Method for providing diagnosis result by utilizing in vitro diagnosis reading model based on deep learning, and apparatus therefor

Abstract

The method for providing a diagnosis result by utilizing an in vitro diagnosis reading model based on deep learning, by a device comprising a processor and memory, according to an embodiment of the present invention, comprises: (a) a first step for receiving a learning data set for training an in vitro diagnosis reading model; (b) a second step for training the in vitro diagnosis reading model by using the received learning data set; (c) a third step for receiving specimen data from a user terminal; and (d) a fourth step for outputting a diagnosis result by inputting the received specimen data to the in vitro diagnosis reading model for which training has been completed, wherein the learning data set comprises one or more pieces of voice learning data that are specimen data for a person who does not have a disease / illness to be diagnosed through the in vitro diagnosis reading model.

Description

Method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model and device therefor

[0001] The present invention relates to a method and device for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model. More specifically, the present invention relates to a method and device for using an in vitro diagnostic interpretation model to analyze sample data collected from a user in real time, thereby providing easy and accurate diagnosis results for specific diseases / conditions.

[0002] As artificial intelligence technology advances, its underlying machine learning / deep learning models are being widely applied in various fields, and the medical field, which handles medical data related to human health, is no exception.

[0003] Meanwhile, in order for machine learning / deep learning models to achieve excellent performance, a large amount of training data and a learning process using the data are essential. However, in the case of medical data, not only is it difficult to collect due to its nature, but positive data for cases corresponding to a specific disease / illness is rarer and harder to obtain than negative data for cases that do not correspond to the disease / illness. This characteristic is a major obstacle to the trend of actively utilizing artificial intelligence technology in the medical field.

[0004] Nevertheless, the application of AI technology to the medical field is expected to yield countless benefits, and the relevant industry is focusing on research and development in this area. The present invention relates to this topic.

[0005] The technical problem to be solved by the present invention is to provide a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model capable of effectively providing a diagnostic result for a specific disease / disease by effectively training a machine learning / deep learning model despite the difficulty in collecting positive data for a specific disease / disease, and a device therefor.

[0006] Another technical challenge to be solved by the present invention is to provide a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model, which can provide an accurate diagnostic result by fine-tuning the in vitro diagnostic reading model when positive data for a specific disease / illness that is difficult to collect during an actual service process is obtained after the primary learning of the in vitro diagnostic reading model, and a device therefor.

[0007] Another technical challenge that the present invention seeks to address is to provide a method and device for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model that can provide real-time diagnostic results for specific diseases / illnesses even to users who are in an environment where network connectivity is impossible or who have difficulty using devices, such as the elderly.

[0008] The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0009] In order to achieve the above technical problem, a method for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model by a device including a processor and a memory according to an embodiment of the present invention comprises (a) a first step of receiving a training data set for training the in vitro diagnostic reading model, (b) a second step of training the in vitro diagnostic reading model using the received training data set, (c) a third step of receiving specimen data from a user terminal, and (d) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which training has been completed to output a diagnosis result, wherein the training data set includes at least one negative training data which is specimen data for a person who does not have a disease / illness to be diagnosed through the in vitro diagnostic reading model.

[0010] According to one embodiment, the learning data set may further include one or more generated negative learning data, which are sample data generated by adjusting the concentration of a reagent for detecting the disease / disease within a predetermined range for sample data of a person who does not have the disease / disease to be diagnosed through the in vitro diagnostic reading model.

[0011] According to one embodiment, the second step may include a second-1 step of individually calculating the SE (Squared Error) of the input and output to the in vitro diagnostic reading model for each voice learning data included in the learning data set, and a second-2 step of setting the maximum value among the SEs for each of the calculated voice learning data as the first threshold value.

[0012] According to one embodiment, the fourth step may include a step 4-1 of calculating the SE of input and output to the in vitro diagnostic reading model for the received specimen data, a step 4-2 of determining whether the SE calculated in the step 4-1 exceeds the set first threshold value, and a step 4-3 of outputting the diagnosis result of the received specimen data as positive if it exceeds the threshold value, and as negative if it does not.

[0013] According to one embodiment, between the steps 4-2 and 4-3, if the calculated SE exceeds the first threshold value as a result of the determination of the step 4-2, a step 4-2.5 is further included to determine whether it exceeds a second threshold value within a predetermined range from the first threshold value. In this case, a positive diagnosis result of the step 4-3 may be output if the SE exceeds the second threshold value as a result of the determination of the step 4-2.5, and a negative diagnosis result may be output if the SE does not exceed the second threshold value as a result of the determination of the step 4-2.5.

[0014] According to one embodiment, if the diagnosis result is output as positive in the step 4-3, the method may further include a fifth step, which includes a step 5-1 of calculating the SE (Squared Error) of the input and output to the in vitro diagnostic reading model for the specimen data that output the diagnosis result as positive, and a step 5-2 of retraining the in vitro diagnostic reading model in a direction that maximizes the calculated SE for the specimen data that output the diagnosis result as positive.

