Apparatus and method for providing sinus image and learning method thereof

Abstract

The apparatus for providing a maxillary sinus image according to the embodiments includes a first artificial neural network model for separating a region of interest of a portion of a coronal plane from an image of a coronal plane of a head, a second artificial neural network model for extracting a two-dimensional maxillary sinus image in a region of interest of a horizontal plane corresponding to a position of the region of interest of the coronal plane with respect to a horizontal image of the head, and an image processing unit for generating a three-dimensional sinus image by using the two-dimensional sinus image.

Description

Technical Field

[0001] This disclosure relates to an apparatus for providing images of the maxillary sinus, a method for providing images of the maxillary sinus performed by the apparatus, and a learning method for the apparatus. Background Technology

[0002] As is well known, dental implants are dental prostheses used to restore damaged teeth by placing an implant (fixture) in the maxilla or mandible and fixing the artificial tooth to the implant. For a successful implantation, it is crucial that the implant is firmly fixed to the bone. Therefore, when the bone tissue at the intended implantation site is thin or insufficient in thickness, the implantation surgery is performed after strengthening the bone tissue through methods such as autologous bone grafting, artificial bone grafting, or allogeneic bone grafting.

[0003] However, when performing implantation surgery below the maxillary sinus, in many cases, the bone thickness is insufficient relative to the implant length, making implant placement difficult. Here, when implantation surgery is performed with insufficient bone thickness, problems such as implant dislodgement during chewing or fracture of the surrounding bone tissue may occur because the placed implant cannot be adequately supported by the surrounding bone tissue.

[0004] To prevent the aforementioned problems, maxillary sinus lift is performed when implantation surgery is carried out below the maxillary sinus in cases where the maxillary bone thickness is insufficient. Maxillary sinus lift involves elevating the maxillary sinus membrane, which forms on the inner side of the maxillary sinus, and then grafting bone onto the sinus to ensure space for implant placement. However, maxillary sinus lift is a highly skilled procedure, and in recent years, the application of image processing technology in digital dentistry to perform maxillary sinus lifts has presented significant challenges. Summary of the Invention

[0005] Technical issues

[0006] The embodiments relate to an apparatus and method for providing maxillary sinus images, which are capable of distinguishing axial plane regions of interest from axial plane images of the head based on the location of regions of interest in the coronal plane distinguished from coronal plane images of the head, and then determining the maxillary sinus region within the distinguished axial plane regions of interest.

[0007] The embodiments also relate to a learning method for a device for providing maxillary sinus images, the learning method allowing a first artificial neural network model and a second artificial neural network model to learn such that the first artificial neural network model distinguishes a coronal region of interest from a coronal image of the head, and the second artificial neural network model distinguishes an axial region of interest from an axial image of the head, and determines the maxillary sinus region within the axial region of interest.

[0008] The purpose of this disclosure is not limited to the purposes described above, and other unmentioned purposes should be clearly understood by one of ordinary skill in the art to which this disclosure pertains based on the following description.

[0009] Technical solutions

[0010] A first aspect provides an apparatus for providing maxillary sinus images, the apparatus comprising: a first artificial neural network model configured to distinguish a portion of a coronal region of interest from a coronal image of the head; a second artificial neural network model configured to extract a two-dimensional maxillary sinus image from an axial region of interest corresponding to the location of the coronal region of interest for an axial image of the head; and an image processor configured to generate a three-dimensional maxillary sinus image using the two-dimensional maxillary sinus image.

[0011] The second aspect provides a method for providing a maxillary sinus image performed by a device for providing a maxillary sinus image, the method comprising: distinguishing a portion of a coronal region of interest from an input coronal image of the head; for an axial image of the head, extracting a two-dimensional maxillary sinus image from an axial region of interest corresponding to the location of the coronal region of interest; and generating a three-dimensional maxillary sinus image using the two-dimensional maxillary sinus image.

[0012] A third aspect provides a computer-readable recording medium storing a computer program, wherein the computer program includes instructions that, when executed by a processor, allow the processor to perform a method for providing a maxillary sinus image, the method comprising: distinguishing a portion of a coronal region of interest from an input coronal image of the head; for an axial image of the head, extracting a two-dimensional maxillary sinus image from an axial region of interest corresponding to the location of the coronal region of interest; and generating a three-dimensional maxillary sinus image using the two-dimensional maxillary sinus image.

