A jaw cyst identification method and system based on artificial intelligence
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QINGDAO MEDICON DIGTAL ENG CO LTD
- Filing Date
- 2026-04-30
- Publication Date
- 2026-06-26
Smart Images
Figure CN122289802A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for identifying jaw cysts based on artificial intelligence. Background Technology
[0002] Jawbone cysts are fluid-filled cystic lesions within the jawbone. They can occur anywhere on the jawbone, initially without any symptoms. As they grow slowly, the cyst can compress and expand the surrounding bone. Jawbone cysts are one of the most common diseases in oral and maxillofacial surgery, characterized by high incidence, insidious nature, and high misdiagnosis rate. Many patients with jawbone cysts are diagnosed with significant bone destruction at the time of diagnosis. Continued disease progression or improper treatment can lead to gingival retraction, tooth displacement, malocclusion, and in severe cases, maxillofacial deformities, significantly impacting the patient's quality of life.
[0003] Currently, the identification of jaw cysts is usually done by dentists manually reviewing CBCT images, lacking automated identification and detection methods. Summary of the Invention
[0004] In view of the above problems, the present invention is proposed to provide an artificial intelligence-based method and system for identifying jaw cysts that overcomes the above problems.
[0005] This invention provides an artificial intelligence-based method for identifying jaw cysts, the method comprising: Identify jawbone cyst lesion areas in oral CBCT images to be processed; 3D connected component analysis was performed on the jawbone cyst lesion area, and the connected component with the largest volume was taken as the region of interest of the cyst lesion. Image features of the region of interest of the cyst lesion were extracted. Identify the location regions of each tooth in the oral cavity within the oral CBCT image; The target tooth is determined based on the positional relationship between the location of each tooth and the area of the jawbone cyst lesion. A first spatial feature is constructed for the jaw cyst lesion region, which is used to characterize whether the coordinates of the center point of the jaw cyst lesion region overlap with the coordinates of the target tooth location region; Identify the functional region of interest (FROI) at the root apex of the target tooth and extract the image features of the FROI. A second spatial feature is constructed for the jawbone cyst lesion region, which is used to characterize whether the jawbone cyst lesion region overlaps with the midline region of the oral CBCT image; The image features of the region of interest for cystic lesions, the image features of the region of interest for root apical function, and the first and second spatial features of the jawbone cystic lesion region are concatenated to generate a joint feature vector. The joint feature vector is input into a pre-trained machine learning model for classification and identification to determine the lesion nature of jawbone cysts.
[0006] Further, the jawbone cyst lesion region in the oral CBCT images to be processed was identified, including: A pre-trained 3D convolutional segmentation model was used to identify the oral CBCT images to be processed, in order to identify the jaw cyst lesion area in the oral CBCT images; The training steps for a 3D convolutional segmentation model include: The annotation tool is used to annotate the outline region of the cyst lesion on the CT image samples in the preset oral CBCT image set, and an annotated volume data file is generated. The pixel value of 0 in the annotated volume data file represents the background area, and the pixel value of 1 represents the corresponding coordinate position in the image as the jaw cyst lesion area. A 3D convolutional segmentation model was trained using labeled volume data files as training data.
[0007] Furthermore, image features of the region of interest in the cystic lesion are extracted, including: The morphological, grayscale, and texture features of the region of interest in the cystic lesion are extracted. The morphological features include volume, surface area, sphericity, elongation, flatness, and surface area to volume ratio. The grayscale features include the mean, median, entropy, skewness, and kurtosis of the pixel grayscale values. The texture features include the grayscale co-occurrence matrix, the grayscale size region matrix, and the grayscale run length matrix.
[0008] Furthermore, the target teeth are determined based on the positional relationship between the location of each tooth and the area of the jawbone cyst lesion, including: Determine whether the coordinates of the center point of the jawbone cyst lesion overlap with the coordinates of the location areas of the teeth in the oral cavity; If there is overlap, the tooth whose center point coordinates overlap with the center point coordinates of the jaw cyst lesion area is selected as the target tooth; otherwise, the distance between the center point coordinates of the jaw cyst lesion area and the center point coordinates of the location areas of each tooth in the oral cavity is calculated, and the tooth corresponding to the shortest distance is selected as the target tooth.
[0009] Furthermore, the region of interest at the root apex of the target tooth is identified, including: Obtain the tooth number and corresponding pulp number of the target tooth; Based on the tooth number and pulp number of the target tooth, the sagittal plane is sliced layer by layer along the X-axis in the volume data of the oral CBCT image, and the pixel area of the target tooth in each layer is calculated to obtain the target sagittal plane image with the largest pixel area. Identify the region of interest at the root apex of the target tooth in the target sagittal image.
