Personal identification system and personal identification method using oral x-ray image

Abstract

Position data defining first enclosed regions (CA1 to CA6) of a plurality of known patients is accumulated in a database (20). The first enclosed regions are regions set on the basis of first landmarks (LM) for respective alveoli of each patient and each enclosed by a group of coordinates of the first landmarks belonging to one of a plurality of jaw sections obtained by dividing each of upper and lower jaws according to curvature positions along the horseshoe shape of the jaw. At the time of personal identification of a subject such as a corpse, second landmarks (LM) are set for respective alveoli of natural teeth in each of the upper and lower jaws of the subject in an X-ray image of the subject. On the basis of the second landmarks, a plurality of second enclosed regions (CA1' to CA6') are set, each of which is enclosed by a group of coordinates of the second landmarks belonging to one of a plurality of jaw sections obtained by dividing each of the upper and lower jaws according to curvature positions along the horseshoe shape of the jaw. The degree of similarity between at least one of the plurality of second enclosed regions and a first enclosed region in the database that corresponds to the at least one second enclosed region in terms of position is analyzed on a per-enclosed region basis. On the basis of the degree of similarity, the age or age range of the subject, or identification information of the subject is estimated.

Description

Personal identification system and method using oral x-ray images

[0001] The present invention relates to a personal identification system and method that can estimate age information and personal information of an unidentified subject based on morphological changes over time in the alveolar bone of the subject, as information indicating one aspect of personal identification, from an oral X-ray image (panoramic X-ray image or CT image) of the subject.

[0002] Hideko Fujimoto, one of the inventors of the present application, has previously published papers on the methods described in Non-Patent Documents 1 and 2. The personal identification methods described in these papers are also proposed as a method of personal identification in Patent Documents 1 and 2.

[0003] The idea behind this personal identification is based on the unique shape of the alveolar bone, which is largely influenced by not only congenital factors but also acquired factors such as infection with caries-causing bacteria and periodontal disease bacteria. Patent Document 2 therefore presents a method for improving objective accuracy so that it can also be used in periodontal disease testing. Specifically, the method involves having a computer acquire images of the subject's upper and lower jaws in chronological order from a modality. Furthermore, the computer interactively or automatically identifies a set of coordinates of multiple landmarks, including the alveolar crest and alveolar floor, for each of the acquired upper and lower jaw images. Furthermore, the computer outputs information regarding the progression of alveolar bone morphology based on the set of coordinates of the multiple landmarks identified for each upper and lower jaw image.

[0004] A specific example of the configuration of this information output is as follows: The computer divides the landmark coordinate group for each of the upper and lower jaw images into multiple regions and calculates the similarity between the multiple upper and lower jaw images for each region using Procrustes analysis. Furthermore, the computer calculates the average similarity based on the similarity calculated for each region and identifies the alveolar bone morphology for each region. The alveolar bone condition identified for each region is output via a display or the like in a manner corresponding to the respective progress status. This method is called the IDOL method (dental personal identification method).

[0005] The IDOL method described in the above-mentioned Patent Document 2 and Non-Patent Document 2 has attracted attention as a method that suggests that it is possible to provide information that estimates the age (or age range) of an unidentified individual, such as an unidentified person, based on an index of the average secular changes in the alveolar bone of the entire Japanese population.

[0006] Patent application 2019-201595 Patent application 2022-168622

[0007] "A novel method for landmark-based personal identification on panoramic dental radiographic and computed tomographic images" H. Fujimoto, T. Hayashi, M. Iino; Journal of Forensic Radiology and Imaging 7 21-27 December 2016 "Implementation of a personal identification system using alveolar bone images" H. Fujimoto, K. Kimura-kataoka, H. Kanayama, K. Kitamori, Y. Kurihara, D. Zangpo, H. Takeshita; Forensic science international 343(11548) February 2023

[0008] However, even with the IDOL method disclosed in Patent Document 2 and Non-Patent Document 2, no specific method for estimating age (or its range) has yet been proposed.

[0009] The present invention has been made in consideration of the above circumstances, and its main purpose is to provide a personal identification system using oral X-ray images that can provide information indicating one aspect of a subject's personal identification, such as information indicating future predictions related to morphological changes in the subject's alveolar bone over time, or conversely, can estimate the age information of a subject whose age is unknown.

[0010] To achieve the above object, in one aspect, there is provided a personal identification system based on oral X-ray images that can acquire information specific to a subject whose age is known from information on X-ray images that capture the upper and lower jaws of the subject. This personal identification system may be functionally installed in a dental examination system that uses X-ray images (CT images, panoramic images, etc.).

[0011] In one aspect, the personal identification system includes a first landmark setting means for setting a first landmark for each alveolus of natural teeth in each of the upper and lower jaws of a subject in an X-ray image of the subject, a first closed area setting means for setting, based on the first landmark for each alveolus, a plurality of first closed areas enclosed by a group of coordinates of the first landmarks belonging to each of a plurality of jaw regions into which each of the upper and lower jaws is divided according to the position of the horseshoe-shaped curve of the jaw, a database that can be referenced for personal identification and a database storage means for storing position data defining the plurality of first closed areas in the database, and a database storage means for storing position data defining the plurality of first closed areas in the database when acquiring information unique to the subject. a second landmark setting means for setting second landmarks for each of the tooth alveoli; a second closed area setting means for setting, based on the second landmarks for each of the tooth alveoli, a plurality of second closed areas that are closed by a group of coordinates of the second landmarks that belong to each of a plurality of jaw regions into which each of the upper and lower jaws is divided according to the position of the horseshoe-shaped curvature of the jaw; a similarity analysis means for analyzing, for each closed area, the similarity between at least one of the plurality of second closed areas obtained from the subject and a first closed area in the database that is positionally equivalent to the at least one second closed area; and an estimation means for estimating, based on the similarity, the age or age range of the subject or the identification information of the subject as the unique information.

[0012] According to another aspect, a personal identification system includes a first segmentation means for segmenting each tooth of the upper and lower jaws of a subject in an X-ray image of the subject, a first spectrum analysis means for analyzing the relationship between pixel value information of each tooth of the upper and lower jaws segmented by the first segmentation means and the position of each tooth in the upper and lower jaws as a first spectrum for each region of the upper and lower jaws, a database that can be referenced for personal identification and for constructing the first spectrum for each region analyzed by the first spectrum analysis means in the database, and a database construction means for constructing the first spectrum for each region of the upper and lower jaws of the subject in the X-ray image of the subject when acquiring information unique to the subject. a second spectrum analysis means for analyzing the relationship between pixel value information of each tooth of the upper and lower jaws segmented by the second segmentation means and the position of each tooth in the upper and lower jaws as a second spectrum for each part of the upper and lower jaws; a search means for screening the first spectra stored in the database to search for the first spectrum that shows the highest matching rate with the second spectrum; and an estimation means for estimating personal information of the subject as the unique information from the personal identification state of the subject that shows the first spectrum that shows the highest matching rate searched for by the search means.

[0013] According to the various aspects described above, it is possible to estimate a subject's unique information, such as age or age range, through a similarity analysis between closed regions using landmarks set in the alveoli based on morphological changes in the alveoli over time. Furthermore, it is possible to estimate personal information through a spectrum similarity analysis based on pixel value information for each of the subject's teeth. Therefore, by further developing the IDOL method described in Patent Document 2 and Non-Patent Document 2, it is possible to provide a system using oral X-ray images that can more precisely estimate personal information.

[0014] In the accompanying drawings: Fig. 1 is a block diagram showing an overview of an inspection system using oral X-ray images equipped with a personal identification system according to an embodiment of the present disclosure. Fig. 2 is a flowchart showing selection from three personal identification functions executed by a CPU (processor) of the inspection system. Fig. 3 is a flowchart explaining database updating for estimating a subject's age or age range as one piece of personal identification information in the selection process shown in Fig. 2. Fig. 4 is a flowchart explaining processing when estimating a subject's age or age range as one piece of personal identification information is specified in the selection process shown in Fig. 2. Fig. 5 is an explanatory diagram for explaining the processing for estimating a subject's age or age range. Fig. 6 is a table explaining some of the items stored in a database used for estimating a subject's age or age range. Fig. 7 is a table explaining the estimation of a subject's age or age range. Fig. 8 is a flowchart explaining database updating for estimating a subject's personal information itself (such as name) as one piece of personal identification information in the selection process shown in Fig. 2. Fig. 9 is a flowchart explaining processing when estimating a subject's personal information itself (such as name) as one piece of personal identification information is specified in the selection process shown in Fig. 2. FIG. 10 is a schematic diagram for explaining the estimation of the subject's personal information itself (such as name). FIG. 11 is a flowchart outlining the processing of a verification example performed to verify the accuracy of the estimation of the subject's personal information itself (such as name). FIG. 12 is a diagram illustrating mutually compared panoramic images and similarity score information as part of the verification example. FIG. 13 is a diagram illustrating mutually compared panoramic images and similarity score information as another part of the verification example. FIG. 14 is a block diagram showing an overview of an examination system using oral X-ray images according to a modified example. FIG. 15 is a flowchart showing an overview of a diagnostic process (including AI generation process) according to a modified example. FIG. 16 is a diagram schematically illustrating the state of a periodontal pocket.