[0013] A fourth aspect provides a learning method for a device for providing maxillary sinus images, the learning method comprising: distinguishing a portion of a coronal region of interest from an input coronal image of the head, causing a first artificial neural network model to learn using the learned coronal image of the head as first learning input data and using a masked image in which the coronal region of interest in the learned coronal image of the head is masked as first label data; and extracting a two-dimensional maxillary sinus image from a cropped image including an axial image of the head that corresponds to the location of the coronal region of interest, causing a second artificial neural network model to learn using the cropped image as second learning input data and using a masked image in which the maxillary sinus region in the cropped image is masked as second label data.

[0014] Beneficial effects

[0015] According to an embodiment, an axial plane region of interest is distinguished from a head axial plane image based on the location of the coronal plane region of interest distinguished from the coronal plane image, and then the maxillary sinus region is determined from the distinguished axial plane region of interest. According to this embodiment of the present disclosure, since cells present in the head axial plane image are accurately distinguished and automatically excluded based on the location of the coronal plane region of interest, such as maxillary sinus-like cells located outside the maxillary sinus region, it is fundamentally preventable that similar cells are included in the final provided maxillary sinus image. Attached Figure Description

[0016] Figure 1 and Figure 2 An example is shown of a set of learning data learned by an artificial neural network model in a maxillary sinus image providing device for distinguishing maxillary sinus regions from axial view images of the head, wherein, Figure 1 It is the head axis surface image that is used as input data for this set of learning data, and Figure 2 It is a masked image that serves as the label data for this set of learning data, in which the maxillary sinus region is masked.

[0017] Figure 3 This includes those already used Figure 1 and Figure 2 An exemplary view of a maxillary sinus image provided by an artificial neural network model that learns from a set of learning data, which can generate a three-dimensional maxillary sinus image based on an input axial plane image of the head.

[0018] Figure 4 It is through the use of Figure 1 and Figure 2 An exemplary view of two-dimensional maxillary sinus images distinguished from axial plane images of the head by an artificial neural network model that learns from a set of training data.

[0019] Figure 5 This is a block diagram of a maxillary sinus image providing device according to one embodiment of the present disclosure.

[0020] Figure 6 It is used to describe Figure 5 The flowchart illustrates the learning method of the first and second artificial neural network models in the maxillary sinus image providing device. This learning method allows the maxillary sinus image providing device to accurately distinguish two-dimensional maxillary sinus images using coronal and axial images of the head.

[0021] Figure 7 It is used to describe by Figure 5 The flowchart illustrates the method for providing maxillary sinus images performed by the shown maxillary sinus image providing device.

[0022] Figures 8 to 11 It is a cross-sectional image or a three-dimensional image of the head used to describe a maxillary sinus image providing method performed by a maxillary sinus image providing device according to one embodiment of the present disclosure. Detailed Implementation

[0023] The advantages and features of this disclosure, as well as methods for achieving these advantages and features, will become apparent from the following detailed description of embodiments with reference to the accompanying drawings. However, this disclosure is not limited to the embodiments disclosed below and can be implemented in various different forms. The embodiments provided herein are intended to complete this disclosure and fully inform those skilled in the art of the scope of this disclosure. Furthermore, this disclosure is limited only by the scope of the claims.

[0024] After a brief description of the terminology used herein, this disclosure will be described in detail.

[0025] As used herein, the terminology selected has been chosen to reflect the functionality of this disclosure, and is generally widely used. However, these terms may vary depending on the intent or customary practice of those skilled in the art, the emergence of new technologies, etc. Furthermore, in some cases, the terminology may be arbitrarily chosen by the applicant, and in such cases, the meaning of the term will be described in detail in the relevant section of this disclosure. Therefore, the terminology used herein should be defined based on its meaning and its context throughout this disclosure, rather than simply on its name.

[0026] Throughout the specification, when a component is described as "including" another component, unless otherwise specified, that component may also include another component, rather than excluding other components.

[0027] Furthermore, the term "component" as used herein refers to a software or hardware component, such as a Field-Programmable Gate Array (FPGA) or Application-Specific Integrated Circuit (ASIC), and a "component" plays a certain role. However, the meaning of "component" is not limited to software or hardware. A "component" can be configured to reside in addressable memory or to operate one or more processors. Therefore, examples of "components" include components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, subroutines, program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided in components and "components" can be combined into a smaller number of components and "components," or it can be separated into additional components and "components."

