AI system for estimating maxillofacial biological age based on panoramic X-ray images

Abstract

We estimate useful information other than the presence or absence of lesions from dental panoramic X-ray images. [Solution] The AI ​​program or AI estimation method causes a computer 110 included in an AI system 100 to execute an image input module 11 that receives at least one dental panoramic X-ray image, a feature extraction module 12 that extracts at least one feature related to the jawbone, including the alveolar bone, from the at least one dental panoramic X-ray image using CNN and outputs at least one numerical value of the at least one feature, and an age estimation module 13 that estimates at least one biological age from the at least one numerical value using regression models 14 and 15 based on DNN.

Description

[Technical Field]

[0001] The present invention relates to an AI system for estimating maxillofacial biological age based on dental panoramic X-ray images. [Background technology]

[0002] Traditionally, dental panoramic X-ray images are often taken first when a patient visits a dental clinic for the first time to comprehensively assess the condition of the entire oral cavity. Dental panoramic X-ray images are also known to be useful for early detection of tooth or jawbone defects caused by caries or periodontal disease, or intra-jawbone lesions, including inflammatory lesions of osteitis or osteomyelitis associated with caries or periodontal disease, during regular dental checkups.

[0003] For example, claim 1 of Patent Document 1 discloses a dental examination AI system comprising: "a first deep learning unit that detects lesion locations from dental panoramic X-ray images and determines the name of the disease; a second deep learning unit that performs semantic segmentation of the dental panoramic X-ray images to identify the tooth number and tooth area of ​​each tooth; and a dental diagnosis unit, wherein the dental diagnosis unit identifies the presence or absence of a lesion and the name of the disease for each tooth number by matching the lesion locations detected by the first deep learning unit with the tooth areas identified by the second deep learning unit."

[0004] For example, claim 1 of Patent Document 2 discloses a "dental analysis system including a deep learning unit that detects lesion locations from dental panoramic X-ray images and determines the name of a disease, and a display unit that displays the lesion locations and disease names identified by the deep learning unit on the dental panoramic X-ray images, wherein the deep learning unit includes a YOLO (you only look once) system." [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2024-013413 [Patent Document 2] Patent Publication No. 6830082 Summary of the Invention [Problem to be solved by the invention]

[0006] However, conventional AI systems or analysis systems such as those described in Patent Documents 1 and 2 merely diagnose diseases by examining the presence, type, size, shape, location, etc. of lesions, similar to the diagnosis made by dentists. On the other hand, the inventors thought that because the images obtained from dental panoramic X-ray images cover a wide range, useful information other than the characteristics that form the basis of diagnosis could be obtained.

[0007] One object of the present invention is to estimate useful information other than the presence or absence of a lesion from a dental panoramic X-ray image (hereinafter also referred to as a "dental panoramic X-ray image" or "panoramic X-ray image"). Other objects of the present invention will become apparent to those skilled in the art by referring to the following exemplary aspects and best modes, as well as the accompanying drawings. [Means for solving the problem]

[0008] In order to facilitate an understanding of the outline of the present invention, the following examples are given of embodiments according to the present invention.

[0009] In an embodiment according to the present invention, the program or method is configured to cause a computer included in an AI system to: an image input module for receiving at least one dental panoramic x-ray image; a feature extraction module that extracts at least one feature related to the jaw bone including the alveolar bone from the at least one dental panoramic X-ray image using a convolutional neural network (CNN) and outputs at least one numerical value of the at least one feature; an age estimation module that estimates at least one biological age from the at least one numerical value using a machine learning, e.g., a trained regression model using a deep neural network (DNN); In other words, the AI ​​module includes at least the image input module, feature extraction module, and age estimation module.

[0010] According to an embodiment of the present invention, jawbone (typically the mandible) features (e.g., the angle of the jaw angle (commonly called "square jaw"), the thickness of the bone cortex, etc.) are extracted from a dental panoramic X-ray image. Furthermore, according to an embodiment of the present invention, biological age, which provides useful information, can be estimated using a regression model that reflects the natural law that the extracted features change with age. Here, the regression model may be a conventional machine learning model, and a trained model using DNN is particularly preferable. That is, while traditional machine learning models such as linear regression, polynomial regression, ridge regression, Lasso regression, and support vector regression can be used, a trained regression model using DNN can represent complex nonlinear relationships and can be obtained by using a large amount of training data (numerical values ​​of features extracted by CNN and calendar age) can be handled to improve the accuracy of biological age estimation.

