KR20230135404A - Medical image processing method for diagnosing supernumerary tooth in radiographic images - Google Patents

Abstract

According to an embodiment of the present disclosure, there is a medical image processing method performed by a computing device, the method comprising: determining an image type of a medical image; inputting the medical image into a diagnostic model determined according to the image type; and obtaining a diagnostic image including disease information regarding the medical image from the diagnostic model.

Description

Medical image processing method for diagnosing supernumerary teeth in radiological images {MEDICAL IMAGE PROCESSING METHOD FOR DIAGNOSING SUPERNUMERARY TOOTH IN RADIOGRAPHIC IMAGES}

The present disclosure relates to a method and device for processing medical images, and more specifically, to a method and device for processing various types of medical images for diagnosis of supernumerary teeth.

Electronic devices typically used in the medical field include medical image acquisition devices - for example, CT, MRI, CBCT, etc. - to determine a person's condition in a non-invasive way. In particular, recently, CBCT images have been widely used for dental diagnosis, treatment, and surgical planning. CBCT imaging offers advantages over regular CT imaging, such as lower radiation dose to the patient, shorter acquisition time, and higher resolution.

Meanwhile, medical images obtained in this way are read by skilled doctors and used as an auxiliary means of diagnosing diseases. The act of reading medical images is performed depending on the subjective judgment of doctors, that is, people. Accordingly, errors may occur in the diagnosis of disease depending on the condition or skill level of the doctor who judges the medical image.

For example, supernumerary teeth disease is a disease in which there are more teeth than the normal number of teeth. It occurs in the maxillary anterior region and can be difficult to detect. However, if these excess teeth are left in place due to diagnostic errors, there is a risk of causing various complications such as delayed eruption of adjacent permanent teeth, displacement, diastema, crowding, and tooth root resorption.

Accordingly, in order to enable objective and accurate interpretation of medical images, there is a need for technology that identifies diseases from various image types of medical images and automatically provides disease information such as disease diagnosis, disease location, and disease type. There is a need in the industry.

The problem to be solved is to provide a method to accurately infer information about diseases present in medical images by processing medical images based on an artificial intelligence model optimized for each image type. In addition to the above tasks, it can be used to achieve other tasks not specifically mentioned.

According to some embodiments of the present disclosure, there is a medical image processing method performed by a computing device, the method comprising: determining an image type of a medical image; inputting the medical image into a diagnostic model determined according to the image type; and obtaining a diagnostic image including disease information regarding the medical image from the diagnostic model.

The medical image may include a dental radiology image, and an image type of the medical image may include at least one of a panoramic image type and a periapical image type.

The diagnostic model may be at least one diagnostic model optimized for the determined image type, among a plurality of diagnostic models, using a training medical image corresponding to the determined image type.

The diagnostic model may be trained to infer the diagnostic image from the input medical image based on training data consisting of a pair of the training medical image and a labeling image in which diseases in the training medical image are manually marked.

The learning data may include one or more medical images for learning in which at least some of the age value of the subject and the gender value of the subject are different from each other.

The diagnostic model is a deep learning model based on an object detection model, and the object detection model may include a YOLO model that uses a Darknet-based deep structure as a backbone.

The disease information may include at least some of whether the disease is identified in the medical image, the location of the identified disease, the type of the identified disease, and the diagnosis name of the identified disease.

The diagnostic image may be an image in which at least part of the disease information is overlaid on the medical image.

The diagnostic image is overlaid on an image area corresponding to the location of the disease in the medical image, and may include a bounding box formed to have a shape corresponding to the shape of the disease.

The diagnosis of the disease may include excess teeth.

According to some embodiments of the present disclosure, a medical image processing method performed by a computing device, the method comprising: determining an image type of a medical image; at least one diagnostic model optimized for the determined image type among a plurality of diagnostic models; automatically determining a diagnostic model, and inputting the medical image into the automatically determined diagnostic model, thereby obtaining a diagnostic image including disease information regarding the medical image.