[0028] In the following, embodiments of this disclosure will be described in detail with reference to the accompanying drawings to allow those skilled in the art to readily implement this disclosure. Furthermore, in the drawings, portions irrelevant to the description have been omitted to clearly illustrate the disclosure.

[0029] Maxillary sinus lift involves elevating the sinus membrane located medially to the maxillary sinus and then grafting bone onto the sinus to ensure space for implant placement. It is a highly skilled procedure. To support maxillary sinus lift surgery, images of the maxillary sinus obtained by differentiating the sinus region from medical head imaging are required.

[0030] Meanwhile, among various methods for generating and providing maxillary sinus images based on input medical head images, the following method can be considered, in which the maxillary sinus image providing device is configured to include an artificial neural network model, which is pre-learned so that the artificial neural network model can accurately distinguish the maxillary sinus region from the input medical head image, and to provide the maxillary sinus image through the learned maxillary sinus image providing device.

[0031] Figure 1 and Figure 2 An example of a set of learning data for distinguishing maxillary sinus regions from axial images of the head is shown in a maxillary sinus image providing device, learned by an artificial neural network model. Figure 1 It is a head axis surface image that is used as input data for learning, and Figure 2 It is a masked image of the labeled data as a set of learning data, in which the maxillary sinus region (20) is masked.

[0032] According to including pre-learning Figure 1 and Figure 2 A device for providing maxillary sinus images to an artificial neural network model based on a set of learning data can be used to infer that the artificial neural network model can output a two-dimensional maxillary sinus image obtained by distinguishing the maxillary sinus region from the input axial plane image of the head, and can provide image processing such as overlaying multiple two-dimensional maxillary sinus images. Figure 3 The image shown is a three-dimensional image of the maxillary sinus.

[0033] However, in the human head, there are multiple maxillary sinus-like units in regions other than the maxillary sinus region. Specifically, such as... Figure 4 As shown, when capturing axial plane images of the head across the entire head, and the size of the axial plane images is large while the area of ​​the maxillary sinus region is relatively small, noise (N) beyond the maxillary sinus region can be included at multiple sites in the two-dimensional maxillary sinus image generated by the artificial neural network model from the input axial plane images of the head. In this case, as... Figure 3 As in the example, the three-dimensional maxillary sinus image that can be ultimately provided by the maxillary sinus image providing device may also include various types of noise (N) other than the maxillary sinus region (30).

[0034] An embodiment of the maxillary sinus image providing apparatus according to this disclosure distinguishes axial plane regions of interest from a head axial plane image based on the location of a coronal plane region of interest distinguished from a coronal plane image of the head, and then determines the maxillary sinus region within the distinguished axial plane region of interest. In this way, noise present in the head axial plane image, such as maxillary sinus-like units located outside the maxillary sinus region, can be accurately distinguished and automatically excluded based on the location of the coronal plane region of interest.

[0035] Figure 5 This is a block diagram of a maxillary sinus image providing device (500) according to one embodiment of the present disclosure.

[0036] Reference Figure 5 The maxillary sinus image providing device (500) includes a first artificial neural network model (510), a second artificial neural network model (520), and an image processor (530). Furthermore, the maxillary sinus image providing device (500) may also include a data receiver (540) and / or a data provider (550).

[0037] The data receiver (540) is a device for receiving images, and receives a coronal image of the head including at least a portion of the maxillary sinus region or an axial image of the head including at least a portion of the maxillary sinus region.

[0038] A coronal image of the head is a cross-sectional image in the coronal plane direction, and an axial image of the head is a cross-sectional image in the axial plane direction. As an example, both coronal and axial images of the head can be images captured by a radiographic imaging device using the same sagittal-axial-coronal coordinate system. For instance, coronal and axial images of the head can be images of teeth captured using the same cone-beam computed tomography (CBCT) scanning technique.

[0039] The data receiver (540) can receive coronal and axial images of the head from another module or device, but this disclosure is not limited thereto. As an example, the data receiver (540) can receive CBCT images. Coronal and axial images of the head can be generated from the CBCT images.

