Method for automatically detecting whether dental medical data is duplicated and computer-readable recording medium on which program for executing same in computer is recorded

Abstract

This method for automatically detecting whether dental medical data is duplicated comprises the steps of: collecting dental medical data on a dental condition of a patient; preprocessing the dental medical data to generate training data; training an artificial intelligence neural network by using the training data; and determining whether the dental medical data is duplicate of stored data by using the artificial intelligence neural network.

Description

A computer-readable recording medium having recorded thereon a method for automatically detecting duplication of dental medical data and a program for executing the method on a computer.

[0001] The present invention relates to a method for automatically detecting duplication in dental medical data and a computer-readable recording medium having recorded thereon a program for executing the method on a computer. More specifically, the present invention relates to a method for automatically detecting duplication in dental medical data using deep learning and a computer-readable recording medium having recorded thereon a program for executing the method.

[0002] Artificial intelligence (AI) technology is being utilized in diverse fields, including generation, classification, detection, diagnosis, and prediction. The quality of data used for learning, the availability of high-quality data, and the efficiency of data management can all determine the performance of AI-based neural networks. Accordingly, research is underway to utilize AI neural networks to manage and process data, including detecting duplicates and similarities, across a wide range of fields, from healthcare to manufacturing, industry, and gaming. In particular, the importance of technologies that effectively detect and remove duplicate data within large volumes of data, including unstructured data, is increasingly recognized.

[0003] Meanwhile, when training AI neural networks, excessively increasing the number of training sessions to improve learning accuracy can lead to overfitting, a phenomenon in which the AI ​​neural network over-adapts to the learning process and degrades test performance. Consequently, research is ongoing into technologies to prevent overfitting and improve the learning performance of AI neural networks.

[0004] The purpose of the present invention is to provide a method for automatically detecting duplication of dental medical data, which can efficiently determine whether data is duplicated within a large amount of data.

[0005] However, the problem to be solved by the present invention is not limited to the problem mentioned above, and may be expanded in various ways without departing from the spirit and scope of the present invention.

[0006] In order to achieve the above-described object of the present invention, a method for automatically detecting whether dental medical data is duplicated according to an embodiment of the present invention may include a step of collecting dental medical data on a patient's dental condition, a step of preprocessing the dental medical data to generate learning data, a step of training an artificial intelligence neural network using the learning data, and a step of determining whether the dental medical data is duplicated with stored data using the artificial intelligence neural network.

[0007] In one embodiment, the step of training the artificial intelligence neural network may include a step of training using first data within a training group among a plurality of groups of the training data, a step of training using a first portion of the first data and second data within a verification group among the plurality of groups, and a step of testing using a second portion of the first data and third data within a test group among the plurality of groups.

[0008] In one embodiment, the first part of the first data used in the step of learning using the second data can be completed through the step of learning using the first data.

[0009] In one embodiment, the second portion of the first data used in the step of testing using the third data may be completed through the step of learning using the first data.

[0010] In one embodiment, the total number of the first data may be greater than the total number of the second data.

[0011] In one embodiment, the total number of the second data may be greater than the total number of the third data.

[0012] In one embodiment, the number of the first portions of the first data used and the number of the second data used during the step of testing using the second data may be substantially equal to each other.

[0013] In one embodiment, the number of the second portions of the first data used and the number of the third data used during the step of testing using the third data may be substantially equal to each other.

[0014] In one embodiment, the step of training the artificial intelligence neural network may set a new data set for training among the training data.

[0015] In one embodiment, in the step of learning the artificial intelligence neural network, all of the previous data for which learning has been completed may be included in the new data set through the step of learning using the first data, the step of learning using the second data, and the step of testing using the third data, thereby repeating learning and testing.

[0016] In one embodiment, in the step of learning the artificial intelligence neural network, a part of the previous data for which learning has been completed may be included in the new data set through a step of learning using the first data, a step of learning using the second data, and a step of testing using the third data, thereby repeating learning and testing.

[0017] In one embodiment, in the step of training the artificial intelligence neural network, the new data set may not overlap with previous data that have been trained through the step of training using the first data, the step of training using the second data, and the step of testing using the third data.

[0018] In one embodiment, when the dental medical data includes mesh data having meshes, the step of generating the learning data may include a step of remeshing the mesh data so that each of the meshes has a uniform size, and a step of aligning the mesh data in a direction of a predetermined coordinate system at an origin of a predetermined coordinate system.

[0019] In one embodiment, when the dental medical data includes medical image data, the step of generating the learning data may include the step of resizing the medical image data so that the medical image data has a specific resolution and the step of performing histogram equalization on the resized medical image data.

[0020] In one embodiment, the dental medical data may include unstructured data having an unstructured structure.

[0021] A method for automatically detecting whether dental medical data overlaps with stored data according to one embodiment of the present invention includes the steps of: collecting dental medical data on a patient's dental condition to generate learning data; training an artificial intelligence neural network using first data in a training group among a plurality of groups of the learning data; training the artificial intelligence neural network using a first portion of the first data and second data in a verification group among the plurality of groups; testing the artificial intelligence neural network using a second portion of the first data and third data in a test group among the plurality of groups; and determining whether the dental medical data overlaps with stored data using the artificial intelligence neural network. Through the step of learning using the first data, the step of learning using the second data, and the step of testing using the third data, at least a portion of previous data for which learning has been completed can be included in a new data set to repeat learning and testing.