[0010] Further, identifying the region of interest (ROI) at the root apex of the target tooth in the target sagittal image includes: The tooth contour and corresponding pulp contour of the target tooth are obtained from the target sagittal image; Identify the root apex of the tooth in the tooth profile of the target tooth, and identify the coordinate point of the pulp closest to the root apex in the pulp profile of the target tooth. Calculate the coordinates of the first midpoint between the root apex of the tooth and the point in the pulp closest to the root apex. Then, draw a circular region centered on the first midpoint and with the distance from the first midpoint to the root apex as the radius, and use this circular region as the functional region of interest.
[0011] Furthermore, prior to constructing the second spatial features of the jawbone cyst lesion region, the method further includes: Identify the left maxillary sinus region, right maxillary sinus region, and pharyngeal region in the oral CBCT images; In the transverse section of the oral CBCT image, find the 2D section image with the largest total outline area of the left and right maxillary sinuses, and use it as the target 2D section image. The coordinates of the center point of the left maxillary sinus, the center point of the right maxillary sinus, and the center point of the pharynx were identified in the target 2D cross-sectional image. Calculate the coordinates of the second midpoint between the center point of the left maxillary sinus contour and the center point of the right maxillary sinus contour; The line connecting the coordinates of the second midpoint and the coordinates of the center point of the pharynx contour is taken as the 2D midline of the target 2D cross-sectional image; The 2D midline is expanded into a three-dimensional midline region in the volume data corresponding to the oral CBCT image and extended vertically along the Z-axis to generate a 3D midline region that runs through the volume contour of the volume data, thus obtaining the midline region of the oral CBCT image.
[0012] Furthermore, after generating the joint feature vector, the method further includes: The joint feature vector is normalized. The normalized joint feature vector is subjected to T-test feature selection, Pearson correlation coefficient feature selection, and Lasso regression coefficient feature selection in sequence to shrink the regression coefficients of redundant features to 0.
[0013] Another aspect of the present invention provides an artificial intelligence-based jawbone cyst identification system, the system comprising: The lesion area identification module is used to identify jaw cyst lesion areas in oral CBCT images to be processed; The first feature extraction module is used to perform 3D connected component analysis on the jawbone cyst lesion area, and to extract the image features of the cyst lesion region of interest by taking the largest connected component as the region of interest of the cyst lesion. The oral cavity tooth position recognition module is used to identify the positional regions of each tooth in the oral cavity in the oral CBCT image; The selection module is used to determine the target tooth based on the positional relationship between the location of each tooth and the jawbone cyst lesion area. The first feature generation module is used to construct the first spatial feature of the jaw cyst lesion area. The first spatial feature is used to characterize whether the coordinates of the center point of the jaw cyst lesion area overlap with the coordinates of the target tooth location area. The second feature extraction module is used to identify the functional region of interest (FROI) at the root apex of the target tooth and extract the image features of the FROI. The second feature generation module is used to construct the second spatial feature of the jaw cyst lesion area. The second spatial feature is used to characterize whether the jaw cyst lesion area overlaps with the midline area of the oral CBCT image. The feature stitching module is used to stitch together the image features of the region of interest of the cystic lesion, the image features of the region of interest of the root apex function, and the first and second spatial features of the jawbone cystic lesion region to generate a joint feature vector. The lesion nature identification module is used to input the joint feature vector into a pre-trained machine learning model for classification and identification, so as to identify the lesion nature of jawbone cyst lesions.
[0014] Furthermore, the system also includes: The feature optimization module is used to normalize the joint feature vector after the feature splicing module generates it; and to perform T-test feature selection, Pearson correlation coefficient feature selection and Lasso regression coefficient feature selection on the normalized joint feature vector in sequence, so as to shrink the regression coefficient of redundant features to 0.
[0015] The artificial intelligence-based method and system for identifying jaw cysts provided by this invention can effectively improve the accuracy of identifying and assessing the nature of jaw cysts by constructing an intelligent machine learning model that can automatically extract features from labeled medical image data and classify and identify the nature of cyst lesions. This has significant clinical application value.
[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a flowchart of an artificial intelligence-based method for identifying jaw cysts according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the target sagittal image in an embodiment of the present invention; Figure 3 This is a schematic diagram of the region of interest for apical function in an embodiment of the present invention. Figure 4 This is a schematic diagram of the centerline region in an embodiment of the present invention; Figure 5 This is a structural block diagram of an artificial intelligence-based jaw cyst identification system according to an embodiment of the present invention. Detailed Implementation
[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined.
[0020] Figure 1 The flowchart illustrating an artificial intelligence-based method for identifying jaw cysts provided by an embodiment of the present invention is shown. Figure 1 As shown, the artificial intelligence-based jaw cyst identification method proposed in this invention includes the following steps: S11. Identify jawbone cyst lesion areas in the oral CBCT images to be processed.
[0021] S12. Perform 3D connected component analysis on the jawbone cyst lesion area, and take the largest connected component as the region of interest of the cyst lesion to extract the image features of the region of interest of the cyst lesion.