[0015] First Embodiment A first embodiment of a personal identification system and method based on an oral cavity X-ray image of a subject according to the present invention will be described with reference to FIGS.

[0016] 1 shows an examination system 10 as a personal identification system that functionally includes this personal identification system and performs a personal identification method using the functions of this system. This examination system 10 is a system that examines oral diseases of a subject based on X-ray images of the subject's oral cavity, and an overview of this system is shown in FIG.

[0017] This examination system 10 has a personal identification function that identifies (estimates) information related to the individual subject. Of course, this examination system 10 also has diagnostic functions such as individual diagnosis and screening, which examines the subject's oral diseases from oral X-ray images, as is commonly performed, and provides the results to medical professionals such as dentists or doctors. This individual diagnosis also includes the diagnostic configuration described in the developed example below. In other words, this examination system 10 is provided as an integrated system that includes a personal identification function while being a modality that assumes such diagnostic functions.

[0018] Therefore, in the following explanation, the test system 10 will be described mainly with respect to its personal identification function, but it may of course be provided as a personal identification device specialized for the personal identification function.

[0019] In this embodiment, the subject refers to a patient undergoing a diagnosis in the case of typical individual diagnosis or screening, and often refers to a deceased person's corpse for identification in the case of postmortem examination. Furthermore, the information targeted for personal identification is information that can lead to the identification of such a corpse, such as name, gender, and age range (or age range). In the case of personal identification, it is, of course, best to be able to identify the name, as this information can lead to crucial information for identifying the individual, such as address and place of origin. On the other hand, at the scene of a crime or accident, or even the scene of a natural disaster, even knowing an age range, such as 30s or 40s, is considered extremely effective for final identification. Therefore, in the testing system 10 (functionally including a personal identification system and a personal identification method) according to this embodiment, the personal identification (identification) of a corpse and age (or range) estimation are collectively referred to as personal identification.

[0020] Furthermore, the reason for using oral X-ray images as the basis for personal identification is that many people have panoramic X-ray images or CT images of their oral cavity taken for dental treatment or other purposes before death. In other words, because medical institutions have a large collection of oral X-ray images (panoramic images, CT images), it is easy to construct reference data from before death. This reference data is stored in a database updated via a communication network during regular medical treatment. The larger the amount of data in the database, the higher the estimation accuracy and matching accuracy when referencing the X-ray images of the subject to be identified in the database. In particular, in this embodiment, at least a portion of the identification process is performed using machine learning (e.g., deep learning) using an AI (artificial intelligence) learning model, so the data used to train the learning model can be data stored in a database during regular medical treatment.

[0021] Another advantage of using X-ray images of the oral cavity is that the jaws of the oral cavity tend to remain in the same condition as at the time of death more often than other parts of the body. Of course, this varies depending on the cause of death and the time that has passed since death.

[0022] The inspection system 10 (personal identification system) according to this embodiment is configured to be able to selectively perform one or more of the following three personal identification functions. When using this inspection system 10, the personal identification functions may be selectively performed, or these functions may be selected to be performed in combination, making this inspection system 10 also an integrated personal identification system. This selection function is shown in the flowchart of FIG. 2, which will be described later.

[0023] - Estimating the age of a corpse (subject) using the alveolar bone that is visible in an X-ray image - Identifying the individual of a corpse (subject) using the alveolar bone that is visible in an X-ray image - Identifying the individual of a corpse (subject) using each tooth that is visible in an X-ray image.

[0024] Of course, depending on the condition of the body, only one or two of the identifications may be selectively performed.

[0025] The X-ray images used for personal identification of corpses are those obtained by postmortem panoramic X-ray photography or CT photography using an X-ray CT scanner. In the case of CT photography, the whole body, including the head, is often photographed, and the 3D image of the oral cavity is reprojected onto a 2D X-ray image.

[0026] DETAILED DESCRIPTION OF EMBODIMENTS An inspection system 10 (having the function of a personal identification system) shown in FIG. 1 will be described below with reference to the accompanying drawings.

[0027] As shown in FIG. 1, the inspection system 10 includes an image processing device 12, an operating device 14 communicably connected to the image processing device 12, a display device 16, a report output device 18, and a database 20.

[0028] The image processing device 12 is connected to a medical modality 22 that performs X-ray imaging via a communication line such as the Internet, to which DICOM is applied. Therefore, this medical modality 22 is a medical device that performs panoramic X-ray imaging and X-ray CT imaging in a medical facility, and these X-ray images are transmitted to the image processing device 12 via a communication line such as the Internet. The X-ray imaging data, i.e., X-ray image data, along with accompanying information such as patient information (patient ID, patient name (if the patient's identity is unknown, an indication to that effect)), examination information (including examination ID, examination date and time, examination area, and type of modality), are digitally stored in a database 20 via the image processing device 12 each time they are transmitted.

[0029] Of course, the inspection system 10 may be used in a stand-alone format, and the X-ray image data may be stored in a recording medium and provided to the inspection system 10. Also, if the X-ray image data taken in a medical facility is in analog quantities, i.e., X-ray film, it is converted into digital quantities by a digitizer and provided to the image processing device 12. This digital quantity conversion may be performed in the image processing device 12. The database 20 may be separately constructed on a communication line such as the Internet.

[0030] The image processing device 12 has a CPU (Central Processing Unit) 30 which is responsible for the core of the calculation function. This CPU 30 is connected via an internal bus 32 to an input / output interface 34, a temporary storage memory 36 consisting of RAM for temporarily storing data, a read-only main storage device 38 consisting of ROM, and an auxiliary storage device 40 which is also stored in a writable and readable RAM.

[0031] A program for personal identification processing according to this embodiment is stored in the form of source code in the main memory device 38 (recording medium). Therefore, upon startup, the CPU 30 as a processor reads various predetermined programs from the main memory device 38 into a work area (not shown) of the CPU 30. As shown in the example of FIG. 1, these programs include a normal diagnostic processing program, a personal identification processing program, and, in cases where some of these programs are implemented by AI (artificial intelligence), a machine learning model generation program and an AI processing program.

[0032] When the personal identification processing program is read, the CPU 30 executes calculations for each step of the source code written in the personal identification processing program, thereby performing processes such as reading, diagnosing, and outputting reports on patient X-ray images, as well as processing for collecting information related to the personal identification of unidentified bodies.

[0033] The "normal" part of the normal diagnostic processing program refers to the processing of patients who have come for dental treatment or consultation, rather than examinations that target corpses.

[0034] In this embodiment, the former normal diagnostic process can be performed in the same manner as existing processes, and therefore the following description will focus on the latter process of identifying the individual body.

[0035] <Personal Identification Processing> The personal identification processing will be described with reference to the accompanying drawings.

[0036] As mentioned above, this personal identification processing has three types of modes: - Processing to estimate the age of a corpse using the alveolar bone reflected in an X-ray image - Processing to identify the individual of a corpse using the alveolar bone reflected in an X-ray image (1) - Processing to identify the individual of a corpse using each tooth reflected in an X-ray image (2), and is configured to be able to selectively perform these processes.

[0037] Specifically, the CPU 30 performs the selection process shown in FIG. 2 . First, the CPU 30 interactively interacts with the user via the operating device 14 to determine whether to perform a "cadaver age estimation process using alveolar bone (hereinafter, referred to as "age estimation process" as needed)" (FIG. 2, step S101). If the determination is YES, i.e., if the CPU 30 determines that the age estimation process should be performed, the CPU 30 interactively determines whether to update the database 20 for the age estimation process (step S103). Note that the database 20 is set so that the age estimation process is not performed for subjects in the primary dentition and mixed dentition stages (step S103, NO). If the determination in step S103 is YES, the CPU 30 executes a database update process as a subroutine (step S105), as described below.