[0011] Those skilled in the art will easily understand that the exemplified embodiments according to the present invention can be further modified without departing from the spirit of the present invention. [Brief explanation of the drawings]

[0012] [Figure 1] 1 shows an example of the overall configuration of an AI system according to an embodiment. [Figure 2] 10 shows an example of the configuration of an AI module according to a second embodiment. [Figure 3] FIG. 3(a) shows the region including the jaw angle of the mandible in a dental panoramic X-ray image, and FIG. 3(b) shows a schematic diagram illustrating the principle of estimating biological age from the angle of the jaw angle using a regression model. [Figure 4]Figure 4(a) shows an area including the mandibular cortex in a dental panoramic X-ray image, and Figure 4(b) shows a schematic diagram of the principle of estimating biological age from the cortical thickness using a regression model. [Figure 5] An example of the output from the output module in Figure 1 is shown below. [Figure 6] 10 shows an example of the configuration of an AI module according to a second embodiment. [Figure 7] 10 shows an example of the configuration of an AI module according to a third embodiment. [Figure 8] An example of the display of the output module of Figure 1 is shown. DETAILED DESCRIPTION OF THE INVENTION

[0013] The best mode described below is used to facilitate understanding of the present invention, and therefore, those skilled in the art should be aware that the present invention is not unduly limited by the embodiments described below.

[0014] FIG. 1 shows an example of the overall configuration of an AI system according to an embodiment. The AI ​​(artificial intelligence) system 100 typically includes a computer 110 and a dental panoramic X-ray imaging device (hereinafter also referred to as "panoramic X-ray imaging device") 120. The panoramic X-ray imaging device 120 is disclosed in, for example, Patent Document 1 and Patent Document 2, and is well known to those skilled in the art, so a detailed description thereof will be omitted in this specification. In addition, the structure and hardware of the computer 110 are also well known to those skilled in the art, so the configuration and functional blocks of the computer 110 will be briefly described below, and the parts of the present invention will be described in detail.

[0015] The panoramic X-ray imaging apparatus 120 may be equipped with a dedicated device or a computer such as a personal computer, either internally or externally. In the example of Fig. 1, it is controlled by a computer 110, which is, for example, a personal computer. In other words, dental panoramic X-ray images (hereinafter also referred to as "panoramic X-ray images") taken by the panoramic X-ray imaging apparatus 120 based on the operation of an operator such as a dentist, dental hygienist, doctor, or nurse are input to and stored in a first storage unit 40 via an input unit / communication unit 20.

[0016] The control unit 30 can control the AI ​​module 10, the input unit / communication unit 20, the first memory unit 40, the second memory unit 50, and the output unit / communication unit 60, and inputs data from input devices such as the panoramic X-ray imaging device 120 and the keyboard / mouse 130, generates information from the data, and outputs the information to output devices such as the monitor / printer 140.

[0017] The AI ​​module 10 typically includes an image input module 11, a feature extraction module 12, and an age estimation module 13. The AI ​​module 10 preferably includes a comment generation module 16 and / or an output module 17.

[0018] Additionally, the AI ​​module 10 may include other modules, which are functional blocks not shown, to perform other calculations. For example, the AI ​​module 10 may include a diagnostic module such as those disclosed in Patent Documents 1 and 2. That is, the AI ​​module 10 may estimate only useful information other than the presence or absence of a lesion from a panoramic X-ray image, or may also estimate the presence or absence of a lesion.

[0019] FIG. 2 shows an example configuration of an AI module according to a first embodiment. The AI ​​module 10 includes an image input module 11 that receives a panoramic X-ray image from a panoramic X-ray imaging device 120, a feature extraction module 12 that extracts features (e.g., jaw angle, bone cortex, etc.) related to a jawbone (e.g., a mandible, etc.) including an alveolar bone from the panoramic X-ray image using CNN and outputs numerical values ​​of the features (e.g., angle, thickness, etc.), and an age estimation module 13 that estimates biological age from the numerical values ​​(e.g., the angle of the mandibular jaw angle, the thickness of the mandibular bone cortex, etc.) using a trained regression model (hereinafter simply referred to as a "regression model") based on DNN machine learning. The inventors have discovered that the angle of the mandibular jaw angle and the thickness of the mandibular bone cortex are particularly suitable for estimating biological age because changes due to aging can be easily detected from the features.