[0022] In one embodiment, a program for executing a method for automatically detecting duplication of dental medical data on a computer can be recorded on a computer-readable recording medium.

[0023] In a method for automatically detecting duplication of dental medical data according to embodiments of the present invention, an artificial intelligence neural network can be trained using the training data. The training process can be performed so that the artificial intelligence neural network experiences overfitting, so that the artificial intelligence neural network can be trained so that the artificial intelligence neural network can effectively memorize the training data. Accordingly, after collecting a large amount of unstructured data through the method for automatically detecting duplication of dental medical data, duplication of the unstructured data can be efficiently and accurately detected.

[0024] Additionally, in each of the steps of training the AI ​​neural network using a validation group and testing the AI ​​neural network using a test group, some of the data included in the training group can be mixed into each of the validation group and test group to conduct training and testing. Accordingly, the AI ​​neural network can be trained to better memorize the training data.

[0025] However, the effects of the present invention are not limited to the effects described above, and may be expanded in various ways without departing from the spirit and scope of the present invention.

[0026] FIG. 1 is a flowchart illustrating a method for automatically detecting duplication of dental medical data according to one embodiment of the present invention.

[0027] FIG. 2 is a drawing for explaining a method for automatically detecting whether dental medical data of FIG. 1 is duplicated.

[0028] Figure 3 is a flowchart showing an example of a step of preprocessing dental medical data of Figure 1 to generate learning data.

[0029] FIG. 4a is a drawing showing an example of the dental medical data of FIG. 1, FIG. 4b is a drawing showing an example of a process in which a step of performing remeshing on the mesh data of FIG. 3 is performed, and FIG. 4c is a drawing showing an example of a process in which a step of aligning the mesh data of FIG. 3 is performed.

[0030] FIG. 5a is a drawing showing another example of the dental medical data of FIG. 1, FIG. 5b is a drawing showing another example of the process in which the step of performing remeshing on the mesh data of FIG. 3 is performed, and FIG. 5c is a drawing showing another example of the process in which the step of aligning the mesh data of FIG. 3 is performed.

[0031] FIG. 6a is a diagram showing another example of the dental medical data of FIG. 1, FIG. 6b is a diagram showing another example of a process in which a step of performing remeshing on the mesh data of FIG. 3 is performed, and FIG. 6c is a diagram showing another example of a process in which a step of aligning the mesh data of FIG. 3 is performed.

[0032] Fig. 7 is a flowchart showing another example of a step of preprocessing dental medical data of Fig. 1 to generate learning data.

[0033] Figure 8 is a flowchart showing an example of a step for training an artificial intelligence neural network using the learning data of Figure 1.

[0034] Figure 9 is a diagram for explaining the steps of distributing the learning data of Figure 8 into three groups.

[0035] Fig. 10 is a diagram for explaining a learning step using the first data within the training group of Fig. 8.

[0036] FIG. 11 is a diagram for explaining a learning step using the first part of the first data of FIG. 8 and the second data within the verification group.

[0037] FIG. 12 is a drawing for explaining a step of testing using the second part of the first data of FIG. 8 and the third data within the test group.

[0038] Figure 13 is a diagram for explaining the steps of setting up a new data set that includes all data for which learning in Figure 8 has been completed.

[0039] Figure 14 is a flowchart showing another example of a step for training an artificial intelligence neural network using the learning data of Figure 1.

[0040] Figure 15 is a diagram for explaining a step of setting up a new data set including some of the data for which learning in Figure 14 has been completed.

[0041] Figure 16 is a flowchart showing another example of the steps for training an artificial intelligence neural network using learning data.

[0042] Figure 17 is a diagram for explaining a step of setting up a new data set that does not overlap with the data for which learning has been completed in Figure 16.

[0043] Figures 18 and 19 are drawings for comparing an artificial intelligence neural network that performs a method for automatically detecting duplication of dental medical data of Figure 1 with a conventional artificial intelligence neural network.

[0044] With respect to the embodiments of the present invention disclosed in the text, specific structural and functional descriptions are merely illustrative for the purpose of explaining the embodiments of the present invention, and the embodiments of the present invention may be implemented in various forms and should not be construed as being limited to the embodiments described in the text.

[0045] The present invention is susceptible to various modifications and takes various forms. Specific embodiments are illustrated in the drawings and described in detail herein. However, this is not intended to limit the present invention to specific disclosed forms, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and technical scope of the present invention.

[0046] While terms like "first" and "second" may be used to describe various components, these components should not be limited by these terms. These terms may be used to distinguish one component from another. For example, without departing from the scope of the present invention, a first component could be referred to as a "second component," and similarly, a second component could also be referred to as a "first component."

[0047] When a component is referred to as being "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components in between. Conversely, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between. Other expressions that describe the relationship between components, such as "between" and "directly between" or "adjacent to" and "directly adjacent to", should be interpreted similarly.

[0048] The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, it should be understood that the terms "comprise" or "have" indicate the presence of a described feature, number, step, operation, component, part, or combination thereof, but do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0049] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by those of ordinary skill in the art to which this invention pertains. 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 shall not be construed in an idealized or overly formal sense unless explicitly defined herein.

[0050] Meanwhile, if a particular embodiment can be implemented differently, the functions or operations specified within a particular block may occur in a different order than specified in the flowchart. For example, two consecutive blocks may actually be executed substantially simultaneously, or, depending on the related functions or operations, the blocks may be executed in reverse order.