[0015] According to one embodiment, between the steps 5-1 and 5-2, the method may further include a step 5-1.3 of sending the specimen data that outputs the diagnosis result as positive to a specialized medical institution server, receiving the diagnosis result for the specimen data from the specialized medical institution server, and a step 5-1.6 of labeling the specimen data with the diagnosis result-positive if the diagnosis result received from the specialized medical institution server is positive, and labeling the specimen data with the diagnosis result-negative if the diagnosis result received is negative.

[0016] According to one embodiment, after the fifth step, a sixth step of re-sharing the re-learned in vitro diagnostic reading model to the user terminal according to a request from the user terminal may be further included.

[0017] According to one embodiment, the second and third steps may include training the in vitro diagnostic reading model in a direction that minimizes the calculated SE for each of the above voice learning data.

[0018] According to one embodiment, between the second and third steps, a second and fifth step is further included for sharing the learned in vitro diagnostic reading model with the user terminal at the request of the user terminal, and the user terminal can download the learned in vitro diagnostic reading model and output a diagnostic result for the specimen data.

[0019] According to one embodiment, the user terminal has its own reading model other than the downloaded in vitro diagnostic reading model, and can firstly calculate a diagnostic result for the specimen data through the self-reading model, and if the calculated diagnostic result is positive, secondarily calculate a diagnostic result through the in vitro diagnostic reading model to output a final diagnostic result.

[0020] In order to achieve the above technical task, an apparatus for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to an embodiment of the present invention comprises one or more processors, a network interface, a memory for loading a computer program executed by the processor, and a storage for storing large-capacity network data and the computer program, wherein the computer program executes, by the one or more processors, (A) a first step of receiving a training data set for training the in vitro diagnostic reading model, (B) a second step of training the in vitro diagnostic reading model using the received training data set, (C) a third step of receiving specimen data from a user terminal, and (D) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which training has been completed and outputting a diagnostic result, wherein the training data set includes one or more negative training data, which is specimen data for a person who does not have a disease / illness to be diagnosed through the in vitro diagnostic reading model.

[0021] According to an embodiment of the present invention for achieving the above technical task, a computer program stored in a computer-readable medium executes (AA) a first step of receiving a training data set for training the in vitro diagnostic reading model, (BB) a second step of training the in vitro diagnostic reading model using the received training data set, (CC) a third step of receiving specimen data from a user terminal, and (DD) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which training has been completed and outputting a diagnosis result, wherein the training data set includes at least one negative training data which is specimen data for a person who does not have a disease / illness to be diagnosed through the in vitro diagnostic reading model.

[0022] According to the present invention as described above, an in vitro diagnostic reading model is trained using negative specimen data rather than positive specimen data, and positive specimen data is diagnosed using SE, so that despite the difficulty in collecting positive specimen data, there is an effect of being able to effectively provide accurate diagnosis results for a specific disease / illness.

[0023] In addition, in diagnosing positive specimen data, the positive specimen data is finally diagnosed through a two-stage verification process by setting not only the first threshold but also the second threshold, which has the effect of improving the reliability of the diagnosis result provision.

[0024] In addition, it provides an in vitro diagnostic reading model trained not only on the server but also on the in vitro diagnostic device corresponding to the user terminal, thereby implementing a small network. This has the effect of providing real-time diagnosis results for specific diseases / illnesses to users who are in an environment where network connection is not possible or who have difficulty using the device, such as the elderly.

[0025] Additionally, when positive specimen data is received during the service provision process, this is used to retrain the in vitro diagnostic interpretation model to fine-tune it. This has the effect of improving the accuracy of the diagnostic results as more users use the service.

[0026] In addition, before using the specimen data diagnosed as positive specimen data for retraining, the diagnosis result is verified by experts such as doctors, and whether or not to retrain is decided based on the result, so that the risk of any possible error in the diagnosis result can be fundamentally eliminated, and it can also have the effect of contributing to improving the performance of the in vitro diagnostic interpretation model.

[0027] The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

[0028] FIG. 1 is a drawing showing the overall configuration of a device that provides diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a first embodiment of the present invention.

[0029] FIG. 2 is a diagram illustrating the overall system structure for providing a method of providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0030] FIG. 3 is a flowchart showing representative steps of a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0031] FIG. 4 is a diagram exemplarily illustrating the structure of an in vitro diagnostic reading model in a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0032] FIG. 5 is a flowchart detailing the second step of training an in vitro diagnostic reading model in a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0033] FIG. 6 is a flowchart illustrating a first embodiment of a fourth step of outputting a diagnostic result in a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0034] FIG. 7 is a flowchart illustrating a second embodiment of the fourth step of outputting a diagnostic result in a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0035] FIG. 8 is a flowchart detailing the fifth step of retraining a deep learning reading model in a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0036] FIG. 9 is a flowchart illustrating a method for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention, in which a step of verifying a diagnosis result for specimen data is added from a specialized medical institution server.

