KR20240041199A - Learning method for identifying model of pediatric orbit fractures using radiographic preprocessing - Google Patents

Abstract

This specification discloses a method for learning a model that can identify pediatric orbital fractures from radiographs. The method for learning a pediatric orbital fracture identification model according to the present specification includes the steps of: (a) storing, by a processor, radiographs of the skull of children under 20 years of age in memory; (b) storing, by a processor, an identification value for a photo file in which an orbital fracture is identified among radiographs stored in the memory; (c) storing, by a processor, in the memory a partial image of the orbital portion in the radiograph stored in the memory; and (d) the processor inputting the radiograph and the partial image of the orbital portion stored in the memory into a neural network model to train the orbital fracture identification model.

Description

Method for learning pediatric orbital fracture identification model using radiograph preprocessing {LEARNING METHOD FOR IDENTIFYING MODEL OF PEDIATRIC ORBIT FRACTURES USING RADIOGRAPHIC PREPROCESSING}

The present invention relates to artificial intelligence models, and more specifically, to a method for training an artificial intelligence model.

The content described in this section simply provides background information on the embodiments described in this specification and does not necessarily constitute prior art.

Pediatric orbital fractures are the most common of all facial fractures, with an incidence of up to 45% reported. In particular, raised fractures, which commonly occur in children, require emergency surgery unlike adults.

Most pediatric patients suspected of having an orbital fracture in the emergency room are diagnosed with a CT scan, but orbital fractures are confirmed on CT in only 20% of cases. For children, who are more sensitive to radiation than adults, CT scanning is unnecessary radiation, and has a particularly significant impact on lenses that are not surrounded by bone.

Among radiographs, orbital rim views can be helpful in diagnosis, but there are limitations to accurate diagnosis due to the orbital structure.

Public Patent Publication No. 10-2019-0087472

The purpose of this specification is to provide a method for learning a model that can identify pediatric orbital fractures from radiographs.

This specification is not limited to the above-mentioned tasks, and other tasks not mentioned will be clearly understood by those skilled in the art from the description below.

The method for learning a pediatric orbital fracture identification model according to the present specification to solve the above-described problem includes the steps of: (a) the processor storing radiographs of the skull of children under 20 years of age in memory; (b) storing, by a processor, an identification value for a photo file in which an orbital fracture is identified among radiographs stored in the memory; (c) storing, by a processor, in the memory a partial image of the orbital portion in the radiograph stored in the memory; and (d) the processor inputting the radiograph and the partial image of the orbital portion stored in the memory into a neural network model to train the orbital fracture identification model.

According to an embodiment of the present specification, step (c) may be a step in which the processor generates the partial image using an orbital portion identification algorithm in a radiograph of the skull and stores it in the memory.

The method for learning a pediatric orbital fracture identification model according to the present specification includes the steps of: (e) a processor storing a radiograph stored in the memory and a partial image of the orbital portion as one set of data; And (f) a step of the processor inputting the set data into the orbital fracture identification model to improve the accuracy of the orbital fracture identification model.

According to an embodiment of the present specification, step (e) is a step in which the processor determines the similarity between the radiograph stored in the memory and the partial image of the orbital portion and stores images with high similarity as set data. You can.

The pediatric orbital fracture identification model learning method according to the present specification may be implemented in the form of a computer program written to perform each step of the pediatric orbital fracture identification model learning method on a computer and recorded on a computer-readable recording medium.

Other specific details of the invention are included in the detailed description and drawings.

According to one aspect of the present specification, pediatric orbital fracture patients requiring emergency surgery can be identified using radiographs alone.

According to another aspect of the present specification, unnecessary CT scans can be reduced, thereby reducing exposure of pediatric patients to radiation.

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.

Figure 1 is a reference diagram for the diagnostic process of a typical pediatric orbital fracture patient.
Figure 2 is a flowchart of a method for learning a pediatric orbital fracture identification model according to an embodiment of the present specification.
Figure 3 is a flowchart of a method for learning a pediatric orbital fracture identification model according to another embodiment of the present specification.
Figure 4 is a reference diagram for the development of the skull and paranasal sinuses according to growth.

