Periodontal key structure detection method and system, electronic device and readable storage medium

Abstract

The application provides a periodontal key structure detection method, including: obtaining three-dimensional body data of a target tooth, and determining a tooth long axis; intercepting a plurality of two-dimensional slices passing through the tooth long axis at a set angle along the tooth long axis, the plurality of two-dimensional slices covering at least two directions of mesial, distal, buccal and lingual; determining an enamel cementum junction point according to an extreme value of an enamel cementum junction thermal map of each two-dimensional slice; determining an alveolar crest point according to an extreme value of an alveolar crest thermal map of each two-dimensional slice, and still determining the alveolar crest point according to the extreme value of the thermal map at a position where the tooth and the alveolar bone are no longer adjacent; determining an apical point according to an extreme value of a three-dimensional root apex thermal map; and outputting a local enamel cementum junction curve, a local alveolar crest curve and the apical point of at least one tooth surface. The application also provides a periodontal key structure detection system, an electronic device and a readable storage medium.

Description

Technical Field

[0001] This invention relates to a method and system for detecting critical periodontal structures. Background Technology

[0002] Locating the cementoenamel junction, alveolar ridge, and root apex on cone-beam computed tomography (CBCT) is a prerequisite for subsequent bone loss measurements. Errors in locating these three structures will propagate to subsequent distance and proportion measurements.

[0003] One approach involves first performing three-dimensional segmentation of the enamel, then intersecting it with a tooth mask to obtain the cementoenamel junction, and then intersecting it with an alveolar bone mask to obtain the alveolar ridge. For the root apex, the extreme region of the tooth mask along the long axis of the tooth is used. On cone-beam computed tomography (CBCT), the thickness of the enamel layer is often close to or smaller than the voxel size, resulting in weak contrast between enamel and dentin, and thin-slice segmentation is inherently unstable. When two unstable masks intersect, the junctional burrs will become curved jitter. Holes caused by metal artifacts will create false points at the junction. In cases of bone dehiscence or bone fenestration, the outer surface of the tooth may not have adjacent alveolar bone, and the junction may be an empty set in that circumferential position, or it may jump to a distant bone wall.

[0004] Another approach involves detecting two feature points on a single maximum buccal-lingual cross-section, calculating the planar distance, and comparing it to a spacing threshold to determine the presence of bone dehiscence. However, a single cross-section cannot reliably provide the cementoenamel junction, alveolar ridge, and apex in multiple directions (mesial, distal, buccal, and lingual). Furthermore, the output is not a set of curves and points suitable for subsequent tooth-surface measurements.

[0005] Therefore, the current methods for locating key anatomical structures still need improvement due to issues such as unstable masking boundaries, insufficient information from single sections, and the possibility of no boundary at bone fissure locations. Under conditions of artifacts, thin bone plates, and bone fissures, the cementoenamel boundary, alveolar ridge, and apical points on multiple tooth surfaces are still difficult to reliably determine, and missing or false points may still appear. Summary of the Invention

[0006] To address one of the aforementioned technical problems, this invention provides a method and system for detecting key periodontal structures.

[0007] According to one aspect of the present invention, a method for detecting critical periodontal structures includes: acquiring three-dimensional volume data of a target tooth and determining the long axis of the tooth; cutting multiple two-dimensional slices along the long axis of the tooth at a predetermined angle, the multiple two-dimensional slices covering at least two directions among mesial, distal, buccal, and lingual; determining the cementoenamel boundary point based on the extreme values ​​of the cementoenamel boundary thermogram of each two-dimensional slice; determining the alveolar ridge point based on the extreme values ​​of the alveolar ridge thermogram of each two-dimensional slice, and still determining the alveolar ridge point based on the extreme values ​​of the thermogram at positions where the tooth and alveolar bone are no longer adjacent; determining the apex point based on the extreme values ​​of the three-dimensional apical thermogram; and outputting a local cementoenamel boundary curve, a local alveolar ridge curve, and the apex point of at least one tooth surface.

