embodiment, the information providing device (200) may determine that the subject is not a lung cancer patient but a normal patient (e.g., BEAS-2B) and determine that the subject is a lung cancer patient, and also, among the lung cancer patients, determine that the subject is a patient with non-small cell lung cancer having at least a normal epidermal growth factor receptor (EGFR), preferably a patient with non-small cell lung cancer having a normal epidermal growth factor receptor (EGFR) but including a KRAS mutation (e.g., the output value is A549), determine that the subject is a patient with non-small cell lung cancer having a single genetic mutation in epidermal growth factor receptor (EGFR) (e.g., the output value is PC9), or determine that the subject is a patient with non-small cell lung cancer having a drug-resistant mutation in epidermal growth factor receptor (EGFR) (e.g., the output value is PC9 / GR). The result information may be derived as output data.

[0064] In addition, the output data of the information providing device (200) may be provided together with the basis for deriving the determination result in the case where the subject is determined to be normal, or to have lung cancer, or to have a specific category of lung cancer. According to the embodiment of Fig. 2a, the information providing device (200) may provide the basis for deriving the determination result through one class activation map (CAMc) that displays the position information in a feature map stacked in a three-dimensional direction. According to the embodiment of Fig. 2b, the information providing device (200) may provide the basis for deriving the determination result through class activation maps (CAM1 to CAM5) the same number as the number of input data, which displays the position information in a feature map stacked in a three-dimensional direction for each of the input images (CH1 to CH5). In the case of the parallel convolutional neural network-based algorithm according to the embodiment of Fig. 2b, there is an effect that it is possible to confirm which specific feature of which input image is reflected in the determination result.

[0065] The information providing device (200) can prepare training data including surface characteristic images of extracellular vesicles extracted from cancer cells in order to build and learn an artificial intelligence-based diagnostic model. According to one embodiment, the information providing device (200) can collect a first surface characteristic image of extracellular vesicles from normal lung epithelial cells, a second surface characteristic image of extracellular vesicles extracted from non-small cell lung cancer cells having normal epidermal growth factor receptor (EGFR) but including a KRAS mutation, a third surface characteristic image of extracellular vesicles extracted from non-small cell lung cancer cells having one genetically mutant epidermal growth factor receptor (EGFR), and a fourth surface characteristic image of extracellular vesicles extracted from non-small cell lung cancer cells having a mutant epidermal growth factor receptor (EGFR) that has drug resistance. Here, the extracellular vesicles extracted from non-small cell lung cancer cells having a mutant epidermal growth factor receptor (EGFR) with drug resistance may be extracellular vesicles extracted from non-small cell lung cancer cells having two genetically mutant epidermal growth factor receptors (EGFR). Each of the first to fourth surface characteristic images may include a topographic image, a stiffness image, a Young's modulus image obtained from the image generating device (100), and an image of the binding force between an antibody introduced into the probe of the atomic force microscope and a specific protein on the surface of the extracellular vesicle. Meanwhile, the extracellular vesicles may include exosomes (about 50 to 200 nm in size), microvesicles (about 200 to 1,000 nm in size), apoptotic bodies (about 800 to 5,000 nm in size), etc., and according to one embodiment of the present invention, the extracellular vesicles of the surface characteristic images prepared as learning data may be exosomes.

[0066] Lung cancer can include small cell lung cancer and non-small cell lung cancer. Non-small cell lung cancer (NSCLC) comprises all types of epithelial lung cancer, excluding small cell lung cancer (SCLC), and accounts for a significant proportion, approximately 80% to 85%, of all lung cancers. NSCLC has various causes, but it is classified as a significant subtype of lung cancer caused by mutations in the epidermal growth factor receptor (EGFR), a transmembrane protein that acts as a receptor for members of the epidermal growth factor (EGF) family of extracellular protein ligands. This disease is more common in women than in men and is less associated with smoking. In addition, it is known that lung cancer caused by mutant EGFR is effective in primary treatment with EGFR inhibitors alone or in combination, unlike lung cancer caused by normal EGFR. In addition, the type of inhibitor or combination therapy with anticancer drugs or other agents should be prescribed differently depending on whether the mutant EGFR is caused by a single genetic mutation or a mutation that conferres drug resistance. Therefore, it is important to distinguish whether the non-small cell lung cancer is caused by mutant EGFR or a mutation in mutant EGFR that conferres drug resistance, as this greatly affects the cure rate and survival rate of the subject.Accordingly, the information providing device (200) of the present invention can strictly distinguish the surface characteristics of extracellular vesicles according to the presence or absence of a mutant epidermal growth factor receptor (EGFR) and the surface characteristics of extracellular vesicles according to whether the mutant epidermal growth factor receptor (EGFR) has a genetic mutation or a drug-resistant mutation. When constructing an artificial intelligence-based machine learning model, in addition to the first surface characteristic image of the extracellular vesicle from a normal lung epithelial cell, the second surface characteristic image of the extracellular vesicle extracted from a non-small cell lung cancer cell that has a normal epidermal growth factor receptor (EGFR) but a KRAS mutation, the third surface characteristic image of the extracellular vesicle extracted from a non-small cell lung cancer cell that has one genetic mutation of the epidermal growth factor receptor (EGFR), and the fourth surface characteristic image of the extracellular vesicle extracted from a non-small cell lung cancer cell that has a drug-resistant mutant epidermal growth factor receptor (EGFR) are classified and labeled and collected, and this is used as learning data.