[0051] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. Identical components in the drawings are designated by the same reference numerals, and redundant descriptions of identical components are omitted.

[0052] FIG. 1 is a flowchart illustrating a method for automatically detecting duplication of dental medical data according to one embodiment of the present invention. FIG. 2 is a diagram illustrating a method for automatically detecting duplication of dental medical data of FIG. 1.

[0053] Referring to FIGS. 1 and 2, a method for automatically detecting whether dental medical data is duplicated according to an embodiment of the present invention may include a step of collecting dental medical data (S100), a step of preprocessing the dental medical data to generate learning data (S200), a step of training an artificial intelligence neural network (100) using the learning data (S300), and a step of determining whether the dental medical data is duplicated using the artificial intelligence neural network (100) (S400).

[0054] The method for automatically detecting whether the above dental medical data is duplicated can be performed by a computing device.

[0055] The dental medical data collected in the step of collecting dental medical data (S100) may be data including the patient's oral condition, dental condition, etc. for performing dental treatment. For example, the dental medical data may include 3D oral scan data obtained by scanning teeth and the oral cavity or a model or reconstruction thereof using an oral scanner, CT data obtained by photographing the patient's teeth, MRI data, etc. Specifically, the 3D oral scan data may be mesh data including 3D points (vertexes) and triangular or rectangular faces created by connecting the points. In addition, the 3D oral scan data may be image data captured using a 3D scanner. There is no limitation on the file extension of the 3D oral scan data, and may be, for example, one of ply, obj, and stl. In addition, each of the CT data and the MRI data may include 2D image data and 3D volume data, respectively.

[0056] In one embodiment, the dental medical data may include unstructured data. For example, the unstructured data may be data having an unstructured structure that is not identifiable by a user terminal or other external device because it does not follow a predefined data model. In one embodiment, the dental medical data may include unstructured dental medical image data, 3D model data, etc. In addition, the dental medical data may further include structured data. For example, the structured data and the unstructured data may be mixed within the dental medical data.

[0057] However, the types of dental medical data according to embodiments of the present invention are not limited thereto. In addition, the data used in the method for automatically detecting whether there is duplication in the dental medical data is not limited to the dental medical data for dental treatment, and the method for automatically detecting whether there is duplication in the dental medical data may use various types of unstructured data that can be used for various purposes. In addition, the types of the unstructured data are not limited thereto, and the unstructured data may include various types of data such as voice, text, and video.

[0058] The collected dental medical data may be stored in a database. In the step of collecting the dental medical data (S100), data including the atypical data may be collected, and among the collected data, data that requires a determination of whether it overlaps with the data stored in the database may be selected and stored in the database. In other words, the dental medical data stored in the database may be data that has been selected based on a determination of whether it overlaps with the stored data.

[0059] In the step (S200) of preprocessing the dental medical data to generate the learning data, the dental medical data stored in the database can be received. Preprocessing including normalization, standardization, etc. can be performed on the received dental medical data. The step (S200) of preprocessing the dental medical data to generate the learning data can be performed to uniformly convert size information and value information of the data collected in the step (S100) of collecting the dental medical data. In addition, the step (S200) of preprocessing the dental medical data to generate the learning data can be performed to provide data of the same or similar format to the step (S300) of training the artificial intelligence neural network (100) using the learning data.

[0060] In the step (S300) of training the artificial intelligence neural network (100) using the above learning data, the artificial intelligence neural network (100) can be trained so that the artificial intelligence neural network (100) memorizes the learning data. For example, the artificial intelligence neural network (100) can learn a process of determining whether the learning data is data that is seen for the first time or data that has already been seen. The input of the artificial intelligence neural network (100) may be the learning data. The output of the artificial intelligence neural network (100) may be a determination of whether the dental medical data is duplicated. Specifically, through the step (S300) of training the artificial intelligence neural network (100) using the learning data, the process of determining whether the learning data is data that is seen for the first time or data that has already been seen can be learned, thereby determining whether the dental medical data is duplicated with data stored in the database. That is, a step (S400) of determining whether the dental medical data is duplicated using the learned artificial intelligence neural network (100) through a step (S300) of learning the artificial intelligence neural network (100) using the learning data can be performed.

[0061] In the step (S400) of determining whether the dental medical data is duplicated using an artificial intelligence neural network (100), the artificial intelligence neural network (100) can determine whether the data stored in the database and the dental medical data are duplicate data.

[0062] In one embodiment, in the step (S400) of determining whether the dental medical data is duplicated using the artificial intelligence neural network (100), the artificial intelligence neural network (100) may classify the dental medical data as duplicate data or new data. Specifically, if the dental medical data includes content that overlaps with the stored data, the artificial intelligence neural network (100) may classify the dental medical data as duplicate data. Accordingly, the artificial intelligence neural network (100) may temporarily store the dental medical data classified as duplicate data in garbage. Data stored in garbage may be automatically deleted if there is no access for a certain period of time. In addition, if the dental medical data does not include content that overlaps with the stored data, the artificial intelligence neural network (100) may classify the dental medical data as new data. The new data may be stored in the database.

[0063] In one embodiment, in the step (S400) of determining whether the dental medical data is duplicated using an artificial intelligence neural network (100), the artificial intelligence neural network (100) may classify the dental medical data into duplicate data, new data, or similar data. Specifically, if the dental medical data includes some content that overlaps with the stored data or is data with different content for the same patient, the artificial intelligence neural network (100) may classify the dental medical data as similar data. The similar data may be tagged with a category that is identical to or similar to the category of the data that provided the criteria for determining that the data is similar. The tagged similar data may be stored in the database.