[0020] The AI ​​program for executing the AI ​​module 10, which includes the image input module 11, the feature extraction module 12, the age estimation module 13, etc., is stored, for example, in the second memory unit 50 of FIG. 1, and the control unit 30 can execute the AI ​​program to configure the AI ​​module 10.

[0021] According to the AI ​​module 10 of FIG. 2, biological age, which is useful information, can be estimated using a regression model that reflects the natural law that jawbone (typically mandible) features (e.g., jaw angle, bone cortex thickness, etc.) extracted by CNN from panoramic X-ray images change with age. In addition, a regression model trained by DNN as machine learning can represent complex nonlinear relationships, and can also be used to input chronological age using the keyboard 130 and generate a large amount of learning data (the numerical values ​​of the features extracted by CNN and the like). calendar age) to improve the accuracy of biological age estimation in regression models.

[0022] The mandibular angle typically falls within the ranges of 130-140°, 110-130°, and 120-140° for children (6-12 years old), adults (20-40 years old), and the elderly (65 years old and above), respectively.

[0023] Infants and young children can be excluded from regression models because their chewing function is underdeveloped and the morphology of the mandible itself is immature. Furthermore, from childhood to adulthood, as the jaw skeleton and dentition mature, the masticatory muscles also develop further, forming the jaw angle. Furthermore, in elderly people, tooth loss leads to a decline in the function of the masticatory muscles and progressive jawbone resorption, resulting in changes in mandibular morphology and a larger jaw angle. Excluding children and adolescents, the angle of the mandibular jaw increases with age. Therefore, the regression model can be used to estimate biological age by limiting the target age to 20 years and older, or 20 to 65 years old.

[0024] The thickness of the mandibular cortical bone increases with skeletal maturation due to physical growth and the development of the masticatory muscles. Subsequently, the thickness of the mandibular cortical bone decreases with age. In particular, in elderly people, when teeth are lost, the mechanical stimulation from chewing decreases, leading to progressive resorption of the mandible, resulting in thinner cortical bone. In some cases, distinct cortical structures may even disappear. Similarly, the regression model can be used to estimate biological age by limiting the target age range to 20 years and older, or 20 to 65 years.

[0025] Figure 3(a) shows the area including the mandibular angle of the mandible in a panoramic X-ray image, and Figure 3(b) shows a schematic diagram of the principle of estimating biological age using a regression model from the angle of the mandibular angle. As indicated by the tip of the arrow labeled "Mandibular" in Figure 3(a) (on the right side of Patient X), the mandibular angle of the mandible is the angle at the intersection of the posterior border of the ramus of the mandible and the lower border of the body of the mandible, known as the jaw line. The internal angle formed at the point where a tangent drawn to the posterior border of the ramus (posterior tangent) intersects with a tangent drawn to the lower border of the body of the mandible (lower tangent) can be measured.

[0026] The jaw angle of the mandible may be only on the right side of the patient, only on the left side of the patient, or both on the right and left sides of the patient (for example, an average value).

[0027] The image input module 11 inputs a panoramic X-ray image of patient X from a panoramic X-ray imaging device 120 to the feature extraction module 12 via the input unit and communication unit 20. The CNN used by the feature extraction module 12 is, for example, a trained mandibular angle CNN (a CNN for the jaw angle of the mandible), and the mandibular angle CNN of the feature extraction module 12 automatically extracts coordinates of reference landmarks that define the jaw angle of the mandible of patient X from the panoramic X-ray image of patient X. The feature extraction module 12 uses the coordinates of the extracted landmarks to calculate the posterior edge tangent and the inferior edge tangent, and calculates the angle between the two tangents.

[0028] The mandibular angle CNN can reduce computational costs by processing only a region of interest (ROI) instead of the entire panoramic X-ray image. The ROI can be set manually. Alternatively, a simple contour detection can be used to identify the approximate location of the mandible, and an ROI can be automatically set around it.

[0029] The ROI may be on only the right side of the patient, only the left side of the patient, or both the right and left sides of the patient.