[0067] In this regard, the information providing device (200) can label and distinguish the first to fourth surface characteristic images based on the diagnostic code information assigned to the surface characteristic images of extracellular vesicles collected as learning data, information on cultured cells from which the corresponding extracellular vesicles were extracted, and clinical content data of patients evaluated during cell culture. For example, the first surface characteristic image can be labeled as BEAS-2B, the second surface characteristic image can be labeled as A549, the third surface characteristic image can be labeled as PC9, and the fourth surface characteristic image can be labeled as PC9 / GR.

[0068] Figure 3 is a drawing for explaining an image processing process for processing an image of the surface characteristics of extracellular vesicles.

[0069] Referring to FIG. 3, the information providing device (200) can clarify the shape of the extracellular vesicle of interest through an image processing process, and determine quantitative information, location, and relationship with the extracellular vesicle of the epidermal growth factor receptor (EGFR), thereby helping to determine what kind of relationship the epidermal growth factor receptor (EGFR) has with the subtypes of lung cancer.

[0070] Specifically, the information providing device (200) can acquire a first processed image with an extracellular vesicle boundary line added from a topographic image, and a second processed image with an indication of a binding force above a predetermined standard added from a binding force image. In addition, the first processed image and the second processed image can be overlaid on a stiffness image or a Young's modulus image to generate the processed surface characteristic image.

[0071] FIG. 4 and FIG. 5 are conceptual diagrams for explaining a process in which an information providing device (200) according to one embodiment provides result data by an artificial intelligence-based diagnostic model.

[0072] The artificial intelligence-based diagnostic model of the information providing device (200) according to one embodiment may be an algorithm based on a convolutional neural network (CNN), but is not limited thereto, and it goes without saying that various artificial intelligence-based algorithms that are already known or may be developed in the future can be applied to the machine learning model disclosed herein.

[0073] In addition, according to one embodiment, optimization can be performed by applying the Adaptive Moment Estimation Optimizer (ADAM optimizer) to an analysis model of an algorithm type based on a convolutional neural network (CNN), and cross entropy can be used as a loss function, but is not limited thereto.

[0074] The AI-based diagnostic model disclosed herein is a two-dimensional (2-D) CNN, and can use a ResNet (Residual Network) model or a DenseNet (Dense Convolutional Network) model as a classification algorithm, but is not limited thereto. The diagnostic model can include a convolution layer, a pooling layer, and a fully-connected layer. The convolution layer can perform an inner product operation of input variables through each convolution kernel to output a feature map. At this time, each convolution kernel can extract information for a local receptive field of the extracellular vesicle surface characteristic data. This local feature extraction can identify features of the entire data. In addition, the local receptive field can be expanded by using multiple convolution layers and pooling layers.

[0075] The pooling layer of the diagnostic model receives the output data (feature map) of the convolutional layer as input and selects a representative value from the characteristics of the input data. The selection method and application area can be set, with selection methods such as the maximum or average. To pass the data to the fully connected layer corresponding to the classifier, the pooling layer's application area can be converted to a one-dimensional vector, with the entire input as the application area.

[0076] Next, the surface properties of extracellular vesicles acquired from the convolutional layer and / or pooling layer are transferred to a fully connected layer structure, and a classifier including a dropout layer and a fully connected layer in order to prevent overfitting outputs disease information including prediction results for normal and lung cancer by multiple categories.