[0037] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. The advantages and features of the present invention, and methods for achieving them, will become clear with reference to the embodiments described in detail below together with the attached drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

[0038] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in their common sense to those of ordinary skill in the art to which the present invention pertains. Furthermore, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0039] Meanwhile, the terms used in this specification are intended to describe embodiments and are not intended to limit the present invention. In this specification, singular forms also include plural forms, unless specifically stated otherwise.

[0040] As used herein, the terms “comprises” and / or “comprising” do not exclude the presence or addition of one or more other components, steps, operations and / or elements.

[0041] FIG. 1 is a drawing showing the overall configuration of a device (100) that provides diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a first embodiment of the present invention.

[0042] However, this is only a preferred embodiment for achieving the purpose of the present invention, and some components may be added or deleted as needed, and the role performed by one component may be performed by another component as well.

[0043] A device (100) for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to a first embodiment of the present invention may include a processor (10), a network interface (20), a memory (30), a storage (40), and a data bus (50) connecting them, and it will be understood that it may further include additional components required to achieve the purpose of the present invention.

[0044] The processor (10) controls the overall operation of each component. The processor (10) may be any one of a CPU (Central Processing Unit), an MPU (Micro Processor Unit), an MCU (Micro Controller Unit), or a processor of a type widely known in the technical field to which the present invention pertains, and may be implemented as an artificial intelligence model processor, such as a machine learning model processor or a deep learning model processor. In addition, the processor (10) may perform operations for at least one application or program for performing a method of providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention.

[0045] The network interface (20) supports wired and wireless Internet communication of the device (100) that provides diagnostic results using the deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention, and may also support other known communication methods. Accordingly, the network interface (20) may be configured to include a corresponding communication module.

[0046] The memory (30) stores various information, commands and / or information, and can load one or more computer programs (41) from the storage (40) to perform a method of providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention. In Fig. 1, RAM is illustrated as one of the memories (30), but it goes without saying that various storage media can be used as the memory (30).

[0047] Storage (40) can non-temporarily store one or more computer programs (41) and large-capacity network information (42). This storage (40) can be any one of non-volatile memory such as Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), flash memory, a hard disk, a removable disk, or any type of computer-readable recording medium widely known in the art to which the present invention pertains.

[0048] A computer program (41) is loaded into a memory (30) so that one or more processors (10) can execute (A) a first step of receiving a training data set for training the in vitro diagnostic reading model, (B) a second step of training the in vitro diagnostic reading model using the received training data set, (C) a third step of receiving specimen data from a user terminal, and (D) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which training has been completed and outputting a diagnosis result.

[0049] The operations performed by the computer program (41) briefly mentioned above can be viewed as a function of the computer program (41), and a more detailed description will be provided later in the description of a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention.

[0050] The data bus (50) serves as a path for transferring commands and / or information between the processor (10), network interface (20), memory (30), and storage (40) described above.

[0051] The device (100) for providing a diagnosis result by utilizing the in vitro diagnostic reading model based on deep learning according to the first embodiment of the present invention, which has been briefly described above, may be in the form of an independent device, for example, an electronic device or a server (including a cloud), and the electronic device may be not only a desktop PC or server device that is fixedly installed and used in one place, but also a portable device that is easy to carry, such as a smart phone, tablet PC, notebook PC, PDA, PMP, etc. Any electronic device that has a CPU corresponding to a processor (10) installed therein and only a network function may be used.

[0052] Hereinafter, assuming that the device (100) for providing a diagnostic result utilizing a deep learning-based in vitro diagnostic reading model according to a first embodiment of the present invention is in the form of a "server" among the electronic devices that are independent devices, a process for providing a method for providing a diagnostic result utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention through a dedicated application installed on a user terminal (200) of a user who wishes to receive a diagnostic result for a specific disease / illness described later will be described with reference to FIGS. 2 to 8.

[0053] FIG. 2 is a diagram illustrating the overall system structure for providing a method of providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0054] Referring to FIG. 2, the entire system for providing a method for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention may include a device (100) for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention corresponding to a server, an in vitro diagnostic (POCT, Point Of Care Testing) device corresponding to a user terminal (200), and in some cases, a server (300) of a specialized medical institution such as a hospital, and the method for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention relates to a service provided to a user against the background of the system illustrated in FIG. 3, in which the device (100) for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention becomes the subject of the service provision.

[0055] FIG. 3 is a flowchart showing representative steps of a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0056] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0057] In addition, each step is assumed to be performed by a device (100) that provides a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention, and the implementation form is a server form, and in the case of a user terminal (200), it may include not only an in vitro diagnostic device having a network function, for example, a professional device that can collect and analyze a specimen such as saliva, blood, urine, etc., but also a smartphone, a smart watch, smart glasses, a laptop computer, a tablet computer, a PDA, a PDP, a PMP, etc. depending on the type of biomarker collected or collected from the user. However, the following description will continue on the assumption that the user terminal (200) is an in vitro diagnostic device.