[0030] Next, the mandibular angle CNN can efficiently extract the mandibular contour by applying, for example, Canny edge processing only within the ROI (or, if both right and left ROIs are set, the left and right ROIs or each of the divided ROIs). The mandibular angle CNN can detect a series of points that form the mandibular contour (maximum contour) from the edges obtained by Canny edge processing. The mandibular angle CNN can apply an approximation polygon (ApproxPolyDP) to the detected mandibular contour to extract three points (landmarks) that form the intersections of the posterior and lower edge tangents. The mandibular angle CNN can use the coordinates of the three extracted landmarks to define two vectors that form the posterior and lower edge tangents and calculate the vector angle (arccos dot product method).

[0031] In Figure 3(b), regression model 14 is calendarThis shows the relationship between age and the jaw angle of the mandible (mandibular angle). The regression model 14 shown in Figure 3(b) is shown as a straight line to make the explanation easier to understand, but in the case of DNN, it is a complex curve with the same tendency as a straight line. In the simplified regression model 14 of Figure 3(b), calendar As age increases, the angle of the mandibular angle (the numerical value of the feature) increases. When the measured value of the angle of the mandibular angle of patient X is B, the age of the regression model 14 corresponding to the measured value B is the biological age α of patient X.

[0032] In addition, patient X calendar The standard value of the regression model 14 corresponding to age (e.g., based on keyboard 130 input) is Q, and when measurement value B > standard value Q, the biological age α estimated from measurement value B > the chronological age of patient X.

[0033] Regression model 14 is calendar The relationship between age and the jaw angle of the mandible (mandibular angle) is shown, and is machine-learned in advance using, for example, a DNN. When learning the regression model 14 using a DNN, it is preferable to use data from more than 100 people. calendar The regression model 14 is determined in advance based on the relationship between age and mandibular angle. Using a DNN, a highly accurate regression model 14 can be constructed with 100 samples, and the accuracy of the regression model 14 may be further improved with more than 100 samples.

[0034] The following describes a method for training the regression model 14. The first storage unit 40 stores training panoramic X-ray images of at least 100 patients. The image input module 11 inputs the training panoramic X-ray images of at least 100 patients to the feature extraction module 12.

[0035] When the first memory unit 40 stores training panoramic X-ray images of 101 or more first-time patients, the image input module 11 may input all of the training panoramic X-ray images to the feature extraction module 12, or may select only a predetermined number of training panoramic X-ray images, for example, randomly, and input them to the feature extraction module 12.

[0036] Next, the feature extraction module 12 sequentially processes each of the learning panoramic X-ray images of, for example, 100 first-time patients using the mandibular angle CNN, and generates the angles (angle values) of the mandibular angles of the 100 first-time patients.

[0037] Next, the feature extraction module 12 extracts, for example, 100 angle values ​​for 100 first-time patients, calendar The data is stored in association with the age. The storage destination may be any location, such as the feature extraction module 12 or the second storage unit 50.

[0038] Here, for learning calendar The ages of the 100 patients who first visited the hospital when the panoramic X-ray images were taken were calendar Age: 18+ calendar Age (at the time of first examination) calendar The age) is typically input via the keyboard 130, and the input timing may be before or after the generation of the angle value, or before or after the initial imaging. calendar Age (at the time of first examination) calendar age) may be included.

[0039] For example, patient data in which the birth date of a patient is associated with the patient ID is stored in the patient database, and a learning panoramic X-ray image is associated with the patient ID and stored in the first storage unit 40 together with the shooting date. calendar The age may be calculated automatically.

[0040] Next, the feature extraction module 12 extracts, for example, 100 training values ​​associated with, for example, 100 angle values ​​(training values). calendar The age is output to the age estimation module 13. In response to this, the age estimation module 13 outputs, for example, 100 pairs of angle values ​​(learning values) and learning values. calendar Age is input. The age estimation module 13 uses DNN to learn a regression model 14 based on this data, and saves the learned regression model 14. Here, the saving destination can be any location, such as the age estimation module 13 or the second storage unit 50.

[0041] In addition, dropout may be performed during DNN training to prevent overfitting and improve the generalization performance of the regression model 14.

[0042] Figure 4(a) shows the area containing the mandibular cortex (also called "cortical bone") in a panoramic X-ray image, and Figure 4(b) shows a schematic diagram of the principle of estimating biological age from cortical bone thickness using a regression model. As indicated by the tip of the arrow labeled "cortical bone" in Figure 4(a) (on the left side of Patient X), cortical bone thickness is the thickness of the hard, dense bone tissue (the hard, dense part) that covers the outermost part of the mandible. The width (linear distance) of the cortex in the center of the left side of the mandibular body can be measured.