[0040] For example, the data receiver (540) may include a communication module that can receive image data through a communication channel or a serial interface that can receive image data as input through a communication port, and may receive coronal and / or axial images of the head through the communication module or the serial interface.

[0041] The image processor (530) is a processor for generating new images based on images provided from the data receiver (540). For example, the image processor (530) may include a central processing unit (CPU), a graphics processing unit (GPU), a microcontroller unit (MCU), or a dedicated processor for performing methods according to embodiments of this disclosure.

[0042] The image processor (530) according to embodiments of this disclosure can use multiple artificial neural network models to obtain 3D images with minimized noise. The multiple artificial neural network models include a first artificial neural network model (510) and a second artificial neural network model (520). The first artificial neural network model (510) is an artificial neural network model for generating coordinate information corresponding to a region of interest based on a coronal plane image of the head, and the second artificial neural network model (520) is an artificial neural network model for obtaining a masked axial plane image using the generated coordinate information. The first artificial neural network model (510) uses cross-sectional images in the coronal plane direction as learning data, and the second artificial neural network model (520) uses cross-sectional images in the axial plane direction as learning data. Furthermore, the multiple data provided to the input layers of the two artificial neural network models differ in terms of cross-sectional direction attributes. Additionally, the first artificial neural network model (510) and the second artificial neural network model (520) can be learned using convolutional neural networks.

[0043] More specifically, the first artificial neural network model (510) can distinguish at least one coronal region of interest from the input head coronal image. The first artificial neural network model (510) is pre-learned to distinguish the coronal region of interest from the head coronal image. For example, the first artificial neural network model (510) is pre-learned using a first set of learning data, and the first set of learning data has a head coronal image as input data and an image incorporating (annotated) the coordinate information of the region of interest as label data.

[0044] The first artificial neural network model (510), which has already been trained, can respond to the input of a head coronal image and support the generation of coordinate information of the region of interest corresponding to the input head coronal image. For example, the first artificial neural network model (510) can extract multiple candidate regions that can be distinguished as coronal regions of interest from the head coronal image, and can determine at least one candidate region as a coronal region of interest based on the comparison of the positions of the coronal coordinate values ​​of the multiple candidate regions or the comparison of the positions of the central axes of the axial plane.

[0045] The coordinate information of the region of interest generated using the first artificial neural network model (510) as described above is used to provide input data for the second artificial neural network model (520). As an example, the image processor (530) can use the coordinate information of the region of interest generated using the first artificial neural network model (510) to generate a cropped image. The cropped image is a cross-sectional image in the axial plane direction, and the normal direction of the cropped image is not parallel to the image used as input data for the first artificial neural network model, and in some cases, orthogonal to it. Here, the coordinate information used to generate the cropped image is the sagittal-axial-coronal coordinate information of the coronal region of interest.

[0046] The maxillary sinus image providing device (500) can obtain coordinate information for each coronal image of the head from multiple coronal images, and the image processor (530) can use this coordinate information to obtain multiple cropped images from head CT data. Each cropped image has a normal in the axial plane direction.

[0047] More specifically, the image processor (530) utilizes a first artificial neural network model (510) to output sagittal-axial-coronal coordinate information of the coronal region of interest distinguished from the head coronal image. Furthermore, the image processor (530) can obtain the sagittal and axial coordinate values ​​of the axial region of interest based on the sagittal and coronal coordinate values ​​of the coronal region of interest, and can distinguish the axial region of interest from the head axial image. The axial region of interest forms the basis for generating the cropped image.

[0048] In this way, the image processor (530) generates a cropped image including the axial plane region of interest in the head axial plane image based on the coordinate information of the coronal plane region of interest distinguished by the first artificial neural network model (510), and provides the generated cropped image to the second artificial neural network model (520).

[0049] The image processor (530) generates a three-dimensional maxillary sinus image using a two-dimensional maxillary sinus image provided by a second artificial neural network model (520), and provides the generated three-dimensional maxillary sinus image to a data provider (550). The image processor (530) can generate a three-dimensional maxillary sinus image by superimposing multiple two-dimensional maxillary sinus images output by the second artificial neural network model (520) through image processing.