[0038] Furthermore, if the result of step S101 is NO, if the result of step S103 is NO, or if the processing of step S105 is completed, the CPU 30 enters "processing for personal identification of the corpse using the alveolar bone reflected in the X-ray image (hereinafter referred to as personal identification process (1) as necessary)." This personal identification process (1) will be described later, and its characteristic is that it obtains information for identifying the identity of the corpse (such as name and age) based on the changes over time of the alveolar bone, that is, from the morphology of the alveolar bone on X-ray images taken while the corpse was alive to the changes on X-ray images taken after death.

[0039] First, the CPU 30 interactively determines whether or not to perform the personal identification process (1) (step S109). Note that the personal identification process (1) is set not to be performed for subjects in the primary dentition stage and the mixed dentition stage (step S109, NO). If the determination in step S109 is YES, the CPU 30 then determines whether or not to update the database 20 for the personal identification process (1) (step S111). If this determination is again YES, the database 20 is updated as described below (step S113). On the other hand, if the determination in step S111 is NO, the CPU 30 proceeds to step S115, where the personal identification process (1) itself is performed as described below.

[0040] Furthermore, when the determination in step S109 is NO, the determination in step S111 is NO, or the processing in step S113 is completed, the CPU 30 enters "processing for personal identification of the corpse using each tooth reflected in the X-ray image (hereinafter referred to as personal identification process (2) as necessary)." This personal identification process (2) will be described later, and its characteristic is that it obtains information for identifying the identity of the corpse (such as name and age) based on the spectrum of X-ray transmittance (or similarity score) of each tooth growing in the upper and lower jaws.

[0041] Therefore, the CPU 30 first interactively determines whether or not to execute the personal identification process (2) in step S117, and then determines whether or not to update the database 20 for the personal identification process (2) in step S119. If the determinations in both steps S117 and S119 are YES, the CPU 30 updates the database 20 (step S121).

[0042] In contrast, if the determination in step S117 is YES, but the determination in step S119 is NO, it is determined that the user wishes to execute the personal identification process (2) itself, such as the user's name, and the CPU 30 executes the personal identification process (2) (step S123).

[0043] When step S123 is completed, when a NO determination is made in step S117, or when the processing of step S121 ends, the CPU 30 further interactively determines whether or not to terminate the series of processing shown in Fig. 2 (step S125). If the determination is YES, the selection processing shown in Fig. 2 is terminated, but if a NO determination is made, the CPU 30 returns the processing to step S101 and repeatedly executes each of the above-mentioned procedures.

[0044] 2 interactively with the user, the CPU 30 can select one of the above-mentioned age estimation process, personal identification process (1), and personal identification process (2) and instruct the execution of the process. Furthermore, when it seems that a satisfactory result has not been obtained with one estimation / identification process, or when the result is different from what was expected, the user can select another process and instruct the execution of that process (step S125). The selection steps of steps S101, S109, and S117 constitute an identification function selection means when configuring a personal identification integrated system.

[0045] For example, if the personal identification process (2) is first performed to collect information for identifying the identity of the corpse (such as name and age), but the screening information does not yield any hits and the matching rate is low, the personal identification process (1) can be performed. This allows for age estimation of the corpse based on morphological changes in the alveolar bone if the screening information and / or matching rate for identifying the identity of the corpse (such as name and age) based on morphological changes in the alveolar bone are poor. In this way, for unidentified corpses (subjects), the search can be expanded to include personal identification information that clearly identifies the individual, such as the name, as well as other potentially relevant personal identification information, such as the age or age group. In other words, it becomes possible to approach the personal identification of the same corpse (subject) using different approaches during the identification process.

[0046] <1A: Database update process for age estimation process and personal identification process (1)> In the flow of FIG. 2 described above, if a YES determination is made in step S103 or a YES determination is made in step S111, the database update process shown in FIG. 3 for this age estimation process or personal identification process (1) is initiated.

[0047] In this age estimation process or personal identification process (1), the only difference between the update processes for the database 20 is the type of updated data, so they will be described using the same flow.

[0048] First, since X-ray images transmitted from the medical modality 22 have already been stored in the database 20, the CPU 30 interactively determines whether or not to subject some or all of the stored images to an update process for age estimation or personal identification process (1) (step S151). This update process involves setting landmarks for each alveolar cavity in the stored upper and lower jaw images of the patient, dividing the landmark groups into multiple groups, and adding and storing the position data of the landmark groups in the database. As will be described later, the position data of the multiple landmark groups stored in this database 20 is compared with position data of similar multiple landmark groups obtained from the target corpse at the time of the estimation process, forming the basis for age estimation or personal identification.

[0049] If the CPU 30 judges YES in step S151 (update), the CPU 30 reads the first oral and maxillofacial image data to be updated into its work area (not shown) and displays the image data on the display 16 (step S153).

[0050] Next, the CPU 30 sets upper and lower limiting lines LU and LL for limiting the processing to the upper and lower alveolar bone portions of the displayed intra-oral and maxillofacial image IM (step S155). These limiting lines LU and LL may be set interactively between the CPU 30 and the operator, or may be set using machine learning (deep learning) using an AI (artificial intelligence) model.

[0051] For this reason, the landmark setting process, which will be described later, only needs to search the range between the upper and lower limiting lines LU and LL. A specific method for setting these limiting lines may be, for example, the method shown in FIG. 15 of Japanese Patent No. 6437914. By providing these limiting lines LU and LL, the amount of calculation required for processing data can be reduced. This has significant benefits whether landmark setting is performed by comparing the density values ​​of each pixel or by machine learning using an AI model that embodies AI (artificial intelligence).

[0052] Instead of the limiting lines LU and LL, the user may manually set a marker indicating the range to be calculated on the image displayed on the display 16. Of course, the entire image may be processed without performing the process in step S155.

[0053] Next, the CPU 30 sets landmarks LM for determining the shape of the alveolar bone at the alveolar crest and alveolar floor on the distal side of the alveolus of each tooth in the X-ray image IM currently displayed as shown in Figure 5(A) as reference points (reference positions) (step S157). These landmarks LM may be set manually by the user or through AI (artificial intelligence) processing. The manner in which these landmarks are set is shown in Figure 5(B). The square markers and white circle markers are examples of landmarks.

[0054] The alveolar crest and alveolar floor of the alveolus of an artificial tooth (implant, etc.) or a third molar (wisdom tooth) shall not be used as a landmark. In addition, for missing teeth or teeth extracted during orthodontic treatment, if the alveolus remains, the alveolar crest and alveolar floor of the relevant alveolus may be used as a landmark.

[0055] This method of setting landmarks is described in detail in Japanese Patent Application Laid-Open No. 2022-168622, proposed by Hideko Fujimoto, one of the inventors of the present invention.

[0056] The reason for referencing the alveolar crest and alveolar floor of each tooth's alveolus is that the arrangement of the alveolus is unique to each individual, and changes in that arrangement, i.e., changes in the alveolar bone over time, are correlated with the individual's age. For this reason, capturing the alveolar crest and alveolar floor as feature points is useful for estimating an individual's age and predicting changes in periodontal disease.

[0057] Next, the CPU 30 interactively sets six first closed areas CA1-CA6 based on the landmarks LM for the alveoli of the upper and lower jaws set as described above (step S159). That is, three first closed areas CA1-CA3 are set based on the landmarks for the alveoli of the natural teeth of the upper jaw, and three first closed areas CA4-CA6 are set based on the landmarks for the natural teeth of the lower jaw (see FIG. 5C). The three first closed areas CA1-CA3 and CA4-CA6 for the upper and lower jaws are set so as to overlap the alveolar portions of the left and right canines, respectively. Canines typically have longer roots than other permanent teeth, making them more likely to remain. Therefore, their alveoli are easier to identify and function well as landmarks when multiple closed areas are set.

[0058] This method of setting the closed regions is also described in detail in Japanese Patent Application Laid-Open No. 2022-168622, proposed by Hideko Fujimoto, one of the inventors of the present application. As described in the publication, the number of closed regions is not limited to three for each of the upper and lower jaws, and two or four or more closed regions may be set for each of the upper and lower jaws.

[0059] Furthermore, it is also possible to process both the landmark setting step and the closed area setting step using AI.