[0043] The cortical bone thickness is not limited to the width of the cortex in the central part of the left side of the mandibular body, but may be, for example, the width of the cortex at the intersection of a perpendicular line drawn from the mental foramen (near the tip of the root of the second premolar (the fifth tooth from the front) of the mandible) to the lower margin tangent line. Alternatively, the cortical bone thickness may be the width (average, median, maximum, etc.) of the cortex of the entire mandibular body.

[0044] The cortical bone thickness may be measured on only the left side of the patient, only the right side of the patient, or both the left and right sides of the patient (for example, an average value).

[0045] The image input module 11 inputs a panoramic X-ray image of patient X from a panoramic X-ray imaging device 120 to the feature extraction module 12 via the input unit and communication unit 20. The CNN used by the feature extraction module 12 is, for example, a trained mandibular cortical bone CNN (a CNN for the mandibular cortex), and the mandibular cortical bone CNN of the feature extraction module 12 automatically extracts the coordinates of landmarks that serve as references for defining the mandibular cortex of patient X from the panoramic X-ray image of patient X. The feature extraction module 12 calculates the width of the cortex at the center of the left side of the mandibular body using the coordinates of the extracted landmarks.

[0046] Instead of processing the entire panoramic X-ray image, the mandibular cortical bone CNN can reduce computational costs by processing only the ROI, which can be just the left side of the patient, just the right side of the patient, or both the left and right sides of the patient.

[0047] Next, the mandibular cortical bone CNN can efficiently extract the mandibular contour by applying, for example, Canny edge processing only within the ROI on the lower side of the panoramic X-ray image. From the edges obtained by Canny edge processing, the mandibular cortical bone CNN can detect a predetermined region (specifically, a series of points) that forms a grayscale brightness (specifically, a set of points) equal to or greater than a predetermined value (threshold for mandibular cortex extraction) within the mandibular contour. The mandibular cortical bone CNN can calculate the width (cortical bone thickness) of the center of the detected predetermined region.

[0048] The mandibular cortical bone CNN may calculate a value corresponding to bone density from brightness values ​​such as the average, median, and maximum grayscale values ​​of the detected predetermined region. Here, the brightness value may be based on the distribution (e.g., histogram value) of the grayscale values ​​of the detected predetermined region.

[0049] In Figure 4(b), regression model 15 is calendar This shows the relationship between age and mandibular cortical bone thickness. The regression model 15 shown in Figure 4(b) is shown as a straight line for ease of understanding, but in the case of DNN, it is a complex curve with the same tendency as a straight line. In the simplified regression model 15 in Figure 4(b), calendar As age increases, cortical bone thickness decreases. When the measured cortical bone thickness of patient X is C, the age in regression model 15 corresponding to measurement C is patient X's biological age β.

[0050] In addition, patient X calendar The standard value of the regression model 15 corresponding to age is R, and when measurement value C<standard value R, biological age β estimated from measurement value C>patient X's calendar It becomes age.

[0051] Regression model 15 is calendar The relationship between age and cortical bone thickness may be shown and trained in advance using a DNN. When training a regression model 15 using a DNN, it is preferable to train a regression model using data from more than 100 people. calendar A regression model 15 is determined in advance based on the relationship between age and cortical bone thickness. Using a DNN, a highly accurate regression model 15 can be constructed with 100 samples, and the accuracy of the regression model 15 can be further improved with more than 100 samples.

[0052] The regression model 15 can be trained in the same manner as the regression model 14, and therefore a detailed description of the regression model 15 training method will be omitted.

[0053] 1 estimates biological age α from the angle of the jaw of patient X using regression model 14, and estimates biological age β from the thickness of the cortical bone of patient X using regression model 15. Age estimation module 13 can calculate the average value of a pair of biological ages α and β of patient X.

[0054] It should be noted that the age estimation module 13 may estimate only the biological age α or only the biological age β. However, estimating the biological ages α and β can more accurately estimate a set of biological ages of patient X. The average value of the biological ages α and β has the advantage of being easy to calculate.

[0055] The age estimation module 13 estimates the biological age of patient X (e.g., 0.5(α+β)). calendar The biological age and the age can be output to the comment generation module 16. The comment generation module 16 calendar Age and nature words Comments can be generated using word processing AI.