[0077] When the process of constructing and learning the artificial intelligence-based machine learning model described above is completed, the information providing device (200) can acquire surface characteristic data of extracellular vesicles obtained from the body fluid of a specific diagnostic subject (subject). For example, the information providing device (200) can receive a surface characteristic image of an extracellular vesicle of the diagnostic subject from the image generating device (100). Here, the extracellular vesicle in the surface characteristic image may be an exosome.

[0078] In addition, the surface characteristic image of the extracellular vesicle of the diagnostic target may include a topographic image, a stiffness image, a Young's modulus image, and an image of the binding force between an antibody introduced into a probe of the atomic force microscope and a specific protein on the surface of the extracellular vesicle obtained from an atomic force microscope, and the information providing device (200) may additionally process at least two or more of the surface characteristic images obtained through an image processing process to further generate a processed surface characteristic image in which the boundary line of the extracellular vesicle and the epidermal growth factor receptor (EGFR) are indicated.

[0079] The input data thus obtained can be input as a 3D quantitative image, which is a 3D concatenation of a topographic image, a stiffness image, a Young's modulus image, and a bonding force image obtained from an image generating device (100), and a processed surface characteristic image, as illustrated in FIG. 4. In this case, the information providing device (200) of the present invention can input a 3D quantitative image of a diagnosis target as input data, and output a discrimination result and a class activation map. For example, the information providing device (200) inputs the input 3D quantitative image to a convolutional layer and / or a pooling layer to output feature maps, outputs a pooled feature map whose size is reduced for a single 3D feature map through the pooling layer, and outputs a discrimination result of the diagnosis target based on the pooled feature map and the weights of the classifier. In addition, the information providing device (200) can output a class activation map for the 3D quantitative image, and display the basis for inferring the discrimination result as location information of the feature map.

[0080] However, the present invention is not limited thereto, and each image may be input in parallel as illustrated in FIG. 5. In this case, the information providing device (200) of the present invention may take each image as input data and output a discrimination result and a class activation map. For example, the information providing device (200) may individually input each image into a parallel convolutional layer and / or a pooling layer to output respective feature maps, output pooled feature maps in which the size of each feature map is reduced through the pooling layer, and output a discrimination result of the diagnosis target by considering the features of all input images based on the weights in the classifier and the pooled feature maps corresponding to each input image. In addition, the information providing device (200) may output respective class activation maps for each image, and may visually display which surface properties among the topography, stiffness, Young's modulus, and bonding force of the extracellular vesicle of the diagnosis target, and which location of the image among the surface properties, has an influence on inferring the discrimination result.

[0081] The information providing device (200) can derive disease information related to lung cancer of a subject corresponding to the surface characteristic data of the extracellular vesicles to be diagnosed through the constructed artificial intelligence-based diagnostic model. Specifically, the information providing device (200) can derive disease information that determines that the subject is normal (BEAS-2B), determines that the subject is a patient with non-small cell lung cancer (A549) having normal epidermal growth factor receptor (EGFR) but including a KRAS mutation, determines that the subject is a patient with non-small cell lung cancer (PC9) having a single genetic mutation in epidermal growth factor receptor (EGFR), or determines that the subject is a patient with non-small cell lung cancer (PC9 / GR) having a drug-resistant mutation in epidermal growth factor receptor (EGFR) through the artificial intelligence-based machine learning model.

[0082] According to another embodiment, the information providing device (200) can provide, in addition to the discrimination result, the basis for inference from the discrimination result as additional information. The additional information can be expressed by visualizing the feature that is the basis of the discrimination result as location information (e.g., x, y) of a feature map, and the additional information can be provided as a class activation map (CAM) that displays the features for the input data. The class activation map can obtain information about which nanomechanical properties among the surface characteristic data of the extracellular vesicles were utilized for the classification result, and can explain which spatial information led to the corresponding discrimination result.

[0083] Referring to FIG. 4, the information providing device (200) provides n two-dimensional feature maps (F) which are CNN results of one three-dimensionally combined input data. i , i= 1,2,...,n) and the weights (w) between the diagnostic (classification) results and the pooled feature maps. i , i=1,2,...,n) and the result of multiplying and adding ( , where i = 1,2,...,n) can generate one class activation map. Here, the pooled feature map is a representative value extracted from the feature map, and can be, for example, the maximum or average value in the feature map.

[0084] In another embodiment, referring to FIG. 5, for each input data, n two-dimensional feature maps (F) which are CNN results k,i , k= 1,2,3,4,5, i= 1,2,...,n) and the weights (w) between the diagnostic (classification) results and the pooled feature maps k,i , k = 1,2,3,4,5, i=1,2,...,n) multiplied and added results ( , where i = 1,2,...,n) can be generated to generate k class activation maps. Here, the pooled feature map is a representative value extracted from the feature map, and can be, for example, the maximum or average value in the feature map.

