Learning method for implants, determination method, and implant determination system therefor

Abstract

A learning method for implants, a determination method, and an implant determination system therefor are disclosed. The present invention applies an X-ray simulator to a 3D implant image to convert the 3D implant image into a 2D implant image, applies various image effects to augment the 2D implant image, and generates various implant learning models. In addition, the present invention determines placed implant model information from an actual patient image by using an implant learning model.

Description

Implant learning method, identification method, and the implant identification system

[0001] The present invention relates to image data processing technology, and more specifically, to technology for analyzing and processing image data.

[0002] Since the introduction of dental treatment procedures using implants, various implant models have been developed. Dental prostheses consist of a crown, which functions as a tooth, and an implant, which is embedded in the alveolar bone below the gums and acts as a tooth root. Furthermore, an abutment to support the crown and a fixing screw to connect the abutment may be additionally included. In other words, since the implant is embedded in the alveolar bone, dentists cannot directly inspect it with the naked eye and must visually recognize the implant through X-ray images.

[0003] Due to the lack of standardization for implant-related components, implants with varying specifications are currently being developed by different manufacturers. Consequently, interoperability between implant models is poor, requiring continuous awareness and retention of implant information for the long-term maintenance of implanted devices. It is difficult for patients to retain information about their implanted implant model over extended periods, and there are frequent instances where the clinic where the procedure was performed closes down or related medical records are lost. As a result, patients who received implants in the past frequently struggle with identifying implant model information, leading to difficulties in maintaining and repairing their implants. While this issue is addressed through the experience and trial-and-error of medical professionals, it requires a high level of expertise; therefore, there are limitations for medical practitioners with limited clinical experience to properly grasp this information.

[0004] Furthermore, since implants are intended for long-term use, periodic monitoring is necessary to ensure comfortable use over the long term. To monitor implants periodically, a method is currently used to assess the degree of bone loss in the surrounding bone using radiographs. However, due to the lack of automated tools to measure surrounding bone loss from radiographs, researchers and clinicians are currently measuring the extent of bone loss manually after taking radiographs, utilizing graphic programs within dedicated radiograph viewers.

[0005] In the case of such manual measurements, there is a problem in that the accuracy of surrounding bone loss is low due to measurement errors caused by the difference in ratio between the radiographic subject and the actual object. Furthermore, it is difficult to consider the measurement of the actual surrounding bone loss as objective due to errors arising from the operator's visual perception and manipulation skills. In other words, it is difficult to accurately measure the amount of surrounding bone loss because reproducibility and consistency are low depending on the operator's proficiency.

[0006] According to one embodiment, the present invention proposes an implant learning method and an implant discrimination system capable of generating an implant learning model by reflecting an X-ray simulator in a 3D implant image to convert it into a 2D implant image and applying various image effects to the 2D implant image.

[0007] In addition, we propose an implant identification method and an implant identification system that utilize an implant learning model including augmented implant images to more accurately identify implant model information and detect implant abnormalities such as peri-implantitis, regardless of the patient's shooting angle and image quality.

[0008] An implant learning method according to one embodiment includes the steps of converting a 3D implant image into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image, generating an augmented image by applying image effects to the 2D implant image, and generating an implant learning model for determining implant model information by learning at least one of the 2D implant image and the augmented image through an artificial neural network.

[0009] Image effects may include at least one of a light source vector, a direction vector, a pixel size, blurring, a gamma effect, a brightness effect, a contrast effect, a rotation effect, and a clipping effect.

[0010] The implant learning method may further include the step of generating a training image by changing at least one of the shooting angle and shooting position for a virtual phantom model.

[0011] The implant learning method may further include the step of segmenting a missing tooth region in a training image and the step of replacing a numbering region corresponding to the segmented missing tooth region with a 2D implant image generated by reflecting an X-ray simulator in a 3D implant image.

[0012] The implant learning method further includes a step of learning a generated implant learning model, and the learning step may include a step of segmenting a plurality of augmented 2D implant images, a step of extracting a plurality of implant feature vectors from the segmented images of the plurality of 2D implant images through a feature extraction model, a step of clustering the extracted plurality of implant feature vectors by implant type, a step of subdividing each of the clustered plurality of implant types into a plurality of detailed types, and a step of performing implant matching using the classified implant feature vectors.

[0013] The implant identification method includes the step of obtaining an implant learning model generated by learning a converted 2D implant image through an artificial neural network by reflecting an X-ray simulator into a 3D implant image, the step of inputting an actual patient image into the implant learning model, and the step of identifying the implant learning model from the actual patient image through the implant learning model and outputting implant model information.