[0022] Specifically, the image features of the region of interest for the cystic lesion are extracted, including: morphological features, grayscale features, and texture features of the region of interest for the cystic lesion; among which, morphological features include: volume, surface area, sphericity, elongation, flatness, and surface area to volume ratio; grayscale features include: mean, median, entropy, skewness, and kurtosis of pixel grayscale; texture features include: grayscale co-occurrence matrix, grayscale size region matrix, and grayscale run length matrix.
[0023] In this embodiment of the invention, the identified jawbone cyst lesion area is first processed. Specifically, the mask region represents different segmented regions, with pixel value 0 = background and pixel value 1 = cyst. Then, through 3D connected component analysis, only the largest connected component is retained, and falsely detected small fragments are removed. The largest connected component is extracted as the region of interest (ROI) of the cyst lesion within the cyst cavity. Specifically, the PyRadiomics library can be used to extract the morphological features, grayscale features, and texture features of the cyst lesion ROI.
[0024] S13. Identify the location regions of each tooth in the oral cavity in the oral CBCT image.
[0025] S14. Determine the target tooth based on the positional relationship between each tooth and the jawbone cyst lesion area. Specifically, select the tooth closest to the jawbone cyst lesion area based on the positional relationship of each tooth and use it as the target tooth.
[0026] S15. Construct a first spatial feature of the jawbone cyst lesion region, wherein the first spatial feature is used to characterize whether the coordinates of the center point of the jawbone cyst lesion region overlap with the coordinates of the location region of the target tooth.
[0027] In this embodiment, if the coordinates of the center point of the jawbone cyst lesion area overlap with the coordinates of the target tooth location area, the first spatial feature is set to "overlapping coordinates of cyst and tooth root" = 1, and if they do not overlap, "overlapping coordinates of cyst and tooth root" = 0.
[0028] S16. Identify the functional region of interest (GROUP) at the root apex of the target tooth and extract the image features of the GROUP. The image features of the GROUP are extracted using the same method as those of the GROUP for cystic lesions, and will not be described again here.
[0029] S17. Construct a second spatial feature of the jawbone cyst lesion area. The second spatial feature is used to characterize whether the jawbone cyst lesion area overlaps with the midline area of the oral CBCT image.
[0030] In this embodiment, if the jawbone cyst lesion area overlaps with the midline area, the second spatial feature is set to "cyst overlaps with midline area" = 1, and if they do not overlap, "cyst overlaps with midline area" = 0.
[0031] S18. The image features of the region of interest of the cystic lesion, the image features of the region of interest of the root apex function, and the first and second spatial features of the jawbone cystic lesion region are concatenated to generate a joint feature vector.
[0032] S19. The joint feature vector is input into a pre-trained machine learning model for classification and identification to determine the nature of the jawbone cyst lesions. The nature of the jawbone cyst lesions includes: apical cysts, odontogenic keratocysts, and non-odontogenic cysts.
[0033] In this embodiment, machine learning models such as XGBoost, RF, and SVM can be trained to input oral CBCT images of a patient with a jaw cyst. The image features of the region of interest of the cyst lesion, the image features of the region of interest of the periapical function, and the first and second spatial features of the jaw cyst lesion area can be extracted and spliced to generate a joint feature vector. By inputting the joint feature vector into the machine learning model, the lesion nature of the periapical cyst, keratotic cyst, and non-odontogenic cyst can be determined.
[0034] The artificial intelligence-based method for identifying jaw cysts provided in this invention identifies the lesion area of jaw cysts and the location areas of various teeth in the oral cavity through AI. Further feature analysis and extraction are performed on the lesion area and the location areas of various teeth to construct a series of radiomics features, including image features of the region of interest of the cyst lesion, image features of the region of interest of the periapical function, and multi-dimensional features such as the first spatial features and the second spatial features of the jaw cyst lesion area. Based on an intelligent machine learning model, the multi-dimensional radiomics features are classified and identified to determine the nature of the jaw cyst lesion, effectively improving the accuracy of identification and assessment of the nature of jaw cyst lesions, and has significant clinical application value.
[0035] In this embodiment of the invention, step S11, identifying the jawbone cyst lesion region in the oral CBCT image to be processed, specifically includes: using a pre-trained 3D convolutional segmentation model to identify the oral CBCT image to be processed, so as to identify the jawbone cyst lesion region in the oral CBCT image.
[0036] The training steps for the 3D convolutional segmentation model include: The annotation tool is used to mark the contour regions of cystic lesions on CT image samples from a pre-defined oral CBCT image set, generating an annotated volumetric data file. In the annotated volumetric data file, a pixel value of 0 represents the background area, and a pixel value of 1 indicates that the corresponding coordinate position in the image is the jawbone cystic lesion area. Specifically, a pre-defined oral CBCT image set is obtained by collecting oral CBCT images of patients with jawbone cysts. The 3DSlicer annotation tool is then used to mark the contour regions of cystic lesions on the CT images in the image set, generating an annotated volumetric data file.