[0050] Here, the second artificial neural network model (520) extracts a two-dimensional maxillary sinus image from a cropped image including an axial region of interest provided by the image processor (530), and provides the extracted two-dimensional maxillary sinus image to the image processor (530). The second artificial neural network model (520) can be pre-learned to distinguish maxillary sinus regions from the cropped image including the axial region of interest. For example, the second artificial neural network model (520) can be pre-learned using a second set of learning data, which may include the cropped image including the axial region of interest as learning input data and may include a masked image in which the maxillary sinus region in the cropped image is masked as label data. The data provider (550) provides the three-dimensional maxillary sinus image generated by the image processor (530) to the outside. For example, the data provider (550) may include a communication module that can transmit three-dimensional maxillary sinus image data through a communication channel, a serial interface that can output three-dimensional maxillary sinus image data through a communication port, or a display device that can visualize the three-dimensional maxillary sinus image.

[0051] Figure 6 It is used to describe in Figure 5 The flowchart illustrates the learning methods of the first artificial neural network model (510) and the second artificial neural network model (520) in the maxillary sinus image providing device (500), which are used to allow the maxillary sinus image providing device (500) to accurately distinguish two-dimensional maxillary sinus images using coronal and axial images of the head.

[0052] The first artificial neural network model (510) and / or the second artificial neural network model (520) can be pre-learned using any computing device. For learning, a first set of learning data required for learning the first artificial neural network model (510) and a second set of learning data required for learning the second artificial neural network model (520) are provided.

[0053] The first set of learning data may include learned coronal images of the head as learning input data, and head coronal images incorporating (annotated) regions of interest from the learned coronal images of the head as label data. As an example, the label data used for learning the first artificial neural network model (510) may have coordinate information of the coronal regions of interest, and the image processor (530) may obtain the coordinate information based on CT data according to the output of the first artificial neural network model (510). The image processor (530) may use the coordinate information obtained based on the coronal images to determine the regions of interest in the axial plane images.

[0054] The second set of learning data may include cropped images containing axial plane regions of interest as part of the head axial plane image as learning input data, and may include masked images in which the maxillary sinus region in the cropped images is masked as label data.

[0055] By utilizing a first artificial neural network model and a second artificial neural network model, a maxillary sinus image providing device according to one embodiment of the present disclosure predetermines the region of interest from the coronal plane image instead of immediately masking the image in the axial plane direction, and thus has the advantage of preventing noise caused by masking unnecessary regions.

[0056] Reference Figure 6 The maxillary sinus image providing device (500) enables the first artificial neural network model (510) to learn using the prepared first set of learning data to distinguish the coronal region of interest from the head coronal images (S610).

[0057] Using the first artificial neural network model (510), the maxillary sinus image providing device (500) can later obtain the coordinate information of the coronal plane region of interest. The coordinate information of the coronal plane region of interest can also be used as the coordinate information of the axial plane region of interest corresponding to the coronal plane region of interest.

[0058] The maxillary sinus image providing device (500) enables the second artificial neural network model (520) to learn using the prepared second set of learning data to distinguish the maxillary sinus region from the cropped image of the head axial plane image (S620).

[0059] Figure 7 It is used to describe by Figure 5 The flowchart illustrates the maxillary sinus image provisioning method performed by the maxillary sinus image provisioning device (500). Figures 8 to 11This refers to a cross-sectional image or a three-dimensional image of the head used to describe a maxillary sinus image providing method performed by a maxillary sinus image providing device (500) according to one embodiment of the present disclosure. Hereinafter, the maxillary sinus image providing method performed by the maxillary sinus image providing device (500) according to an embodiment will be described in detail with reference to the accompanying drawings.

[0060] The maxillary sinus image providing device (500) uses a first artificial neural network model (510) to distinguish a portion of the coronal plane region of interest from the head coronal plane image and generates sagittal-axial-coronal coordinate information of the distinguished coronal plane region of interest (S710).

[0061] The maxillary sinus region is included in the coronal plane region of interest distinguished by the first artificial neural network model (510), which has been pre-learned to distinguish the coronal plane region of interest that includes the maxillary sinus region. However, since there are many units similar to the maxillary sinus in regions other than the maxillary sinus region in the human head, in order to accurately distinguish the maxillary sinus region, the maxillary sinus region and similar regions can be accurately distinguished from each other.