[0014] The model information of the implant may include at least one of the manufacturer of the implant, the manufacturer's system, the length and diameter of the implant, and the number and spacing of the implant threads.

[0015] The step of outputting implant model information may include a step of matching the diameter, length, and rotation information of the implant between the implant segmented in the actual patient image and the 2D implant image, a step of extracting feature points of the two implants between the implant segmented in the actual patient image and the 2D implant image, a step of matching the shape of the external threads of the two implants using the extracted feature points, and a step of matching the internal shape of the two implants when the shape of the external threads of the two implants is the same.

[0016] In the step of outputting implant model information, the length and diameter of the implant in the actual patient image can be calculated using the magnification / reduction ratio between the implant in the actual patient image and the implant in the 2D implant image.

[0017] The implant identification method may further include a step of identifying abnormalities in the implant based on whether there is contact between the level seen in the actual patient image and the implant.

[0018] The step of determining abnormalities in the implant may include a step of determining it as normal if the implant is not in contact with the level seen in the actual patient image, and a step of determining it as abnormal and calculating the height of bone loss if the implant is in contact with the level seen and deviates by more than a predetermined angle.

[0019] The implant identification method may further include a step of recommending a tool for removing the implant, tailored to the implant manufacturer and system, if, in the actual patient image, the contact area between the bone level and the implant is located below the area where the abutment is attached from the top of the implant.

[0020] The implant identification method may further include a step of recommending antibiotic therapy or surgical treatment based on the height of bone loss if, in actual patient images, the contact area between the bone level and the implant is located above the area where the abutment is attached from the top of the implant.

[0021] The implant identification method may further include a step of providing implant model information to the patient management program by linking with the patient management program of the display device.

[0022] An implant identification system according to another embodiment includes: a learning device that converts a 3D implant image into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image and generates an augmented image by applying image effects to the converted 2D implant image, and generates an implant learning model for identifying implant model information by learning at least one of the 2D implant image or the augmented image through an artificial neural network; an identification device that inputs an actual patient image into the implant learning model that has completed training, identifies implant model information implanted in the actual patient image through the implant learning model, and outputs implant model information; and a display device that receives and displays implant model information from the identification device.

[0023] The present invention can generate an implant learning model for identifying various implants by converting a 3D implant image into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image, and by augmenting the 2D implant image by applying various image effects.

[0024] The present invention utilizes an implant learning model including an augmented implant image to more accurately determine implant model information and identify implant abnormalities such as peri-implantitis, regardless of the patient's shooting angle and image quality.

[0025] FIG. 1 is a diagram illustrating the configuration of an implant identification system according to one embodiment of the present invention.

[0026] FIG. 2 is a diagram illustrating the flow of an implant learning method according to an embodiment of the present invention.

[0027] FIG. 3 is a diagram illustrating the flow of an implant identification method according to an embodiment of the present invention.

[0028] FIG. 4 is a drawing illustrating a screen for generating a 2D implant image by reflecting an X-ray simulator in a 3D implant image according to an embodiment of the present invention.

[0029] FIG. 5 is a drawing illustrating a screen for generating an augmented image through various image processing according to an embodiment of the present invention.

[0030] FIG. 6 is a drawing illustrating a screen dividing an anatomical structure according to an embodiment of the present invention.

[0031] FIG. 7 is a drawing illustrating a screen for allocating numbering areas to teeth and implants of a learning image according to an embodiment of the present invention.

[0032] FIGS. 8 and 9 are drawings illustrating a screen for generating a learning video according to various embodiments of the present invention.

[0033] FIG. 10 is a drawing illustrating a screen for replacing a loss region of a learning image with a 2D implant image according to an embodiment of the present invention.

[0034] FIG. 11 is a diagram illustrating a 2D implant image for implant matching when training an implant learning model according to an embodiment of the present invention.

[0035] FIG. 12 is a diagram illustrating a screen for clustering a plurality of implant feature vectors by a plurality of implant types when training an implant learning model according to an embodiment of the present invention.

[0036] FIG. 13 is a drawing illustrating a screen that subdivides each of the clustered multiple implant types into multiple detailed types when training an implant learning model according to an embodiment of the present invention.

[0037] FIG. 14 is a diagram illustrating implant feature vectors during implant learning model training according to an embodiment of the present invention.