The medical image may include a dental radiology image, and an image type of the medical image may include at least one of a panoramic image type and a periapical image type.

Each of the plurality of diagnostic models generates the diagnostic image from the input medical image based on learning data consisting of a pair of a learning medical image of a corresponding image type and a labeling image in which the disease in the learning medical image is manually marked. It can be learned to reason.

The at least one diagnostic model is a deep learning model based on an object detection model, and the object detection model may include a YOLO model that uses a Darknet-based deep structure as a backbone.

The disease information may include at least some of whether the disease is identified in the medical image, the location of the identified disease, the type of the identified disease, and the diagnosis name of the identified disease.

The diagnostic image is overlaid on an image area corresponding to the location of the disease in the medical image, and may include a bounding box formed to have a shape corresponding to the shape of the disease.

The diagnosis of the disease may include excess teeth.

According to some embodiments of the present disclosure, a medical image processing method performed by a computing device, the method comprising: determining an image type of a medical image; if the image type is a panoramic image type, selecting the panoramic image type; Inputting the medical image into an optimized diagnostic model, and if the image type is a periapical image type, inputting the medical image into a diagnostic model optimized for the periapical image type, and from the diagnostic model, of the medical image It may include obtaining a diagnostic image containing information about the supernumerary tooth.

The diagnostic model is trained to infer the diagnostic image from the input medical image based on training data consisting of a pair of a training medical image of the corresponding image type and a labeling image in which the disease in the training medical image is manually marked. It can be.

The step of acquiring the diagnostic image includes, when supernumerary tooth disease is identified in the medical image, a bounding box that is overlaid on an image area corresponding to the location of the supernumerary tooth disease and has a shape corresponding to the shape of the supernumerary tooth disease. , acquiring the diagnostic image, and when the supernumerary tooth disease is not identified in the medical image, acquiring the medical image as the diagnostic image.

According to some embodiments of the present disclosure, by processing medical images based on an artificial intelligence model optimized for each image type, information about diseases present in the medical images can be accurately and easily inferred.

According to some embodiments of the present disclosure, by processing medical images based on an artificial intelligence model optimized for each image type, it is possible to quickly and accurately obtain a diagnostic image that intuitively displays the location and form of hypertooth disease present in the medical image. You can.

1 is a block diagram showing a medical image processing device according to some embodiments of the present disclosure.
FIG. 2 is a diagram illustrating an example of training data for a diagnostic model according to some embodiments of the present disclosure.
FIG. 3 is a diagram illustrating an example of a diagnostic model optimized for a panoramic image type according to some embodiments of the present disclosure.
Figure 4 is a diagram illustrating an example of a diagnostic model optimized for a periapical image type according to some embodiments of the present disclosure.
5 is a flowchart of a medical image processing method according to some embodiments of the present disclosure.
FIG. 6 is a block diagram illustrating a computing device that provides a medical image processing method according to some embodiments of the present disclosure.

Below, with reference to the attached drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily practice them. However, the present disclosure may be implemented in many different forms and is not limited to the embodiments described herein. In order to clearly explain the present disclosure in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

In the present disclosure, when a part “includes” a certain component, this means that it may further include other components rather than excluding other components, unless specifically stated to the contrary. The devices that make up the network may be implemented as hardware, software, or a combination of hardware and software.

1 is a schematic diagram showing a medical image processing device according to some embodiments of the present disclosure.

Referring to FIG. 1, the medical image processing device 100 according to the present disclosure includes an image receiving unit 110, a type determining unit 120, a diagnostic model unit 130, a model learning unit 140, and an image output unit 150. ) may include. However, the above-described configuration is not essential for implementing the medical image processing device 100 according to the present disclosure, and the medical image processing device 100 according to the present disclosure may include more or less configurations than the above-described configuration. there is.

For example, the medical image processing device 100 according to the present disclosure stores a received medical image, an output value obtained by processing the medical image (e.g., a diagnostic image, etc., which will be described later), or training data for a diagnostic model, etc. It may further include a storage unit (not shown). However, it is not limited to this.