[0037] Annotated volumetric data files are used as training data to train a 3D convolutional segmentation model. In this embodiment, the open-source NNUnet-V2 algorithm architecture can be used to train the 3D convolutional segmentation model. This allows the model to infer and identify the corresponding volumetric data segmentation file from a patient's CBCT volumetric image. Consistent with the annotated data, a pixel value of 0 in the identified volumetric data file represents the background region, and a pixel value of 1 represents the corresponding coordinate location in the image as a jawbone cyst lesion.
[0038] In this embodiment of the invention, step S13, identifying the location regions of each tooth in the oral cavity in the oral CBCT image, specifically includes: using a pre-trained second 3D convolutional segmentation model to identify the oral CBCT image to be processed, so as to identify the location regions and key parts of each tooth in the oral cavity in the oral CBCT image.
[0039] Specifically, a second 3D convolutional segmentation model was trained using the open-source labeled dataset ToothFairy3 Dataset. The ToothFairy3 Dataset has the following categories: "labels": { "background": 0, "Mandible": 1, "Maxilla": 2, "Left inferior alveolar nerve canal": 3, "Right inferior alveolar nerve canal": 4, "Left maxillary sinus": 5, "Right maxillary sinus": 6, "Pharynx": 7, "Dental bridge": 8, "Dental crown": 9, "Implants": 10, "Maxillary right central incisor": 11, "Right maxillary lateral incisor": 12, "Right maxillary canine": 13, "Maxillary right first premolar": 14, "Maxillary right second premolar": 15, "Maxillary right first molar": 16, "Maxillary right second molar": 17, "Maxillary right third molar (wisdom tooth)": 18, "Maxillary left central incisor": 19, "Left maxillary lateral incisor": 20, "Left maxillary canine": 21, "Maxillary left first premolar": 22, "Maxillary left second premolar": 23, "Maxillary left first molar": 24, "Maxillary left second molar": 25, "Maxillary left third molar (wisdom tooth)": 26, "Left mandibular central incisor": 27, "Left lateral incisor of the lower jaw": 28, "Left mandibular canine": 29, "Left mandibular first premolar": 30, "Left mandibular second premolar": 31, "Left mandibular first molar": 32, "Left mandibular second molar": 33, "Lower left third molar (wisdom tooth)": 34, Mandibular right central incisor: 35, "Right lateral incisor of the lower jaw": 36, "Right lower canine": 37, "Right mandibular first premolar": 38, "Right mandibular second premolar": 39, "Right mandibular first molar": 40, "Right lower second molar": 41, "Right lower third molar (wisdom tooth)": 42, "Left mandibular incisor canal": 43, "Right mandibular incisor canal": 44, "Lingual tube": 45, "Maxillary right central incisor pulp": 46, "Maxillary right lateral incisor pulp": 47, "Pulp of the right maxillary canine": 48, "Pulp of the first premolar on the right maxillary side": 49, "Pulp of the second premolar on the right side of the maxilla": 50, "Pulp of the first maxillary right molar": 51, "Pulp of the second maxillary molar on the right side": 52, "Pulp of the maxillary right third molar (wisdom tooth): 53, "Maxillary left central incisor pulp": 54, "Maxillary left lateral incisor pulp": 55, "Pulp of the left maxillary canine": 56, "Pulp of the maxillary left first premolar": 57, "Pulp of the maxillary left second premolar": 58, "Pulp of the left maxillary first molar": 59, "Pulp of the maxillary left second molar": 60, "Pulp of the maxillary left third molar (wisdom tooth): 61, "Pulp of the left mandibular central incisor": 62, "Pulp of the left mandibular lateral incisor": 63, "Pulp of the left mandibular canine": 64, "Pulp of the left mandibular first premolar": 65, "Pulp of the second premolar on the left side of the mandible": 66, "Pulp of the left mandibular first molar": 67, "Pulp of the second mandibular molar on the left side": 68, "Pulp of the left mandibular third molar (wisdom tooth)": 69, "Pulp of the right mandibular central incisor": 70, "Implantable pulp of the right lateral incisor of the mandible": 71, "Pulp of the right mandibular canine": 72, "Pulp of the first premolar on the right side of the mandible": 73, "Pulp of the second premolar on the right side of the mandible": 74, "Pulp of the right mandibular first molar": 75, "Pulp of the right mandibular second molar": 76, "Pulp of the right mandibular third molar (wisdom tooth)": 77 } Based on the above dataset, a second 3D convolutional segmentation model was trained using the open-source nnUnet-V2 algorithm structure. This model takes a patient's CBCT volumetric image as input and uses the segmentation model to infer and identify the corresponding volumetric data segmentation file, obtaining the location regions of each tooth, the location regions of key parts, and their corresponding numbers.