[0038] FIG. 15 is a drawing illustrating a screen for matching identified implant model information to an actual patient image according to an embodiment of the present invention.

[0039] FIG. 16 is a drawing illustrating a screen for determining an implant abnormality according to an embodiment of the present invention.

[0040] FIG. 17 is a drawing illustrating a screen providing recommendation information based on whether there is an abnormality in an implant according to one embodiment of the present invention.

[0041] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0042] In describing the embodiments of the present invention, if it is determined that a detailed description of known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description will be omitted. Furthermore, the terms described below are defined considering the functions in the embodiments of the present invention, and these may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification.

[0043] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. However, the embodiments of the present invention exemplified below may be modified in various different forms, and the scope of the present invention is not limited to the embodiments described below. The embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art to which this invention pertains.

[0044] FIG. 1 is a diagram illustrating the configuration of an implant identification system according to one embodiment of the present invention.

[0045] Referring to FIG. 1, the implant identification system (1) includes a learning device (11), a identification device (12), and a display device (13). The learning device (11) and the identification device (12) may be computing devices.

[0046] The learning device (11) generates an implant learning model for determining implant model information based on a learning image. For example, the learning device (11) can generate an implant learning model by converting a 3D implant image into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image, and by augmenting the 2D implant image by applying various image effects to the 2D implant image. The implant learning method of the learning device (11) will be described later with reference to FIG. 2.

[0047] The discrimination device (12) obtains an implant learning model that has been learned by the learning device (11) and uses the obtained implant learning model to determine implant model information in an actual patient image. Furthermore, it can determine whether there is an abnormality in the implant. The method of determining the implant of the discrimination device (12) will be described later with reference to FIG. 3.

[0048] When the learning device (11) generates an implant learning model through learning on a learning image, the discrimination device (12) can obtain the implant learning model from the learning device (11). The implant learning model may be used while stored in the learning device (11) or the discrimination device (12), or it may be stored in a separate external device such as an AI device.

[0049] The learning device (11) and the discrimination device (12) may each include a processor and a database. The learning device (11) and the discrimination device (12) may be in the form of a server.

[0050] A processor executes a program. A processor is, for example, a CPU (Central Processing Unit), a processing unit, an arithmetic unit, a processor, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor).

[0051] The database can store information necessary for learning and implant learning models generated based on the learning, and can provide them upon user request.

[0052] The display device (13) is an electronic device capable of running a patient management program. Electronic devices include PCs (Personal Computers), laptops, tablets, smartphones, etc. Patient management programs include viewer programs, image editing programs, etc. Additionally, it can be reflected in other general image processing programs.

[0053] The display device (13) displays information on the screen of the patient management program. For example, the display device (13) can receive and display implant model information identified using an implant learning model generated according to the learning of the learning device (11) from the identification device (12).

[0054] FIG. 2 is a diagram illustrating the flow of an implant learning method according to one embodiment of the present invention.

[0055] Referring to FIGS. 1 and 2, the learning device (11) converts the 3D implant image into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image (S210). An example of 2D implant image conversion will be described later with reference to FIG. 4. Here, the 3D implant image may include all training images generated to train the implant learning model, in addition to images taken of actual patients.

[0056] Next, the learning device (11) generates an augmented image by applying an image effect to a 2D implant image (S220). The image effect may include at least one of a light source vector, a direction vector, a pixel size, blurring, a gamma effect, a brightness effect, a contrast effect, a rotation effect, and a clipping effect. An example of generating an augmented image will be described later with reference to FIG. 5.

[0057] Next, the learning device (11) learns at least one of the 2D implant image and the augmented image through an artificial neural network to generate an implant learning model (S230). An example of generating an implant learning model will be described later with reference to FIGS. 12 to 14.

[0058] Furthermore, the learning device (11) can learn the generated implant learning model (S240).

[0059] The training of the implant learning model may include image segmentation, feature extraction, clustering, implant matching, etc. For example, in the implant learning model training step (S240), the learning device (11) may segment a plurality of augmented 2D implant images and extract a plurality of implant feature vectors from the segmented images of the plurality of 2D implant images through a feature extraction model. In addition, the learning device (11) may cluster the extracted plurality of implant feature vectors by implant type and subdivide each of the clustered plurality of implant types to classify them into a plurality of detailed types. Subsequently, the learning device (11) may perform implant matching using the classified implant feature vectors.

[0060] FIG. 3 is a diagram illustrating the flow of an implant identification method according to one embodiment of the present invention.