[0056] Here, nature wordsWord processing AI is a technology that enables computers to understand and generate words like humans, and it is possible to use generative AI chat functions such as ChatGPT, Gemini, and Claude. The comment generation module 16 uses the API of basic models such as OpenAI's "GPT-3.5 Turbo" and "GPT-4" to generate "biological age" and " calendar It is possible to generate simple comments according to the user's age.

[0057] The comment is, for example, a medical treatment comment of 400 to 800 characters, and the medical treatment comment may be for the patient only, for the dentist only, or for both the patient and the dentist.

[0058] The feature extraction module 12 calculates brightness values ​​corresponding to bone density based on a panoramic X-ray image of a patient, for example, using the mandibular cortical bone CNN and mandibular angle CNN. The age estimation module 13 estimates biological age and bone density. The comment generation module 16 generates comments based on the "biological age" and " calendar The comment generation module 16 may generate a clinical comment based only on the biological age β (cortical bone thickness) and bone density (brightness value) that indicate the bone condition.

[0059] Additionally, the comment generation module 16 may input dentist-entered comments based on, for example, keyboard 130 input, and generate practice comments that are also based on the dentist-entered comments.

[0060] The feature extraction module 12 or the mandibular cortical bone CNN or other well-known CNN may extract image diagnosis information (presence or absence of alveolar bone resorption, presence or absence of wisdom teeth, number of teeth) based on, for example, a panoramic X-ray image of a patient. The comment generation module 16 may generate an image comment based on the image diagnosis information. The comment generation module 16 may also generate a medical comment based on the image diagnosis information or the image comment.

[0061] The comment generating module 16 can store, for example, CSV format data (image name / estimated age / actual age / gender / bone cortex brightness / mandibular angle (left and right) / comments) in the comment generating module 16, the second storage unit 50, or the like.

[0062] Fig. 5 shows an example of output from the output module 17 in Fig. 1. The output module 17 can output the numerical values ​​extracted or estimated by the feature extraction module 12 or the age estimation module 13, and the comments generated by the comment generation module 16.

[0063] The output module 17 can output the saved numerical values ​​and / or comments in a predetermined format such as Word format, PDF format, etc. As shown in FIG. 5, the output module 17 can output, for example, (1) the biological age of the person indicated by the reference numeral 147 (51 years old indicated by the reference numeral 148, with an error of ±2 years) and (2) (a) the biological age of the person indicated by the reference numeral 142. calendar Output of age (50 years old as indicated by code 142) (3) Output of comments based on the panoramic X-ray image (medical comment 150 (estimated age 56 years in 3 years as indicated by code 149) which is a future prediction, and image comments 143 such as alveolar bone resorption as indicated by code 144, no wisdom teeth as indicated by code 145, number of teeth 26 as indicated by code 146) are displayed or printed, for example, on a monitor / printer 140.

[0064] In addition to or instead of the output of (2)(a), the output module 17 may output biological age and calendar The difference with age may be output.

[0065] As shown in Figure 5, the patient information, for example, reference numeral 141, output from the output module 17 may include, for example, dentist-input comments (lifestyle-related diseases, osteoporosis, etc.), for example, reference numeral 151, and gender (female), for example, reference numeral 151.

[0066] Fig. 6 shows an example of the configuration of an AI program according to Example 2. In Fig. 1, the computer 110 inputs panoramic X-ray images from the panoramic X-ray imaging device 120. In Example 2 of Fig. 6, the computer 210 inputs panoramic X-ray images from the panoramic X-ray imaging device 120 and stores them in the storage unit 40.

[0067] In other words, Clinic A may introduce, for example, a computer 110 (AI terminal), and Clinic B may introduce a computer 210 (storage terminal). The computer 210 (storage terminal) may have a first storage unit 40 that stores panoramic X-ray images and a communication unit 260 (first communication unit) that transmits the panoramic X-ray images to the communication unit 20 (second communication unit), and the computer 110 (AI terminal) may have a communication unit 20 (second communication unit) that receives the panoramic X-ray images from the communication unit 260 (first communication unit) and a control unit 30 that causes the image input module 11 to accept the panoramic X-ray images.

[0068] Here, the panoramic X-ray image may be a panoramic X-ray image of a patient, or may be a part or all of the learning panoramic X-ray images of at least 100 first-time patients.