[0060] In this way, changes over time in the alveolar crest and alveolar floor of each tooth's alveolus can be perceived as changes in the shape of the three first closed areas CA1 to CA3 and CA4 to CA6 set for each of the upper and lower jaws as described above.

[0061] Next, the CPU 30 converts the position data indicating the distance from the center of gravity M of each of the shapes CA1 to CA6 of the six first closed areas illustrated in Fig. 5(D) to a number (e.g., CSV data) (step S161), and stores the converted data as average values ​​by age in the database 20 as schematically illustrated in Fig. 6 (step S163). Note that, unlike the planar figure illustrated in Fig. 5(D), the six first closed areas CA1 to CA6 are actually processed as closed areas in space, and the distance from the center of gravity M to each corner CN is also processed as a distance in space.

[0062] As an example, the ages of the viewers are divided into 14 to 29 years old, 30 to 39 years old, 40 to 49 years old, 50 to 59 years old, 60 to 69 years old, 70 to 79 years old, and 80 years old or older (see Figure 7 described below), and the average value of the position data of the first closed areas CA1 to CA6 for each age group is calculated and saved.

[0063] The above-described process is repeated until there is no more image data to update the database 20 (step S165).

[0064] As a result, as shown schematically in FIG. 6 , the database 20 is updated and recorded with patient information including the patient ID, name (blank), cause of death, circumstances of death including place of death (in the case of a corpse), type of X-ray image, and numerical values ​​defining the closed area (average value of position data).

[0065] <1B: Age Estimation Processing and Personal Identification Processing (1)> In the judgment flow shown in Fig. 2, if the answer is YES in step S101 and NO in step S103, or if the answer is YES in step S109 and NO in step S111, the CPU 30 proceeds to the age estimation processing or personal identification processing (1) of the corpse. An overview of this age estimation processing and personal identification processing (1) is shown in Fig. 4.

[0066] First, when the age estimation process is started (step S171, YES), an X-ray image of the subject is read and displayed on the display 16 (step S173). This X-ray image is an X-ray image of the oral cavity that has been two-dimensionally reconstructed from a CT image of the deceased body, or a two-dimensional X-ray image obtained by panoramic photography. In other words, this X-ray image may be an X-ray image of the oral cavity that has been captured or captured and converted in real time during age estimation, or may be an X-ray image that has been captured at another medical facility and stored in the database 20.

[0067] In order to reduce the amount of data calculation, the CPU 30 sets limit lines, if necessary, on the X-ray image of the oral cavity of the subject corpse displayed on the display, as described above (step S175).

[0068] Next, the CPU 30 sets second landmarks LM' at the natural alveoli (alveolar crest and alveolar floor) of each of the upper and lower jaws on the X-ray image of the subject, similar to step S157 described above (FIG. 5B: step S177). These landmarks LM' are used when executing the age estimation process or personal information identification (1) and are of the same nature as those used in the database update process described above (see FIGS. 3 and 5). However, since the subjects are different, the landmarks are labeled as first and second to distinguish them. That is, for example, a subject for a normal diagnosis is referred to as the first landmark LM, and a subject for a cadaver identification is referred to as the second landmark LM' to distinguish them.

[0069] This second landmark LM' is also set interactively by AI (e.g., deep learning) or manual operation by an operator. Once this is complete, the CPU 30 sets second closed areas CA1' to CA6' based on the position of the set second landmark LM', similar to step S159 described above (FIG. 5C: step S179). The names of these second and first closed areas are also intended to differentiate them in the same way as the landmarks LM and LM' described above.

[0070] In this case, it is desirable that the number of second closed areas CA1' to CA6' for each of the upper and lower jaws is the same as the number of first closed areas CA1 to CA6. For each of these second closed areas CA1' to CA6', position data indicating the spatial distance from the center of gravity M to each corner is converted into numerical values ​​(e.g., CSV data) (step S181), and the converted values ​​are stored in the database 20 ( FIG. 6 : step S183).

[0071] Next, the CPU 30 analyzes the similarity of each of the second closed areas CA1' to CA6' obtained from the subject with the average first closed areas CA1 to CA6 of the corresponding positions (top right, top center, top left, bottom right, bottom center, bottom right) of the viewer by age group stored in the database 20 by calculating the Procrustes distance between them (S185).

[0072] This Procrustes distance is the sum of squares of the error between the shape of one closed region and the shape of the other closed region when the shape of one closed region is enlarged, reduced, translated, or rotated. Therefore, the sum of squares of the error that minimizes the Procrustes distance is calculated as the similarity (Procrustes distance). The smaller this similarity value, the more similar the two closed regions are. This method of using the Procrustes distance as the similarity between alveolar bones is described in detail in Japanese Patent Application Laid-Open No. 2022-168622, proposed by Hideko Fujimoto, one of the inventors of the present application.

[0073] This similarity is calculated for each corresponding body part (upper right, upper center, upper left, lower right, lower center, and lower right) for all referees stored in the database 20, or for all referees for whom a doctor or the like has narrowed the search range (for example, only males). The calculated similarity is stored in the database 20.

[0074] Instead of the Procrustes distance, the similarity may be calculated by calculating the correlation coefficient between the two closed regions, or by performing machine learning using AI. In the latter case, the first closed region, i.e., the spatial position data defining each closed region, is provided to a machine learning model for each alveolar bone region for machine learning, and when the second closed region is input for each region, the machine learning model outputs the corresponding reference and its similarity. This similarity is preferably determined for all regions of the upper and lower jaws, i.e., the upper right, upper center, upper left, lower right, lower center, and lower right regions as a whole. For example, it may be determined by the average similarity for each region (each closed region). Examples of AI algorithms for estimating similarity include, but are not limited to, CNN (Convolution Neural Network).

[0075] Next, the CPU 30 estimates the age of the subject from the similarity calculated as described above (step S187). This estimation method will be explained with reference to the table in Fig. 7. Fig. 7 shows some of the results of an age estimation simulation performed by Hideko Fujimoto, one of the inventors of the present invention.

[0076] In Fig. 7, the cause of death or condition at the time of death for each subject is plotted, and the horizontal axis shows the average value of the first closed regions CA1 to CA6 (if the gender is known, the average value for the gender can also be calculated) for subjects by age group (14 to 29 years old, 30 to 39 years old, 40 to 49 years old, 50 to 59 years old, 60 to 69 years old, 70 to 79 years old, 80 years old or older) and the similarity (Procrustes distance) between the second closed regions CA1' to CA6' of the subject. Note that the gender of the subject may or may not be known at the time of death.

[0077] In the example of Figure 7, for the subject in the first row, the second closed areas CA1' to CA6' (the average value of the six closed areas) are most similar (= 0.16810) to the first closed areas CA1 to CA6 in the 30-39 age range in the database 20. In other words, because the Procrustes distance is smallest, this subject is estimated to be 34 years old, taking the median of the age range from 30 to 39 years old. Similarly, for the subject in the second row, the median of the age range from 14 to 29 years old, which has the smallest Procrustes distance (= 0.16997), is taken to be 25 years old. Using a similar algorithm, the ages of the third and subsequent subjects are also estimated based on the shape of their alveolar bones at the time of death. The gender and age of the subjects listed in this table were known for the simulation, but the degree of agreement with the estimated ages was high. This age range affects the accuracy of age estimation and is set in consideration of the performance required for age estimation. Furthermore, even if the Procrustes distances belong to the same age group, it is not necessary to use the median value of that age group as the estimated age, but it can be set appropriately in consideration of the amount of data stored and the required estimation accuracy.

[0078] After completing the age estimation in this manner, the CPU 20 further performs personal identification (executing screening and calculating a matching rate) comparable to the above-described age estimation of the subject with the data stored in the database 20 (step S189). That is, the CPU 20 performs screening to narrow down a predetermined number of top candidates in order to search for data indicating the first areas CA1 to CA6 of the subject during his / her lifetime among the data stored in the database 20, and calculates the matching rate (probability that the subject is the subject).

[0079] When calculating this matching rate, the shape of the first areas CA1 to CA6 before the subject's death is taken into consideration. Therefore, in step S189, an AI deep learning model is trained, and candidates with a high degree of matching with the subject, i.e., personal information, can be output from the learning model.