[0061] Referring to FIGS. 1 and FIGS. 3, the discrimination device (12) obtains an implant learning model generated by learning a converted 2D implant image through an artificial neural network by reflecting an X-ray simulator to a 3D implant image (S310).

[0062] Next, the discrimination device (12) inputs the actual patient image into the implant learning model that has completed training (S320).

[0063] Next, the discrimination device (12) identifies implant model information from an actual patient image through an implant learning model and outputs the identified implant model information (S330). The implant model information may include at least one of the manufacturer of the implant, the manufacturer's system, the length and diameter of the implant, and the number and spacing of the implant threads.

[0064] In the implant model information output step (S330), the discrimination device (12) can calculate the length and diameter of the implant in the actual patient image using the enlargement / reduction ratio between the implant in the actual patient image and the implant in the 2D implant image.

[0065] Next, the determination device (12) can determine an abnormality of the implant based on whether the level seen in the actual patient image is in contact with the implant (S340). At this time, the determination device (12) can determine it to be normal if the level seen in the actual patient image is not in contact with the implant. If the level seen and the implant are in contact and deviate by more than a predetermined angle, it can determine it to be abnormal / abnormal and calculate the bone loss height. An example of determining an implant abnormality will be described later with reference to FIGS. 16 and FIGS. 17.

[0066] Furthermore, the determination device (12) can recommend a tool for removing the implant in accordance with the implant manufacturer and system if, in the actual patient image, the contact area between the level and the implant is located below the area where the abutment is attached from the top of the implant.

[0067] Furthermore, the determination device (12) can recommend antibiotic therapy or surgical treatment according to the bone loss height if, in the actual patient image, the contact area between the level and the implant is located above the area where the abutment is attached from the top of the implant.

[0068] Furthermore, the discrimination device (12) can provide implant model information to the patient management program by linking with the patient management program of the display device (13).

[0069] FIG. 4 is a diagram illustrating a screen that generates a 2D implant image by reflecting an X-ray simulator into a 3D implant image according to one embodiment of the present invention.

[0070] Referring to FIGS. 1 and 4, the learning device (11) can generate a 2D implant image (420) by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image (410). The X-ray simulator is a tool that allows the shape of the implant or the shape of the abutment attached to the implant to be seen in the 2D implant image (420) by virtually irradiating X-rays (430) onto the 3D implant image (410) to generate the 2D implant image (420).

[0071] FIG. 5 is a diagram illustrating a screen that generates an augmented image through various image processing according to an embodiment of the present invention.

[0072] Referring to FIGS. 1 and FIGS. 5, the learning device (11) can generate an augmented image by applying image effects to an implant model of a 2D implant image, for example, the first model or the second model of FIG. 5. For example, the learning device (11) can apply a light source vector, a direction vector, pixel size adjustment, a blurring effect, a gamma effect, brightness, an effect, a contrast effect, a rotation effect, and a clipping effect. Only one vector and one effect may be provided, or they may be provided in combination.

[0073] For example, regarding the application of image effects, reference numeral 510 and reference numeral 520 represent augmented images processed by applying different values ​​to the light source vector, detector vector, gamma effect, and rotation effect, while the pixel size, blurring effect, brightness effect, and contrast effect are the same.

[0074] By applying various image effects in this way, a large amount of augmented data can be generated, allowing one to capture panoramic or oral sensor images without being constrained by various shooting angles or image quality.

[0075] FIG. 6 is a drawing illustrating a screen dividing an anatomical structure according to one embodiment of the present invention.

[0076] Referring to FIGS. 1 and FIGS. 6, the learning device (11) acquires a learning image (60) and segments anatomical structures in the acquired learning image (60). The anatomical structures may be teeth, bones, implants, etc. The learning image (60) may be a 2D panoramic image, a periapical radiographic image, etc.

[0077] The learning device (11) can divide anatomical structures using an implant learning model generated through AI learning on a learning image (60). For example, the learning device (11) can divide the implant (620), tooth (610), and bone level (630) in the learning image (60). Then, the learning device (11) can display the divided implant (620), tooth (610), and bone level (630) by masking each, and can also display the implant (620), tooth (610), and bone level (630) after integrating them and then masking them.

[0078] AI learning can be performed through the computation of artificial neural networks. Artificial neural networks may include Conventional Neural Networks (CNN), Recurrent Neural Networks (RNN), Deep Belief Networks, Restricted Boltzmann Machines, etc.