[0064] To more accurately and efficiently perform the method for automatically detecting duplication of the above dental medical data, the artificial intelligence neural network (100) may be periodically updated. For example, transfer learning may be performed using the trained artificial intelligence neural network (100). Specifically, after training the artificial intelligence neural network (100) with CT data, 3D oral scan data having a different format from the CT data may be input into the artificial intelligence neural network (100) and retrained. Accordingly, the artificial intelligence neural network (100) may perform additional training to improve its suitability for a specific type of data. However, the transfer learning method of the artificial intelligence neural network (100) according to embodiments of the present invention is not limited thereto.

[0065] FIG. 3 is a flowchart showing an example of a step of preprocessing the dental medical data of FIG. 1 to generate learning data. FIG. 4a is a diagram showing an example of the dental medical data of FIG. 1, FIG. 4b is a diagram showing an example of a process in which a step of performing remeshing on the mesh data of FIG. 3 is performed, and FIG. 4c is a diagram showing an example of a process in which a step of aligning the mesh data of FIG. 3 is performed. FIG. 5a is a diagram showing another example of the dental medical data of FIG. 1, FIG. 5b is a diagram showing another example of a process in which a step of performing remeshing on the mesh data of FIG. 3 is performed, and FIG. 5c is a diagram showing another example of a process in which a step of aligning the mesh data of FIG. 3 is performed. FIG. 6a is a diagram showing another example of the dental medical data of FIG. 1, FIG. 6b is a diagram showing another example of a process in which a step of performing remeshing on the mesh data of FIG. 3 is performed, and FIG. 6c is a diagram showing another example of a process in which a step of aligning the mesh data of FIG. 3 is performed.

[0066] Referring to FIGS. 1 to 6c, in the step (S200) of preprocessing the dental medical data to generate the learning data, as described above, the dental medical data may include mesh data having meshes in which three-dimensional points are connected. The step (S200) of preprocessing the dental medical data to generate the learning data may include a step (S210) of remeshing the mesh data and a step (S220) of aligning the mesh data.

[0067] Through the step (S210) of performing remesh on the mesh data, each of the meshes of the mesh data can have a uniform size. Specifically, when the mesh data includes a jaw and teeth, the combination and arrangement of each of the three-dimensional points constituting each of the meshes can be reset so that each of the meshes constituting the jaw and the teeth have substantially the same size. In the step (S220) of aligning the mesh data, the mesh data on which the remeshing process has been performed can be aligned to the origin of a predetermined coordinate system in the direction of a predetermined coordinate system. For example, the position and direction of the mesh data on which the remeshing process has been performed can be determined so that the mesh data faces a specific direction using principle component analysis (PCA). Specifically, the y-axis can be determined using the direction in which the front and back of the teeth located at the center of the jaw included in the mesh data face, the z-axis can be determined using the direction in which the antagonistic teeth arranged in the jaw face each other, and the x-axis can be determined using the position and direction of both ends of the jaw. However, the alignment and axis determination criteria of the mesh data according to embodiments of the present invention are not limited thereto, and various criteria can be set so that the mesh data faces a specific direction.

[0068] Fig. 7 is a flowchart showing another example of a step of preprocessing dental medical data of Fig. 1 to generate learning data.

[0069] Referring to FIGS. 1, 2, and 7, in the step (S200A) of preprocessing the dental medical data to generate the learning data, as described above, the dental medical data may include medical image data such as CT data and MRI data. The step (S200A) of preprocessing the dental medical data to generate the learning data may include a step (S210A) of resizing the medical image data, and a step (S220A) of performing histogram equalization on the resized medical image data. Unlike the step (S200) of preprocessing the dental medical data to generate the learning data of FIG. 3, the step (S200A) of preprocessing the dental medical data to generate the learning data of FIG. 7 may not additionally require a process of aligning the dental medical data.

[0070] In the step (S210A) of resizing the medical image data, the resolution of the medical image data may be adjusted so that the medical image data has a specific resolution. For example, if the medical image data has a resolution higher than the specific resolution, the resolution of the medical image data may be lowered. After the step (S210A) of resizing the medical image data, histogram equalization may be performed on the resized medical image data, thereby converting the medical image data into a form suitable for input into an artificial intelligence neural network (100).

[0071] FIG. 8 is a flowchart illustrating an example of a step of training an artificial intelligence neural network using the training data of FIG. 1. FIG. 9 is a diagram for explaining a step of distributing the training data of FIG. 8 into three groups. FIG. 10 is a diagram for explaining a step of training using first data in the training group of FIG. 8. FIG. 11 is a diagram for explaining a step of training using a first part of the first data of FIG. 8 and second data in the verification group. FIG. 12 is a diagram for explaining a step of testing using a second part of the first data of FIG. 8 and third data in the test group. FIG. 13 is a diagram for explaining a step of setting a new data set including all data on which training has been completed of FIG. 8.

[0072] Referring to FIGS. 8 to 13, the step (S300) of training an artificial intelligence neural network (100) using the training data may include a step (S310) of distributing the training data into a plurality of groups, a step (S320) of training using first data (DAT1) in a training group (TRG), a step (S330) of training using a first part (P1) of the first data (DAT1) and second data (DAT2) in a verification group (VAG), a step (S340) of testing using a second part (P2) of the first data (DAT1) and third data (DAT3) in a test group (TEG), and a step (S350) of setting a new data set (ND) including all previous data (TCD) for which training has been completed.