[0062] Figure 8 This is an example of a coronal image captured by a low-dose CBCT imaging device, and Figure 9 Examples show the maxillary sinus region (A) and a similar region (B) of the mandibular region containing units similar to the maxillary sinus in a coronal image. The first artificial neural network model (510) can extract both the similar region (B) and the maxillary sinus region (A) from the head coronal image, and can select only the maxillary sinus region (A) while discarding the similar region (B). For example, the first artificial neural network model (510) can extract multiple candidate regions (A) and (B) that can be distinguished as regions of coronal interest from the head coronal image, and determine at least one candidate region as a region of coronal interest (A) based on a positional comparison of the coronal coordinate values ​​of the multiple candidate regions (A) and (B) or a positional comparison based on the central axis of the axial plane. For example, when performing a positional comparison based on coronal coordinate values, two candidate regions with relatively large coronal coordinate values ​​can be determined as regions of coronal interest (A). Alternatively, when performing a position comparison based on the central axis of the axial plane, candidate regions located below the central axis of the axial plane in multiple candidate regions (A) and (B) can be excluded, and candidate regions located above the central axis of the axial plane can be identified as coronal region of interest (A).

[0063] The image processor (530) can determine whether there are any undifferentiated coronal regions of interest in the head coronal images provided by the data receiver (540) to the first artificial neural network model (510) of the image processor (530) (S720). If there are undifferentiated coronal regions of interest in the head coronal images, the image processor (530) can repeat step (S710) until the coronal regions of interest in all the head coronal images provided by the data receiver (540) to the first artificial neural network model (510) of the image processor (530) are differentiated. Then, if the coronal regions of interest in all the head coronal images provided by the data receiver (540) to the first artificial neural network model (510) of the image processor (530) are differentiated, the image processor (530) can perform step (S730).

[0064] In this manner, the image processor (530) receives an axial plane image of the head and receives sagittal-axial-coronal coordinate information indicating the location of the coronal region of interest from a first artificial neural network model (510). The image processor (530) distinguishes the axial plane region of interest in the head axial plane image based on the three-dimensional coordinate values ​​of the coronal region of interest distinguished by the first artificial neural network model (510) (S730).

[0065] The image processor (530) generates a cropped image including the distinguished axial plane regions of interest, and provides the generated cropped image to the second artificial neural network model (520) (S740). For example, the image processor (530) may provide the second artificial neural network model (520) with... Figure 10 The image shown is a cropped image of the axial region of interest (ROI).

[0066] Here, the image processor (530) can obtain the sagittal and axial coordinate values ​​of the axial region of interest from the sagittal and coronal coordinate values ​​of the coronal region of interest provided by the first artificial neural network model (510). Since the head coronal and axial images are captured by a radiographic imaging device using the same sagittal-axial-coronal coordinate system, when a predetermined location in the head is included in both the coronal and axial regions of interest, the sagittal coordinate values ​​of the predetermined location in the coronal region of interest are the same as those in the axial region of interest, and the axial coordinate values ​​of the predetermined location in the coronal region of interest are the same as those in the axial region of interest. Therefore, the image processor (530) can distinguish the axial region of interest in the head axial image based on the location information of the coronal region of interest and can generate a cropped image that includes the distinguished axial region of interest.

[0067] The second artificial neural network model (520) distinguishes the maxillary sinus region from a cropped image including the region of interest in the axial plane of the head provided by the image processor (530), and provides a two-dimensional maxillary sinus image associated with the distinguished maxillary sinus region to the image processor (S750).

[0068] The image processor (530) can determine whether there are any cropped images in the cropped images provided to the second artificial neural network model (520) that do not generate a two-dimensional maxillary sinus image (S760). If there are cropped images that do not generate a two-dimensional maxillary sinus image, the image processor (530) can repeat steps (S730) to (S750) for all cropped images until a two-dimensional maxillary sinus image is generated. Then, for all cropped images provided to the second artificial neural network model (520) of the image processor (530), if a two-dimensional maxillary sinus image is generated, the image processor (530) can execute step (S770).

[0069] Then, the image processor (530) uses the two-dimensional maxillary sinus image provided by the second artificial neural network model (520) to generate a three-dimensional maxillary sinus image. For example, the image processor (530) can generate a three-dimensional maxillary sinus image by performing image processing and superimposing multiple two-dimensional maxillary sinus images output by the second artificial neural network model (520). Furthermore, the image processor (530) provides the generated three-dimensional maxillary sinus image to the data provider (550) (S770).