[0079] FIG. 7 is a drawing illustrating a screen for allocating numbering areas to teeth and implants of a learning image according to one embodiment of the present invention.

[0080] Referring to FIGS. 1 and FIGS. 7, numbering boxes (701, 702) can be generated by numbering individual teeth and implants in a learning image (70) in which teeth and implants are segmented. The numbering boxes (701, 702) represent areas in which the longest regions based on width and height for the segmented individual teeth and implants are respectively represented as rectangular prisms in the data.

[0081] FIGS. 8 and 9 are drawings illustrating a screen for generating a learning video according to various embodiments of the present invention.

[0082] Referring to FIGS. 1, 8 and 9, the learning device (11) can not only acquire a learning image but also virtually generate a learning image and then divide anatomical structures including teeth, bones, and implants based on the generated learning image.

[0083] The learning device (11) can generate various training images by changing the shooting angle or shooting position on a virtual phantom model. The training images may be dental images such as virtual panoramic data or periapical radiographic data.

[0084] For example, the learning device (11) can produce an anterior reduction effect by virtually generating an image (810) in which the patient bites the bite block deeply or places the virtual phantom model in front, as shown in FIG. 8. Alternatively, it can produce an anterior enlargement effect by virtually generating an image (820) in which the patient bites the bite block thinly or places the virtual phantom model behind.

[0085] As another example, the learning device (11) can generate a V-shaped image (910) by applying the effect of lowering the virtual phantom model as shown in FIG. 9. Alternatively, it can obtain a straight image (920) by applying the effect of raising the virtual phantom model as shown in FIG. 9.

[0086] As another example, the learning device (11) can obtain an image tilted to one side, either left or right, by giving the effect of tilting the virtual phantom model to take a picture.

[0087] In addition, the learning device (11) can generate various learning images by changing the tube voltage (kVp), tube current (mA), etc., for a virtual phantom model.

[0088] The learning device (11) can further augment data by adjusting image effects on the learning image, in addition to the acquired learning image. At this time, the image effects include brightness (Windowing Level), contrast (Windowing Width), black and white inversion (Invert), sharpness (Sharpen), etc.

[0089] FIG. 10 is a diagram illustrating a screen that replaces the loss region of a learning image with a 2D implant image according to one embodiment of the present invention.

[0090] Referring to FIG. 1 and FIG. 10, the training image (111) may have a missing tooth region. The training device (11) may divide the missing tooth region for the training image (111) and then replace the missing tooth region with a 2D implant image obtained by reflecting an X-ray simulator to a 3D implant image.

[0091] According to the method of replacing with a 2D implant image, the learning device (11) assigns a numbering area (1110) to the missing value area of ​​the learning image (111). Subsequently, the assigned numbering area (1110) is replaced with the same 2D implant image (1120) to generate a learning image (112) to which the 2D implant image (1120) has been added. For example, as shown in FIG. 10, the learning device (11) can create a horizontal*vertical rectangular model of the same horizontal*vertical '.png format 2D implant image (1120)' in the horizontal*vertical numbering area (1110) corresponding to the missing value area of ​​the learning image (110), and then add the 2D implant image (1120) to the numbering area (1110).

[0092] The learning device (11) can learn from various angles or positions by using a virtual learning image generated from a virtual phantom model and a 2D implant image, and by making the director vector or rotation vector different. In addition, since the correct answer regarding which implant is being learned is already known, the learning model can be created, so there is no need to determine the implant model information in the learning image of a patient with an actual implant, thereby eliminating the time consumption and improving the accuracy rate.

[0093] FIG. 11 is a diagram illustrating a 2D implant image for implant matching when training an implant learning model according to an embodiment of the present invention.

[0094] Referring to FIG. 1 and FIG. 11, the learning device (11) performs implant matching using feature points. For example, the learning device (11) can match two implants by using feature points of two implants between a training image and a 2D implant image generated by adding a 2D implant image to a virtual phantom image.

[0095] The learning device (11) can divide a 2D implant image into a box shape with a preset shape. For example, the learning device (11) can distinguish between an image with only an implant (1210) and an image with an abutment attached to the implant (1220, 1230), and divide the image into a rectangular box shape according to the size of the image.

[0096] FIG. 12 is a diagram illustrating a screen for clustering a plurality of implant feature vectors by a plurality of implant types when training an implant learning model according to an embodiment of the present invention.