[0040] In this embodiment of the invention, step S14, determining the target tooth based on the positional relationship between the location regions of each tooth and the jawbone cyst lesion region, includes the following steps: Determine whether the coordinates of the center point of the jaw cyst lesion area overlap with the coordinates of the location areas of each tooth in the oral cavity. If they overlap, select the tooth whose coordinates overlap with the center point of the jaw cyst lesion area as the target tooth. Otherwise, calculate the distance between the coordinates of the center point of the jaw cyst lesion area and the coordinates of the center points of the location areas of each tooth in the oral cavity, and select the tooth with the shortest distance as the target tooth.
[0041] Specifically, as one of the criteria for determining "dentine cysts", the coordinates of the center point of the cyst lesion are first obtained through the first 3D convolutional segmentation model, and the coordinates of 32 teeth are obtained through the second 3D convolutional segmentation model. If the center point of the cyst lesion is included in the coordinates of the above 32 teeth, it means that the coordinates of the cyst and the tooth overlap; otherwise, they do not overlap.
[0042] This embodiment determines whether the center of the cyst lesion overlaps with the coordinates of the tooth root by "determining whether the coordinates of the cyst lesion center overlap with the coordinates of the tooth root". If they overlap, the overlapping tooth is designated as the tooth with the "smallest distance", which is the target tooth. If they do not overlap, the coordinates of the center points of 32 tooth regions are calculated. Based on the coordinates of the center point of the cyst lesion, the shortest distance is calculated with the coordinates of the center points of these 32 tooth regions, and the tooth with the "smallest distance" is identified as the target tooth.
[0043] In this embodiment of the invention, step S15 constructs the first spatial feature of the jawbone cyst lesion area. That is, when the coordinates of the center point of the jawbone cyst lesion area overlap with the coordinates of the position area of the teeth in the oral cavity, a feature value is set: "Cyst and tooth root coordinates overlap" = 1; if they do not overlap, the feature value is set: "Cyst and tooth root coordinates overlap" = 0.
[0044] In this embodiment of the invention, the "periapical functional region" corresponding to the tooth adjacent to the cyst is a key area for determining a "periapical cyst". In this embodiment, identifying the periapical functional region of interest of the target tooth in step S16 specifically includes the following steps (not shown in the accompanying drawings): S161. Obtain the tooth number and corresponding pulp number of the target tooth. After obtaining the target tooth, based on the recognition results of the second 3D convolutional segmentation model, the tooth number and outline of the target tooth, as well as the corresponding pulp number and outline, can be obtained.
[0045] S162. Based on the tooth number and pulp number of the target tooth, cut the sagittal plane layer by layer along the X-axis in the volume data of the oral CBCT image, calculate the pixel area of the target tooth in each layer, and obtain the target sagittal plane image with the largest pixel area.
[0046] Each tooth number has a corresponding independent label in the volumetric data segmentation file. Based on the recognition results, the sagittal plane is sliced layer by layer along the X-axis from the volumetric data of the original image. The tooth pixel area of the target tooth in each layer is calculated, and the sagittal slice with the largest area is found. The sagittal image of that layer is then exported. Figure 2 As shown.
[0047] S163. Identify the apical functional region of interest (RFI) of the target tooth in the target sagittal image. Specifically, this includes: acquiring the tooth contour and corresponding pulp contour of the target tooth in the target sagittal image; identifying the root apex of the tooth in the tooth contour and identifying the coordinate point of the pulp closest to the root apex in the pulp contour; calculating the coordinates of a first midpoint between the root apex and the coordinate point of the pulp closest to the root apex; and using the first midpoint coordinates as the center and the distance from the first midpoint coordinates to the root apex as the radius as the circular region of interest.
[0048] In this embodiment, the root apex orientation of a tooth can be determined from its known tooth number; the root apex of a maxillary tooth points upwards, and the root apex of a mandibular tooth points downwards. The method for finding the region of interest (ROI) at the root apex of the tooth closest to the cyst, i.e., the target tooth, is as follows: Given the mask of the target tooth, calculate the set of coordinate points of the outer contour, and take the point with the smallest Y coordinate in the set of contour points (for mandibular teeth, the point with the largest Y coordinate) as the root tip, coordinate point A; Given the pulp mask corresponding to the target tooth, calculate the set of coordinate points of the outer contour, and take the point with the smallest Y coordinate in the contour point set (for mandibular teeth, the point with the largest Y coordinate) as the coordinate point B closest to the root apex of the pulp. Calculate the coordinates C of the center point between coordinate points A and B; Calculate the distance from coordinate point C to coordinate point A as the radius, and draw a mask region as the region of interest (ROI) with coordinate C as the center point. Figure 3 As shown.