[0085] FIGS. 6A to 6C are tables and graphs showing evaluation results of the discrimination performance of an artificial intelligence-based machine learning model for lung cancer, as verification experiment examples related to an information providing device (200) according to one embodiment of the present invention.

[0086] Referring to FIG. 6a, an information providing device (200) according to one embodiment exhibits an accuracy of 65% or more when using a ResNet model as a classification algorithm, and referring to FIG. 6b, it was confirmed that it has excellent discrimination performance, with an AUCROC (area under the Receive Operation Characteristic (ROC) curve) of 0.86 for normal, 0.97 for non-small cell lung cancer having normal epidermal growth factor receptor (EGFR) but including KRAS mutation, 0.85 for non-small cell lung cancer having one genetically mutated epidermal growth factor receptor (EGFR), and 0.84 for non-small cell lung cancer having a drug-resistant mutated epidermal growth factor receptor (EGFR). Referring again to FIG. 6a, the information providing device (200) according to one embodiment exhibits an accuracy of 79% or more when using the DenseNet model as a classification algorithm, and referring to FIG. 6c, it was confirmed to have excellent discrimination performance, with an AUCROC (area under the Receive Operation Characteristic (ROC) curve) of 0.86 for normal, 0.99 for non-small cell lung cancer having normal epidermal growth factor receptor (EGFR) but including KRAS mutation, 0.95 for non-small cell lung cancer having one genetically mutated epidermal growth factor receptor (EGFR), and 0.92 for non-small cell lung cancer having a drug-resistant mutated epidermal growth factor receptor (EGFR).

[0087] Figure 7 is a schematic diagram of a cancer diagnosis information providing device (200) using extracellular vesicle surface characteristic data according to one embodiment of the present invention.

[0088] Referring to FIG. 7, the information providing device (200) may include a data collection unit (210), an image processing unit (220), a machine learning unit (230), a determination unit (240), and a basis derivation unit (250). The information providing device (200) includes at least one processor. Accordingly, each component may be driven and its operations may be performed by at least one processor (not shown).

[0089] The data collection unit (210) can prepare training data including surface characteristic images of extracellular vesicles extracted from cancer cells. In detail, the data collection unit (210) can collect a first surface characteristic image of extracellular vesicles from normal lung epithelial cells, a second surface characteristic image of extracellular vesicles extracted from non-small cell lung cancer cells having normal epidermal growth factor receptor (EGFR) but including a KRAS mutation, a third surface characteristic image of extracellular vesicles extracted from non-small cell lung cancer cells having one genetically mutant epidermal growth factor receptor (EGFR), and a fourth surface characteristic image of extracellular vesicles extracted from non-small cell lung cancer cells having a mutant epidermal growth factor receptor (EGFR) that has drug resistance. Each of the first to fourth surface characteristic images may include a topography image, a stiffness image, a Young's modulus image, and an image of the binding force between an antibody introduced into the probe of the atomic force microscope and a specific protein on the surface of the extracellular vesicles obtained from the image generating device (100).

[0090] The image processing unit (220) processes the surface characteristic images collected as learning data through an image processing process to generate additional processed surface characteristic images. The image processing unit (220) obtains a first processed image with an extracellular vesicle boundary line added from a topographic image, obtains a second processed image with an indication of a binding force above a predetermined standard added from a binding force image, and then overlays the first processed image and the second processed image on a stiffness image or a Young's modulus image to generate a processed surface characteristic image.

[0091] When a characteristic image from a data collection unit (210) and a processed surface characteristic image from an image processing unit (220) are input as learning data, the machine learning unit (230) can construct an artificial intelligence-based machine learning model that outputs disease information related to lung cancer for a subject related to the surface characteristic data of an extracellular vesicle to be diagnosed that is input later, based on the collected learning data, using predetermined features linked to the input data.

[0092] The discrimination unit (240) can obtain surface characteristic data of the extracellular vesicles to be diagnosed, and derive disease information of the subject (subject) through the artificial intelligence machine learning model constructed by the machine learning unit (230). In addition to determining that the subject is normal and not lung cancer (e.g., BEAS-2B) and a lung cancer patient, the discrimination unit (240) can also determine that the subject is a non-small cell lung cancer patient with normal epidermal growth factor receptor (EGFR) but a KRAS mutation (e.g., the output value is A549), determine that the subject is a non-small cell lung cancer patient with a single genetic mutation epidermal growth factor receptor (EGFR) (e.g., the output value is PC9), or determine that the subject is a non-small cell lung cancer patient with a mutant epidermal growth factor receptor (EGFR) that has drug resistance (e.g., the output value is PC9 / GR).