The advantages and features of the invention disclosed in this specification and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present specification is not limited to the embodiments disclosed below and may be implemented in various different forms, and the present embodiments are merely intended to ensure that the disclosure of the present specification is complete and to provide a general understanding of the technical field to which the present specification pertains. It is provided to fully inform those skilled in the art of the scope of this specification, and the scope of rights of this specification is only defined by the scope of the claims.

The terms used in this specification are for describing embodiments and are not intended to limit the scope of this specification. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context. As used in the specification, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other elements in addition to the mentioned elements.

Like reference numerals refer to like elements throughout the specification, and “and/or” includes each and every combination of one or more of the referenced elements. Although “first”, “second”, etc. are used to describe various components, these components are of course not limited by these terms. These terms are merely used to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may also be a second component within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings commonly understood by those skilled in the art to which this specification pertains. Additionally, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless clearly specifically defined.

Figure 1 is a reference diagram for the diagnostic process of a typical pediatric orbital fracture patient.

Referring to Figure 1, when a pediatric patient is admitted to the hospital with an injury around the eyes, a radiograph (aka: X-ray photograph) is first taken. Radiographs of the orbit (eye socket) are taken overlapping with various structures such as the skull because the three-dimensional orbit is taken in two dimensions. Therefore, based on the doctor's clinical judgment, additional computed tomography (CT) is performed to more accurately diagnose orbital fractures. However, as previously mentioned, orbital fractures are identified on CT in only 20% of cases. For pediatric patients who are more sensitive to radiation than adults, CT scanning is unnecessary radiation, and has the disadvantage of having a particularly large effect on lenses that are not surrounded by bone. However, because radiographs are two-dimensional images of the three-dimensional orbit, it is difficult for a doctor to identify an orbital fracture using radiographs alone due to the phenomenon of overlapping with various structures such as the skull, and it is risky to rely only on the doctor's clinical judgment. The reality is that it is even more difficult to expect from doctors with little clinical experience.

Accordingly, the present inventor recognized the need for an artificial intelligence model that can identify orbital fractures in children using only radiographs. However, it was not possible to derive an artificial intelligence model that could identify orbital fractures in children using a commonly known artificial intelligence learning method, that is, a deep learning method. First, because the radiograph was a two-dimensional image of the three-dimensional orbit, the accuracy of the artificial intelligence model's recognition of the orbital area did not improve beyond a certain level. Additionally, because other structures were overlapped in the orbital area in the radiographs, the accuracy for fractures and non-fractures could not reach a reliable level.

The present inventor came up with a learning method that can solve the above problem and generate a pediatric orbital fracture identification model with high accuracy. Hereinafter, a method for learning a pediatric orbital fracture identification model according to the present specification will be described with reference to the accompanying drawings.

Meanwhile, terms used in this specification are defined in advance. In this specification, “model” means a neural network model. Artificial Neural Network (ANN) implements artificial intelligence by connecting artificial neurons that mathematically model the neurons that make up the human brain. For example, an artificial neural network includes deep learning technology using a convolutional neural network (CNN). In this specification, “radiography” means a photograph taken using radiation. The radiographs can be distinguished from tomography images, unlike CT and MRI scans. In particular, the radiograph in this specification is a photograph of the skull and includes the orbital region. In this specification, “learning” refers to the process of changing parameter values so that an artificial intelligence model has the ability to output a specific value. Therefore, once the learning of the artificial intelligence model in this specification is completed, the existing artificial neural network can become a model that can identify whether a child has an orbital fracture.