The image receiving unit 110 of the medical image processing device 100 according to the present disclosure may receive a medical image from an external server or terminal (not shown). In this case, the external server or terminal may be, for example, a server of a medical institution, an image acquisition device used in a medical institution, or a user terminal of a user who wishes to perform diagnosis and treatment based on medical images.

The type determination unit 120 of the medical image processing device 100 according to the present disclosure may determine the image type of the received medical image. That is, the medical image in the present disclosure may include any type of image captured by any method for any body part, and the type determination unit 120 may determine the medical image received through the image receiver 110. You can determine which video type it belongs to.

In the present disclosure, the medical image is a dental radiology image, and in particular, the image type of the medical image is described as an example including at least one of a panoramic view type and a periapical view type. However, it is not limited to this, and the medical image according to the present disclosure is a maxillary occlusal radiographic image type, CBCT (Cone-beam CT). It may be an imaging type, a cephalometric x-ray type, a bitewing type, or a magnetic resonance imaging (MRI) type.

The image type of a medical image may be determined on a rule basis based on the properties of each medical image (eg, image size ratio, image file metadata, etc.). However, it is not limited to this, and the image type of the medical image can be classified by pre-trained classification models such as a clustering model, SVM (Support Vector Machine) model, or CNN (Convolutional Neural Network) model using features extracted from medical images. may be determined, or may be determined directly by the user.

The diagnostic model unit 130 of the medical image processing device 100 according to the present disclosure may include a plurality of diagnostic models for inputting and processing medical images. The diagnostic model unit 130 according to the present disclosure may determine at least one diagnostic model among a plurality of diagnostic models based on the image type determined by the type determination unit 120 and input a medical image into the determined diagnostic model. . In this case, the diagnostic model can infer a diagnostic image including disease information about the medical image from the input medical image.

Each of the plurality of diagnostic models existing in the diagnostic model unit 130 may be a model optimized for each different image type. Accordingly, based on the image type determined by the type determination unit 120, the diagnostic model unit 130 selects at least one diagnostic model optimized for the corresponding image type from among the plurality of diagnostic models as a model for processing medical images. You can decide.

Meanwhile, in order to optimize the diagnostic model, the model learning unit 140 of the medical image processing device 100 according to the present disclosure converts the diagnostic models existing in the diagnostic model unit 130 into medical images for learning of each optimized image type. You can learn using . Here, the learning medical image can be paired with a labeling image in which the disease in the learning medical image is manually marked to form learning data.

As an example, in the case of a diagnostic model optimized for a panoramic image type (hereinafter, “panoramic image diagnosis model”), a medical image for learning including one or more panoramic images, and the diagnosis name, type, and form of the disease present in each panoramic image. It can be learned to infer a diagnostic image from a panoramic image type medical image based on learning data consisting of a pair of labeled images with manually marked positions, etc.

As another example, in the case of a diagnostic model optimized for the periapical image type (hereinafter, “periapical image diagnosis model”), a medical image for learning including one or more periapical images, and the diagnosis name and form of the disease present in each periapical image. Based on learning data consisting of pairs of labeled images with manually marked positions, etc., it can be learned to infer a diagnostic image from a medical image of the periapical image type.

Meanwhile, the learning data may be configured to include one or more medical images for learning in which at least some of the age of the subject and the gender of the subject are different from each other. That is, medical images for learning according to the present disclosure can be collected for subjects of various ages and genders, from the deciduous dentition period to the mixed dentition period and permanent dentition period (or from infants to adults), and can be collected for subjects of a specific age and/or gender. Can be evenly selected for various ages and genders to prevent bias.

However, it is not limited to this, and medical images for learning may be selected to include more medical images for subjects of a specific age and/or gender. For example, in the case of supernumerary tooth disease, diagnosis is generally more difficult in the deciduous and mixed dentition stages than in the permanent dentition stage. Accordingly, medical images for learning may be selected to include more medical images of the deciduous dentition period and the mixed dentition period to improve the diagnostic accuracy of the diagnosis model's supernumerary teeth.