[0070] The data provider (550) provides an external three-dimensional maxillary sinus image generated by the image processor (530). For example, the data provider (550) can transmit the three-dimensional maxillary sinus image data via a communication channel, output the three-dimensional maxillary sinus image data via a communication port, or visualize the three-dimensional maxillary sinus image on a display device. For example, the data provider (550) can make... Figure 11 The three-dimensional image of the maxillary sinus is shown in the visualization.

[0071] Figure 3 This is an example of a 3D maxillary sinus image generated using only axial plane images of the head, and Figure 11 This is a three-dimensional maxillary sinus image generated using coronal and axial images of the head, according to embodiments of this disclosure. By comparison... Figure 3 and Figure 11 It can be seen that, with Figure 3 In comparison, Figure 11 The noise located outside the maxillary sinus region (30) was significantly less.

[0072] Meanwhile, each step in the learning method of the maxillary sinus image providing method and / or maxillary sinus image providing device according to the above embodiments can be implemented in a computer-readable recording medium that records a computer program including instructions for performing the steps.

[0073] As described above, according to one embodiment of this disclosure, an axial plane region of interest is distinguished from a head axial plane image based on the location of the region of interest in the coronal plane from the coronal plane image, and then the maxillary sinus region is determined within the distinguished axial plane region of interest. According to this embodiment of the disclosure, since noise present in the head axial plane image, such as maxillary sinus-like units located outside the maxillary sinus region, is accurately distinguished and automatically excluded based on the location of the coronal plane region of interest, it is possible to fundamentally prevent similar units from being included in the final provided maxillary sinus image.

[0074] The combination of steps in each flowchart appended herein can be executed by computer program instructions. Since the computer program instructions can be embedded in the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate means for performing the functions described in the steps of each flowchart. Since the computer program instructions can also be stored in a computer-usable or computer-readable recording medium that can support a computer or other programmable data processing apparatus to implement the functions in a particular manner, the instructions stored in the computer-usable or computer-readable recording medium can also produce an article of manufacture containing instruction means for performing the functions described in the steps of each flowchart. Since the computer program instructions can also be embedded in a computer or other programmable data processing apparatus, the processes executed by the computer through a series of operational steps performed in the computer or other programmable data processing apparatus, and the instructions executed by the computer or other programmable data processing apparatus, can also provide steps for performing the functions described in the steps of each flowchart.

[0075] Furthermore, each step may represent a portion of a module, segment, or code comprising one or more executable instructions for performing a specific logic function. It should also be noted that in some alternative implementations, the mentioned functions may be executed in a different order. For example, two steps shown as being executed consecutively may in reality be executed substantially simultaneously, or in some cases, they may be executed in reverse order depending on their respective functions.

[0076] The above description is merely an illustrative description of the technical spirit of this disclosure, and those skilled in the art can make various modifications and changes without departing from the basic nature of this disclosure. Therefore, the embodiments disclosed herein are used to describe, not limit, the technical spirit of this disclosure, and the scope of the technical spirit of this disclosure is not limited by the embodiments. The scope of protection of this disclosure should be interpreted by the following claims, and all technical spirit within the scope of the claims should be interpreted as included within the scope of the rights of this disclosure.

Claims

1. An apparatus for providing images of the maxillary sinus, the apparatus comprising: The first artificial neural network model is configured to distinguish partial coronal regions of interest from head coronal images; The second artificial neural network model is configured to: extract a two-dimensional maxillary sinus image from a cropped image for a head axial plane image, the cropped image including an axial plane region of interest corresponding to the location of the coronal plane region of interest; as well as An image processor configured to generate a three-dimensional maxillary sinus image using the two-dimensional maxillary sinus image. The image processor is configured to: provide the cropped image to the second artificial neural network model based on the three-dimensional coordinates of the coronal region of interest, the cropped image including the axial region of interest in the head axial image; and The coronal image and the axial image of the head are captured by a radiographic imaging device using the same sagittal-axial-coronal coordinate system. The first artificial neural network model is configured to output sagittal-axial-coronal coordinate information of the coronal region of interest distinguished from the coronal image of the head; and The image processor is configured to: obtain the sagittal and axial coordinate values ​​of the axial region of interest from the sagittal and coronal coordinate values ​​of the coronal region of interest, and distinguish the axial region of interest from the head axial image.

