[go: up one dir, main page]

WO2020027481A1 - Système basé sur l'intelligence artificielle pour prédire la densité osseuse à l'aide de radiographies dentaires, et procédé de prédiction de la densité osseuse - Google Patents

Système basé sur l'intelligence artificielle pour prédire la densité osseuse à l'aide de radiographies dentaires, et procédé de prédiction de la densité osseuse Download PDF

Info

Publication number
WO2020027481A1
WO2020027481A1 PCT/KR2019/009060 KR2019009060W WO2020027481A1 WO 2020027481 A1 WO2020027481 A1 WO 2020027481A1 KR 2019009060 W KR2019009060 W KR 2019009060W WO 2020027481 A1 WO2020027481 A1 WO 2020027481A1
Authority
WO
WIPO (PCT)
Prior art keywords
bone density
bone
dental
unread
score
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/KR2019/009060
Other languages
English (en)
Korean (ko)
Inventor
이기선
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Korea University Research and Business Foundation
Original Assignee
Korea University Research and Business Foundation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from KR1020190087472A external-priority patent/KR102256315B1/ko
Application filed by Korea University Research and Business Foundation filed Critical Korea University Research and Business Foundation
Publication of WO2020027481A1 publication Critical patent/WO2020027481A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/50—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications
    • A61B6/51—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications for dentistry
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation

Definitions

  • the present invention relates to a bone density prediction system using artificial intelligence-based dental radiographs and to a method for predicting bone density. Specifically, the bone density score and the degree of bone density (normal, osteopenia, osteoporosis or severe osteoporosis) are predicted from dental radiographs. In addition, the present invention relates to a bone density prediction system using an artificial intelligence-based dental radiograph that can pre-screen osteoporosis-related diseases, and a method for predicting bone density by the same.
  • Osteoporosis is a systemic skeletal disease that increases bone fragility and is easily fractured due to a decrease in bone mineral density.
  • the osteoporosis is a disease that is expected to decrease the quality of life, increase mortality and increase medical expenses due to elderly fractures.
  • the medical diagnosis of osteoporosis is based on bone mineral density measurement using densitometry as a standard of WHO and the International Osteoporosis Foundation.
  • Conventional methods for measuring bone density include dual energy X-ray absorptiometry (DEXA), quantitative computed tomogram (QCT), or quantitative ultrasound (QUS).
  • dental radiography which is used for basic dental examinations, it is a relatively low-cost basic examination radiograph that can help to treat dental diseases because the entire jawbone, alveolar bone and entire teeth can be diagnosed in advance.
  • Non-Patent Documents 1 to 4 it is judged to be useful for the pre-screening of osteoporosis patients by analyzing the thickness or characteristic or morphology of the cortical bone of the mandibular bone observed by dental radiographs.
  • dental radiography equipment Since dental radiography equipment is available at any dental clinic, it can be used as a preliminary screening for osteoporosis patients, and as a recommendation tool for patients who are not aware of osteoporosis, to receive additional osteoporosis and physical diagnosis. If so, early detection of osteoporosis can help early treatment.
  • Patent Literature 1 discloses a method for determining whether osteoporosis is caused by the degree of cavities (nodal sites) of cortical bone.
  • Patent Literature 3 and Non-Patent Literature 2 disclose a method for determining osteoporosis according to the thickness of the mandibular cortical bone.
  • Patent Document 1 does not have specific details on the automation and image size of the image segment process for reading.
  • Patent Document 2 mentions the size of the segmented image, the resolution and size of the first photographed image are segmented into a uniform value without considering the matter, depending on the photographing equipment. Therefore, it is estimated that the size of the segmented region does not equally apply to the size of the actual anatomical structure.
  • the size of the panorama image differs depending on the shooting equipment. For example, when the 300X300 size is extracted from the 2000X1000 resolution and the 1000X5000 resolution panorama, the actual reflected size is different.
  • Patent Document 2 has a difficulty in determining whether the osteoporosis quickly because the segment process is estimated by a manual method.
  • Patent Document 3 the bone density is measured based on the thickness of the cortical bone of a specific site, but it is not specified about whether to automate the designation of a specific site, and there is a problem that the reliability of the bone density measurement is low.
  • Non-Patent Document 4 in the case of a patient with osteoporosis, the lower boundary line of the lower margin of the lower cortical bone is clear, but as shown in FIG. 16, the upper boundary line is not clear, so that the value of the cortical bone thickness due to the determination of the measurement site is determined. There is a possibility of large deviations.
  • Figure 16 is an illustration of the morphological characteristics of the mandibular lower cortical bone according to the change in bone density.
  • C1 is the morphological characteristics of the mandibular lower cortical bone of a normal person
  • C2 is the morphological characteristics of the mandible lower cortical bone of osteopenia patients
  • C3 is the morphological characteristics of the mandible lower cortical bone of osteoporosis patients.
  • the characteristic characteristic of the panoramic picture is that the hyoid bone is overlapped with the reading area of the cortical bone for osteoporosis determination, which is an obstacle to reading (see FIG. 5).
  • patent document 1 mentioned above is Unexamined-Japanese-Patent No. 2004-209089
  • patent document 2 is Unexamined-Japanese-Patent No. 2008-36068
  • patent document 3 is international publication 2006/043523.
  • the non-patent document 1 is described in T. Nakamoto, A. Taguchi, A. Asano, M. Ohtsuka, Y. Suei, M. Fujita, M. Sanada, K. Ohama, and K. Tanimoto, "Computer-aided diagnosis of low skeletal bone mass on panoramic radiographs, "presented at the 82nd General Session & Exhibition of the International Association for Dental Research no. 1953, Hawaii, 2004.,
  • Non-Patent Document 2 discloses A. Z. Arifin, A. Asano, A. Taguchi, T. Nakamoto, M. Ohtsuka, and K. Tanimoto, "Computer-aided system for measuring the mandibular cortical width on panoramic radiographs in osteoporosis diagnosis", in Proc. SPIE Med Imaging 2005-Image Processing Conference, San Diego, 2005, pp. 813-821,
  • Non-Patent Document 3 discloses MS Kavitha, SY An, CH An, KH Huh, WJ Yi, MS Heo, SS Lee, SC Choi, "Texture analysis of mandibular cortical bone on digital dental panoramic radiographs for the diagnosis of osteoporosis in Korean women". Oral Surg Oral Med Oral Pathol Oral Radiol. 2015 Mar; 119 (3): 346-356.,
  • Non-Patent Document 4 describes OS Kim, MH Shin, IH Song, IG Lim, SJ Yoon, OJ Kim, YH Lee, YJ Kim, HJ Chung, "Digital panoramic radiographs are useful for diagnosis of osteoporosis in Korean postmenopausal women", Gerodontology. 2016 Jun; 33 (2): 185-192.
  • the present invention is to predict the bone mineral density and bone mineral density (normal, osteopenia or osteoporosis) from the dental radiographs, bone density prediction system using artificial intelligence-based dental radiographs that can pre-screen osteoporosis-related diseases and thereby bone density prediction
  • the purpose is to provide a method.
  • the present invention removes the lower portion of the lower cortex and the hyoid bone in the dental radiography, artificial intelligence-based dental radiographs that can eliminate the possibility of reading errors due to the overlap of the lower cortex and lower bone when predicting bone density score
  • An object of the present invention is to provide a bone density prediction system using the same and a method for predicting bone density by the same.
  • the dental radiograph with bone density information on the bone density degree according to the bone density score and bone density score is input as bone density learning data, bone density learning Bone density learning unit through the deep learning to input data, the bone density of the jaw joint deep learning, the bone density prediction model is generated; And an unread dental radiograph is input, it is preferable to include an unread photograph evaluation unit for predicting the bone density score and bone density degree from the unread dental radiograph through the bone density prediction model.
  • the dental radiograph with the hyoid bone portion is specified as the first learning data, the deep learning to input the first learning data Osteotomy shadow learning unit through which the hyoid bone shadow processing model is generated; And a dental radiograph in which the mandible lower margin is designated as the second learning data, and through the deep learning using the second learning data, the contour learning unit for generating the mandible lower marginal contour detection model is further included.
  • the unread picture evaluation unit, the hyoid bone shadow processing unit for removing the hyoid bone shadow from the unread dental radiographs through the hyoid bone shadow processing model; Contour detection unit for detecting the contour of the lower mandibular margin from the unread dental radiograph through the contour detection model; Image preprocessing unit that recognizes the area between the mandibular vertebra and the mandible nodule as a detection site by using a predetermined detection site recognition algorithm in the unread dental radiography, and generates a detection image for reading the detection site. ; And a bone density prediction unit predicting a bone density score and a bone density degree according to the bone density score from the detection image for reading through the bone density prediction model.