[0097] Referring to FIGS. 1 and FIGS. 12, the learning device (11) generates a feature extraction model by contrast learning a plurality of augmented 2D implant images. The feature extraction model may be an artificial neural network-based model for extracting implant feature vectors of segmented images of a plurality of augmented 2D implant images. 'Contrastive learning' is to learn so that the distance between feature vectors becomes closer (similar) for implant images of the same type, and so that the distance between feature vectors becomes farther for implant images of different types.

[0098] The learning device (11) can cluster multiple implant feature vectors extracted using a feature extraction model generated through learning according to multiple implant types (see FIG. 12). Furthermore, the learning device (11) can subdivide each of the clustered multiple implant types to classify them into multiple detailed types (see FIG. 13).

[0099] More specifically, when a feature extraction model is generated, implant feature vectors of the same type of implant images have a similar distribution, while implant feature vectors of different types of implant images may have different distributions.

[0100] For example, as illustrated in FIG. 12, the learning device (11) can extract implant feature vectors for each of a plurality of segmented implant images (a, b, c, d, e, f) through a feature extraction model. The implant feature vectors corresponding to each of the first implant image (a: TS3), the second implant image (b: TS4), the third implant image (c: KS3), the fourth implant image (d: SS3), the fifth implant image (e: US2), the sixth implant image (f: US3), etc., can be distributed in the feature space (1300).

[0101] Next, the learning device (11) can group the first implant image (a: TS3) and the second implant image (b: TS4), which are implants of the same type, into a first group; the fifth implant image (e: US2) and the sixth implant image (f: US3), which are implants of different types and each have feature vectors of similar distribution, into a third group; and the third implant image (c: KS3) into a second group. That is, the learning device learns to have different distributions for each type (for example, implant types TS and KS are of the same type, while SS and US are of different types). Multiple implant feature vectors can be clustered by type according to distance (e.g., K-Means algorithm) or density (e.g., DBSCAN method).

[0102] FIG. 13 is a diagram illustrating a screen that subdivides each of the clustered multiple implant types into multiple detailed types when learning an implant learning model according to an embodiment of the present invention.

[0103] Referring to FIG. 1 and FIG. 13, the learning device (11) can cluster by classifying into detailed types with similar patterns from the implant feature vector input, since different patterns exist even within the same type. For example, as shown in FIG. 13, implant types TS3(a) and TS4(b) can be classified even within the first group.

[0104] The learning device (11) can cluster images with similar patterns through an image clustering engine or classify images by extracting features of images through an image classification engine including an image feature extraction network. At this time, at least one of the image clustering engine or the image classification engine including an image feature extraction network may be composed of an artificial neural network in which at least a portion is learned according to a machine learning algorithm.

[0105] FIG. 14 is a diagram illustrating implant feature vectors during implant learning model training according to one embodiment of the present invention.

[0106] Referring to FIG. 1 and FIG. 14, the learning device (11) can extract an implant feature vector for each of a plurality of segmented implant images. As shown in FIG. 14, the implant feature vector may include an implant external thread shape (1510), an implant external thread spacing (number) (1520), an implant external apex shape (1530), a distance from the top of the implant to the implant internal thread (1540), a distance from the top of the implant to the abutment attachment site (1550), an implant internal thread diameter (1560), an implant internal angle (1570), an implant external angle (1580), etc.

[0107] FIG. 15 is a diagram illustrating a screen that matches identified implant model information to an actual patient image according to one embodiment of the present invention.

[0108] Referring to FIGS. 1 and FIGS. 15, the discrimination device (12) can identify implant model information in actual patient images by using an implant learning model that clusters feature extraction vectors through virtual data augmentation and contrast learning. The implant model information may include the implant manufacturer, system, length and diameter of the implant.

[0109] For example, the identification device (12) can input an actual patient image (1600) into an implant learning model to obtain and output implant model information. Specifically, the implant learning model can divide anatomical structures including teeth, bones, and implants (1610) in the actual patient image (1600), and then determine implant model information including the manufacturer and system of the divided implants (1610) and output the determined implant model information.

[0110] Next, the discrimination device (12) matches two implants in the actual patient image (1600) and the 2D implant image (1620) as illustrated in FIG. 15. For example, the discrimination device (12) sets a segmented implant (1610) of the actual patient image (1600) as a region of interest and matches two implants in the 2D implant image (1620) corresponding to the implant model information output through the implant learning model with the set region of interest. When matching, the discrimination device (12) can match the diameter, length, rotation information, etc. of the two implants between the segmented implant (1610) of the actual patient image (1600) and the 2D implant image (1620).