[0073] In the step (S310) of distributing the above learning data into a plurality of groups, the preprocessed learning data can be distributed into a training group (TRG), a verification group (VAG), and a test group (TEG). The training group (TRG) may be a group composed of first data (DAT1) distributed to train an artificial intelligence neural network (100). The verification group (VAG) may be a group composed of second data (DAT2) to supplement a learning process performed only by the training group (TRG). The test group (TEG) may be a group composed of third data (DAT3) to test an artificial intelligence neural network (100) trained through the training group (TRG) and the verification group (VAG).

[0074] In one embodiment, the total number of first data (DAT1) may be greater than the total number of second data (DAT2). In one embodiment, the total number of first data (DAT1) may be greater than the total number of third data (DAT3). In one embodiment, the total number of second data (DAT2) may be greater than the total number of third data (DAT3). However, the relationship between the numbers of the second data (DAT2) and the third data (DAT3) according to embodiments of the present invention is not limited thereto, and the total number of second data (DAT2) and the total number of third data (DAT3) may be equal to each other.

[0075] In one embodiment, the ratio of the number of first data (DAT1), second data (DAT2), and third data (DAT3) used for one-time learning may be 7:2:1. However, the ratio of the number of first data (DAT1), second data (DAT2), and third data (DAT3) used for one-time learning according to embodiments of the present invention is not limited thereto, and may have various ratios of numbers for learning so that the artificial intelligence neural network (100) can well remember the learning data. For example, the ratio of the number of first data (DAT1), second data (DAT2), and third data (DAT3) used for one-time learning may be 6:3:1.

[0076] In the step (S320) of learning using the first data (DAT1) in the training group (TRG), the first data (DAT1) in the training group (TRG) can be input into the artificial intelligence neural network (100) to perform learning. In the step (S320) of learning using the first data (DAT1) in the training group (TRG), the artificial intelligence neural network (100) can learn a process of determining whether the first data (DAT1) is data that is seen for the first time or data that has already been seen. For example, in the step (S100) of collecting dental medical data, the dental medical data collected for the first time can be determined as data that the artificial intelligence neural network (100) has seen for the first time. If the artificial intelligence neural network (100) determines that the dental medical data is data that has already been seen, it corresponds to a case of an incorrect judgment, and the artificial intelligence neural network (100) can calculate a loss and reset the weights by reflecting the loss.

[0077] In the step (S330) of learning using the first part (P1) of the first data (DAT1) and the second data (DAT2) within the verification group (VAG), the artificial intelligence neural network (100) can learn using the second data (DAT2) and the first part (P1) of the first data (DAT1). The first part (P1) of the first data (DAT1) may be a part of the data (TCD) for which learning has already been completed in the step (S320) of learning using the first data (DAT1) within the training group (TRG). That is, in the process of learning using the verification group (VAG), the artificial intelligence neural network (100) can be trained by mixing the first part (P1) of the first data (DAT1) within the training group (TRG) with the second data (DAT2). Accordingly, the artificial intelligence neural network (100) can be trained so that the artificial intelligence neural network (100) can well remember the above learning data.

[0078] In one embodiment, during the step (S330) of learning using the first part (P1) of the first data (DAT1) and the second data (DAT2) within the verification group (VAG), the number of the first parts (P1) of the first data (DAT1) used and the number of the second data (DAT2) used may be substantially the same. Accordingly, in the above process of learning using the verification group (VAG), the artificial intelligence neural network (100) may perform balanced learning of the training group (TRG) and the verification group (VAG). However, the relationship between the number of the first part (P1) of the first data (DAT1) and the second data (DAT2) used in the step (S330) of learning using the second data (DAT2) according to embodiments of the present invention is not limited thereto, and the first part (P1) of the first data (DAT1) and the second data (DAT2) may have different numbers.

[0079] In the step (S340) of testing using the second part (P2) of the first data (DAT1) and the third data (DAT3) within the test group (TEG), the artificial intelligence neural network (100) can be tested using the third data (DAT3) and the second part (P2) of the first data (DAT1). The testing process of the artificial intelligence neural network (100) may be part of a learning process to induce an overfitting phenomenon in the artificial intelligence neural network (100). The second part (P2) of the first data (DAT1) may be part of the data (TCD) on which learning has already been completed in the step (S320) of learning using the first data (DAT1) within the training group (TRG). That is, in the process of testing using the test group (TEG), the second part (P2) of the first data (DAT1) within the training group (TRG) can be mixed with the third data (DAT3) to perform testing using the artificial intelligence neural network (100). Accordingly, the artificial intelligence neural network (100) can be tested so that the artificial intelligence neural network (100) can properly remember the learning data.

[0080] Meanwhile, the number of first parts (P1) of the first data (DAT1) used in the step (S330) of learning using the first part (P1) of the first data (DAT1) and the second data (DAT2) in the verification group (VAG) and the number of second parts (P2) of the first data (DAT1) used in the step (S340) of testing using the second part (P2) of the first data (DAT1) and the third data (DAT3) in the test group (TEG) may be different from each other. For example, the number of first parts (P1) of the first data (DAT1) used in the step (S330) of learning using the second data (DAT2) may be greater than the number of second parts (P2) of the first data (DAT1) used in the step (S340) of testing using the third data (DAT3). However, the number of the first part (P1) and the second part (P2) according to embodiments of the present invention is not limited thereto. For example, the number of each of the first part (P1) and the second part (P2) may be the same, or the number of the first part (P1) may be smaller than the number of the second part (P2).