[0008] According to at least one embodiment of the present invention, a mesiodistal reference direction is determined, wherein the mesiodistal reference direction is adopted as the tangent direction of the dental arch curve or the proximal contact direction, and a set of planes passing through the tooth long axis are generated at a set interval using the mesiodistal reference direction as the angular reference, and the two-dimensional slice is obtained by resampling the planes.

[0009] According to at least one embodiment of the present invention, the voxel resolution of the resampling is determined according to the image spatial resolution, and the tooth region is cropped to a fixed field of view centered on the circumscribed rectangle on the plane, and the coronal side is oriented toward a first side of the slice image and the apical side is oriented toward a second side of the slice image.

[0010] According to at least one embodiment of the present invention, a thermal map of the cementoenamel boundary is output from a thermal map prediction model, local maxima are extracted after smoothing the thermal map of the cementoenamel boundary, and the centroids of the two connected regions with the highest area ranking are determined as a pair of cementoenamel boundary points on the slice.

[0011] According to at least one embodiment of the present invention, during training, a physician marks two points on the slice where the slice intersects with the cementoenamel boundary, and places a two-dimensional Gaussian kernel at the two points to generate a supervised heatmap; the spatial scale of the Gaussian kernel is set according to voxel resolution; and the supervised target is a point-like thermal field, rather than an enamel region segmentation mask.

[0012] According to at least one embodiment of the present invention, the left extreme value and the right extreme value are mapped to tooth surface points on both sides of the slice, with the projection of the long axis of the tooth in the slice as the midline; and the tooth surface points are mapped to the buccal and lingual surfaces, or to the mesial and distal surfaces, according to the direction of the included angle of the slice.

[0013] According to at least one embodiment of the present invention, a heatmap prediction model outputs an alveolar ridge heatmap on the same set of two-dimensional slices; the extreme points are extracted after smoothing the alveolar ridge heatmap; and the alveolar ridge heatmap and the cementoenamel interface heatmap are output in separate channels, or output by two separate models.

[0014] According to at least one embodiment of the present invention, two-dimensional points are mapped to three-dimensional points based on the origin and basis vector of the slice plane; three-dimensional points in the same tooth surface direction are sorted and connected circumferentially to form a local curve of the tooth surface; and the local curve is allowed to be broken and is not forced to be closed into a full circumferential curve.

[0015] According to at least one embodiment of the present invention, if the axial coordinate deviation of the cementoenamel junction point on adjacent angle slices exceeds the deviation threshold, the point is downweighted or eliminated.

[0016] According to at least one embodiment of the present invention, a root tip heat map is output from a three-dimensional heat map prediction model; connected extrema are extracted after smoothing; under single-root conditions, the centroid of the extremum with the largest integral is determined as the root tip; and under multi-root conditions, the extrema exceeding the response threshold are output as the root tip set.

[0017] According to at least one embodiment of the present invention, the maxilla or mandible is determined based on the tooth position number to unify the crown-root direction; and the expected number of roots given by the tooth position number is only an optional constraint and does not divide the set of apical points to a specific tooth surface.

[0018] According to at least one embodiment of the present invention, the outputs of the cementoenamel junction thermal map, alveolar ridge thermal map, and periapical thermal map are spatial thermal fields, rather than the center of the detection frame.

[0019] According to at least one embodiment of the present invention, the local cementoenamel boundary curve, the local alveolar ridge curve, and the apical point are provided to the subsequent measurement module for use.