[0049] Non-odontogenic cysts mostly occur in the midline region, not around the tooth root. Their core imaging characteristic is determining whether the cyst lesion is located within the midline structure through anatomical localization. Because human imaging is subject to "posture inaccuracies," meaning the resolution center of CBCT images is not necessarily perfectly symmetrical to the left and right sides of the body, this embodiment of the invention, before constructing the second spatial feature of the jawbone cyst lesion region, further includes a step of identifying the midline region of the oral CBCT image. The specific implementation of identifying the midline region of the oral CBCT image is as follows: Identify the left maxillary sinus region, right maxillary sinus region, and pharyngeal region in the oral CBCT image; find the 2D cross-sectional image with the largest total contour area of the left and right maxillary sinuses in the transverse section of the oral CBCT image as the target 2D cross-sectional image; identify the contour center point coordinates of the left maxillary sinus, right maxillary sinus, and pharyngeal region in the target 2D cross-sectional image; calculate the second midpoint coordinates between the contour center point coordinates of the left and right maxillary sinuses; connect the second midpoint coordinates with the contour center point coordinates of the pharyngeal region as the 2D midline of the target 2D cross-sectional image; extend the 2D midline into a three-dimensional midline region in the volume data corresponding to the oral CBCT image and extend it vertically along the Z-axis to generate a 3D midline region that runs through the volume contour of the volume data, thus obtaining the midline region of the oral CBCT image.
[0050] Specifically, such as Figure 4 As shown, the "left maxillary sinus" in CBCT images can be identified using the second 3D convolutional segmentation model. Figure 4 (dark green area in the middle), "right maxillary sinus" ( Figure 4 The dark purple area in the middle), the "pharynx" area ( Figure 4 (The brown area in the image), find the 2D cross-sectional image with the largest outline area of "left maxillary sinus" + "right maxillary sinus" from the transverse section, such as... Figure 4 As shown. First, calculate the coordinates of the center point of the contour of the "left maxillary sinus" and the center point of the contour of the "right maxillary sinus". Calculate the center coordinate point D between the two points. Then, calculate the coordinates E of the center point of the contour of the "pharynx" in this 2D cross-sectional image. Connect the two points on the 2D plane with line DE, which is defined as the midline. Draw a straight line along point D and point E as the midline of the current image. Extend this 2D midline to form the three-dimensional midline region in the entire CBCT volume data. Extend this line vertically outward along the Z-axis (head-to-foot direction) to generate a 3D midline region that runs through the entire volume contour of the CBCT volume data, thus obtaining the midline region of the oral CBCT image.
[0051] In this embodiment of the invention, after generating the joint feature vector, the method further includes: normalizing the joint feature vector; and sequentially performing T-test feature selection, Pearson correlation coefficient feature selection, and Lasso regression coefficient feature selection on the normalized joint feature vector to shrink the regression coefficients of redundant features to 0.
[0052] In this embodiment, the image features of the region of interest for cystic lesions, the image features of the region of interest for apical function, and the first and second spatial features of the jawbone cystic lesion region are directly concatenated in the feature dimension to form a longer joint feature vector. After feature normalization, T-test feature selection, Pearson correlation coefficient feature selection, and Lasso regression coefficient feature selection, the regression coefficients of some unimportant or redundant features can be shrunk to 0, thereby achieving automatic feature selection, reducing feature dimension and collinearity, and thus ensuring the accuracy of feature recognition results.
[0053] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.
[0054] This invention provides an artificial intelligence-based jaw cyst identification system, the system including functional modules for implementing the artificial intelligence-based jaw cyst identification method as described in any of the preceding claims. Figure 5 The schematic diagram illustrates the structure of an artificial intelligence-based jaw cyst identification system according to an embodiment of the present invention. (Refer to...) Figure 5 The system described in this embodiment of the invention includes: The lesion area identification module 501 is used to identify the jawbone cyst lesion area in the oral CBCT image to be processed; The first feature extraction module 502 is used to perform 3D connected component analysis on the jawbone cyst lesion area, take the connected component with the largest volume as the region of interest of the cyst lesion, and extract the image features of the region of interest of the cyst lesion. Oral tooth position recognition module 503 is used to identify the positional regions of each tooth in the oral cavity in the oral CBCT image; Select module 504 to determine the target tooth based on the positional relationship between the location area of each tooth and the jawbone cyst lesion area; The first feature generation module 505 is used to construct a first spatial feature of the jaw cyst lesion area. The first spatial feature is used to characterize whether the coordinates of the center point of the jaw cyst lesion area overlap with the coordinates of the target tooth location area. The second feature extraction module 506 is used to identify the functional region of interest (FROI) of the target tooth and extract the image features of the FROI. The second feature generation module 507 is used to construct a second spatial feature of the jaw cyst lesion area. The second spatial feature is used to characterize whether the jaw cyst lesion area overlaps with the midline area of the oral CBCT image. The feature stitching module 508 is used to stitch together the image features of the region of interest of the cystic lesion, the image features of the region of interest of the root apex function, and the first and second spatial features of the jawbone cystic lesion region to generate a joint feature vector. The lesion nature identification module 509 is used to input the joint feature vector into a pre-trained machine learning model for classification and identification, so as to identify the lesion nature of the jawbone cyst lesion.