[0080] It is also possible to omit the personal identification process (1) in step S189 and perform only the age estimation in step S187. Therefore, when the process of FIG. 4 is executed as the age estimation process, the process of step S189 is omitted. When the process of FIG. 4 is executed as the reverse personal identification process (1), step S187 is omitted. Of course, even when the process of FIG. 4 is started for either of the two processes, both steps S187 and S189 may be executed, and the information displayed / output in step S189, which will be described later, may be specialized for the age estimation or personal identification process (1).

[0081] Finally, the CPU 30 provides the age estimation result obtained in step S187 or the age estimation and personal identification results obtained in both steps S187 and S189 to the display 16 and the report output device 18 in an appropriate manner (step S191). The estimation and identification process described above is repeatedly executed until an instruction to end the process is issued (step S193).

[0082] In the above-mentioned personal identification process (1), i.e., personal identification process using morphological changes in the alveolar bone, Hideko Fujimoto, one of the present inventors, performed a simulation to determine the degree of separation of the Procrustes distance between the same person group and the other person groups when the subject to be identified was the same person. As a result, the same person group and the other person group were separated from each other at a good cutoff value of about 4.978, and it was confirmed that the same person belonged to the same person group. It was also confirmed that a good value could be obtained for the ROC curve (Receiver Operating Characteristic Curve).

[0083] <2A: Database update process for personal identification process (2) based on each tooth> Next, we will first explain the update process of the database 20 executed by the CPU 30 for personal identification process (2) of a corpse using each tooth reflected in an X-ray image.

[0084] In the flow of FIG. 2 described above, if a YES determination is made in step S117, the database update process shown in FIG. 8 is initiated to perform personal identification process (2) from the pixel values ​​of each tooth of the subject reflected in the X-ray image.

[0085] When this update process is started, the CPU 30 determines, reads, and displays the X-ray image of the oral cavity of the subject, as described above (FIG. 8, steps S301 and S303), and then automatically or manually sets limit lines LU and LL that limit the image processing range, as described above (steps S305 and S307). Of course, the process of setting these limit lines may be omitted.

[0086] As shown in Fig. 10(A), an X-ray image IM (a parametric image or a two-dimensional image of the oral cavity reconstructed from X-ray CT data) of a subject's oral cavity is read and displayed. When the limiting lines LU and LL are set on the image, upper and lower limiting lines LU and LL are depicted in the upper and lower jaws, respectively, at a certain distance from the upper and lower tooth rows, as shown in Fig. 10(A).

[0087] Therefore, the CPU 30 performs a segmentation process to extract the edges EG of the teeth (including natural and treated teeth) constituting each of the upper and lower dental rows based on the pixel data of the dental row portion between the restriction lines LU and LL (step S309). In this embodiment, this segmentation process is performed using deep learning, a type of AI-based machine learning. For example, the YOLO (You Only Look Once) method is used as the deep learning algorithm. The concept of a neural network to which this algorithm is applied is shown by reference numeral 60 in FIG. 15 (described later) as an example, although it is described in a different embodiment. FIG. 10B (a) shows an example of an image IM in which semantic segmentation has been applied to the entire set of teeth, and FIG. 10B (b) shows an example of an image IM in which instance segmentation has been applied to each tooth detected by object detection. While one of the segmentation methods is selected in consideration of the overlap between teeth, the pixel data for each tooth is determined.

[0088] However, the segmentation process is not necessarily limited to a process using AI, and a process of discriminating pixel values ​​or matching with tooth shape patterns may also be used.

[0089] It is determined whether the process in step S309 has been completed for all teeth, and if there are any missing or missing teeth, appropriate complementation processing is performed (steps S311, S313, S315). This complementation processing may be deep learning using AI, or pattern matching processing using adjacent teeth and corresponding teeth in the upper and lower dental rows.

[0090] After all the segmentation processes are completed, the CPU 30 calculates a histogram of the pixel values ​​of the pixels that make up each segmented tooth (step S317). This histogram is, for example, a graph in which the horizontal axis represents pixel density and the vertical axis represents the frequency of each density (see FIG. 10C).

[0091] After the histogram calculation is completed, the CPU 30 calculates the median value CE of the density representative of the curve showing the histogram, sets this as information (value) representative of the density of the pixels of each tooth, and stores it in the RAM 36, for example (step S319).

[0092] Next, the CPU 30 calculates a spectrum with the median density CE, which is the density representative value of the pixels on the image constituting each tooth calculated as described above, on the vertical axis and the position of each tooth constituting the upper and lower tooth rows on the horizontal axis (step S321). An example of this spectrum is shown in FIG. 10 (D1). As an example, the horizontal axis is set to represent the positions of each tooth in the upper tooth row and the lower tooth row, with numbers 1 to 28. The vertical axis represents the density level according to the tooth position. If the median density CE, which serves as the density representative value, is plotted for each tooth at the tooth position, an example density spectrum is drawn, as shown schematically in FIG. 10 (D1). Note that the spectra shown in FIG. 10 (D1) to (D3) are explanatory diagrams that schematically show the similarity between spectra.

[0093] In the example of Figure 10 (D1), density spectra for five reference individuals, A, B, C, D, and E, are depicted schematically. The shape of the spectrum reflecting the median pixel density CE of each tooth in an X-ray image is highly likely to represent personal information for people aged 14 or older, and is thought to be an index for personal identification. This is because the second molars mature between the ages of 14 and 16.

[0094] The spectrum data thus obtained is stored in a predetermined area of ​​the database 20 (see FIG. 6: step S323) and is subjected to matching with that of the subject.

[0095] The above-described update process is performed for each patient (step S325), and reference spectrum data is stored in the database 20.

[0096] 2B: Personal Identification Processing Using Each Tooth (2) Next, an outline of the personal identification processing (2) based on the spectrum data of each of the patient's teeth, which is executed under the control of the CPU 30, will be described with reference to FIG. 9.

[0097] When the determination in step S121 in FIG. 2 is YES, the CPU 30 starts the personal identification process (2) shown in FIG.

[0098] The personal identification process in Figure 9 is a process performed on a corpse whose identity is unknown, and steps S351 to S371 are the same as steps S301 to S321 in Figure 8. The only difference, however, is that the subject is a corpse whose identity is to be identified. As a result, when the process in step S371 is completed, a spectrum reflecting pixel information for each tooth reflected in the X-ray image of the subject's oral cavity is calculated and displayed on the display 16 (see Figure 10 (D2)).

[0099] Therefore, the CPU 30 uses the spectrum of the subject to screen the spectra of each referee stored in the database 20 (which may contain information about the subject before his / her death) to narrow down to a predetermined number of spectra, calculates the pattern matching rate (steps S373 and S375), and determines the spectrum with the highest matching rate among these comparisons (step S377). For example, the spectrum SP of the subject shown in FIG. 10(D2) Ｔ is the spectrum SP of the references A to E stored in the database 20. Ａ ~SP Ｅ (See FIG. 10(D1))), as shown in FIG. 10(D3), the spectrum SP Ａ It can be judged that it has the highest matching rate.

[0100] This allows us to conclude that there is an extremely high possibility that the subject (corpse) we wish to identify is Reference Person A. Since the X-ray images of Reference Person A's oral cavity are panoramic images or CT images taken before the subject was alive, the name and other identifying information of the subject is almost always known. Therefore, the inference result indicates that there is a high possibility that the subject (corpse) is Reference Person A, and it is possible to obtain personal information such as the name of the corpse.

[0101] The results of this screening and the matching rate are presented to appropriate users (dentists, doctors, police officers, etc.) (step S379). Furthermore, the above-mentioned process is repeated until the spectrum similarity determination for all subjects is completed (step S381).

[0102] <Verification Example> A verification example conducted by the inventors of the present application (Fujimoto, Sakurai, Tanigawa, and Kawai) to verify the validity of the above-described personal identification process (2) will be described below. In this verification, since most X-ray images of the oral cavity taken after death are CT images, a similarity score was calculated as the correlation between multiple panoramic X-ray reconstructed images (subject) reconstructed from CT images and multiple panoramic X-ray images (references) taken during the subject's lifetime and whose personal information was known. This calculation was performed by a computer. The verification process and estimation results will be described with reference to FIGS. 11 to 13.

[0103] Specifically, as shown in Figure 11, Step (F1) Reconstruction of panoramic images from CT (axial) images: First, postmortem CT image data (hereinafter referred to as CT images) of the oral cavity of six subjects were obtained, and panoramic X-ray images (hereinafter referred to as panoramic images) were manually created from the CT scan tomographic images using XelisDental (software manufactured by INFINITT). At this time, a panoramic image along the dentition was reconstructed using the MPR (Multi-Planar Reconstruction) method.