[0069] In addition, computer 110 (AI terminal) may also input panoramic X-ray images (training panoramic X-ray images) from panoramic X-ray imaging device 120 or other dental panoramic X-ray imaging device, and regression model 14 and / or regression model 15 may be trained from many training panoramic X-ray images input and stored in computer 110 and / or computer 210.

[0070] Fig. 7 shows an example of the configuration of an AI program according to Example 3. In Fig. 1, the computer 110 is equipped with AI modules 10 such as an image input module 11, a feature extraction module 12, and an age estimation module 13. In Example 3 of Fig. 7, a cloud computer 310 can be equipped with the image input module 11, the feature extraction module 12, the age estimation module 13, etc.

[0071] In other words, the computer that operates the panoramic X-ray imaging device 120 or the computer built into the panoramic X-ray imaging device 120 does not need to implement the AI ​​module 10, and can input the panoramic X-ray image into the cloud computer 310 and receive biological age, etc. from the cloud computer 310.

[0072] The cloud computer 310 (AI platform cloud server) is, for example, Google Colab, which is equipped with a development environment based on Jupyter Notebook. The cloud computer 310 or Google Colab is specialized in machine learning and data analysis using Python, and Python code can be executed immediately on the cloud computer 310 with a browser on a local computer. In addition, the cloud computer 310 is equipped with high-performance GPUs and TPUs required for machine learning, making it possible to train large-scale models and perform complex calculations at high speed.

[0073] The feature extraction module 12 in Figures 1, 2, 6 and 7 can store features such as jaw angle, bone cortex, etc. in association with age and health condition. calendar The age (or chronological age) and / or biological age. The health condition may be based, for example, on keyboard 130 input or, alternatively, stored in a patient database. The comment generation module 16 may generate a comment (e.g., a medical comment) based on the age and health condition.

[0074] Here, health status may include, for example, physical health (various numerical values ​​measured in health checkups, etc.), mental health, social health, and the like.

[0075] 8 shows an example of the display of the output module 17 in FIG. 1. As shown in FIG. 8, the output module 17 outputs, for example, (1) the biological age of patient X (reference numeral 147) (36 years old, reference numeral 148) and (2) (a) the biological age of patient X (reference numeral 142). calendar(b) Age output (code 142's actual age 34 years old) and (b) biological age calendar The output of the difference with age (aging 2 years (+2 years) indicated by reference numeral 153) and (3) the output of the panoramic X-ray image (photographic image indicated by reference numeral 152) can be displayed or printed, for example, on a monitor / printer 140.

[0076] 1, 2, 6, and 7 can associate features such as jaw angle and cortical bone with treatment details and store them. The treatment details may be based on, for example, keyboard 130 input or may alternatively be stored in a patient database. The comment generation module 16 may generate comments (e.g., clinical comments) based on the treatment details.

[0077] Here, the treatment content may be, for example, the content of subsequent dental treatments including the initial consultation (including, for example, preventive dentistry, conservative treatment, periodontal disease treatment, restorative / prosthetic treatment, oral surgery, orthodontics, and cosmetic dentistry), and preferably includes periodontal disease treatment, root canal treatment, and re-root canal treatment, which are particularly related to aging.

[0078] The present invention is not limited to the above exemplary embodiments, and those skilled in the art will be able to easily modify the above exemplary embodiments to the extent that they fall within the scope of the claims. For example, the feature extraction module 12 improves the accuracy of estimating biological age by limiting the feature extraction module 12 to first-time patients. However, if the feature extraction module 12 also targets patients after the first visit in association with the treatment details, it can also estimate the biological age of the first-time patient after receiving a specific treatment. [Explanation of symbols]

[0079] 10...AI module, 11...image input module, 12...feature extraction module, 13...age estimation module, 14, 15...regression model, 16...comment generation module, 17...output module, 20...communication unit, 30...control unit, 40, 50...memory unit, 60...communication unit, 100...AI system, 110...computer (personal computer, cloud computer), 120...keyboard / mouse, 140...monitor / printer, 210...computer, 260...communication unit

Claims

1. 1. An artificial intelligence (AI) program for estimating biological age based on a plurality of dental panoramic radiographs, comprising: an image input module for receiving at least one dental panoramic x-ray image; a feature extraction module that extracts at least one feature related to the jaw bone including the alveolar bone from the at least one dental panoramic X-ray image using a convolutional neural network (CNN) and outputs at least one numerical value of the at least one feature; an age estimation module that estimates at least one biological age from the at least one numerical value using a trained regression model through machine learning; An AI program to execute the above.