The pediatric orbital fracture identification model learning method, which will be described below, uses processors, application-specific integrated circuits (ASICs), other chipsets, logic circuits, and registers known in the art to store, calculate, and execute various processing logic. , communication modem, data processing device, etc. Additionally, when the above-described logic is implemented as software, the processor, etc. may be implemented as a set of program modules. At this time, the program module may be stored in the memory device and executed by the processor. Hereinafter, the method for learning a pediatric orbital fracture identification model according to the present specification will be described based on how it is executed by a processor. However, the method for learning a pediatric orbital fracture identification model according to the present specification is not limited to the examples described below.

The computer program is C/C++, C#, JAVA, or Python that the processor (CPU) of the computer can read through the device interface of the computer in order for the computer to read the program and execute the methods implemented in the program. , may include code encoded in a computer language such as machine language. These codes may include functional codes related to functions that define the necessary functions for executing the methods, and include control codes related to execution procedures necessary for the computer's processor to execute the functions according to predetermined procedures. can do. In addition, these codes may further include memory reference-related codes that indicate at which location (address address) in the computer's internal or external memory additional information or media required for the computer's processor to execute the above functions should be referenced. there is. In addition, if the computer's processor needs to communicate with any other remote computer or server in order to execute the above functions, the code uses the computer's communication module to determine how to communicate with any other remote computer or server. It may further include communication-related codes regarding whether communication should be performed and what information or media should be transmitted and received during communication.

The storage medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, examples of the storage medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc., but are not limited thereto. That is, the program may be stored in various recording media on various servers that the computer can access or on various recording media on the user's computer. Additionally, the medium may be distributed to computer systems connected to a network, and computer-readable code may be stored in a distributed manner.

Figure 2 is a flowchart of a method for learning a pediatric orbital fracture identification model according to an embodiment of the present specification.

Referring to FIG. 2, in step S100, the processor may store radiographs of the skull of a child under 20 years of age in memory. In the next step S110, the processor may store an identification value for a photo file in which an orbital fracture is identified among the radiographs stored in the memory. The presence or absence of an orbital fracture among the above radiographs refers to cases in which an orbital fracture is confirmed through the results of existing CT scans. The identification value may be added by a doctor or administrator. Through step S110, each photo file can be divided into photos of patients with orbital fractures and photos of patients without fractures.

In the next step S120, the processor may store a partial image of the orbital portion in the radiograph stored in the memory in the memory. The partial image is a so-called crop image of the orbital area. According to one embodiment, the cropping operation may be performed by a person extracting the orbital region from each radiograph. According to another embodiment, the cropping operation may be performed by a processor using an orbital region identification algorithm in a radiograph of the skull to generate the partial image and store it in the memory. At this time, the partial image may include the same identification value for orbital fracture included in the original image.

In the next step S130, the processor may input the radiograph and the partial image of the orbital portion stored in the memory into a neural network model and train it as an orbital fracture identification model. The radiographs and partial images contain identification values for orbital fractures in each file, so the model can be learned using a so-called supervised learning method. It is obvious that the learning in step S130 can be repeated until a result identical to the identification value for orbital fracture can be derived through a plurality of radiographs and partial images, that is, learning data.

Figure 3 is a flowchart of a method for learning a pediatric orbital fracture identification model according to another embodiment of the present specification.

Referring to FIG. 3, it can be seen that steps S100 to S130 of FIG. 2 are the same, and step S140 and step S150 were added thereafter. In step S140, the processor may store the radiograph and the partial image of the orbital portion stored in the memory as one set of data. In other words, the orbital area of the partial image is matched with the corresponding original image. The matching of the radiograph and the partial image can be done by using the file name of the original radiograph image in the process of generating the partial image in step S120 to distinguish which images should be matched. In another embodiment, the processor may determine the similarity between the radiograph stored in the memory and the partial image of the orbital region and store images with high similarity as set data. As an example, the processor can calculate the cosine similarity between a radiograph and a partial image to match photos with a low cosine similarity value.