[0093] The basis derivation unit (250) provides the basis for the discrimination derived from the discrimination unit (240) as a class activation map. The basis derivation unit (250) can use the feature map used for discrimination in the discrimination unit (240) to the surface characteristic data (image) of the extracellular vesicles to be diagnosed input to the machine learning unit (230), and generate class activation maps equal to the number of data (images) input to the discrimination unit (240).

[0094] Figure 8 is a schematic flowchart illustrating a method for providing cancer diagnostic information using the surface characteristics of extracellular vesicles according to one embodiment. Figure 9 is an explanatory diagram illustrating extracellular vesicles extracted from normal lung cells and lung cancer cells. Figure 10 is a conceptual diagram illustrating the probe of an image generation device (100).

[0095] The information providing method illustrated in FIG. 8 can be performed by the information providing device (200) described above, and can be equally applied to the information providing method even if there is omitted content compared to the information providing device (200).

[0096] In step 100, the information providing device (200) collects learning data including surface characteristic images of extracellular vesicles extracted from cancer cells.

[0097] Cancer cells can be prepared based on cells extracted from the body fluid or organ tissue of a lung cancer patient, and the extracellular vesicles extracted from the cancer cells can include, as shown in Fig. 9, extracellular vesicles of non-small cell lung cancer (A549) cells having normal epidermal growth factor receptor (EGFR) but including KRAS mutation, extracellular vesicles of non-small cell lung cancer (PC9) cells having one genetically mutant epidermal growth factor receptor (EGFR), and extracellular vesicles of non-small cell lung cancer (PC9 / GR) cells having a mutant epidermal growth factor receptor (EGFR) with drug resistance. In addition to cancer cells, the training data can also include surface characteristic images of extracellular vesicles extracted from lung epithelial cells (BEAS-2B) of a normal person without lung cancer. Meanwhile, the extracellular vesicles in the training data may be exosomes. Meanwhile, the drug here may be an anticancer agent that selectively inhibits the tyrosine kinase of the epidermal growth factor receptor (EGFR), for example, gefitinib. In addition, the extracellular vesicles of non-small cell lung cancer (PC9) cells with a single genetic mutation in the epidermal growth factor receptor (EGFR) have a mutation due to Exon 19 deletion, as shown in Fig. 9. In addition, the extracellular vesicles of non-small cell lung cancer (PC9 / GR) cells with a drug-resistant mutant epidermal growth factor receptor (EGFR) have both a mutation due to Exon 19 deletion and a T790M mutation, as shown in Fig. 9. Meanwhile, P in Fig. 9 represents phosphate, indicating that phosphorylation occurs due to the mutation.Extracellular vesicles with normal epidermal growth factor receptor (EGFR) undergo phosphorylation in response to signaling after epidermal growth factor (EGF) binding, but extracellular vesicles with mutant epidermal growth factor receptor (EGFR) continue to undergo signaling even when EGF is not bound.

[0098] As illustrated in FIG. 10, a surface characteristic image can be generated by an atomic force microscope (AFM) with an epidermal growth factor receptor (EGFR) antibody attached to a probe. In this case, the surface characteristic image can be a plurality of multi-channel images including a morphology image, a stiffness image, a Young's modulus image, and an adhesion image. According to one embodiment of the present invention, by utilizing the adhesion image, the presence and qualitative analysis of the epidermal growth factor receptor (EGFR) distributed in the extracellular vesicles can be performed. In addition, by utilizing the stiffness image and the Young's modulus image, it is possible to utilize the differences in the components of the lipid membrane of the extracellular vesicles depending on the presence or cause of lung cancer in the cell, and thus the differences in the stiffness and elasticity of the lipid membrane can be utilized in the diagnostic results. In addition, by removing errors in the stiffness image and utilizing the normalized Young's modulus image, more accurate identification results can be obtained.