Returning to the diagnostic model unit 130, the diagnostic model may be a deep learning model based on an object detection model, and the object detection model is YOLO (You Only Look Once) using a Darknet-based deep structure as a backbone. Can include models.

For example, the panoramic image diagnosis model inputs medical images by adjusting them to a size of 512x256, and detects objects from feature maps of three different sizes, such as 128x64x18, 64x32x18, and 32x16x18, through the Darknet-53 architecture (e.g., It may include a YOLOv3 model that performs disease identification).

Alternatively, the periapical imaging model inputs medical images by adjusting them to a size of 512x512, and uses a YOLOv3 model that performs object detection from feature maps of three different sizes, such as 64x64x18, 32x32x18, and 16x16x18, through the Darknet-53 architecture. It can be included.

However, it is not limited to this, and the diagnostic model may be implemented using any algorithm and/or neural network structure available for object detection, such as CNN or SSD (Single Shot Detector).

The image output unit 150 of the medical image processing device 100 according to the present disclosure may output a diagnostic image generated from a specific diagnostic model of the diagnostic model unit 130. The diagnostic image may include disease information related to the medical image. In other words, the diagnostic image may include information about the disease identified in the medical image.

Here, the disease information is information inferred based on the medical image from the above-described diagnostic model, including whether or not the disease is identified in the medical image, the location of the disease identified in the medical image, the type of disease identified in the medical image, and It may include at least some of the diagnoses of the identified disease. The diagnostic image may be an image in which at least some of the above-described disease information is overlaid on the medical image.

Regarding whether the disease is identified, if the disease is not identified by the diagnostic model in the input medical image, the image output unit 150 may output the medical image itself as a diagnostic image. However, it is not limited to this, and the image output unit 150 may output a diagnostic image in which arbitrary graphics or text indicating that a disease is not identified in the medical image is overlaid on the medical image. Conversely, when a disease is identified by a diagnostic model in an input medical image, the image output unit 150 may output a diagnostic image in which at least part of the disease information of the identified disease is overlaid on the medical image.

Regarding the diagnosis name of the disease, the image output unit 150 may output a diagnosis image including the name of the disease identified by the diagnosis model in the medical image. In this case, the diagnosis of the disease may be displayed, for example, in an area on the diagnostic image adjacent to the location of the identified disease, or may be displayed in a preset area on the diagnostic image, such as the upper left corner. However, it is not limited to this.

If the medical image is a dental radiology image, as explained in this disclosure as an example, the diagnosis name of the disease may be, for example, “excess teeth.” Alternatively, the diagnosis of the disease may be a diagnosis of a disease that can be identified from dental radiological images such as dental caries, periodontal disease, or other diseases that occur in areas such as teeth and surrounding tissue, jawbone, temporomandibular joint, and maxillary sinus. However, it is not limited to this, and the diagnosis name for any disease that can be identified from the medical image may be included in the diagnostic image and output.

The location and/or form of the disease may be displayed in the form of a bounding box on the diagnostic image. Specifically, the diagnostic image is overlaid on the image area corresponding to the location of the disease in the medical image and may include a bounding box formed to have a shape corresponding to the shape of the disease. For example, when the diagnosis of a disease identified in a medical image is supernumerary teeth, the diagnostic image is overlaid on the location of the identified supernumerary tooth and may include a bounding box formed to include at least a portion of the shape of the identified supernumerary tooth. However, it is not limited to this.

FIG. 2 is a diagram illustrating an example of training data for a diagnostic model according to some embodiments of the present disclosure. Specifically, Figure 2 shows an example of a labeling image constituting training data for a periapical imaging diagnosis model.

The medical image processing apparatus 100 according to the present disclosure may include a plurality of diagnostic models for inputting and processing medical images. The medical image processing apparatus 100 according to the present disclosure may determine at least one diagnostic model among a plurality of diagnostic models based on the image type of the medical image and input the medical image into the determined diagnostic model. In this case, the diagnostic model can infer a diagnostic image including disease information about the medical image from the input medical image.