[0111] Next, the discrimination device (12) extracts features of two implants between the segmented implant (1610) of the actual patient image (1600) and the 2D implant image (1620), and searches for corresponding feature point pairs by comparing the extracted features. Next, the discrimination device (12) uses the searched feature point pairs to primarily match the shape of the external screw threads (screw spacing, degree of protrusion) of the two implants, and if the shape of the external screw threads is the same, it can secondarily match the internal shape of the two implants.

[0112] Next, the determination device (12) can calculate the accurate length and diameter of the implant (1610) by using the magnification / reduction ratio (1630) between the divided implant (1610) of the actual patient image (1600) and the two implants of the 2D implant image (1620). For example, the determination device (12) can calculate the accurate length and diameter of the implant (1610) by matching the two implants by applying the magnification / reduction ratio of the actual patient image to the 2D implant image.

[0113] When a 3D implant is captured in a 2D panoramic view, the vertex direction of the implant may not be implanted parallel to the y-axis but may be tilted toward the z-axis. In contrast, the present invention learns a 2D implant image (1620) generated by reflecting an X-ray simulator at various angles onto a 3D implant image, so the implant can be accurately matched in 2D, and the accurate length and diameter of the implant can be determined through the magnification / reduction ratio (1630).

[0114] The determination device (12) can calculate the distance from the top of the implant to the part where the abutment is attached, using matching information as well as information on the manufacturer, system, diameter, and length of the implant.

[0115] FIG. 16 is a drawing illustrating a screen for determining an implant abnormality according to one embodiment of the present invention.

[0116] Referring to FIG. 1 and FIG. 16, the determination device (12) can determine abnormalities in the implant and surrounding teeth using the divided bone level after determining implant model information in the actual patient image.

[0117] The determination device (12) can determine the contact area between the bone level (1710) and the implant (1610) after dividing the bone and the implant (1610) in the actual patient image (1600). The bone level (1710) must have a gentle curve with respect to the implant (1610). If it deviates by more than a predetermined angle, the determination device (12) can determine that there is bone loss with respect to the implant (1610) and calculate the contact area between the implant (1610) and the bone level (1710). In the case of consecutive implants, for example, in the order of tooth, implant 1, implant 2, implant 3, and tooth, the determination device (12) can determine that implants 1 and 2 have bone loss if the bone level deviates by more than a certain angle from implant 1, maintains a straight line up to implant 2, and deviates by more than a certain angle from implant 3.

[0118] As shown in (a), if the level (1710) and the implant (1610) are not in contact in the actual patient image (1600), the determination device (12) determines it to be normal and does not calculate the bone loss height. In contrast, as shown in (b), if the level (1710) and the implant (1610) are in contact and deviate by more than a predetermined angle, it determines it to be abnormal / abnormal with a possibility of peri-implantitis and calculates the bone loss height.

[0119] FIG. 17 is a drawing illustrating a screen providing recommendation information based on whether there is an abnormality in an implant according to one embodiment of the present invention.

[0120] Referring to FIGS. 1, 16 and 17, as shown in (a), the detection device (12) can recommend a tool to remove the implant (1610) in accordance with the implant manufacturer and system if, in an actual patient image, the contact area between the present level (1710) and the implant (1610) is located below the abutment attachment area (1820) from the top of the implant (1810). This is because if the present level (1710) is located below the abutment attachment area (1820) from the top of the implant (1810), there is a risk that the implant will fracture.

[0121] When recommending an implant removal tool, the determination device (12) can recommend an implant removal tool based on the information value of the height of the part where the abutment is attached at the top of the implant.

[0122] If, as illustrated in (b), the level (1710) is located above the abutment attachment site (1820) from the top of the implant (1810), the determination device (12) calculates the bone loss height using enlargement / reduction ratio information and detected implant length information. Subsequently, if the calculated bone loss height is a predetermined height, for example, 2 mm or less, the determination device (12) may recommend antibiotic therapy.

[0123] In contrast, if the bone loss height is greater than a predetermined height and the level (1710) is located above the abutment attachment site (1820) from the top of the implant (1810), the determining device (12) may recommend surgical treatment (smoothing the surface of the exposed implant and bone grafting).

[0124] Furthermore, the identification device (12) can provide tool information for removing the implant, bone loss height information, and implant model information (manufacturer, system, length, diameter) in conjunction with the patient management program of the display device (13).