2. The apparatus according to claim 1, wherein, The first artificial neural network model uses learned head coronal images as learning input data and uses a masked image combining the coronal region of interest in the learned head coronal images as label data for learning.

3. The apparatus according to claim 1, wherein, The second artificial neural network model uses a cropped image including the axial plane region of interest as learning input data and a masked image in which the maxillary sinus region in the cropped image is masked as label data for learning.

4. The apparatus according to claim 1, wherein, The first artificial neural network model extracts multiple candidate regions that can be distinguished as the coronal region of interest from the head coronal image, and determines at least one candidate region as the coronal region of interest based on the comparison of the positions of the coronal coordinate values ​​of the multiple candidate regions or the comparison of the positions of the central axis of the axial plane.

5. A method for providing a maxillary sinus image, performed by a device for providing a maxillary sinus image, the method comprising: The first artificial neural network model is used to distinguish a portion of the coronal region of interest from the input head coronal image and output the sagittal-axial-coronal coordinate information of the coronal region of interest. For the axial plane image of the head, a two-dimensional maxillary sinus image is extracted from the cropped image using a second artificial neural network model, the cropped image including the axial plane region of interest corresponding to the position of the coronal plane region of interest; as well as A three-dimensional maxillary sinus image is generated using the two-dimensional maxillary sinus image. The method further includes: providing the cropped image to the second artificial neural network model based on the three-dimensional coordinate values ​​of the coronal region of interest, wherein the cropped image includes the axial region of interest in the head axial image. The coronal image and the axial image of the head are captured by a radiographic imaging device using the same sagittal-axial-coronal coordinate system. The method further includes: obtaining the sagittal and axial coordinate values ​​of the axial region of interest from the sagittal and coronal coordinate values ​​of the coronal region of interest, and distinguishing the axial region of interest from the head axial image.

6. A computer-readable recording medium for storing a computer program, wherein, The computer program includes instructions that, when executed by a processor, enable the processor to perform a method for providing maxillary sinus images, the method comprising: The first artificial neural network model is used to distinguish a portion of the coronal region of interest from the input head coronal image and output the sagittal-axial-coronal coordinate information of the coronal region of interest. For the axial plane image of the head, a two-dimensional maxillary sinus image is extracted from the cropped image using a second artificial neural network model. The cropped image includes an axial plane region of interest corresponding to the location of the coronal region of interest. A three-dimensional maxillary sinus image is generated using the two-dimensional maxillary sinus image. The method further includes: providing the cropped image to the second artificial neural network model based on the three-dimensional coordinate values ​​of the coronal region of interest, wherein the cropped image includes the axial region of interest in the head axial image. The coronal image and the axial image of the head are captured by a radiographic imaging device using the same sagittal-axial-coronal coordinate system. The method further includes: obtaining the sagittal and axial coordinate values ​​of the axial region of interest from the sagittal and coronal coordinate values ​​of the coronal region of interest, and distinguishing the axial region of interest from the head axial image.

7. A learning method for a device for providing images of the maxillary sinus, the learning method comprising: From an input head coronal image, a portion of the coronal region of interest is distinguished. A first artificial neural network model uses the learned head coronal image as first learning input data and a masked image of the coronal region of interest in the learned head coronal image as first label data for learning. The first artificial neural network model is trained to output the sagittal-axial-coronal coordinate information of the coronal region of interest distinguished from the head coronal image. A two-dimensional maxillary sinus image is extracted from a cropped image of the axial plane region of interest, which corresponds to the location of the coronal plane region of interest in the head axial plane image. The second artificial neural network model uses the learned cropped image as second learning input data and uses the masked image of the maxillary sinus region in the cropped image as second label data for learning. The coronal image and the axial image of the head are captured by a radiographic imaging device using the same sagittal-axial-coronal coordinate system. The cropped image is generated by obtaining the sagittal and axial coordinate values ​​of the axial region of interest from the sagittal and coronal coordinate values ​​of the coronal region of interest, and distinguishing the axial region of interest from the head axial image.

Images