[0055] In this embodiment of the invention, the system further includes a midline recognition module. This module is used to identify the left maxillary sinus region, right maxillary sinus region, and pharyngeal region in the oral CBCT image before the second feature generation module constructs the second spatial features of the jaw cyst lesion region. It searches for the 2D cross-sectional image with the largest total contour area of the left and right maxillary sinuses in the transverse section of the oral CBCT image, using it as the target 2D cross-sectional image. It identifies the contour center point coordinates of the left and right maxillary sinuses and the pharynx in the target 2D cross-sectional image. It calculates the second midpoint coordinates between the contour center point coordinates of the left and right maxillary sinuses. It uses the line connecting the second midpoint coordinates and the contour center point coordinates of the pharynx as the 2D midline of the target 2D cross-sectional image. It extends the 2D midline into a three-dimensional midline region in the volume data corresponding to the oral CBCT image and extends it vertically along the Z-axis to generate a 3D midline region that penetrates the volume contour of the volume data, thus obtaining the midline region of the oral CBCT image.
[0056] In this embodiment of the invention, the system further includes a feature optimization module, which is used to normalize the joint feature vector after the feature splicing module generates the joint feature vector; and to perform T-test feature selection, Pearson correlation coefficient feature selection and Lasso regression coefficient feature selection on the normalized joint feature vector in sequence, so as to shrink the regression coefficient of redundant features to 0.
[0057] The artificial intelligence-based method and system for identifying jaw cysts provided by this invention can effectively improve the accuracy of identifying and assessing the nature of jaw cysts by constructing an intelligent machine learning model that can automatically extract features from labeled medical image data and classify and identify the nature of cyst lesions. This has significant clinical application value.
[0058] As the system implementation is basically similar to the method implementation, the description is relatively simple, and relevant parts can be found in the description of the method implementation.
[0059] Furthermore, another embodiment of the present invention provides a computer program product storing a computer program, which, when executed by a processor, implements the steps described in the above embodiment of the artificial intelligence-based jaw cyst identification method, for example... Figure 1 Steps S11-S19 are shown.
[0060] Furthermore, another embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps described in the above embodiment of the artificial intelligence-based jaw cyst identification method, for example... Figure 1 Steps S11-S19 are shown.
[0061] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, any of the claimed embodiments can be used in any combination.
[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying jaw cysts based on artificial intelligence, characterized in that, The method includes: Identify jawbone cyst lesion areas in oral CBCT images to be processed; 3D connected component analysis was performed on the jawbone cyst lesion area, and the connected component with the largest volume was taken as the region of interest of the cyst lesion. Image features of the region of interest of the cyst lesion were extracted. Identify the location regions of each tooth in the oral cavity within the oral CBCT image; The target tooth is determined based on the positional relationship between the location of each tooth and the area of the jawbone cyst lesion. A first spatial feature is constructed for the jaw cyst lesion region, which is used to characterize whether the coordinates of the center point of the jaw cyst lesion region overlap with the coordinates of the target tooth location region; Identify the functional region of interest (FROI) at the root apex of the target tooth and extract the image features of the FROI. A second spatial feature is constructed for the jawbone cyst lesion region, which is used to characterize whether the jawbone cyst lesion region overlaps with the midline region of the oral CBCT image; The image features of the region of interest for cystic lesions, the image features of the region of interest for root apical function, and the first and second spatial features of the jawbone cystic lesion region are concatenated to generate a joint feature vector. The joint feature vector is input into a pre-trained machine learning model for classification and identification to determine the lesion nature of jawbone cysts.
2. The method according to claim 1, characterized in that, Identify jawbone cystic lesion regions in oral CBCT images to be processed, including: A pre-trained 3D convolutional segmentation model was used to identify the oral CBCT images to be processed, in order to identify the jaw cyst lesion area in the oral CBCT images; The training steps for a 3D convolutional segmentation model include: The annotation tool is used to annotate the outline region of the cyst lesion on the CT image samples in the preset oral CBCT image set, and an annotated volume data file is generated. The pixel value of 0 in the annotated volume data file represents the background area, and the pixel value of 1 represents the corresponding coordinate position in the image as the jaw cyst lesion area. A 3D convolutional segmentation model was trained using labeled volume data files as training data.
3. The method according to claim 1, characterized in that, Extracting image features from the region of interest of the cystic lesion, including: The morphological, grayscale, and texture features of the region of interest in the cystic lesion are extracted. The morphological features include volume, surface area, sphericity, elongation, flatness, and surface area to volume ratio. The grayscale features include the mean, median, entropy, skewness, and kurtosis of the pixel grayscale values. The texture features include the grayscale co-occurrence matrix, the grayscale size region matrix, and the grayscale run length matrix.