  • the image preprocessing unit is a width (W) of the entire lower surface of the lower mandible between the mandrel and jaw nodule on the unread dental radiograph using a predetermined detection site recognition algorithm. It is preferable to recognize the area including the mandibular angle recess and the chin tip nodule as the detection site by setting a predetermined ratio of the total height H of the read dental radiograph.
  • the image preprocessing unit may generate a detection image for reading through noise removal and sharpening of the detection region according to a predetermined image processing algorithm.
  • the bone density prediction unit through the predetermined bone density classification algorithm, if the bone density score is -1.0 standard deviation or more "normal”, if the bone density score is within the range of -1.0 to -2.5 standard deviation, "osteopenia”, It is desirable to classify bone density as "osteoporosis” if the bone density score is below -2.5 standard deviations, or "severe osteoporosis” if the bone density score is below -2.5 standard deviations and has one or more fatal injuries.
  • the bone density learning unit learns bone density scores, which are output data of the unread photograph evaluation unit, and dental radiographs in which the degree of bone density is predicted as bone density learning data.
  • Bone density prediction method using an artificial intelligence-based dental radiograph according to an embodiment of the present invention, (A) using a dental radiograph with bone density information on the bone density degree according to the bone density score and bone density score as bone density learning data , Deep running the bone density of the jaw joint, generating a bone density prediction model; (B) inputting an unread dental radiograph; And (C) a step of predicting a bone density score from the unread dental radiograph and the degree of bone density according to the bone density score based on the bone density prediction model.
  • Step (A) includes (A1) generating a hyoid bone shadow processing model through deep learning, in which dental radiography having a hyoid bone shadow portion is designated as first learning data, and using first learning data as input; And (A2) it is preferable to further include the step of generating a mandibular lower marginal contour detection model through the deep learning that the dental radiograph with the lower mandible lower contour is specified as the second learning data, the second learning data as input .
  • step (C) comprises: (C1) removing the hyoid bone from the unread dental radiograph through the hyoid bone treatment model; (C2) detecting the contour of the lower mandibular margin from the unread dental radiograph through the contour detection model; (C3) recognizing a region including the mandibular angle recession and the mandible nodule in the lower mandible from the unread dental radiography through a preset detection region recognition algorithm as a detection region, and generating a detection image for reading the detection region; And (C4) it is preferable to include a step of predicting the bone density score and the degree of bone density from the detection image for reading through the bone density prediction model.
  • step (C3) is a width (W) of the entire lower surface of the lower mandible between the mandrel and jaw nodule on the unread dental radiograph using a predetermined detection site recognition algorithm.
  • step (C3) it is preferable to generate a detection image for reading through the noise removal and sharpening process for the detection region through a predetermined image processing algorithm.
  • step (C4) through a predetermined bone density classification algorithm, if the bone density score is more than -1.0 standard deviation "normal”, if the bone density score is within the range of -1.0 to -2.5 standard deviation " It is desirable to classify bone density as "osteoporosis”, “osteoporosis” if the BMD score is below -2.5 standard deviations, or "severe osteoporosis” if the BMD score is below -2.5 standard deviations and has one or more fatal injuries.
  • step (C) after step (C), according to the bone density prediction model, it is preferable that the dental radiographs of which the BMD and BMD are predicted are learned with BMD learning data.
  • the present invention can predict the bone mineral density and bone mineral density (normal, osteopenia or osteoporosis) from the dental radiograph, to pre-screen osteoporosis-related diseases.
  • the present invention includes a technique for removing the overlapping areas of the lower cortex and the hyoid bone from the dental radiograph, and at the same time, the artificial intelligence is automatically applied to the high-contrast contour areas that are not affected by other shadow images among the lower cortex.
  • Recognition of the BMD score and the stochastic prediction based on the WHO BMD classification can exclude reading errors and possibilities due to the overlap of mandibular cortex and hyoid bone and other shadow images. It can increase.
  • the present invention detects the detection region corresponding to the same anatomical size without being affected by the resolution and the image size, rather than being acquired by a predetermined pixel range because the resolution and image size of the image are different depending on the radiation device. can do.
  • the present invention compares the similarity of the texture-like Gaussian image of the lower jaw cortex and its periphery and the similarity of the boundary image with multiple deep learning-based similar image comparison methods, and thus, based on the existing cortical bone thickness.
  • this method it is possible to avoid the risk that the upper boundary of the lower margin of the mandibular cortical bone, which may occur in determining osteoporosis, may become more severe as osteoporosis progresses.
  • the present invention is not only to distinguish between normal and abnormal bone density disease, but also deep bone-based bone mineral density from the radiographs on the basis of bone mineral density by specifying a normal, osteopenia, osteoporosis or severe osteoporosis to provide bone density disease information can do.
  • the present invention can continuously improve the precision of the BMD score of the system by increasing additional learning data.
  • Figure 1 (a) is a schematic view of the jaw joint viewed from the side
  • Figure 1 (b) is a schematic view of the jaw joint viewed from the front.
  • Figure 2 schematically shows the configuration of a bone density prediction system using artificial intelligence based dental radiography according to an embodiment of the present invention.
  • FIG. 3 schematically illustrates a flowchart of a method for predicting bone density using artificial intelligence-based dental radiography according to an embodiment of the present invention.
  • Figure 4 is a flow chart for explaining the process of processing the hyoid bone in the dental radiography in one embodiment of the present invention.
  • FIG. 5 is an illustration of an image in which the hyoid bone overlaps with the mandible lower cortex.
  • 6 and 7 are examples of the dental radiograph before and after removal of the hyoid bone in the learning dental radiography.
  • FIG. 8 is an exemplary photograph for explaining a process of removing a hyoid bone shadow from an unread dental radiograph by the hyoid bone shadow processing unit according to an embodiment of the present invention.
  • FIG. 9 is a diagram for describing a process of generating a contour detection model in the contour learning unit and detecting a lower mandible lower contour from an unread dental radiograph by applying the contour detection model to the contour detection unit.
  • FIG. 10 is an exemplary view for explaining a process of automatically detecting a detection site in the mandible lower cortex bone in one embodiment of the present invention.
  • FIG. 11 is an exemplary view for explaining a process of extracting a detection image for reading from an unread dental radiograph of a hyoid bone.
  • FIG. 13 illustrates a process in which a bone density prediction model is generated in a bone density learning unit, and a bone density prediction model is applied to a bone density prediction unit, thereby predicting a bone density score and a bone density score from an unread dental radiograph in an embodiment of the present invention. It is a figure for following.