In the next step S150, the processor may improve the accuracy of the orbital fracture identification model by inputting the set data into the orbital fracture identification model. The step S150 is an additional learning process. Even though the primary learning is completed in step S130, the secondary learning is performed using the same partial image of the radiographer. The model that has completed the previous primary learning process outputs an identification value for orbital fractures when a radiograph is input. At this time, it is unknown whether the output value is a value determined by determining whether there is a fracture in the orbital region, or a result derived by chance from the characteristic values of the image existing in another region of the skull. Therefore, in order to train the model for which primary learning has been completed into a model that can identify whether the orbital region is fractured, the processor may additionally execute the learning process according to step S150. Secondary learning is characterized in that it is carried out using data stored as a set of radiographs and partial images of the skull in step S140. At this time, since the partial image is an image in which only the orbital region exists, the result of whether or not there is a fracture is naturally the result of whether or not the orbital region is fractured. In addition, for radiographs saved as a set, learning continues until the same result as the partial image is output. When learning is completed, the probability that the result for fractures in the radiograph is also a result derived based on the orbital area is high. increases significantly. In other words, although existing learning methods and artificial intelligence models are unexplainable models, the pediatric orbital fracture identification model learning method according to the present specification is an explainable model learning method that determines whether or not there is a fracture in the orbital region.

Meanwhile, unlike adults, the most common fracture site in children under 7 years of age is the superior region. The reason why fractures in the upper region are frequent is because the sinuses are not yet fully developed at young ages (under 7 years old), the orbital wall is thick, the bone itself is elastic, and the midface is proportionally short and flat. As the sinuses develop, the bones become thinner, and similar to adults, fractures in the lower region occur in the majority.

Figure 4 is a reference diagram for the development of the skull and paranasal sinuses according to growth. In the age group of 7 years or older, like adults, fractures in the inferior region account for the most, but unlike adults, a significant number of raised fractures are seen. A raised fracture refers to a structure in which internal structures come out at the fracture site, muscles or fat within the eye come out through the fracture site, and then the fracture site is closed by the elasticity of the bone. It occurs due to the elasticity of the bones in children, and unlike general orbital fractures, it has been reported that the sooner the surgery is performed, the less likely it is to cause eye movement disorders. Therefore, the pediatric orbital fracture identification model learning method according to the present specification may be classified according to the patient's fracture site. In step S100 of FIGS. 2 and 3, the radiographs can be divided into radiographs of the skull of children under 7 years of age and radiographs of the skulls of children over 7 years of age and under 20 years of age, and each can be learned. In other words, it can be trained separately into a model for children under 7 years of age and a model for children over 7 years of age.

Although the embodiments of the present specification have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention can be implemented in other specific forms without changing the technical idea or essential features. You will be able to understand it. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive.

Claims (5)

(a) causing the processor to store radiographs of the skull of a child under 20 years of age in a memory;
(b) a processor storing an identification value for a photo file in which an orbital fracture is identified among radiographs stored in the memory;
(c) storing, by a processor, in the memory a partial image of the orbital portion in the radiograph stored in the memory; and
(d) the processor inputting the radiograph and the partial image of the orbital portion stored in the memory into a neural network model to learn the orbital fracture identification model; a pediatric orbital fracture identification model learning method comprising a. In claim 1,
The step (c) is a step in which the processor generates the partial image using an orbital partial identification algorithm in a radiograph of the skull and stores it in the memory. A pediatric orbital fracture identification model learning method. In claim 1,
(e) storing, by a processor, the radiograph and the partial image of the orbital portion stored in the memory as one set of data; and
(f) the processor inputting the set data into the orbital fracture identification model to improve the accuracy of the orbital fracture identification model; further comprising a method for learning a pediatric orbital fracture identification model. In claim 3,
In step (e),
A method for learning a pediatric orbital fracture identification model, wherein the processor determines the similarity between the radiograph stored in the memory and the partial image of the orbital portion and stores images with high similarity as set data. A computer program written to perform each step of the pediatric orbital fracture identification model learning method according to any one of claims 1 to 4 on a computer and recorded on a computer-readable recording medium.