4. The method according to claim 1, characterized in that, The target teeth are determined based on their positional relationship with the jawbone cyst lesion area, including: Determine whether the coordinates of the center point of the jawbone cyst lesion overlap with the coordinates of the location areas of the teeth in the oral cavity; If there is overlap, the tooth whose center point coordinates overlap with the center point coordinates of the jaw cyst lesion area is selected as the target tooth; otherwise, the distance between the center point coordinates of the jaw cyst lesion area and the center point coordinates of the location areas of each tooth in the oral cavity is calculated, and the tooth corresponding to the shortest distance is selected as the target tooth.
5. The method according to claim 1, characterized in that, Identify the functional region of interest at the root apex of the target tooth, including: Obtain the tooth number and corresponding pulp number of the target tooth; Based on the tooth number and pulp number of the target tooth, the sagittal plane is sliced layer by layer along the X-axis in the volume data of the oral CBCT image, and the pixel area of the target tooth in each layer is calculated to obtain the target sagittal plane image with the largest pixel area. Identify the region of interest at the root apex of the target tooth in the target sagittal image.
6. The method according to claim 5, characterized in that, Identifying the region of interest (ROI) at the root apex of the target tooth in the target sagittal image includes: The tooth contour and corresponding pulp contour of the target tooth are obtained from the target sagittal image; Identify the root apex of the tooth in the tooth profile of the target tooth, and identify the coordinate point of the pulp closest to the root apex in the pulp profile of the target tooth. Calculate the coordinates of the first midpoint between the root apex of the tooth and the point in the pulp closest to the root apex. Then, draw a circular region centered on the first midpoint and with the distance from the first midpoint to the root apex as the radius, and use this circular region as the functional region of interest.
7. The method according to claim 1, characterized in that, Prior to constructing a second spatial feature of the jawbone cyst lesion region, the method further includes: Identify the left maxillary sinus region, right maxillary sinus region, and pharyngeal region in the oral CBCT images; In the transverse section of the oral CBCT image, find the 2D section image with the largest total outline area of the left and right maxillary sinuses, and use it as the target 2D section image. The coordinates of the center point of the left maxillary sinus, the center point of the right maxillary sinus, and the center point of the pharynx were identified in the target 2D cross-sectional image. Calculate the coordinates of the second midpoint between the center point of the left maxillary sinus contour and the center point of the right maxillary sinus contour; The line connecting the coordinates of the second midpoint and the coordinates of the center point of the pharynx contour is taken as the 2D midline of the target 2D cross-sectional image; The 2D midline is expanded into a three-dimensional midline region in the volume data corresponding to the oral CBCT image and extended vertically along the Z-axis to generate a 3D midline region that runs through the volume contour of the volume data, thus obtaining the midline region of the oral CBCT image.
8. The method according to claim 1, characterized in that, After generating the joint feature vector, the method further includes: The joint feature vector is normalized. The normalized joint feature vector is subjected to T-test feature selection, Pearson correlation coefficient feature selection, and Lasso regression coefficient feature selection in sequence to shrink the regression coefficients of redundant features to 0.
9. A jawbone cyst identification system based on artificial intelligence, characterized in that, The system includes: The lesion area identification module is used to identify jaw cyst lesion areas in oral CBCT images to be processed; The first feature extraction module is used to perform 3D connected component analysis on the jawbone cyst lesion area, and to extract the image features of the cyst lesion region of interest by taking the largest connected component as the region of interest of the cyst lesion. The oral cavity tooth position recognition module is used to identify the positional regions of each tooth in the oral cavity in the oral CBCT image; The selection module is used to determine the target tooth based on the positional relationship between the location of each tooth and the jawbone cyst lesion area. The first feature generation module is used to construct the first spatial feature of the jaw cyst lesion area. The first spatial feature is used to characterize whether the coordinates of the center point of the jaw cyst lesion area overlap with the coordinates of the target tooth location area. The second feature extraction module is used to identify the functional region of interest (FROI) at the root apex of the target tooth and extract the image features of the FROI. The second feature generation module is used to construct the second spatial feature of the jaw cyst lesion area. The second spatial feature is used to characterize whether the jaw cyst lesion area overlaps with the midline area of the oral CBCT image. The feature stitching module is used to stitch together the image features of the region of interest of the cystic lesion, the image features of the region of interest of the root apex function, and the first and second spatial features of the jawbone cystic lesion region to generate a joint feature vector. The lesion nature identification module is used to input the joint feature vector into a pre-trained machine learning model for classification and identification, so as to identify the lesion nature of jawbone cyst lesions.
10. The system according to claim 9, characterized in that, The system also includes: The feature optimization module is used to normalize the joint feature vector after the feature splicing module generates it; and to perform T-test feature selection, Pearson correlation coefficient feature selection and Lasso regression coefficient feature selection on the normalized joint feature vector in sequence, so as to shrink the regression coefficient of redundant features to 0.