  • 15 is an exemplary photograph of normal, osteopenia or osteoporosis according to the degree of bone density.
  • Figure 16 is an illustration of the morphological characteristics of the mandibular lower cortical bone according to the change in bone density.
  • 17 is a table classifying the degree of bone density according to the bone mineral density score.
  • the bone density prediction system 100 using artificial intelligence-based dental radiography is the hyoid bone shadow learning unit 111, contour learning unit 113, bone density learning unit 115 ), The unread picture evaluation unit 130 is included.
  • the present invention 100 through the deep learning as a learning data input, the bone density of the jaw joint is deep learning, to generate a bone density prediction model, the bone density score and bone density from the unread dental radiographs through the bone density prediction model It is a technique of predicting the degree in a probabilistic form and providing the user with the bone density score and the bone density degree.
  • the hyoid bone shadow learning unit 111 generates a hyoid bone shadow processing model through deep learning using first learning data as an input.
  • the first learning data is a dental radiograph with designated hyoid bones.
  • the hyoid bone shadow learning unit 111 receives a dental radiography picture for learning (S1), and a hyoid bone shadow portion is designated on the learning dental radiograph (S2), thereby learning the hyoid bone shadow based on deep learning.
  • S3 To create the hyoid bone shadow processing model (S3).
  • the hyoid bone shadow processing model is applied to the hyoid bone shadow processing unit 131.
  • FIG. 5 (b), 6 (b), 7 (b) and 8 (b) are examples of dental radiographs in which the hyoid bone is present.
  • the hyoid bone shadow overlaps the mandible lower margin.
  • the presence of hyoid bone on dental radiography can cause reading errors in bone density prediction.
  • the present invention removes the hyoid bone from the dental radiography, it is possible to rule out the bone density prediction error due to the hyoid bone shadow.
  • the contour learning unit 113 generates a mandible lower edge contour detection model through deep learning using second learning data as an input.
  • the second learning data is a dental radiograph with a mandible lower margin outlined.
  • the contour learning unit 113 is a learning dental radiograph is input (S4), when the mandible lower margin contour is specified on the learning dental radiograph (S5), based on deep learning mandible lower margin contour By learning to generate a contour detection model (S6).
  • the contour detection model is applied to the contour detection unit 133.
  • the bone density learning unit 115 generates a bone density prediction model through deep learning using learning data as an input. Specifically, the bone density learning unit 115 learns the bone density of the jaw joint on the basis of deep learning, when the bone density score and the bone density degree corresponding to the learning dental radiograph and the learning dental radiograph are input (S7, S8), the bone density A predictive model is generated (S9). The bone density prediction model is applied to the bone density prediction unit 137.
  • the image knowing the existing bone density score (T-Score) is linked to the bone density score (T-Score (see FIG. 14)) to determine the corresponding bone density score according to texture characteristics through the deep learning-based AI learning and evaluation process.
  • Predictable BMD model is established.
  • Figure 14 is for the bone density score table associated with the detection image.
  • the detection image is an image generated by extracting only the detection site separately from the unread dental radiograph.
  • Bone density learning unit 115 is a dental radiograph of the normal group (see Fig. 15 (c)), osteopenia patients (see Fig. 15 (b)), osteoporosis patients (see Fig. 15 (c)) as shown in FIG.
  • the image of the detection site of is trained with bone density learning data.
  • the bone density learning unit 115 is a bone density classification algorithm is preset, the degree of bone density according to the bone density score is learned as bone density learning data. Bone mineral density is classified as normal, osteopenia, osteoporosis, or severe osteoporosis, according to the bone density score. Bone mineral density classification criteria are as shown in FIG.
  • the unread photo evaluation unit 130 to which the hyoid bone shadow processing model, the contour detection model, and the bone density prediction model are applied will be described.
  • the unread photo evaluation unit 130 includes the hyoid bone shadow processing unit 131, the contour detection unit 133, the image preprocessing unit 135, and the bone density prediction unit 137.
  • the hyoid bone shadow processing unit 131 removes the hyoid bone shadow from the unread dental radiograph through the hyoid bone shadow processing model.
  • a deep learning-based artificial intelligence device or program automatically By recognizing the superimposed hyoid shadow image and the algorithm to suppress the shadow image, only the lower cortical lower margin image with the hyoid shadow image removed from the unread dental radiography is programmed.
  • the hyoid bone shadow processing unit 131 inputs an unread dental radiography image (see FIG. 8 (a)) to the hyoid bone shadow learning data (for example, dental radiation shown in FIGS. 6 (b) and 7 (b)). Deep learning based on the photo, the hyoid bone area is automatically recognized in the unread dental radiography.
  • the hyoid bone shadow site (refer to FIG. 8 (b)) is recognized in the unread dental radiograph, the hyoid bone shadow site is removed from the unread dental radiograph as shown in FIG. 8 (c).
  • the unread dental radiography from which the hyoid bone image is removed is provided to the contour detection unit 133.
  • the contour detection unit 133 detects the contour of the lower mandible margin from the unread dental radiography image through the contour detection model.
  • the image preprocessing unit 135 recognizes a region between the mandibular vertebra and the mandible nodules in the contour of the lower mandible through a preset detection region recognition algorithm in an unread dental radiograph. Thus, a detection image for reading the detection portion is generated.
  • the detection site is selected in the dental radiographs in which the degeneration of cortical bone is prominent according to the degree of osteoporosis.
  • the detection site is the width (W) of the entire mandible lower edge between the mandibular recess and the jaw tip nodule on the dental radiograph, and the height of the predetermined height of the total height (H) of the unread dental radiograph In other words, the area including the mandibular angulation and the chin tip nodule.
  • the predetermined ratio may be set to a 10% ratio of the unread dental radiographs.
  • the height of the detection site is not necessarily limited to the 10% ratio of the unread dental radiograph, of course, can be adjusted at various ratios.
  • two detection sites are recognized.
  • the area between the left anterior angular notch and the mental tubercle, and the area between the right anterior mandibular and the mandible nodule are recognized as detection sites.
  • the inaccuracy of the measurement of cortical bone thickness due to ambiguity of the upper boundary of the lower margin of mandibular cortex caused by the problem of the existing patent, and the resolution of radiographic equipment and the size of radiographic image are different. It can solve the difference in the actual anatomical size of the detection site.
  • the image preprocessor 135 recognizes the detection site (see FIG. 11 (b)) and extracts the detection site (see FIG. 11 (c)) from an unread dental radiograph (see FIG. 11 (a)). According to an image processing algorithm, a detection image for reading is generated through noise removal and sharpening of the detection portion (see FIG. 11 (d)).
  • the image preprocessing algorithm may correspond to at least one or more image processing algorithms related to histogram equalization algorithm, Gaussian processing algorithm, canny algorithm, or other image resolution enhancement and noise removal.