[0104] Step (F2) Segmentation of Each Tooth: Next, each of the six panoramic images of the subject created in step F1 is compared with six reference panoramic images (including images of the subject before death) prepared in advance. To do this, tooth segmentation is performed for each image using the segmentation method (see FIG. 10(B)) using AI (e.g., AI processing by YOLO, U-Net, etc.) described above. Note that this segmentation can also be performed manually by an operator using a mouse on the screen.

[0105] When performing segmentation using AI processing, as mentioned above, the panoramic image is reconstructed from a CT image, so it is preferable to perform the following processing before performing the segmentation process itself, and this was also performed in this verification.

[0106] - Image size adjustment - Image processing such as gradation (grayscale) processing and smoothing - Processing the image so that the AI ​​can easily recognize the contours of each tooth As part of the preparation work, the segmentation method using the above-mentioned AI (for example, AI processing by YOLO, U-Net, etc.) was also performed in advance on each of the six reference panoramic images (references). Step (F3) Comparison of the target panoramic image (reconstructed image) and the reference panoramic image:

[0107] Next, for each tooth (i.e., each segmented tooth) in the panoramic images (reconstructed images) of the oral cavity of the six subjects (cadavers), the correlation between each tooth and each tooth in the six reference panoramic images was estimated using a brute-force method. The cosine similarity method was used to calculate the similarity score. Cosine similarity measures the degree of similarity between two data rows (numerical vectors) using an angle. The smaller the angle, the more similar the data are considered, and the closer the score is to 1, the more similar the data are estimated to be. Of course, this cosine similarity calculation method belongs to the similarity-based similarity calculation method, and methods such as the Jaccard coefficient or Dice coefficient can be used instead. Similarity can also be determined based on the correlation coefficient between the two images (e.g., calculated using the CORREL function or Pearson function).

[0108] In both calculation methods, the accuracy of the comparison decreases if the data size varies, so normalization was performed as a preprocessing step. This similarity score is quantified in a range from 0 to 1 and is set to have the following meaning, and the similarity is estimated and evaluated using the frequency (count) within that numerical range. The meaning of the numerical range of this similarity score is as follows: ・ Similarity score = 1.0: The two images match perfectly (the content of the images is completely identical) ・ Similarity score = 0.8 to less than 1.0: The two images show a high similarity (the images are almost identical, with only slight differences) ・ Similarity score = 0.5 to less than 0.8: The two images have a moderate similarity (some parts of the images match, but there are also many differences) ・ Similarity score = 0.0 to less than 0.5: The similarity between the two images is low or there is almost no match (there are many differences between the images, and it is highly likely that the images are not the same person)

[0109] <Evaluation Results> Representative examples of the evaluation results using the similarity scores described above are shown in Figures 12 and 13. The example in Figure 12 shows the comparison results for target panoramic image No. 000102 with six reference panoramic images. These six reference panoramic images, including the correct image Nos. 000102, 000110, 000129, 000147, 000154, and 0001555, are included for comparison. Of the six reference panoramic images, image No. 000102 had a similarity score of 0.6 or higher (16 counts), which is higher than the remaining images. Furthermore, it had a similarity score of 0.7 or higher (9 counts) and a similarity score of 0.8 or higher (7 counts), indicating a high degree of similarity between the images as a whole. This demonstrates the strongest correlation between target panoramic image No. 000102, reconstructed from the CT images of the corpse, and reference panoramic image No. 000102 from the individual's lifetime. The dentition shown in this image No. 000102 is typical and is considered to represent normal jaw structure.

[0110] This allowed us to estimate that the personal information (e.g., name, date of birth, gender) of the corpse representing target panoramic image No. 000102 was likely identical to the known personal information of reference panoramic image No. 000102, which was likely taken during the deceased's lifetime. The example in Figure 13 shows the comparison results for target panoramic image No. 000155 with each of six reference panoramic images. For comparison, these six reference panoramic images, similar to those described above, consist of image Nos. 000102, 000110, 000129, 000147, 000154, and 0001555, including the correct image. This shows that, of the six reference panoramic images, image No. 000155 had a count of 4 with a similarity score of 0.6 or greater and a count of 2 with a similarity score of 0.7 or greater, which are counts not included in the other reference panoramic images, indicating a high degree of similarity between the images as a whole. This verified that there was the strongest correlation between target panoramic image No. 000155, reconstructed from the CT images of the corpse, and reference panoramic image No. 000155 taken before the person's death. This allowed us to estimate that the personal information (name, date of birth, gender, etc.) of the corpse that presented target panoramic image No. 000155 would likely match the known personal information in reference panoramic image No. 000155, which was likely taken before the person's death. Note that image No. 000155 shows an example of a person with a small number of remaining teeth, and a strong correlation was confirmed even with this type of dentition configuration.

[0111] Similar estimates were possible for the other images. In this way, the similarity score was used to quantitatively evaluate the similarity between each image, allowing for a detailed analysis of how accurately the reconstructed CT image reflects the original panoramic image.

[0112] The similarity can be calculated by any method that can calculate the correlation coefficient between images, and the median of the histogram (density-frequency) of pixel values ​​for each tooth can be used. In addition to the similarity score as described above, the ranking within the reference data or the matching rate of the individual may also be calculated.

[0113] In some cases, it may be difficult to obtain postmortem CT images, or it may be difficult to reconstruct a panoramic image from those CT images. In such cases, X-ray image data standardized by the DICOM standard can be collected from a DICOM server and used. Furthermore, one of the imaging conditions for a CT scan is the slice thickness used for scanning. The thinner the slice thickness, the better the image quality (resolution, etc.) of the reconstructed image. For this reason, it is desirable to set the slice thickness as thin as possible (e.g., 0.5 to 1.0 mm) when performing a CT scan of a corpse.

[0114] Furthermore, when performing segmentation for each tooth, the pixel values ​​of each tooth may be binarized in advance, and segmentation may be performed from the binarized (black and white) image. If the median of the pixel value histogram is used as the representative value for each tooth, it is expected that the median can be calculated sufficiently from such a binarized image, and the advantage of being able to more accurately depict the contours of each tooth from the binarization is significant.

[0115] As described above, the examination system 10 (personal identification system) according to this embodiment can selectively execute three functions: estimating the age of a corpse using the alveolar bone reflected in X-ray images; identifying a corpse using morphological changes in the alveolar bone reflected in X-ray images (1); and identifying a corpse using pixel values ​​of each tooth reflected in X-ray images converted into a spectrum (2). Therefore, a more appropriate personal identification method (i.e., age / age range estimation, personal identification process (1), and / or personal identification process (2)) can be used depending on the condition of the corpse, providing multiple types of personal identification methods that further develop and expand on the conventional IDOL method (dental personal identification method). Of course, this examination system 10 not only performs personal identification, but can also estimate information indicating future oral disease risk associated with morphological changes in the patient's alveolar bone over time. Therefore, in medical and corpse verification settings, a variety of estimation and identification methods can be selectively used based on antemortem and postmortem dental information. Furthermore, an example in which future prediction of oral diseases, particularly periodontal diseases, is made from morphological changes of the alveolar bone over time will be described later as a modified example.

[0116] The advantages of the IDOL method-based personal identification method according to this embodiment will now be described in more detail. Although the conventional IDOL method is based on morphological changes over time in the upper and lower jaws, it has the drawback of not making full use of the useful information contained in these changes over time.

[0117] For example, in Japan, it is estimated that more than 12 million 2D dental panoramic images, which also serve as antemortem information, are taken annually. Even taking into account the mandatory 2-5 year retention period for medical records under the Medical Care Act, this figure is estimated to be approximately 30 million. In addition, when oral images (3D) taken with CT scans are also included, a vast amount of upper and lower jaw X-ray images are collected. Moreover, both dental panoramic and CT images are taken daily and stored at each medical institution (or at a center that integrates affiliated medical institutions), where the number continues to grow. Therefore, it would be meaningful to effectively utilize these collected upper and lower jaw images by creating a database as a daily updateable population, but such a system has not yet been found.

[0118] For this reason, it is difficult to say that images of the upper and lower jaws collected in the past, i.e., before the person's death, have been used effectively enough for unidentified persons who are the result of incidents or accidents that occur sporadically in different times and regions, or unidentified persons who are frequently seen in closed areas or at different times, such as during disasters.