[0058] First, a learning data set for learning an in vitro diagnostic reading model is received (S310), which is referred to as the first step.

[0059] Here, the in vitro diagnostic reading model is an artificial intelligence deep learning model based on an auto encoder (AE) installed in the device (100), more specifically, a processor (10) included in the device (100). Details will be described later, and the first step can be viewed as a step of collecting learning data for learning the in vitro diagnostic reading model, which is an artificial intelligence deep learning model.

[0060] Here, the training data set may be a medical data set, and more specifically, may be a data set including one or more negative training data, which are various specimen data for people who do not have the disease / illness to be diagnosed through the in vitro diagnostic reading model.

[0061] In this case, the more specimen data there are for people who do not have the disease / disease, the more desirable it would be. However, due to the nature of medical data, which is difficult to collect, there is a high possibility that only data that is not sufficient for training an in vitro diagnostic interpretation model will be collected. Therefore, in the method for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic interpretation model according to the second embodiment of the present invention, a method for preparing a learning data set by augmenting the collected negative learning data is presented.

[0062] As previously explained, the learning data set is called negative learning data, which is various specimen data on people who do not have the disease / disease to be diagnosed. Since it is specimen data, a reagent is required for diagnosis. In the method for providing diagnostic results using a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention, new specimen data can be generated by applying this to adjust the concentration of the reagent for the disease / disease within a predetermined range, thereby enhancing the collected data, and this is called generative negative learning data.

[0063] Here, the reagent for adjusting the concentration can be selected from among various known reagents depending on the type of specimen and the type of disease / disease to be diagnosed, and by generating generative voice learning data only for the diagnosis of a specific type of disease / disease using only a specific type of specimen, training an in vitro diagnostic reading model specialized for the diagnosis of the specific type of disease / disease will be helpful in accurately producing the diagnosis results of the specific type of disease / disease, but in order to provide a general service such as providing diagnosis results for a number of diseases / diseases, it will also be possible to generate as much generative voice learning data as possible for the diagnosis of various types of specimens and various diseases / diseases.

[0064] By generating generative voice learning data as described above, even if only a small amount of sample data is collected, a large learning data set for learning an in vitro diagnostic reading model can be obtained. Accordingly, the reception in the first stage can be considered a broad concept that includes the reception of voice learning data and the generation of generative voice learning data that augments it.

[0065] If a learning data set is received, the device (100) trains an in vitro diagnostic reading model using the received learning data set (S320), which is referred to as the second step.

[0066] Before explaining the second step, let us explain the in vitro diagnostic interpretation model that has been previously explained.

[0067] FIG. 4 is a diagram exemplarily illustrating the structure of an in vitro diagnostic reading model in a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0068] The in vitro diagnostic reading model is a deep learning model based on an autoencoder. When it receives training data (input data), it extracts features through the encoder, and then restores the extracted features (output data) to be as similar as possible to the original training data through the decoder. It is a model that learns how to compress and then effectively restore input data through a type of lossy compression. Accordingly, the encoder is trained to extract important features that can well represent the input data, and the decoder is trained to well restore the original input data from the features.

[0069] In this case, there is bound to be a difference between the input data and the reconstructed output data, which is called the reconstruction error, and the loss function of the in vitro diagnostic reading model is trained to minimize the MSE (Mean Squared Error, the average of the squares of the differences between the input data and the output data), that is, to minimize the reconstruction error.

[0070] When such an in vitro diagnostic reading model is applied to the present invention, the input data, namely, voice learning data, more specifically, voice sample data, is received, features that can represent the data are extracted, data that is as similar as possible to the voice sample data is restored from the extracted features, and learning is performed in a direction that minimizes the difference between the restored data and the initially received voice sample data.

[0071] Based on the above in vitro diagnostic reading model, the second step of learning it will be explained with reference to Figure 5.

[0072] FIG. 5 is a flowchart detailing the second step of training an in vitro diagnostic reading model in a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0073] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0074] First, the device (100) individually calculates the SE (Squared Error) of the input and output to the in vitro diagnostic reading model for each voice learning data included in the learning data set (S320-1), and this is referred to as step 2-1.

[0075] Here, the voice learning data includes generative voice learning data, and the device (100) can individually calculate SE for all voice learning data included in the learning data set.

[0076] Thereafter, among the SEs for each voice learning data produced by the device (100), the maximum value is set as the first threshold value (S320-2), and this is called step 2-2.

[0077] A large SE means that the difference between the input data, which is the voice learning data, and the output data, which is the reconstruction data, output from the same is large. The device (100) sets the largest value among the SEs produced in the learning process as the first threshold value. This is to distinguish positive sample data when it is received, and a detailed explanation will be provided later.