  • the detection region is characterized by the operation of noise removal, leveling, Gaussian processing, boundary recognition, and the like.
  • the detection region is characterized by its operation by noise removal, leveling, Gaussian processing, boundary recognition, and the like.
  • FIG. 12 discloses an original image, a histogram-leveled image, a Gaussian-treated image, and a boundary-recognized image, for detection sites of normal and osteopenic patients.
  • the image processing unit generates a detection image for reading the detection portion in the unread dental radiograph, the reading detection image is provided to the bone density prediction unit 137.
  • the bone density prediction unit 137 predicts a bone density score from a detection image for reading through a bone density prediction model, and calculates a degree of bone density according to the bone density score.
  • the bone density prediction unit 137 matches the learning detection image of any one of the bone density learning data with the bone density prediction model, and the bone density score of the reading detection image from the bone density score of the matched learning detection image through the bone density prediction model. Predict.
  • the present invention compares the similarity of the texture-like Gaussian image of the lower mandible cortex and its surroundings with the similarity of the boundary image by using a deep image-based similar image comparison method.
  • the risk that the upper boundary of the lower margin of the mandibular cortical bone, which may occur in determining osteoporosis, may become more severe as osteoporosis progresses.
  • the bone density prediction unit 137 classifies the degree of bone density into normal, osteopenia, osteoporosis or severe osteoporosis according to the bone density score through a predetermined bone density classification algorithm.
  • the bone density score is calculated from "(measured value of patient-mean value of young population) ⁇ standard deviation".
  • the BMD score is compared with the BMD of the youngest adult who has the highest BMD to indicate the absolute risk of fracture. The lower the BMD, the lower the BMD.
  • the bone density classification algorithm is set according to the WHO bone density classification criteria (see FIG. 17). Referring to Figure 17, if the bone density score is more than -1.0 standard deviation is classified as "normal”. And, bone density classification algorithm is classified as "osteopenia” if the bone density score is within the range of -1.0 to -2.5 standard deviation. The bone density classification algorithm is classified as "osteoporosis” if the bone density score is less than -2.5 standard deviations. The bone density classification algorithm is classified as "severe osteoporosis” if the bone density score is less than -2.5 standard deviations and one or more fatal injuries.
  • the detection image for reading in which the bone density score is predicted may be provided to the bone density learning unit 115 and may be utilized as learning data.
  • the unread dental radiograph is input to the hyoid bone shadow processing unit 131 (S10).
  • the process of removing the hyoid bone shadow from the unread dental radiography is as follows.
  • the hyoid bone shadow processing unit 131 when the hyoid bone shadow is automatically recognized from the unread dental radiograph through the hyoid bone shadow processing model (S21), the hyoid bone shadow portion is removed from the unread dental radiograph (S22). Subsequently, the dental radiography from which the hyoid bone is removed from the unread photo is output (S23).
  • the present invention through step S20, by automatically recognizing the shadow image (shadow image of hyoid bone) on the dental radiography and can be avoided the possibility of error in bone density readings that can occur when overlapping the hyoid bone shadow.
  • a process of detecting the contour of the lower mandible margin from the dental radiography from which the hyoid bone is removed is performed (S30).
  • the contour detection unit 133 detects the contour of the lower mandible margin in the dental radiography picture through the contour detection model (S32).
  • the present invention includes a technique for removing the overlapping portion of the lower cortex and the hyoid bone from the dental radiograph through the step S30, and at the same time, artificially sharpen the contour portion of the lower cortex that is not affected by other shadow images in the lower cortex.
  • the intelligence is automatically recognized, and the prediction of bone density scores can be ruled out in the prediction of BMD and stochastic classification based on the WHO bone mineral density classification, eliminating errors and possibilities of reading of the mandibular cortex and hyoid superimposition and other shadow images. Can increase the accuracy.
  • the image preprocessing unit 135 recognizes a detection site based on the contour of the lower mandible in the dental radiography, and extracts only the detection site from the dental radiography (S41).
  • Step S41 is performed through a detection site recognition algorithm preset in the image preprocessor 135.
  • the image preprocessing unit 135 generates a detection image for reading through image preprocessing for the detection region through a predetermined image processing algorithm (S42).
  • the bone density prediction unit 137 predicts a bone density score through a bone density prediction model, and calculates a bone density degree from the bone density score through a predetermined bone density classification algorithm. (S52). Subsequently, the bone density prediction unit 137 outputs the risk information of the bone disease according to the bone density score and the bone density degree as the bone density prediction result (S53). In addition, the bone density prediction result is provided to the bone density learning unit 115, it can be utilized as learning data.
  • the reading detection image is subdivided into a histogram equalized image, a Gaussian processed image, and a boundary recognition processed image, and compared with a learning detection image processed by a corresponding algorithm in a deep learning method.
  • the learning detection image most similar to the read detection image may be searched for.
  • the thickness of the lower cortex of the mandible cortex is reflected through the specification of the detection site, and the morphological part of the periphery including the cortical bone from the Gaussian-treated image is reflected.
  • the characteristics are reflected, and the image particle density characteristics are reflected from the boundary-recognized image to predict the bone density score of the detection image for reading, thereby improving the accuracy of the predicted bone density score.
  • the present invention can provide a predicted bone density score numerically to recognize the degree of osteoporosis risk, not merely to distinguish between normal and abnormal.
  • the unread dental radiograph is input to the unread photograph evaluator 130, and the input unread dental radiograph has a BMD of 2.0 (T-Score, see FIG. 14). This 85% can provide information in a probabilistic form, as is expected with an 85% chance of osteopenia.
  • the present invention is not merely to distinguish between bone density disease as normal or abnormal, but to provide bone density disease information by specifying the bone density as normal, osteopenia, osteoporosis or severe osteoporosis along with the bone density score from the dental radiography on the basis of deep learning can do.
  • the present invention is based on the deep learning and reinforcement learning algorithm, the precision of the BMD score of the system can be continuously improved by increasing additional learning data.
  • the patient can be guided to the degree of osteoporosis risk that is predicted without a medical examination or examination of osteoporosis specialists by using only the dental panoramic radiation basically taken for dental examination.
  • the present invention can allow a large number of existing unspecified patients who are not aware of osteoporosis to visit the specialty department and receive additional radiological and physical examinations for the diagnosis of specialized osteoporosis and to start early treatment.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Physics & Mathematics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Optics & Photonics (AREA)
  • Pathology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Dentistry (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