[0119] On the other hand, the IDOL method mentioned above is not necessarily applicable to unidentified individuals, but it also suggests that it is possible to predict the current progression of periodontal disease and future conditions by utilizing morphological changes in the upper and lower jaws of each individual patient.

[0120] However, even when looking only at periodontal disease, there was room for improvement in terms of accuracy and reliability in terms of timely predictions based on the entire collection of X-ray images of the upper and lower jaws taken daily across Japan, which are constantly increasing.

[0121] In this regard, the personal identification function of the testing system according to this embodiment is expected to prove useful in identifying individuals, such as unidentified persons, since it can effectively utilize a database that accumulates daily.

[0122] In this embodiment, data representing closed regions (first and second closed regions) defined by dividing the alveolar bone into multiple landmarks is stored as numerical information, such as CSV, as an index showing morphological changes in the alveolar bone over time. Furthermore, pixel data for each tooth is stored as a representative value (numerical value) of a histogram. Compared to a system configuration that directly stores images, this system reduces the amount of information stored, reduces computational load, and is advantageous in terms of handling personal information.

[0123] Furthermore, data can be accumulated in an updatable database during regular medical examinations, making it easy to obtain training data for training a learning model even when part of the process is performed using AI machine learning (deep learning, etc.). In other words, while not limiting each step of the examination to AI processing, it is possible to provide a configuration that is suitable for implementing AI learning. Furthermore, it is possible to provide a system that is easily compatible with cloud computing.

[0124] <Modifications> Next, modifications will be described with reference to FIGS.

[0125] This variant relates to a diagnostic system that more accurately estimates morphological changes over time in the alveolar bone that supports the teeth due to the onset of oral diseases such as periodontal disease, and communicates predictive information about future symptoms to the patient via the dentist.

[0126] Periodontal disease is a typical example of this oral disease. When periodontal disease develops, the alveolar portion of the alveolar bone is absorbed over time, causing the alveolar bone line to recede. In particular, the degree of absorption varies locally, and the periodontal tissue supporting the teeth recede locally toward the tooth root, forming grooves around the periodontal tissue, known as periodontal pockets PK (see Figure 14). If this periodontal pocket PK is left untreated, it will expand, potentially resulting in tooth loss and the risk of bacterial infection via the periodontal tissue.

[0127] Therefore, the purpose of the examination system 10A of this modified example is to enable dentists to quantitatively predict the current disease state and its future state, i.e., the deterioration of the disease in the future, such as six months from now or one year from now, and provide this to patients.

[0128] As shown in Fig. 15, this examination system 10A has elements similar to those of the configuration shown in Fig. 1 described above, and therefore a duplicate explanation will be avoided, but an AI learning program for AI processing of the panoramic image of the periodontal pocket PK has been added to the main storage device 38, and a learning model 60 for machine learning (deep learning) has been constructed in the auxiliary storage device 40. The CPU 30 activates the learning model 60 shown in Fig. 15 in the auxiliary storage device 40.

[0129] 15, this learning model 60 has an input layer, one or more intermediate layers, and an output layer, and is configured to tune the weighting coefficients and conversion coefficients in the intermediate layers using a predetermined algorithm. Here, this learning model 60 is specialized and constructed as weak AI (to perform a specific task), and learning is performed by supervised learning.

[0130] For this reason, image data of periodontal pockets labeled with labels ranging from normal alveoli to alveoli with various stages of disease progression are provided as training data to this learning model 60, along with the labeling information, and the model is tuned (FIG. 16, step S401). This tuning is performed each time training data is provided, so the more training data there is, the higher the tuning accuracy becomes, and the higher the accuracy of the future periodontal pocket predictions output from the output layer also becomes.

[0131] Therefore, at the time of diagnosis, all or part of an X-ray image (e.g., a panoramic image) of the patient's oral cavity is input to the learning model 60 (step S402). In response, the learning model 60 outputs the progression of periodontal disease in the region of interest (one or more regions) from the output layer (step S403). This output result is prediction information on the disease state of periodontal disease three months, six months, one year, etc. in the future.

[0132] The CPU 30 then incorporates the output results into authorized numerical information in the dental community and creates a report with evaluations, comments, future preventive measures, etc. that are easy for dentists and patients to understand (step S404). Examples of such numerical information include the numerical categories described in "Response to the New Classification of Periodontal Disease" published by the Japanese Society of Periodontology on February 21, 2022. According to these numerical categories, "periodontitis stage" is evaluated as Stage I, Stage II, Stage III, and Stage IV, and "periodontitis grade" is evaluated as Grade A (slow progression), Grade B (moderate progression), and Grade C (rapid progression).

[0133] In other words, as an example, it is possible to provide a future prediction such as, "Your periodontal disease condition is currently Stage II, but because it is Grade C, you are expected to lose about four teeth in three months." Based on this, a doctor can suggest treatment advice.

[0134] This is thought to be more convincing to patients than simply stating vague findings such as "your condition will become serious from now on," and to make treatment easier for both patients and doctors.

[0135] The report creation process in step S404 can also be subjected to AI processing together with step S403.

[0136] Furthermore, from the perspective of checking the progression of periodontal disease, if CT images are obtained at two different points in time, the three-dimensional difference between them can be taken and the progression can be measured from the difference in periodontal pocket volume. Of course, a prediction six months later can be evaluated using AI processing based on the progression. Furthermore, indices for checking the progression of periodontal disease can include not only periodontal pocket volume, but also PESA (Periodontal Epithelial Surface Area) and PISA (Periodontal Inflamed Surface Area), or at least one of these can be used in combination with volume. Diversifying the indices in this way further improves the accuracy of checking the progression of periodontal disease.

[0137] As described above, the invention described in the claims is not limited to the configurations described in the above embodiments and variations, and further, configurations that add conventionally known components and processes and exhibit effects equivalent to or greater than those of the present invention are also construed as being included in the present invention as long as they do not deviate from the spirit of the claims.

[0138] 10 Inspection system with personal identification function (personal identification system, personal identification integrated system) 12 Image processing device 14 Operating device 16 Display device 18 Report output device 20 Database 22 X-ray inspection device 30 CPU (processor: forming an essential part of closed area setting means, landmark setting means, similarity analysis means, estimation means, similarity calculation means, age estimation means, segmentation means, spectrum analysis means, database construction means, and search means, and also forming an essential part of an AI model generation device, and also forming an essential part for providing each step of a personal identification method) 32 Internal bus 24 Input / output interface 36 Temporary storage memory 38 Main storage device (recording medium) 40 Auxiliary storage device IM X-ray image LM, LM' Landmarks LU, LL Restriction lines CA1 to CA6, CA1' to CA6' Closed area SP Ａ ~SP Ｅ , SP Ｔ Spectrum EG Edge

Claims

1. A personal identification system for acquiring information specific to a subject to be identified from information on X-ray images that capture the upper and lower jaws of a subject whose age is at least known, comprising: a first landmark setting means for setting a first landmark for each alveolus of natural teeth in each of the upper and lower jaws of the subject in the X-ray image of the subject; a first closed region setting means for setting, based on the first landmark for each alveolus, a plurality of first closed regions that are closed by coordinate groups of the first landmarks belonging to each of a plurality of jaw regions that are divided into each of the upper and lower jaws according to the position of the horseshoe-shaped curve of the jaw; a database storage means for storing position data that defines the plurality of first closed regions in a database that can be referenced for personal identification; and a second landmark setting means for setting a second landmark for each alveolus of natural teeth in each of the upper and lower jaws of the subject in the X-ray image when acquiring the information specific to the subject. a second closed area setting means for setting, based on the second landmarks for each of the alveoli, a plurality of second closed areas that are closed by a group of coordinates of the second landmarks that belong to each of a plurality of jaw regions into which each of the upper and lower jaws is divided according to the horseshoe-shaped curvature position of the jaw; a similarity analysis means for analyzing, for each closed area, the similarity between at least one of the plurality of second closed areas obtained from the subject and a first closed area in the database that is positionally equivalent to the at least one second closed area; and an estimation means for estimating, based on the similarity, the age or age range of the subject, or the identification information of the subject as the unique information.

2. The personal identification system according to claim 1, wherein the estimation means is configured to estimate the subject's age or age range as the unique information.

3. The personal identification system according to claim 1, characterized in that the first and second landmark setting means are configured to set landmarks at least at the alveolar crest and alveolar floor of each of the subject's natural teeth.