2. The machine learning is a deep neural network (DNN). The AI ​​program according to claim 1.

3. the jawbone is a mandible, The at least one numerical value represents the angle of the jaw angle and / or the thickness of the cortical bone; The AI ​​program according to claim 1.

4. The plurality of dental panoramic X-ray images are training dental panoramic X-ray images of at least 100 first-time patients, The feature extraction module stores at least 100 training values ​​of at least 100 training features corresponding to the training jawbones of the at least 100 first-time patients in association with training chronological ages; the age estimation module trains the regression model from at least 100 training chronological ages associated with the at least 100 training numerical values ​​using a DNN as the machine learning; the image input module accepts an estimated panoramic X-ray image; the feature extraction module extracts at least one estimation feature related to the jawbone from the estimation panoramic X-ray image using the CNN, and outputs at least one estimation numerical value of the at least one estimation feature; the age estimation module estimates the at least one biological age from the at least one estimator value using the regression model; The AI ​​program according to claim 1.

5. the jawbone is a mandible, the feature extraction module extracts at least one set of jaw angle features and bone cortical features related to the mandible from the at least one dental panoramic X-ray image using the CNN, and outputs at least one set of angle numerical values ​​and thickness numerical values ​​of the at least one set of jaw angle features and bone cortical features; The age estimation module estimates an average value of at least one set of jaw angle biological age and bone cortical thickness biological age from the at least one set of angle numerical values ​​and thickness numerical values ​​using the regression model. The AI ​​program according to claim 4.

6. The computer includes a comment generation module that generates a comment using natural language processing AI based on the at least one biological age and chronological age. The AI ​​program according to claim 1, further comprising:

7. an output module configured to output to the computer: (1) the at least one biological age; (2) (a) a chronological age and / or (b) a difference between the at least one biological age and the chronological age; and (3) a comment based on the at least one dental panoramic radiograph. The AI ​​program according to claim 1, further comprising:

8. The feature extraction module and the age estimation module are built on an AI platform cloud server. The AI ​​program according to claim 1.

9. the feature extraction module stores the at least one feature in association with age and health status; The AI ​​program according to claim 1.

10. the feature extraction module stores the at least one feature in association with a medical treatment; The AI ​​program according to claim 1.

11. the storage terminal includes a storage unit that stores the at least one dental panoramic X-ray image and a first communication unit that transmits the at least one dental panoramic X-ray image to a second communication unit; The AI ​​terminal including the computer that executes the AI ​​program according to any one of claims 1 to 10, further comprising: the second communication unit that receives the at least one dental panoramic X-ray image from the first communication unit; and a control unit that causes the image input module to accept the at least one dental panoramic X-ray image, An AI system comprising the AI ​​terminal and the storage terminal.

12. 1. An artificial intelligence (AI) method for estimating biological age, comprising: inputting at least one dental panoramic x-ray image into an image input module; extracting at least one feature related to the jaw bone, including the alveolar bone, from the at least one dental panoramic X-ray image using a convolutional neural network (CNN) in a feature extraction module, and generating at least one numerical value of the at least one feature in the feature extraction module; an age estimation module estimating at least one biological age from the at least one numerical value using a machine learning trained regression model; A comment generation module generates a comment using natural language processing AI from the at least one of the biological age and the chronological age. AI method.

13. 1. An artificial intelligence (AI) method for estimating and displaying biological age, comprising: inputting at least one dental panoramic x-ray image into an image input module; extracting at least one feature related to the jaw bone, including the alveolar bone, from the at least one dental panoramic X-ray image using a convolutional neural network (CNN) in a feature extraction module, and generating at least one numerical value of the at least one feature in the feature extraction module; an age estimation module estimating at least one biological age from the at least one numerical value using a machine learning trained regression model; (1) outputting the at least one biological age and (2) (a) a chronological age and / or (b) a difference between the at least one biological age and the chronological age through an output module; (1) displaying the at least one biological age, (2) (a) the chronological age, and / or (b) the difference between the at least one biological age and the chronological age, and (3) the at least one dental panoramic x-ray image on a monitor at an output unit; AI method.

14. The machine learning is a deep neural network (DNN).

14. The AI ​​method according to claim 12 or 13.

Images