Le système basé sur l'intelligence artificielle pour prédire la densité osseuse à l'aide de radiographies, selon un mode de réalisation de la présente invention, comprend de préférence : une unité d'apprentissage de densité osseuse, dans laquelle des radiographies dentaires comportant des informations de densité osseuse relatives à des scores de densité osseuse et des niveaux de densité osseuse sur la base des scores de densité osseuse sont introduites en tant que données d'apprentissage, et qui apprend en profondeur les densités osseuses des articulations temporo-mandibulaires par un apprentissage profond en utilisant les données d'apprentissage en tant qu'entrée, et génère un modèle de prédiction de densité osseuse ; et une unité d'évaluation de photographie non lue, dans laquelle une radiographie dentaire non lue est introduite et qui prédit un score de densité osseuse et un niveau de densité osseuse à partir de la radiographie dentaire non lue, au moyen du modèle de prédiction de densité osseuse.
PCT/KR2019/009060 2018-08-03 2019-07-23 Système basé sur l'intelligence artificielle pour prédire la densité osseuse à l'aide de radiographies dentaires, et procédé de prédiction de la densité osseuse Ceased WO2020027481A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
KR20180090627 2018-08-03
KR10-2018-0090627 2018-08-03
KR1020190087472A KR102256315B1 (ko) 2018-08-03 2019-07-19 인공지능 기반의 치과방사선사진을 이용한 골밀도 예측시스템 및 이에 의한 골밀도 예측 방법
KR10-2019-0087472 2019-07-19