Here, each of the plurality of diagnostic models may be a model optimized for each different image type. Accordingly, based on the determined image type, the medical image processing device 100 may determine at least one diagnostic model optimized for the corresponding image type from among the plurality of diagnostic models as a model for processing the medical image.

To optimize the diagnostic model, the medical image processing device 100 according to the present disclosure may learn diagnostic models using training medical images of each optimized image type. Here, the learning medical image can be paired with a labeling image in which the disease in the learning medical image is manually marked to form learning data.

As an example, in the case of a panoramic image diagnosis model, a learning medical image including one or more panoramic images, and a learning image consisting of a pair of labeling images in which the diagnosis, form, and location of the disease present in each panoramic image are manually marked. Based on the data, it can be learned to infer a diagnostic image from a panoramic image type medical image.

As another example, in the case of a periapical image diagnosis model, a learning medical image including one or more periapical images, and the diagnosis name, form, and location of the disease present in each periapical image are manually marked as shown in Figure 2. Based on training data consisting of pairs of labeling images, it can be learned to infer a diagnostic image from a periapical image type medical image.

Referring to FIG. 2, the labeling image includes a bounding box 210 in which the form and location of the disease present in the given learning periapical image 200 are manually marked, and the coordinate values of the corresponding bounding box 210 (table on the right) 2 to 5 columns) and the diagnosis value of the disease present in the learning periapical image 200 (column 6 of the right table).

Meanwhile, the learning data may be configured to include one or more medical images for learning in which at least some of the age of the subject and the gender of the subject are different from each other. That is, medical images for learning according to the present disclosure can be collected for subjects of various ages and genders, from the deciduous dentition period to the mixed dentition period and permanent dentition period (or from infants to adults), and can be collected for subjects of a specific age and/or gender. Can be evenly selected for various ages and genders to prevent bias.

However, it is not limited to this, and medical images for learning may be selected to include more medical images for subjects of a specific age and/or gender. For example, in the case of supernumerary tooth disease, diagnosis is generally more difficult in the deciduous and mixed dentition stages than in the permanent dentition stage. Accordingly, medical images for learning may be selected to include more medical images of the deciduous dentition period and the mixed dentition period to improve the diagnostic accuracy of the diagnosis model's supernumerary teeth.

FIG. 3 is a diagram illustrating an example of a diagnostic model optimized for a panoramic image type according to some embodiments of the present disclosure.

Referring to Figure 3, the panoramic image diagnosis model inputs a medical image by adjusting it to a size of 512x256, and through the Darknet-53 architecture, it is converted from feature maps of three different sizes, such as 128x64x18, 64x32x18, and 32x16x18. It may include a YOLOv3 model that performs object detection (e.g., identification of disease). However, it is not limited to this, and the panoramic image diagnosis model may be implemented using any algorithm and/or neural network structure available for object detection, such as CNN and SSD (Single Shot Detector).

In response to the image type of the medical image being determined as a panoramic image type, the medical image processing device 100 according to the present disclosure may input the medical image into a panoramic image diagnosis model as shown in FIG. 3, and the panoramic image diagnosis model. Diagnostic images can be obtained from. The diagnostic image may include disease information regarding the disease identified in the medical image input to the model.

In Figure 3, the medical image determined to be a panoramic image type is adjusted to a size of 512x256 and input into the panoramic image diagnosis model (see top left), and a bounding box indicating the location and form of the superfluous disease identified in the medical image is displayed along with the disease name. An example of a diagnostic image overlaid on a medical image is output (see right) is shown.

Figure 4 is a diagram illustrating an example of a diagnostic model optimized for a periapical image type according to some embodiments of the present disclosure.

Referring to Figure 4, the periapical imaging model inputs a medical image by adjusting its size to 512x512, and performs object detection from feature maps of three different sizes, such as 64x64x18, 32x32x18, and 16x16x18, through the Darknet-53 architecture. It can include the YOLOv3 model. However, it is not limited to this, and the periapical imaging diagnosis model may be implemented using any algorithm and/or neural network structure available for object detection, such as CNN or SSD (Single Shot Detector).