[0078] The training of the in vitro diagnostic reading model using the learning data set proceeds in the direction of minimizing the SE calculated in step 2-1, and since the training was conducted only with negative sample data, the in vitro diagnostic reading model that has completed the training will be a reading model that tends to restore the negative pattern as much as possible regardless of the data received.

[0079] Let us return to the description of Figure 3.

[0080] If the in vitro diagnostic reading model has been trained, the device (100) receives specimen data from the user terminal (200) (S330), which is referred to as the third step.

[0081] The third step corresponds to a step in which a method of providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention is actually provided to a user, and accordingly, the specimen data received from the user terminal (200) may be said to be specimen data collected from a user who wishes to receive a diagnosis result for an actual disease / illness.

[0082] Meanwhile, the device (100) receiving the specimen data from the user terminal (200) means that the device (100) corresponding to the server uses the in vitro diagnostic reading model to produce a diagnostic result. However, in some cases, the network environment may be poor, making it impossible to connect to the device (100), or the elderly or other persons may have difficulty using the user terminal (200). In such cases, a step 2.5 (S325) may be performed to share the in vitro diagnostic reading model learned between the second and third steps with the user terminal (200) according to the request (200) of the user terminal in order to provide the diagnostic result. Accordingly, the user terminal (200) may download the learned in vitro diagnostic reading model and output a real-time diagnostic result for the specimen data.

[0083] In this case, each user terminal (200) may be viewed as forming a small network having an in vitro diagnostic reading model, and the user terminal (200) may first calculate a diagnostic result using its own reading model rather than the downloaded in vitro diagnostic reading model for the sample data collected from the user, and only when the calculated diagnostic result is positive may the user secondarily calculate a diagnostic result using the downloaded in vitro diagnostic reading model to output the final diagnostic result. In addition, the user terminal may also calculate and output a diagnostic result using the in vitro diagnostic reading model downloaded from the beginning.

[0084] If the specimen data is received, the device (100) inputs the received specimen data into the in vitro diagnostic reading model for which learning has been completed to output the diagnosis result (S340), which is referred to as the fourth step.

[0085] The diagnostic results output here are the results of whether the user corresponds to a specific disease / illness, and are fundamentally identical to the output method of the in vitro diagnostic reading model described in Step 2 above. However, Step 2 is about learning and only deals with negative learning data, but in Step 3, positive specimen data may be received, so it is necessary to review how the diagnostic results are calculated and output in such cases. This will be explained below with reference to Figure 6.

[0086] FIG. 6 is a flowchart illustrating a first embodiment of a fourth step of outputting a diagnostic result in a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0087] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0088] First, the SE of the input and output to the in vitro diagnostic reading model for the specimen data received by the device (100) is calculated (S340-1), and this is referred to as step 4-1. Then, it is determined whether the SE calculated in step 4-1 exceeds the set first threshold value (S340-2), and this is referred to as step 4-2.

[0089] As mentioned in the explanation of Step 2-2 above, the maximum value among the SEs calculated for each voice learning data is set as the first threshold. This involves calculating the SE of the input and output of the sample data collected from the user and comparing it with the first threshold.

[0090] As a result of the judgment in step 4-2, if the SE calculated in step 4-1 exceeds the first threshold, the device (100) outputs the diagnosis result of the sample data as positive (S340-3-1), and if it does not exceed the threshold, it outputs the result as negative (S340-3-2), which is called step 4-3.

[0091] Since the in vitro diagnostic reading model is trained only with negative learning data, it tends to restore any data it receives to a negative pattern as much as possible. A large SE output means a large difference between the input data and the output data, which means that restoration to a negative pattern is not performed properly. In other words, it means that the features extracted from the input data do not correspond to the features for negative sample data, which means that the sample data may correspond to positive sample data rather than negative sample data.

[0092] However, simply because the calculated SE is large does not mean that the corresponding specimen data can be immediately diagnosed as positive specimen data. Therefore, in the method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention, the largest SE set in the learning process is set as the first threshold, and the positive specimen data is diagnosed only when it is compared to the first threshold and exceeds it. In such a case, it is natural that the learning data set should cover a large amount of negative learning data in order to produce an accurate diagnostic result.

[0093] Meanwhile, while a diagnosis result may be provided as is through comparison with the first threshold, the learning data set may not cover a large amount of voice learning data sets. Therefore, in order to provide accurate diagnosis results even in such cases, a method of providing diagnosis results utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention may set a second threshold. This will be described below with reference to FIG. 7.

[0094] FIG. 7 is a flowchart illustrating a second embodiment of the fourth step of outputting a diagnostic result in a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0095] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0096] In the second embodiment of the fourth step, steps 4-1 and 4-2 of the first embodiment are the same, and detailed descriptions are omitted to avoid redundant description.