Publications (1)

Publication Number Publication Date
WO2020027481A1 true WO2020027481A1 (fr) 2020-02-06

Family

ID=69232252

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2019/009060 Ceased WO2020027481A1 (fr) 2018-08-03 2019-07-23 Système basé sur l'intelligence artificielle pour prédire la densité osseuse à l'aide de radiographies dentaires, et procédé de prédiction de la densité osseuse

Country Status (1)

Country Link
WO (1) WO2020027481A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113869443A (zh) * 2021-10-09 2021-12-31 新大陆数字技术股份有限公司 基于深度学习的颌骨密度分类方法、系统及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004209089A (ja) * 2003-01-07 2004-07-29 Japan Science & Technology Agency パノラマｘ線画像を用いた骨粗鬆症診断支援装置
WO2006043523A1 (fr) * 2004-10-19 2006-04-27 Hiroshima University Independent Administrative Agency Appareil pour faciliter le diagnostic de l’ostéoporose
JP2008036068A (ja) * 2006-08-04 2008-02-21 Hiroshima Univ 骨粗鬆症診断支援装置および方法、骨粗鬆症診断支援プログラム、骨粗鬆症診断支援プログラムを記録したコンピュータ読み取り可能な記録媒体、骨粗鬆症診断支援用ｌｓｉ
KR20170016778A (ko) * 2015-08-04 2017-02-14 재단법인 아산사회복지재단 심층신경망을 이용한 골 연령 산출방법 및 프로그램
KR20180029476A (ko) * 2016-09-12 2018-03-21 주식회사 뷰노 골밀도 추정 방법 및 장치