In response to the image type of the medical image being determined as the periapical image type, the medical image processing device 100 according to the present disclosure may input the medical image into the periapical image diagnosis model as shown in FIG. 4, and the periapical image diagnosis model Diagnostic images can be obtained from. The diagnostic image may include disease information regarding the disease identified in the medical image input to the model.

In Figure 4, the medical image determined as the periapical image type is adjusted to a size of 512x512 and input into the periapical image diagnosis model (see upper left), and a bounding box indicating the location and form of the supernumerary disease identified in the medical image is displayed along with the disease name. An example of a diagnostic image overlaid on a medical image is output (see right) is shown.

5 is a flowchart of a medical image processing method according to some embodiments of the present disclosure.

Referring to FIG. 5, the medical image processing apparatus 100 of the present disclosure may first determine the image type of the medical image (S110).

The medical image processing device 100 of the present disclosure may first receive a medical image from an external server or terminal. In this case, the external server or terminal may be a server of a medical institution, an image acquisition device used in a medical institution, or a user terminal of a user who wishes to perform diagnosis and treatment based on medical images.

Additionally, the medical image processing device 100 according to the present disclosure can determine the image type of the received medical image. That is, the medical image in the present disclosure may include any type of image captured by any method for any body part, and the medical image processing device 100 determines which image type the received medical image belongs to. You can decide whether

In the present disclosure, the medical image is a dental radiology image, and in particular, the image type of the medical image is described as an example including at least one of a panoramic image type and a periapical image type. However, it is not limited to this, and the medical image according to the present disclosure may be a maxillary occlusal radiography image type, a CBCT image type, a head radiography image type, a bitewing image type, or a magnetic resonance imaging type.

The image type of a medical image may be determined on a rule basis based on the properties of each medical image (eg, image size ratio, image file metadata, etc.). However, it is not limited to this, and the image type of the medical image may be determined by a pre-learned classification model such as a clustering model, SVM model, or CNN model that uses features extracted from the medical image, or may be directly determined by the user. there is.

Next, the medical image processing device 100 of the present disclosure may input a medical image into a diagnostic model determined according to the image type (S120).

The medical image processing apparatus 100 according to the present disclosure may include a plurality of diagnostic models for inputting and processing medical images. The diagnostic model may infer a diagnostic image including disease information about the medical image from the input medical image.

Here, each of the plurality of diagnostic models may be a model optimized for each different image type. Accordingly, based on the determined image type, the medical image processing device 100 may determine at least one diagnostic model optimized for the corresponding image type from among the plurality of diagnostic models as a model for processing the medical image.

Next, the medical image processing apparatus 100 of the present disclosure may obtain a diagnostic image including disease information regarding the medical image from the diagnostic model (S130).

Here, the disease information is information inferred based on the medical image from the above-described diagnostic model, including whether or not the disease is identified in the medical image, the location of the disease identified in the medical image, the type of disease identified in the medical image, and It may include at least some of the diagnoses of the identified disease. The diagnostic image may be an image in which at least some of the above-described disease information is overlaid on the medical image.

Regarding whether the disease is identified, if the disease is not identified by the diagnostic model in the input medical image, the medical image processing device 100 may output the medical image itself as a diagnostic image. However, it is not limited to this, and the medical image processing device 100 may output a diagnostic image in which arbitrary graphics or text indicating that a disease is not identified in the medical image is overlaid on the medical image. Conversely, when a disease is identified by a diagnostic model in an input medical image, the medical image processing device 100 may output a diagnostic image in which at least part of the disease information of the identified disease is overlaid on the medical image.

Regarding the diagnosis name of the disease, the medical image processing device 100 may output a diagnosis image including the name of the disease identified by the diagnosis model in the medical image. In this case, the diagnosis of the disease may be displayed, for example, in an area on the diagnostic image adjacent to the location of the identified disease, or may be displayed in a preset area on the diagnostic image, such as the upper left corner. However, it is not limited to this.