[0097] After step 4-2, if the SE calculated by the device (100) as a result of the judgment of step 4-2 exceeds the first threshold, it is determined whether it exceeds the second threshold within a predetermined range from the first threshold (S340-2.5), and this is called step 4-2.5. In this case, a positive diagnosis result in step 4-3 is output when the result of the judgment of step 4-2.5 exceeds the second threshold, and a negative diagnosis result is output when the result of the judgment of step 4-2.5 does not exceed the second threshold.

[0098] The significance of this second embodiment of the fourth stage lies in the fact that it sets the second threshold value further than the first threshold value, thereby improving the accuracy of the diagnostic result output by performing two verification procedures.

[0099] Here, the second threshold value can basically be set by a person with certain authority, such as an administrator of the device (100), but in order to prevent a situation where the diagnosis result is divided into negative or positive due to a slight difference from the first threshold value due to the special nature of medical data, the second threshold value can be set to any one of 10% to 50% of the first threshold value based on the first threshold value. For example, if the first threshold value is 10, the second threshold value can be set to any one of 1, which is 10% of 10, to 5, which is 50% of 10, for example, 2, so that a positive diagnosis result can be output only when the SE exceeds 12.

[0100] Separately, the second threshold value can be set according to the statistical probability distribution of the first threshold value. Assuming that the first threshold value follows the T distribution from the first learning stage in the second stage up to the case where the number of relearning times described later is 30, the second threshold value can be set to a value (p-value) corresponding to the top 95% that the first threshold value can have. If it exceeds 30 times, assuming that the first threshold value follows a normal distribution, the second threshold value can be set to a value (p-value) corresponding to the top 95% that the first threshold value can have.

[0101] Meanwhile, a method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention may further include a fifth step (S350) of fine-tuning, i.e., retraining, the in vitro diagnostic reading model after the first and second embodiments of the fourth step, which will be described below with reference to FIG. 8.

[0102] FIG. 8 is a flowchart detailing the fifth step of retraining a deep learning reading model in a method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention.

[0103] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0104] Step 5 is a step performed only when the diagnosis result in Step 4-3 is positive. After Step 4, the device (100) calculates the SE of input and output to the in vitro diagnostic reading model for the specimen data that output the diagnosis result as positive (S350-1), and this is referred to as Step 5-1. The in vitro diagnostic reading model is retrained in a direction that maximizes the SE calculated for the specimen data that output the diagnosis result as positive (S350-2), and this is referred to as Step 5-2.

[0105] Here, step 5-1 can be used by loading it as it is because the SE for the corresponding sample data was calculated once in step 4-1, and in this case, it can be omitted. However, since it is about retraining the in vitro diagnostic reading model, it would be desirable to recalculate the SE for the corresponding sample.

[0106] Meanwhile, in step 5-2, the in vitro diagnostic reading model is retrained in a direction that maximizes SE, which is different from the learning direction of the in vitro diagnostic reading model described in step 2. The reason for retraining the in vitro diagnostic reading model in a direction that maximizes SE is that the data to be retrained is positive specimen data, and since the in vitro diagnostic reading model has been trained only with negative learning data, it tries to restore the negative pattern as much as possible. Therefore, for the positive specimen data that is rarely received, the SE must be calculated as large as possible to enable effective and accurate diagnosis.

[0107] Retraining of the above in vitro diagnostic reading model may not be performed quickly because the target data is positive specimen data. If various users receive a service that provides diagnostic results by utilizing the deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention, positive specimen data will also be received with a higher probability. This can be understood as a concept similar to a performance update that can occur naturally during the service process.

[0108] Meanwhile, a method for providing a diagnosis result by utilizing a deep learning-based in vitro diagnostic reading model according to a second embodiment of the present invention is as illustrated in FIG. 9, in steps 5-1 and 5-2, the device (100) transmits the specimen data that outputs a positive diagnosis result to a specialized medical institution server (300), and the diagnosis result for the specimen data can be received from the specialized medical institution server (300) (S350-1.3), which is referred to as step 5-1.3, determines whether the diagnosis result received from the specialized medical institution server (300) is positive (S350-1.5), and if positive, labels the specimen data with a diagnosis result of positive and performs re-learning (S350-1.6-1), and if the received diagnosis result is negative, labels the specimen data with a diagnosis result of negative (S350-1.6-2), which is referred to as step 5-1.6.

[0109] In the method of providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention, since the in vitro diagnostic reading model is sufficiently trained using a learning data set in the second step, there is not much need to doubt or separately verify the accuracy of the provision of the diagnostic result. However, since retraining the in vitro diagnostic reading model using positive specimen data is to fine-tune the in vitro diagnostic reading model that has been trained using negative training data post hoc, there may be a need to confirm whether the positive specimen data used for retraining is true positive specimen data.

[0110] To this end, the device (100) transmits the specimen data output as positive specimen data to a specialized medical institution server (300), allowing experts such as doctors to individually diagnose whether the specimen data is positive or not, and then label the results to determine whether to use them for retraining. This eliminates any potential risk of diagnostic error, and can also contribute to improving the performance of the in vitro diagnostic interpretation model.