[0081] In one embodiment, during the step (S340) of testing using the second part (P2) of the first data (DAT1) and the third data (DAT3) within the test group (TEG), the number of the second part (P2) of the first data (DAT1) used and the number of the third data (DAT3) used may be substantially the same. Accordingly, in the above process of testing the artificial intelligence neural network (100) using the test group (TEG), balanced learning of the training group (TRG) and the test group (TEG) can be performed. However, the relationship between the number of second parts (P2) of the first data (DAT1) and the number of third data (DAT3) used in the step (S340) of testing using the third data (DAT3) according to embodiments of the present invention is not limited thereto, and the second parts (P2) of the first data (DAT1) and the third data (DAT3) may have different numbers.

[0082] In the step (S350) of setting up a new data set (ND) including all previous data (TCD) for which learning has been completed, the data (TCD) for which learning has already been completed can be included in the new data set (ND) for new learning and testing in the step (S340) of testing using the second part (P2) of the first data (DAT1) and the third data (DAT3) within the test group (TEG). The artificial intelligence neural network (100) can be trained using the first data (DAT1) included in the new data set (ND), the artificial intelligence neural network (100) can be trained using a part of the first data (DAT1) and the second data (DAT2) included in the new data set (ND), and the artificial intelligence neural network (100) can be tested using a part of the first data (DAT1) and the third data (DAT3) included in the new data set (ND). Afterwards, the process of forming a new data set again and training and testing the artificial intelligence neural network can be repeated until all training data are used. That is, the steps of setting a new data set (ND) and training using the first data (DAT1) in the training group (TRG) using the set new data set (ND) (S320), training using the first part (P1) of the first data (DAT1) and the second data (DAT2) in the verification group (VAG) (S330), and testing using the second part (P2) of the first data (DAT1) and the third data (DAT3) in the test group (TEG) (S340) can be repeatedly performed.

[0083] However, the learning time according to the embodiments of the present invention is not limited thereto, and the learning process using the artificial intelligence neural network (100) can be performed for a sufficient number of epochs so that the artificial intelligence neural network (100) can be sufficiently learned. Since the new data set (ND) includes all data (TCD) for which learning has been completed, the artificial intelligence neural network (100) can more easily perform learning and testing. In addition, by inducing the overfitting phenomenon to occur in the artificial intelligence neural network (100), the artificial intelligence neural network (100) can be trained so that the artificial intelligence neural network (100) can better remember the learning data. Accordingly, after collecting a large amount of unstructured data through a method for automatically detecting whether there is duplication in the dental medical data, whether there is duplication in the unstructured data can be efficiently and accurately detected.

[0084] Since overfitting degrades the performance of artificial intelligence neural networks, conventional artificial intelligence neural networks can be designed to reduce their capacity by not including an excessive number of layers, or to drop out unnecessary information. Accordingly, overfitting can be prevented in conventional artificial intelligence neural networks.

[0085] However, the artificial intelligence neural network (100) according to embodiments of the present invention may be a model using deep learning designed to cause the above-described overfitting phenomenon during the learning process. Unlike the conventional artificial intelligence neural network designed to prevent the overfitting phenomenon, the artificial intelligence neural network (100) may have the values ​​of parameters set within the artificial intelligence neural network (100) adjusted so that the overfitting phenomenon intentionally occurs in the artificial intelligence neural network (100). This will be described later with reference to FIGS. 18 and 19.

[0086] In one embodiment, the artificial intelligence neural network (100) may be based on a deep learning model that processes point cloud data containing three-dimensional points. For example, the deep learning model may include PointNet, PointNet++, DGCNN, etc. However, the type of artificial intelligence neural network (100) according to embodiments of the present invention is not limited thereto.

[0087] Figure 14 is a flowchart illustrating another example of the steps for training an artificial intelligence neural network using the training data of Figure 1. Figure 15 is a diagram illustrating the steps for setting up a new data set including a portion of the data for which training has been completed in Figure 14.

[0088] The step (S300A) of training the artificial intelligence neural network using the learning data described with reference to FIGS. 14 and 15 may be substantially the same as or similar to the step (S300) of training the artificial intelligence neural network using the learning data described with reference to FIGS. 8 to 13, except for the step (S350A) of setting a new data set (NDa) including a portion of the previous data (TCD) for which training has been completed. Hereinafter, any content overlapping with the content described with reference to FIGS. 8 to 13 is omitted or simplified.

[0089] Referring to FIGS. 14 and 15, the step (S300A) of training the artificial intelligence neural network using training data may include a step (S350A) of setting a new data set (NDa) including a portion of the previous data (TCD) for which training has been completed. In the step (S350A) of setting the new data set (NDa) including a portion of the previous data (TCD) for which training has been completed, a portion of the new data set (NDa) may overlap with the data (TCD) for which training has been completed. In other words, when training and testing are performed using the new data set (NDa), a portion of the data (TCD) for which training has been completed may be included in the new data set (NDa) to perform training and testing.

[0090] Figure 16 is a flowchart illustrating another example of the steps for training an artificial intelligence neural network using training data. Figure 17 is a diagram illustrating the steps for establishing a new data set that does not overlap with the data for which training has been completed in Figure 16.