If the medical image is a dental radiology image, as explained in this disclosure as an example, the diagnosis name of the disease may be, for example, “excess teeth.” Alternatively, the diagnosis of the disease may be a diagnosis of a disease that can be identified from dental radiological images such as dental caries, periodontal disease, or other diseases that occur in areas such as teeth and surrounding tissue, jawbone, temporomandibular joint, and maxillary sinus. However, it is not limited to this, and the diagnosis name for any disease that can be identified from the medical image may be included in the diagnostic image and output.

The location and/or form of the disease may be displayed in the form of a bounding box on the diagnostic image. Specifically, the diagnostic image is overlaid on the image area corresponding to the location of the disease in the medical image and may include a bounding box formed to have a shape corresponding to the shape of the disease.

For example, when the diagnosis of a disease identified in a medical image is supernumerary teeth, the diagnostic image is overlaid on the location of the identified supernumerary tooth and may include a bounding box formed to include at least a portion of the shape of the identified supernumerary tooth. However, it is not limited to this.

FIG. 6 is a block diagram illustrating a computing device that provides a medical image processing method according to some embodiments of the present disclosure.

Here, the computing device 10 that provides the medical image processing method may be the medical image processing device 100 described above, or may be a user terminal (not shown) that is communicatively connected to the medical image processing device 100. However, it is not limited to this.

Referring to FIG. 6, the computing device 10 according to the present disclosure includes one or more processors 11, a memory 12 that loads a program executed by the processor 11, and a storage 13 that stores the program and various data. ), and may include a communication interface 14. However, the above-described components are not essential for implementing the computing device 10 according to the present disclosure, so the computing device 10 may have more or less components than the components listed above. For example, the computing device 10 may further include an output unit and/or an input unit (not shown), or the storage 13 may be omitted.

When loaded into the memory 12, the program may include instructions that cause the processor 11 to perform methods/operations according to various embodiments of the present disclosure. That is, the processor 11 can perform methods/operations according to various embodiments of the present disclosure by executing instructions. A program consists of a series of computer-readable instructions grouped by function and executed by a processor.

The processor 11 controls the overall operation of each component of the computing device 10. The processor 11 includes at least one of a Central Processing Unit (CPU), Micro Processor Unit (MPU), Micro Controller Unit (MCU), Graphic Processing Unit (GPU), or any type of processor well known in the art of the present disclosure. It can be configured to include. Additionally, the processor 11 may perform operations on at least one application or program to execute methods/operations according to various embodiments of the present disclosure.

Memory 12 stores various data, instructions and/or information. Memory 12 may load one or more programs from storage 13 to execute methods/operations according to various embodiments of the present disclosure. The memory 12 may be implemented as a volatile memory such as RAM, but the technical scope of the present disclosure is not limited thereto.

Storage 13 can store programs non-temporarily. The storage 13 may be a 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 the like in the technical field to which this disclosure pertains. It may be configured to include any known type of computer-readable recording medium. The communication interface 14 may be a wired/wireless communication module.

The embodiments of the present disclosure described above are not only implemented through devices and methods, but may also be implemented through programs that implement functions corresponding to the configurations of the embodiments of the present disclosure or recording media on which the programs are recorded.

Although the embodiments of the present disclosure have been described in detail above, the scope of the rights of the present disclosure is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concept of the present disclosure defined in the following claims are also possible. It falls within the scope of rights.