[0111] Meanwhile, retraining of the in vitro diagnostic reading model can be performed not only when the specimen data received from the user terminal (200) is positive specimen data, but also when it is negative specimen data. This means that all specimen data collected throughout the entire process of the user using the device (100) to receive the diagnostic result provision service can be used for retraining the in vitro diagnostic reading model, and there is a difference in that in the case of positive specimen data, the SE is maximized, and in the case of negative specimen data, the SE is minimized.

[0112] More specifically, in the case of negative sample data, the first threshold value set in step 2-2 can be reset during the relearning process because the maximum value of SE calculated during the relearning process changes, and the reset of the first threshold value can be set to a weighted average of the maximum SE proportional to the number of negative sample data, rather than simply changing the existing first threshold value by the maximum value of SE calculated during the relearning process.

[0113] For example, in the second stage, when the in vitro diagnostic reading model is trained using 100 negative learning data, and the first threshold value, the maximum SE, is 10, when 100 additional negative specimen data are collected during the diagnostic result provision service process, the maximum SE can be calculated as 9 during the retraining process for a total of 200 negative specimen data, and when 100 additional negative specimen data are collected again, the maximum SE can be calculated as 12 during the retraining process for a total of 300 negative specimen data. In this case, the first threshold value can be reset to the weighted average of the maximum SE, which is (10 * 100 / 600) + (9 * 200 / 600) + (12 * 300 / 600) = 10.67.

[0114] The above example illustrates resetting the first threshold after two retraining cycles. However, it should be noted that the first threshold could be reset after each retraining cycle. For example, after a single retraining cycle, the first threshold could be immediately reset to (10 * 100 / 300) + (9 * 200 / 300) = 9.33.

[0115] So far, a method for providing a diagnostic result utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention has been described. According to the present invention, the in vitro diagnostic reading model is trained using negative specimen data rather than positive specimen data, and positive specimen data is diagnosed using SE, so that accurate diagnostic results for a specific disease / disease can be effectively provided despite the difficulty in collecting positive specimen data. Furthermore, in diagnosing positive specimen data, not only a first threshold but also a second threshold is set, and the positive specimen data is finally diagnosed through a two-stage verification process, so that the reliability of the diagnostic result provision can be improved. In addition, the trained in vitro diagnostic reading model is provided not only for the device (100) but also for the in vitro diagnostic device corresponding to the user terminal (200), thereby implementing a small-scale network. Therefore, even users who are in an environment where network connection is not possible or who have difficulty using the device, such as the elderly, can receive a diagnostic result for a specific disease / disease in real time. Additionally, if positive specimen data is received during the service provision process, this data is used to fine-tune the in vitro diagnostic interpretation model for retraining. As more users utilize the service, the accuracy of the diagnostic results provided can improve. Furthermore, before using specimen data diagnosed as positive for retraining, the diagnostic results are verified by experts such as doctors, and whether or not to retrain is determined based on the results. This eliminates the risk of any potential errors in the diagnostic results, and can also contribute to improving the performance of the in vitro diagnostic interpretation model.

[0116] Finally, the device (100) for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention and the method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention can be implemented as a computer program stored in a medium according to the third embodiment of the present invention that includes all the same technical features, and in this case, it can be combined with a computer device to execute (AA) a first step of receiving a training data set for training the in vitro diagnostic reading model, (BB) a second step of training the in vitro diagnostic reading model by utilizing the received training data set, (CC) a third step of receiving specimen data from a user terminal, and (DD) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which training has been completed and outputting a diagnostic result. In order to avoid redundant description, the device (100) for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention and the method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention described above can be implemented as a computer program stored in a medium according to the third embodiment of the present invention that includes all the same technical features, and in this case, the computer device can execute (AA) a first step of receiving a training data set for training the in vitro diagnostic reading model, (BB) a second step of training the in vitro diagnostic reading model by utilizing the received training data set, (CC) a third step of receiving specimen data from a user terminal, and (DD) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which training has been completed and outputting a diagnostic result. In order to avoid redundant description, the device (100) for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the first embodiment of the present invention and the method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model according to the second embodiment of the present invention described above. It goes without saying that all technical features applicable to the method of providing diagnostic results by utilizing the same can be equally applied to the computer program stored in the medium according to the third embodiment of the present invention.

[0117] Although embodiments of the present invention have been described with reference to the attached drawings, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without altering the technical concept or essential features thereof. Therefore, the embodiments described above should be understood to be illustrative in all respects and not restrictive.