[0099] [Example 1]

[0100] A sample for training data can be prepared by culturing each of a plurality of sample cells, filtering each cell culture medium, extracting extracellular vesicles for each sample cell from each filtered cell culture medium using ultra-high-speed centrifugation technology, and then spreading mica on one side of a plate member and attaching the extracted extracellular vesicles thereon. The sample for training data can be qualitatively evaluated through a biomarker profile. The qualitative evaluation can be performed using a western blot, and the detection of CD63 and CD9 can be used as biomarkers to determine extracellular vesicles. Next, the probe of an atomic force microscope is silanized with an aminosilane (e.g., silanated poly(ethylene glycol)s, silanated PEG5), and then an antibody that recognizes the epidermal growth factor receptor (EGFR) is attached using an amine-NHS (N-hydroxysuccinimide) coupling reaction. And using the probe attached to the antibody, the surface of the sample for learning data is divided into a predetermined size and sequentially scanned, and the probe can approach and retreat at about 20 um / s in a 64 x 64 pixel area with a nanometer resolution, and nanoindentation can be performed with a force of about 220 pN to 270 pN (preferably 250 pN). It may take about 2 to 3 minutes to scan a single extracellular vesicle. The atomic force microscope stores the spring constant and sensitivity of the probe before scanning, and converts the degree of laser deflection according to the bending of the probe during the scanning process into a force in piconewton (pN) units to determine the force-distance relationship according to the distance, acquires force-distance curve data, and generates various images having the surface characteristics of the sample based on this.

[0101] Again, in step 200 of FIG. 8, the information providing device (200) applies an image processing process to the acquired multi-channel image to obtain processed data. Specifically, a first processed image with an extracellular vesicle boundary line added is obtained from a topographic image, a second processed image with an indication of a binding force above a predetermined standard is obtained from a binding force image, and the first processed image and the second processed image are overlaid on a stiffness image or a Young's modulus image to generate processed surface characteristic image data.

[0102] Figure 11 is a diagram illustrating an image processing process for generating processed surface characteristic image data.

[0103] In step 210 of FIG. 11, the information providing device (200) obtains a first processed image. In detail, the information providing device (200) applies a 2D Gaussian filter to the terrain image to remove noise, and then obtains the first processed image by applying a threshold based on all pixels of the terrain image from which the noise has been removed. Here, the threshold can be calculated based on the height of the mica surface to which the extracellular vesicles are not attached. In detail, the threshold can be obtained by selecting the maximum height value among all pixels of the terrain image from which the noise has been removed and setting the height value to a predetermined percentage (for example, about 20% to 30%) lower than the selected maximum height value. Then, a boundary line of the extracellular vesicles is generated based on the obtained boundary value. Here, the boundary line can be derived by comparing the height value of the terrain image pixel and the boundary value while also considering the size of the pixel cluster. In detail, among the pixels of the topographic image, if the cluster size of the cluster is less than the reference size value (e.g., about 30 nm to 50 nm in diameter), it is not regarded as an extracellular vesicle and is excluded, and if the cluster size of the cluster of the pixels of the topographic image, if the cluster size is greater than the reference size value, it is regarded as an extracellular vesicle and a boundary line can be generated. Here, the reference size value can be determined as a size smaller than the minimum size classified as an exosome (e.g., 50 nm or more).

[0104] In step 220, the information providing device (200) obtains a second processed image, which is an avidity image indicating the location of the epidermal growth factor receptor (EGFR). Specifically, a filter is applied to the avidity image provided from the image generating device (100) to remove noise, and pixels having a binding value falling within a reference avidity value range are displayed based on the pixels of the entire avidity image from which noise has been removed. Here, the reference avidity value range may correspond to a binding value between an antibody and a protein (for example, about 30 pN to 120 pN) among the entire pixels of the avidity image. In addition, the reference binding distance value range in which the reference avidity appears may correspond to a binding distance value between an antibody and a protein (for example, about 3 nm to 40 nm) among the entire pixels of the avidity image. The pixels displayed here indicate the location of the epidermal growth factor receptor (EGFR). Since the binding force image is an image showing the binding force between the epidermal growth factor receptor (EGFR) antibody of the probe and the epidermal growth factor receptor (EGFR) present on the surface, by selecting and displaying the part with strong binding force, the location of the epidermal growth factor receptor (EGFR) present in the sample can be indicated.

[0105] In step 230, the information providing device (200) obtains a processed image, which is an overlay image. As previously described in FIG. 3, the information providing device (200) overlays the first processed image on the rigid image (or Young's modulus image). Due to the influence of the diameter of the probe, the area estimated to be the extracellular vesicles is displayed as a wider area in the rigid image (or Young's modulus image) compared to the topographic image. Therefore, by overlaying a boundary line on the rigid image (or Young's modulus image), the area of ​​the extracellular vesicles can be clearly distinguished. Next, the information providing device (200) obtains a processed surface characteristic image by overlaying a second processed image on the rigid image (or Young's modulus image) on which the first processed image is overlaid. As a result, the processed image can display the boundary line and the location of the epidermal growth factor receptor (EGFR) on the rigid image (or Young's modulus image).