[0091] The step (S300B) of training the artificial intelligence neural network using the training data described with reference to FIGS. 16 and 17 may be substantially the same as or similar to the step (S300) of training the artificial intelligence neural network using the training data described with reference to FIGS. 8 to 13, except for the step (S350B) of setting a new data set (NDb) that does not overlap with the data (TCD) for which training has been completed. Hereinafter, any content overlapping with the content described with reference to FIGS. 8 to 13 is omitted or simplified.

[0092] Referring to FIGS. 16 and 17, the step (S300B) of training the artificial intelligence neural network using training data may include a step (S350B) of setting a new data set (NDb) that does not overlap with the data (TCD) for which training has been completed. The new data set (NDb) may not overlap with the data (TCD) for which training has been completed. In other words, when training and testing are performed using the new data set (NDa), training and testing may be performed using the new data set (NDa) composed of data excluding the data (TCD) for which training has been completed.

[0093] Figures 18 and 19 are drawings for comparing an artificial intelligence neural network that performs a method for automatically detecting duplication of dental medical data of Figure 1 with a conventional artificial intelligence neural network.

[0094] Referring to FIGS. 18 and 19, in one embodiment, the artificial intelligence neural network (100) may have a larger value of the first type parameter than the conventional artificial intelligence neural network. For example, the first type parameter may include a learning rate, model complexity, model capacity, etc.

[0095] Specifically, the artificial intelligence neural network (100) may have a greater number of hidden layers, a greater number of neurons, etc. than the conventional artificial intelligence neural network, so that the artificial intelligence neural network (100) may have a greater model complexity than the conventional artificial intelligence neural network. In one embodiment, the number of hidden layers of the artificial intelligence neural network (100) may be greater than the number of hidden layers of the conventional artificial intelligence neural network.

[0096] In addition, the artificial intelligence neural network (100) can have a larger model capacity than the conventional artificial intelligence neural network by setting the kernel size, pooling size, filter size, etc. to be larger than the conventional artificial intelligence neural network.

[0097] In one embodiment, the artificial intelligence neural network (100) may have a smaller value of the second type parameter than the conventional artificial intelligence neural network. For example, the first type parameter may include the strength of normalization, the degree of batch normalization, the degree of weight initialization, the dropout rate, etc.

[0098] In one embodiment, the artificial intelligence neural network (100) may have a lower strength of L1 / L2 regularization than the conventional artificial intelligence neural network. For example, the artificial intelligence neural network (100) may not use L1 / L2 regularization. However, the artificial intelligence neural network (100) according to embodiments of the present invention is not limited thereto, and the artificial intelligence neural network (100) may also use L1 / L2 regularization.

[0099] In one embodiment, the artificial intelligence neural network (100) may have a smaller number of layers to which batch normalization is applied than the conventional artificial intelligence neural network. For example, the artificial intelligence neural network (100) may not include layers to which batch normalization is applied. Accordingly, an internal covariate shift phenomenon may occur within the artificial intelligence neural network (100). However, the artificial intelligence neural network (100) according to embodiments of the present invention is not limited thereto, and the artificial intelligence neural network (100) may include layers to which batch normalization is applied.

[0100] In one embodiment, the AI ​​neural network (100) may utilize a simpler weight initialization method than the conventional AI neural network, or the AI ​​neural network (100) may not utilize a weight initialization method. Accordingly, the stability of the initial stage of learning may be reduced when learning using the AI ​​neural network (100) compared to when learning using the conventional AI neural network.

[0101] In one embodiment, the artificial intelligence neural network (100) may have a lower dropout rate than the conventional artificial intelligence neural network. For example, the dropout rate of the conventional artificial intelligence neural network may be about 50%, and the dropout rate of the artificial intelligence neural network (100) may be about 20%. Accordingly, the interdependence between network neurons in the artificial intelligence neural network (100) may be higher than that of the conventional artificial intelligence neural network. However, the artificial intelligence neural network (100) according to embodiments of the present invention is not limited thereto, and the dropout rate of the artificial intelligence neural network (100) may have various values ​​within a range that is lower than the dropout rate of the conventional artificial intelligence neural network. For example, the dropout rate of the artificial intelligence neural network (100) may be 0%.

[0102] In one embodiment, the number of learning epochs of the artificial intelligence neural network (100) may be increased so that overfitting occurs in the artificial intelligence neural network (100). In one embodiment, the batch size may be adjusted so that overfitting occurs in the artificial intelligence neural network (100). For example, by increasing the batch size of the artificial intelligence neural network (100), a large number of data may be trained using the artificial intelligence neural network (100). In another example, by reducing the batch size of the artificial intelligence neural network (100), a noisy gradient may be generated, causing the artificial intelligence neural network (100) to over-adapt to a specific pattern. In one embodiment, the overlapping of training data may be allowed or the number of training data may be intentionally increased so that the artificial intelligence model (100) may over-adapt to a pattern so that overfitting occurs in the artificial intelligence neural network (100). In one embodiment, to prevent overfitting from occurring in the artificial intelligence neural network (100), the artificial intelligence neural network (100) may use an optimizer that uses a fixed learning rate (e.g., SGD) instead of an optimizer that uses adaptive learning (e.g., Adam).