4. The personal identification system described in claim 1, characterized in that the database is prepared in advance as a reference database containing coordinate data defining the multiple first closed areas of an unspecified number of subjects whose ages and genders are at least known.

5. The personal identification system of claim 4, wherein the similarity analysis means comprises: a similarity calculation means for calculating the degree of geometric similarity between at least one of the plurality of closed areas of the subject and one or more second closed areas in the database that are positionally equivalent to the at least one first closed area; and an age estimation means for estimating the age of the subject from the age of the subject having the closed area that shows the greatest similarity among the one or more similarities calculated by the similarity calculation means.

6. The personal identification system according to any one of claims 1 to 5, further comprising a display means for displaying the subject's identification information estimated by the estimation means.

7. A personal identification system as described in any one of claims 1 to 5, characterized in that the X-ray images include a two-dimensional panoramic image of the subject's oral cavity taken by a dental panoramic imaging device, and a two-dimensional image of the upper and lower jaws converted from a CT image of the subject's oral cavity taken by an X-ray CT device.

8. The personal identification system according to any one of claims 1 to 5, characterized in that at least one of the first landmark setting means, the second closed area setting means, the second landmark setting means, the second closed area setting means, the similarity analysis means, and the estimation means is configured by a machine learning model that processes the data based on an AI (artificial intelligence) algorithm.

9. The personal identification system according to any one of claims 1 to 5, characterized in that the first landmark setting means, the first closed area setting means, the second landmark setting means, the second closed area setting means, the similarity analysis means, and the estimation means are configured by one or more machine learning models that process the first landmark setting means, the first closed area setting means, the second landmark setting means, the second closed area setting means, the similarity analysis means, and the estimation means based on an AI (artificial intelligence) algorithm.

10. The personal identification system described in claim 9, characterized in that the machine learning model is a learning model using deep learning and is composed of a neural network in which multiple neurons are superimposed to functionally construct the second landmark setting means, the second closed area setting means, the similarity analysis means, and the estimation means.

11. A personal identification system according to any one of claims 1 to 5, wherein the first and second landmark setting means are means for interactively setting the landmarks on the X-ray image between the user and the system.

12. A program applicable to the personal identification system according to any one of claims 8 to 11, the program being pre-recorded on a computer-readable recording medium, and causing a computer or a computer-equipped system to functionally construct a machine learning model by having the computer execute the program read from the recording medium.

13. A recording medium on which a program applicable to the personal identification system according to any one of claims 8 to 11 is pre-stored so as to be readable by a computer, and which causes a computer or a computer-equipped system to functionally construct a machine learning model by executing the read program.

14. A personal identification system for acquiring information specific to a subject to be identified from information on an X-ray image that captures the upper and lower jaws of a subject whose age is at least known, comprising: a first segmentation means for segmenting each tooth of the upper and lower jaws of the subject in the X-ray image; a first spectrum analysis means for analyzing the relationship between pixel value information of each tooth of the upper and lower jaws segmented by the first segmentation means and the position of each tooth in the upper and lower jaws as a first spectrum for each region of the upper and lower jaws; a database construction means having a database that can be referenced for personal identification, and constructing the first spectrum for each region analyzed by the first spectrum analysis means in the database; and a second segmentation means for segmenting each tooth of the upper and lower jaws of the subject in the X-ray image when acquiring the information specific to the subject. a second spectrum analysis means for analyzing the relationship between pixel value information of each tooth of the upper and lower jaws segmented by the second segmentation means and the position of each tooth in the upper and lower jaws as a second spectrum for each part of the upper and lower jaws; a search means for screening the first spectra stored in the database to search for the first spectrum that shows the highest matching rate to the second spectrum; and an estimation means for estimating personal information of the subject as the unique information from the personal identification state of the subject that shows the first spectrum that shows the highest matching rate searched for by the search means.

15. The personal identification system according to claim 14, further comprising display means for displaying the specific information estimated by the estimation means together with at least the second spectrum.

16. The personal identification system described in claim 14, characterized in that the X-ray images include a two-dimensional panoramic image of the subject's oral cavity taken by a dental panoramic imaging device, and a two-dimensional image of the upper and lower jaws converted from a CT image of the subject's oral cavity taken by an X-ray CT device.

17. A personal identification system as described in any one of claims 14 to 16, characterized in that at least one of the first segmentation means, the first spectrum analysis means, the second segmentation means, and the second spectrum analysis means is configured by a machine learning model that processes the means based on an AI (artificial intelligence) algorithm.

18. A personal identification method for acquiring information specific to a subject to be identified from information on an X-ray image that captures the upper and lower jaws of a subject whose age is at least known, comprising: setting a first landmark for each alveolus of natural teeth in each of the upper and lower jaws of the subject in the X-ray image of the subject; setting a plurality of first closed regions, each of which is enclosed by a group of coordinates of the first landmarks belonging to each of a plurality of jaw regions obtained by dividing the upper and lower jaws according to the position of the horseshoe-shaped curve of the jaw, based on the first landmarks for each alveolus; storing position data defining the plurality of first closed regions in a database so as to be referenceable for personal identification; while, when acquiring information specific to the subject, setting a second landmark for each alveolus of natural teeth in each of the upper and lower jaws of the subject in the X-ray image of the subject; setting a plurality of second closed regions, each of which is enclosed by a group of coordinates of the second landmarks belonging to each of a plurality of jaw regions obtained by dividing the upper and lower jaws according to the position of the horseshoe-shaped curve of the jaw, based on the second landmarks for each alveolus; A method for identifying individuals, comprising: analyzing, for each second closed area, a similarity between at least one of a plurality of second closed areas obtained from the subject and a first closed area in the database that corresponds in position to the at least one second closed area; and estimating, based on the similarity, the subject's age or age range, or the subject's identification information as the unique information.

19. A method for personal identification for acquiring information specific to a subject to be identified from information on an X-ray image showing the upper and lower jaws of the subject, the age of which is at least known, comprising: segmenting each tooth of the upper and lower jaws of the subject in the X-ray image; analyzing the relationship between pixel value information of each segmented tooth of the upper and lower jaws and the position of each tooth in the upper and lower jaws as a first spectrum for each region of the upper and lower jaws; and constructing the first spectrum for each region in a database that can be referenced for personal identification; while, when acquiring the information specific to the subject, segmenting each tooth of the upper and lower jaws of the subject in the X-ray image; analyzing the relationship between pixel value information of each segmented tooth of the upper and lower jaws and the position of each tooth in the upper and lower jaws as a second spectrum for each region of the upper and lower jaws; screening the first spectrum stored in the database to search for the first spectrum that shows the highest matching rate with the second spectrum; and estimating personal information of the subject as the unique information from the personal identification state of the subject that exhibits the first spectrum that shows the highest matching rate.

20. A personal identification system for acquiring information specific to a subject to be identified from information on X-ray images that capture the upper and lower jaws of a subject whose age is at least known, the system comprising: landmark setting means for setting first landmarks on each tooth of each of the upper and lower jaws of the subject in the X-ray images of the subject; and closed region setting means for setting, based on the landmarks for each alveolus of each of the upper and lower jaws, multiple closed regions that are enclosed by coordinate groups of the landmarks belonging to multiple jaw regions into which each of the upper and lower jaws is divided according to the position of the horseshoe-shaped curve of the jaw; and an AI (artificial intelligence) model generation device installed in the personal identification system, comprising: an AI (artificial intelligence) model capable of machine learning using a predetermined algorithm for personal identification; an AI model learning means for training the AI ​​model so that: - a plurality of position data defining the plurality of closed areas related to a plurality of the subjects is input; - when position data corresponding to any of the closed areas related to a specific subject is given to the AI ​​model, the AI ​​model learns, under the predetermined algorithm, the similarity between the closed area defined by the given position data and the plurality of stored closed areas; and - the AI ​​model outputs the closed area with the highest similarity among the machine-learned similarities.

21. A diagnostic device for oral diseases that predicts the future progression of oral diseases from X-ray images of the patient's oral cavity using AI processing or by comparing multiple X-ray images collected at different times, and presents the progression to the patient along with numerical information.

22. A personal identification integration system comprising an integrated personal identification device according to claim 1 and a personal identification device according to claim 14, and configured to be capable of selectively estimating the subject's age or age range based on the degree of similarity, estimating the subject's identification information, and estimating the subject's personal information from the personal identification state of the subject that exhibits the first spectrum, which is searched by the search means and shows the highest matching rate.

Images