[0125] The present invention has been described above with reference to its embodiments. Those skilled in the art will understand that the present invention may be implemented in modified forms without departing from the essential characteristics of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

Claims

1. In a method for implanting a computing device, A step of converting into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image; A step of generating an augmented image by applying image effects to the above 2D implant image; and A step of generating an implant learning model for determining implant model information by learning at least one of the above 2D implant image and augmentation image through an artificial neural network; An implant learning method characterized by including 2. In claim 1, the image effect An implant learning method characterized by including at least one of a light source vector, a direction vector, a pixel size, blurring, a gamma effect, a brightness effect, a contrast effect, a rotation effect, and a clipping effect.

3. In claim 1, the implant learning method A step of generating a training image by changing at least one of the shooting angle and shooting position for a virtual phantom model; An implant learning method characterized by further including 4. In Paragraph 3, the implant learning method A step of segmenting the missing value region in a training image; and A step of replacing the numbering region corresponding to the segmented missing tooth region with a 2D implant image generated by reflecting an X-ray simulator into a 3D implant image; An implant learning method characterized by further including 5. In claim 1, the implant learning method It further includes a step of training the generated implant learning model; and The learning stage A step of segmenting multiple augmented 2D implant images; A step of extracting a plurality of implant feature vectors from segmented images of the plurality of 2D implant images through a feature extraction model; A step of clustering multiple extracted implant feature vectors by implant type; A step of subdividing each of the clustered multiple implant types to classify them into multiple detailed types; and A step of performing implant matching using classified implant feature vectors; An implant learning method characterized by including 6. In a method for determining implants in a computing device, A step of obtaining an implant learning model generated by training a 2D implant image, converted by reflecting an X-ray simulator into a 3D implant image, through an artificial neural network; A step of inputting actual patient images into a trained implant learning model; and A step of determining implant model information from actual patient images through an implant learning model and outputting the determined implant model information; A method for determining an implant characterized by including 7. In Clause 6, the model information of the implant is A method for identifying an implant characterized by including at least one of the manufacturer of the implant, the manufacturer's system, the length and diameter of the implant, and the number and spacing of the implant threads.

8. In claim 7, the step of outputting implant model information A step of matching the diameter, length, and rotation information of the implant between the segmented implant in the actual patient image and the 2D implant image; A step of extracting feature points of two implants between a segmented implant in an actual patient image and a 2D implant image; A step of matching the shape of the external screw threads of two implants using extracted feature points; and A step of matching the internal shape of two implants when the shape of the external screw threads of the two implants is identical; A method for determining an implant characterized by including 9. In claim 7, the step of outputting implant model information An implant identification method characterized by calculating the length and diameter of an implant in an actual patient image using the magnification / reduction ratio between the implant in the actual patient image and the implant in the 2D implant image.

10. In Clause 7, the method for determining an implant A step of determining implant abnormalities based on whether the level seen in the actual patient image is in contact with the implant; An implant identification method characterized by further including 11. In Clause 10, the step of determining an abnormality of the implant A step of determining normal if the level seen in the actual patient image and the implant are not in contact; and A step of determining abnormality and calculating the bone loss height if the bone level and the implant are in contact and deviate by more than a predetermined angle; A method for determining an implant characterized by including 12. In Clause 10, the method for determining an implant In actual patient images, if the contact area between the bone level and the implant is located below the abutment attachment point from the top of the implant, a step of recommending an implant removal tool tailored to the implant manufacturer and system; An implant identification method characterized by further including 13. In Clause 10, the method for determining an implant In actual patient images, if the contact area between the bone level and the implant is located above the area where the abutment is attached from the top of the implant, a step of recommending antibiotic therapy or surgical treatment depending on the height of bone loss; An implant identification method characterized by further including 14. In Clause 7, the method for determining an implant A step of providing implant model information to a patient management program by linking with a patient management program of a display device; An implant identification method characterized by further including 15. A learning device that converts a 3D implant image, in which an actual implant is placed, into a 2D implant image by reflecting an X-ray simulator that changes the angle or position of X-ray irradiation on the 3D implant image as a learning image, generates an augmented image by applying image effects to the 2D implant image, and generates an implant learning model for determining implant model information by learning at least one of the 2D implant image or the augmented image through an artificial neural network; A determination device that inputs an actual patient image into a trained implant learning model, determines implant model information implanted in the actual patient image through the implant learning model, and outputs implant model information; and A display device that receives and displays implant model information from a determination device; An implant identification system characterized by including

Images