Claims (20)

A medical image processing method performed by a computing device, comprising:
Determining the image type of the medical image,
Inputting the medical image into a diagnostic model determined according to the image type, and
Comprising the step of obtaining a diagnostic image including disease information regarding the medical image from the diagnostic model,
Medical image processing method.
In paragraph 1:
The medical image is,
Includes dental radiology imaging,
The image type of the medical image is,
Containing at least one of a panoramic image type and a periapical image type,
Medical image processing method.
In paragraph 1:
The diagnostic model is,
Among the plurality of diagnostic models, at least one diagnostic model optimized for the determined image type using a training medical image corresponding to the determined image type,
Medical image processing method.
In paragraph 3,
The diagnostic model is,
Learned to infer the diagnostic image from the input medical image based on learning data consisting of a pair of the learning medical image and a labeling image in which the disease in the learning medical image is manually marked,
Medical image processing method.
In paragraph 4:
The learning data is,
Containing one or more medical images for learning in which at least some of the age value of the subject and the gender value of the subject are different from each other,
Medical image processing method.
In paragraph 1:
The diagnostic model is,
It is a deep learning model based on an object detection model,
The object detection model is,
Including the YOLO model using the Darknet-based deep structure as the backbone,
Medical image processing method.
In paragraph 1:
The disease information is,
Containing at least some of the following: whether the disease is identified in the medical image, the location of the identified disease, the type of the identified disease, and the diagnosis of the identified disease,
Medical image processing method.
In paragraph 7:
The diagnostic image is,
An image in which at least some of the disease information is overlaid on the medical image,
Medical image processing method.
In paragraph 7:
The diagnostic image is,
A bounding box is overlaid on an image area corresponding to the location of the disease in the medical image and is formed to have a shape corresponding to the shape of the disease,
Medical image processing method.
In paragraph 7:
The diagnosis of the disease is:
Containing excess values,
Medical image processing method.
A medical image processing method performed by a computing device, comprising:
Determining the image type of the medical image,
Automatically determining at least one diagnostic model optimized for the determined image type among a plurality of diagnostic models, and
Including the step of inputting the medical image into the automatically determined diagnostic model and obtaining a diagnostic image including disease information related to the medical image,
Medical image processing method.
In paragraph 11:
The medical image is,
Includes dental radiology imaging,
The image type of the medical image is,
Containing at least one of a panoramic image type and a periapical image type,
Medical image processing method.
In paragraph 11:
Each of the plurality of diagnostic models is,
Learned to infer the diagnostic image from the input medical image based on learning data consisting of a pair of a training medical image of the corresponding image type and a labeling image in which the disease in the learning medical image is manually marked,
Medical image processing method.
In paragraph 11:
The at least one diagnostic model is:
It is a deep learning model based on an object detection model,
The object detection model is,
Including the YOLO model using the Darknet-based deep structure as the backbone,
Medical image processing method.
In paragraph 11:
The disease information is,
Containing at least some of the following: whether the disease is identified in the medical image, the location of the identified disease, the type of the identified disease, and the diagnosis of the identified disease,
Medical image processing method.
In paragraph 15:
The diagnostic image is,
A bounding box is overlaid on an image area corresponding to the location of the disease in the medical image and is formed to have a shape corresponding to the shape of the disease,
Medical image processing method.
In paragraph 15:
The diagnosis of the disease is:
Containing excess values,
Medical image processing method.
A medical image processing method performed by a computing device, comprising:
Determining the image type of the medical image,
When the image type is a panoramic image type, the medical image is input into a diagnostic model optimized for the panoramic image type, and when the image type is a periapical image type, the medical image is input into a diagnostic model optimized for the periapical image type. Steps to enter , and
Comprising: obtaining a diagnostic image including information about the redundancy of the medical image from the diagnostic model,
Medical image processing method.
In paragraph 18:
The diagnostic model is,
Learned to infer the diagnostic image from the input medical image based on learning data consisting of a pair of a training medical image of the corresponding image type and a labeling image in which the disease in the learning medical image is manually marked,
Medical image processing method.
In paragraph 18:
The step of acquiring the diagnostic image is,
When supernumerary tooth disease is identified in the medical image, acquiring the diagnostic image to include a bounding box that is overlaid on an image area corresponding to the location of the supernumerary tooth disease and has a shape corresponding to the shape of the supernumerary tooth disease; and
When the supernumerary disease is not identified in the medical image, comprising acquiring the medical image as the diagnostic image,
Medical image processing method.