[0106] Returning to FIG. 8 again, in step 300, the information providing device (200) builds an artificial intelligence-based machine learning model based on the collected learning data and processed data, which outputs disease information related to cancer of the subject of the extracellular vesicle surface characteristic data to be input later by using predetermined features associated with the surface characteristic data of the extracellular vesicles. According to one embodiment of the present invention, in addition to the topographic image, the stiffness image, the Young's modulus image, and the binding force image, a processed image in which the boundary line of the extracellular vesicle is introduced into the stiffness image or the elasticity image and the location of the epidermal growth factor receptor (EGFR) is indicated is utilized as learning data, thereby having the effect of distinguishing whether or not it is lung cancer through the stiffness or elasticity of the lipid membrane of the extracellular vesicle and accurately distinguishing whether or not the lung cancer is caused by the intervention of the epidermal growth factor receptor (EGFR).

[0107] In step 400, the information providing device (200) acquires surface characteristic data of the extracellular vesicles to be diagnosed. Since the surface characteristic data of the extracellular vesicles to be diagnosed can be acquired through liquid biopsy, it is quick and simple and can be utilized even when a tissue examination is not possible depending on the condition of the subject or the location or size of the tumor. The surface characteristic data of the extracellular vesicles to be diagnosed includes a multi-channel image, and includes, for example, a morphology image, a stiffness image, a Young's modulus image, and an adhesion image acquired from an image generating device (100) (atomic force microscope), as well as the processed surface characteristic image described above.

[0108] [Example 2]

[0109] The diagnostic sample is extracted from the subject's body fluid. The diagnostic sample is prepared by filtering the extracted body fluid, extracting extracellular vesicles from the filtered fluid using ultra-high-speed centrifugation technology, and then applying mica to one side of a plate member and attaching the extracted extracellular vesicles thereto. Next, the probe of the atomic force microscope is silanized with aminosilane (e.g., silanated poly(ethylene glycol)s, silanated PEG5), and an antibody that recognizes the epidermal growth factor receptor (EGFR) is attached to the probe using an amine-NHS coupling reaction. The surface of the sample to be diagnosed is sequentially scanned by dividing the surface into a predetermined size, and the probe approaches and retracts at about 20 um / s in a 64 x 64 pixel area with a nanometer resolution while nano-indenting with a force of about 220 pN to 270 pN (preferably 250 pN) to acquire force-distance curve data, and based on this, a multi-channel image having the surface characteristics of the sample to be diagnosed is created.

[0110] Returning to FIG. 8 again, in step 500, the information providing device (200) inputs surface characteristic data of the extracellular vesicles to be diagnosed as input data to the machine learning model constructed in the previous step to derive disease information related to lung cancer. The disease information may include a result determined to be normal (BEAS-2B) or a result determined to be lung cancer, and may include a result determined to be related to non-small cell lung cancer (A549) having normal epidermal growth factor receptor (EGFR) but including a KRAS mutation, related to non-small cell lung cancer (PC9) having a single genetically mutant epidermal growth factor receptor (EGFR), or related to non-small cell lung cancer (PC9 / GR) having a mutant epidermal growth factor receptor (EGFR) that is drug resistant.

[0111] The processors referred to in this specification may be implemented using at least one of application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors, and other electrical units for performing functions.

[0112] Various embodiments of the present disclosure may be implemented as software (e.g., a program) including one or more instructions stored on a machine-readable storage medium. For example, a processor of the machine may call at least one instruction among the one or more instructions stored from the storage medium and execute it. This enables the machine to operate to perform at least one function according to the at least one instruction called. The one or more instructions may include code generated by a compiler or code executable by an interpreter. The machine-readable storage medium may be provided in the form of a non-transitory storage medium. Here, 'non-transitory' only means that the storage medium is a tangible device and does not contain a signal (e.g., electromagnetic waves), and this term does not distinguish between cases where data is stored semi-permanently and cases where it is stored temporarily in the storage medium.

[0113] According to one embodiment, the method according to various embodiments of the present disclosure may be provided as included in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., compact disc read-only memory (CD-ROM)), or may be distributed online (e.g., downloaded or uploaded) through an application store (e.g., Play Store™) or directly between two user devices. In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily generated in a machine-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.