Claims

1. A method for providing a diagnostic result by utilizing a deep learning-based in vitro diagnostic reading model by a device including a processor and memory, (a) a first step of receiving a learning data set for learning the above in vitro diagnostic reading model; (b) a second step of training the in vitro diagnostic reading model using the received learning data set; (c) a third step of receiving sample data from a user terminal; and (d) a fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which learning has been completed to output the diagnostic result; Includes, The above learning data set is, Including at least one negative learning data, which is sample data for a person who does not have the disease / illness to be diagnosed through the above in vitro diagnostic reading model. A method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

2. In paragraph 1, The above learning data set is, Further comprising generative voice learning data, which is one or more specimen data generated by adjusting the concentration of a reagent for detecting the disease / disease within a predetermined range for specimen data of a person who does not have the disease / disease to be diagnosed through the above in vitro diagnostic reading model. A method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

3. In paragraph 1, The second step above is, Step 2-1 of individually calculating the SE (Squared Error) of the input and output to the in vitro diagnostic reading model for each voice learning data included in the learning data set; and Step 2-2 of setting the maximum value among the SEs for each of the voice learning data produced above as the first threshold; A method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model including .

4. In paragraph 3, The fourth step above is, Step 4-1 of calculating the SE of input and output to the in vitro diagnostic reading model for the above-described received specimen data; Step 4-2 of determining whether the SE calculated in Step 4-1 above exceeds the first threshold value set above; and Step 4-3, which outputs the diagnosis result of the received sample data as positive if the above judgment result is exceeded, and outputs the diagnosis result as negative if the above judgment result is not exceeded; A method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model including .

5. In paragraph 4, Between steps 4-2 and 4-3 above, Step 4-2.5, which determines whether the SE calculated as a result of the judgment in step 4-2 exceeds the first threshold value set above, exceeds the second threshold value within a predetermined range from the first threshold value; Including more, In this case, the positive diagnosis result of the above step 4-3 is output when the judgment result of the above step 4-2.5 exceeds the second threshold, and the negative diagnosis result is output when the judgment result of the above step 4-2.5 does not exceed the second threshold. A method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

6. In paragraph 4, If the diagnosis result in step 4-3 above is positive, After the above 4th step, Step 5-1 of calculating the SE (Squared Error) of the input and output to the in vitro diagnostic reading model for the specimen data that output the above diagnostic result as positive; and Step 5-2 of retraining the in vitro diagnostic reading model in a direction that maximizes the calculated SE for the specimen data that output the above diagnostic result as positive; Step 5, which includes; A method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model including more.

7. In paragraph 6, Between steps 5-1 and 5-2 above, Step 5-1.3 of sending the specimen data that outputs the above diagnostic result as positive to a specialized medical institution server and receiving the diagnostic result for the specimen data from the specialized medical institution server; and Step 5-1.6 of labeling the specimen data with a diagnosis result-positive if the diagnosis result received from the specialized medical institution server is positive, and labeling the specimen data with a diagnosis result-negative if the diagnosis result received is negative; A method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model including more.

8. In paragraph 6, After the above 5th step, A sixth step of re-sharing the above-mentioned re-learned in vitro diagnostic reading model to the user terminal according to a request from the user terminal; including more, A method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

9. In paragraph 3, Step 2-3 of training the in vitro diagnostic reading model in a direction that minimizes the calculated SE for each of the above voice learning data; A method for providing diagnostic results by utilizing a deep learning-based in vitro diagnostic reading model including .

10. In paragraph 1, Between the second and third steps above, Step 2.5 of sharing the learned in vitro diagnostic reading model with the user terminal according to the request of the user terminal; Including more, The above user terminal can download the learned in vitro diagnostic reading model and output the diagnostic results for the specimen data. A method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

11. In paragraph 10, The above user terminal, We have our own reading model other than the above downloaded in vitro diagnostic reading model, For the above specimen data, the diagnostic result is first calculated through the self-reading model, and if the calculated diagnostic result is positive, the diagnostic result is secondarily calculated through the in vitro diagnostic reading model to output the final diagnostic result. A method for providing diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

12. One or more processors; network interface; A memory that loads a computer program to be executed by the processor; and Including storage for storing large-capacity network data and the above computer program, The above computer program is executed by the one or more processors, (A) A first step of receiving a learning data set for learning the above in vitro diagnostic reading model; (B) A second step of training the in vitro diagnostic reading model using the received learning data set; (C) a third step of receiving sample data from a user terminal; and (D) A fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which learning has been completed and outputting the diagnosis result; and run it, The above learning data set is, Including at least one negative learning data, which is sample data for a person who does not have the disease / illness to be diagnosed through the above in vitro diagnostic reading model. A device that provides diagnostic results using a deep learning-based in vitro diagnostic interpretation model.

13. In combination with a computing device, (AA) A first step of receiving a learning data set for learning the above in vitro diagnostic reading model; (BB) A second step of training the in vitro diagnostic reading model using the received learning data set; (CC) A third step of receiving sample data from a user terminal; and (DD) A fourth step of inputting the received specimen data into the in vitro diagnostic reading model for which learning has been completed and outputting the diagnosis result; and run it, The above learning data set is, Including at least one negative learning data, which is sample data for a person who does not have the disease / illness to be diagnosed through the above in vitro diagnostic reading model. A computer program stored on a computer-readable medium.

Images