[0103] According to one embodiment of the present invention, a computer-readable recording medium having recorded thereon a program for executing a method for automatically detecting duplication of dental medical data according to the above embodiments on a computer may be provided. The above-described method can be written as a program executable by a computer, and can be implemented in a general-purpose digital computer that executes the program using a computer-readable medium. In addition, the structure of the data used in the above-described method can be recorded on a computer-readable medium through various means. The computer-readable medium may include program commands, data files, data structures, etc., either singly or in combination. The program commands recorded on the medium may be those specially designed and configured for the present invention, or may be known and usable by those skilled in the art in the field of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of program instructions include not only machine language codes, such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention.

[0104] Additionally, the method for automatically detecting whether or not the aforementioned dental medical data is duplicated can also be implemented in the form of a computer program or application executed by a computer and stored in a recording medium.

[0105] The present invention relates to a method for automatically detecting whether dental medical data is duplicated and a computer-readable recording medium having recorded thereon a program for executing the method on a computer, which can reduce the effort and time required to detect whether data is duplicated and improve the accuracy and productivity of detection of duplicate dental medical data.

[0106] Although the present invention has been described above with reference to exemplary embodiments thereof, it will be understood by those skilled in the art that various modifications and changes may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below.

Claims

1. Step of collecting dental medical data on the patient's dental condition; A step of preprocessing the above dental medical data to generate learning data; A step of training an artificial intelligence neural network using the above learning data; and A method for automatically detecting whether dental medical data is duplicated, comprising a step of determining whether the dental medical data is duplicated with stored data using the artificial intelligence neural network.

2. In the first paragraph, the step of learning the artificial intelligence neural network is: A step of learning using first data within a training group among multiple groups of the above learning data; A step of learning using the first part of the first data and the second data within the verification group among the plurality of groups; and A method for automatically detecting whether there is duplication in dental medical data, characterized in that it comprises a step of testing using a second part of the first data and a third data within a test group among the plurality of groups.

3. In the second paragraph, the first part of the first data used in the learning step using the second data is, A method for automatically detecting whether there is duplication in dental medical data, characterized in that learning is completed through a learning step using the above first data.

4. In the second paragraph, the second part of the first data used in the step of testing using the third data is A method for automatically detecting whether there is duplication in dental medical data, characterized in that learning is completed through a learning step using the above first data.

5. In the second paragraph, the total number of the first data is A method for automatically detecting whether there is duplication in dental medical data, characterized in that the number is greater than the total number of the above second data.

6. In the second paragraph, the total number of the second data is A method for automatically detecting whether there is duplication in dental medical data, characterized in that the number is greater than the total number of the above third data.

7. A method for automatically detecting whether dental medical data is duplicated, characterized in that the number of the first portion of the first data used and the number of the second data used during the step of testing using the second data in the second paragraph are substantially the same.

8. A method for automatically detecting whether dental medical data is duplicated, characterized in that the number of the second portions of the first data used and the number of the third data used are substantially the same during the step of testing using the third data in the second paragraph.

9. In the second paragraph, the step of learning the artificial intelligence neural network is: A method for automatically detecting whether there is duplication in dental medical data, characterized by setting a new data set for learning among the above learning data.

10. In the step of learning the artificial intelligence neural network in paragraph 9, A method for automatically detecting whether there is duplication in dental medical data, characterized in that all previous data for which learning has been completed are included in the new data set through a learning step using the first data, a learning step using the second data, and a testing step using the third data, and learning and testing are repeated.

11. In the step of learning the artificial intelligence neural network in paragraph 9, A method for automatically detecting whether there is duplication in dental medical data, characterized in that learning and testing are repeated by including some of the previous data for which learning has been completed in the new data set through a learning step using the first data, a learning step using the second data, and a testing step using the third data.

12. In the step of learning the artificial intelligence neural network in paragraph 9, A method for automatically detecting whether there is overlap in dental medical data, characterized in that the new data set does not overlap with previous data for which learning has been completed through a step of learning using the first data, a step of learning using the second data, and a step of testing using the third data.

13. In the first paragraph, if the dental medical data includes mesh data having meshes, The steps for generating the above learning data are: A step of performing a remesh on the mesh data so that each of the meshes has a uniform size; and A method for automatically detecting whether there is duplication in dental medical data, characterized by comprising a step of aligning the mesh data to the origin of a predetermined coordinate system in the direction of a predetermined coordinate system.

14. In the first paragraph, if the dental medical data includes medical image data, The steps for generating the above learning data are: A step of resizing the medical image data so that the medical image data has a specific resolution; and A method for automatically detecting whether there is duplication in dental medical data, characterized in that it comprises a step of performing histogram equalization on the resized medical image data.

15. A method for automatically detecting whether dental medical data is duplicated, characterized in that the dental medical data in the first paragraph includes unstructured data having an unstructured structure.

16. A step of collecting dental medical data on the patient's dental condition to generate learning data; A step of training an artificial intelligence neural network using first data within a training group among multiple groups of the above learning data; A step of training the artificial intelligence neural network using the first part of the first data and the second data within the verification group among the plurality of groups; A step of testing the artificial intelligence neural network using the second part of the first data and the third data within the test group among the plurality of groups; and A step of determining whether the dental medical data overlaps with stored data using the artificial intelligence neural network is included. A method for automatically detecting whether there is duplication in dental medical data, characterized in that learning and testing are repeated by including at least a portion of previous data for which learning has been completed in a new data set through a learning step using the first data, a learning step using the second data, and a testing step using the third data.

17. A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 1 to 16 on a computer.

Images