[0114] Additionally, in this specification, a “part” may be a hardware component such as a processor or circuit, and / or a software component executed by a hardware component such as a processor.

[0115] The scope of the present invention is indicated by the claims described below rather than the detailed description above, and should be interpreted to include all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts.

[0116] The present disclosure provides a method and device for providing information necessary for cancer diagnosis, and is applicable to the healthcare industry, including the medical diagnosis industry, the medical device manufacturing industry, health management and preventive services, data analysis, and artificial intelligence-based healthcare.

[0117]

Claims

1. In a method for providing information necessary for cancer diagnosis using extracellular vesicles, Each step is performed by an information provider, (a) a step of collecting learning data including surface characteristic images of extracellular vesicles extracted from cancer cells; (b) When surface characteristic data of the extracellular vesicle to be diagnosed is input, a step of constructing an artificial intelligence diagnostic model based on the learning data that outputs disease information related to cancer of the subject of the surface characteristic data of the extracellular vesicle to be diagnosed by using features associated with the surface characteristic image data of the extracellular vesicle to be diagnosed; and (c) a step of obtaining surface characteristic data of the extracellular vesicles to be diagnosed and deriving disease information related to the cancer through the artificial intelligence diagnostic model; A method comprising:

2. In paragraph 1, Step (a) above, Surface characteristic images of extracellular vesicles extracted from non-small cell lung cancer cells with normal epidermal growth factor receptor (EGFR) but containing KRAS mutations; Surface characteristic images of extracellular vesicles extracted from non-small cell lung cancer cells with a single genetic mutation in epidermal growth factor receptor (EGFR); and Surface characterization images of extracellular vesicles extracted from non-small cell lung cancer cells harboring a mutant epidermal growth factor receptor (EGFR) that is drug resistant; A method for collecting learning data including:

3. In paragraph 2, In step (a) above The above surface characteristic image is, A method comprising a topographic image, a stiffness image, a Young's modulus image obtained from an atomic force microscope, and an image of the binding force between an antibody introduced into a probe of the atomic force microscope and a specific protein on the surface of an extracellular vesicle.

4. In paragraph 2, Step (c) above, The above cancer is lung cancer, A method for outputting disease information related to the cancer, including at least one of disease information on non-small cell lung cancer having the normal epidermal growth factor receptor (EGFR) of the subject but including a KRAS mutation, disease information on non-small cell lung cancer having the one genetically mutated epidermal growth factor receptor (EGFR) of the subject, and disease information on non-small cell lung cancer having the mutated epidermal growth factor receptor (EGFR) with drug resistance, through the artificial intelligence diagnostic model.

5. In paragraph 1, The above artificial intelligence diagnostic model is a method based on a convolutional neural network (CNN).

6. In paragraph 5, A step of providing a class activate map indicating the basis for deriving disease information related to the above cancer; A method further comprising:

7. In paragraph 3, A method further comprising an image processing process for generating a processed surface characteristic image obtained by processing the surface characteristic image.

8. In paragraph 7, The above image processing process is, Obtaining a first processed image with an extracellular vesicle boundary line added to the above topographic image and a second processed image with an indication of a binding force greater than a predetermined standard added to the above binding force image; A method for generating the processed surface characteristic image by overlaying the first processed image and the second processed image on the stiffness image or the Young's modulus image.

9. In paragraph 7, The step of building the above artificial intelligence diagnostic model based on the above learning data is: A method for learning by inputting a 3D quantitative image obtained by stacking the surface characteristic image and the processed surface characteristic image into the artificial intelligence diagnostic model.

10. In paragraph 7, The step of building the above artificial intelligence diagnostic model based on the above learning data is: The above artificial intelligence diagnostic model is an algorithm based on a parallel convolutional neural network (CNN). A method for learning by inputting the surface characteristic image and the processed surface characteristic image in parallel into the artificial intelligence diagnosis model.

11. In a device that provides information necessary for cancer diagnosis using extracellular vesicles, A data collection unit for collecting learning data including surface characteristic images of extracellular vesicles extracted from cancer cells; When surface characteristic data of an extracellular vesicle to be diagnosed is input, a machine learning unit builds an artificial intelligence diagnostic model based on the learning data that outputs disease information related to cancer of the subject of the surface characteristic data of the extracellular vesicle to be diagnosed by using features associated with the surface characteristic image data of the extracellular vesicle to be diagnosed; and A determination unit that obtains surface characteristic data of the extracellular vesicles to be diagnosed and derives disease information related to the cancer through the artificial intelligence diagnosis model; A device comprising:

Images