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004209089A (ja) * 2003-01-07 2004-07-29 Japan Science & Technology Agency パノラマｘ線画像を用いた骨粗鬆症診断支援装置
WO2006043523A1 (fr) * 2004-10-19 2006-04-27 Hiroshima University Independent Administrative Agency Appareil pour faciliter le diagnostic de l’ostéoporose
JP2008036068A (ja) * 2006-08-04 2008-02-21 Hiroshima Univ 骨粗鬆症診断支援装置および方法、骨粗鬆症診断支援プログラム、骨粗鬆症診断支援プログラムを記録したコンピュータ読み取り可能な記録媒体、骨粗鬆症診断支援用ｌｓｉ
KR20170016778A (ko) * 2015-08-04 2017-02-14 재단법인 아산사회복지재단 심층신경망을 이용한 골 연령 산출방법 및 프로그램
KR20180029476A (ko) * 2016-09-12 2018-03-21 주식회사 뷰노 골밀도 추정 방법 및 장치

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113869443A (zh) * 2021-10-09 2021-12-31 新大陆数字技术股份有限公司 基于深度学习的颌骨密度分类方法、系统及介质

Similar Documents

Publication Publication Date Title
KR102256315B1 (ko) 인공지능 기반의 치과방사선사진을 이용한 골밀도 예측시스템 및 이에 의한 골밀도 예측 방법
JP6830082B2 (ja) 歯科分析システムおよび歯科分析ｘ線システム
JP5763172B2 (ja) パノラマエックス線写真を利用した診断支援システム、及びパノラマエックス線写真を利用した診断支援プログラム
WO2018080194A1 (fr) Système et procédé pour diagnostiquer une lésion de la cavité buccale
KR101684998B1 (ko) 의료영상을 이용한 구강병변의 진단방법 및 진단시스템
TWI830036B (zh) 骨骼特徵值估計方法
Bodade et al. Panoramic radiography for screening postmenopausal osteoporosis in India: a pilot study
JP3964795B2 (ja) パノラマｘ線画像を用いた骨粗鬆症診断支援装置
WO2020027481A1 (fr) Système basé sur l'intelligence artificielle pour prédire la densité osseuse à l'aide de radiographies dentaires, et procédé de prédiction de la densité osseuse
Dias et al. Osteoporosis screening: leveraging EfficientNet with complete and cropped facial panoramic radiography imaging
Top et al. Quantitative level determination of fixed restorations on panoramic radiographs using deep learning
JP2019208832A (ja) 歯科分析システムおよび歯科分析ｘ線システム
Andersen et al. Individual identification by means of conventional bitewing film and subtraction radiography
Hase et al. Sex estimation from maxillofacial radiographs using a deep learning approach
WO2019098399A1 (fr) Procédé d'estimation de la densité minérale osseuse et appareil l'utilisant
Wood et al. Digitized slice interposition in forensic dental radiographic identification: An in vitro study
KR101069026B1 (ko) 치아우식증 분석을 위한 이미지 변환 시스템 및 그 방법
WO2024214853A1 (fr) Procédé basé sur l'apprentissage automatique pour déduire une densité minérale osseuse à l'aide d'une image radiographique d'une articulation de hanche par rayons x
Juliastuti et al. Dental panoramic image analysis for enhancement biomarker of mandibular condyle for osteoporosis early detection
Sivari et al. Deep learning in diagnosis of dental anomalies and diseases: a systematic review. Diagnostics. 2023; 13: 2512
WO2003049615A1 (fr) Procede et appareil permettant d'etablir une mesure de l'osteoporose
Saha et al. Redefining Dental Radiology Using Deep Learning
Velmurugadass et al. Dentalcariesdetectionsystem Using R-CNN Algorithm
Stancioi et al. Image Segmentation based System for Computing the Shifting of a Tooth
Yılmaz et al. Artificial intelligence for detecting dental ankylosis in primary molars using panoramic radiographs-a retrospective study

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19844702

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19844702

Country of ref document: EP

Kind code of ref document: A1