[0020] According to another aspect of the present invention, a periodontal critical structure detection system includes: a data acquisition module for acquiring three-dimensional volume data of a target tooth and determining the tooth's long axis; a slice generation module for extracting multiple two-dimensional slices along the tooth's long axis at a set angle, the multiple two-dimensional slices covering at least two of the mesial, distal, buccal, and lingual directions; a cementoenamel boundary detection module for determining cementoenamel boundary points based on the extreme values ​​of the cementoenamel boundary thermograms of each two-dimensional slice; an alveolar ridge detection module for determining alveolar ridge points based on the extreme values ​​of the alveolar ridge thermograms of each two-dimensional slice, and still determining the alveolar ridge points based on the extreme values ​​of the thermograms at locations where the tooth and alveolar bone are no longer adjacent; a periapical detection module for determining periapical points based on the extreme values ​​of a three-dimensional periapical thermogram; and an output module for outputting a local cementoenamel boundary curve, a local alveolar ridge curve, and a periapical point of at least one tooth surface.

[0021] According to one aspect of the present invention, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform the method as described in any of the preceding claims.

[0022] According to one aspect of the present invention, a readable storage medium is provided, wherein executable instructions are stored therein, which, when executed by a processor, are used to implement the method as described in any of the preceding claims.

[0023] Compared with the prior art, the present invention has the following beneficial effects.

[0024] A set of two-dimensional slices is taken along the long axis of the tooth at a set angle, covering at least two directions: mesial, distal, buccal, and lingual. Positioning is independent of the maximum buccal-lingual section of a single slice. The cementoenamel boundary and alveolar ridge are taken from the extreme values ​​of the slice thermal map. Even when the tooth and alveolar bone are no longer adjacent, the ridge can still be provided, independent of 3D enamel segmentation or the boundary between the tooth mask and the alveolar bone mask. For a single root, the maximum integral extreme value is taken; for multiple roots, the set of points exceeding a threshold is output. The apical position is independent of the extreme voxels along the long axis of the tooth mask. After mapping the two-dimensional points back to 3D, they are connected according to the tooth surface to form a locally breakable curve. Axial coordinates of adjacent angled slices can be constrained for consistency, making it less likely for outliers on a single slice to enter the final curve. The detection results can be provided to subsequent measurement modules or used independently for display and quality control. The output includes local cementoenamel boundary curves, local alveolar ridge curves, and apical points, and can include the thermal response of each point. Attached Figure Description

[0025] The accompanying drawings illustrate exemplary embodiments of the invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification.

[0026] Figure 1 A method for detecting critical periodontal structures according to an embodiment of the present invention is shown.

[0027] Figure 2 A schematic diagram of a set of cross-tooth long axis slices with different included angles is shown according to an embodiment of the present invention.

[0028] Figure 3 A schematic diagram showing the correspondence between the extreme values ​​of a single slice heatmap and the left and right sides of the midline is shown in one embodiment of the present invention.

[0029] Figure 4 A schematic diagram of a local curve of the tooth surface formed after two-dimensional point back mapping is shown in one embodiment of the present invention.

[0030] Figure 5 A schematic diagram of the extreme values ​​of a three-dimensional root tip thermogram is shown in one embodiment of the present invention.

[0031] Figure 6 A comparative schematic diagram is shown, illustrating how the location of bone fracture can still provide an alveolar ridge point, according to one embodiment of the present invention.

[0032] Figure 7 A schematic diagram of a periodontal critical structure detection system according to an embodiment of the present invention is shown.

[0033] Figure 8 A schematic diagram of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.

[0035] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The technical solution of this invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0036] Unless otherwise stated, the exemplary embodiments shown are to be understood as providing exemplary features of various details that provide ways in which the technical concept of the invention can be implemented in practice. Therefore, unless otherwise stated, the features of the various embodiments may be additionally combined, separated, interchanged, and rearranged without departing from the technical concept of the invention.

[0037] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and “including” and variations thereof are used in this specification, they indicate the presence of the stated features, steps, operations, and components, but do not exclude the presence or addition of other features, steps, operations, and components.

[0038] According to one embodiment of the present invention, a method for detecting critical periodontal structures is provided.

[0039] Figure 1 A method for detecting critical periodontal structures according to an embodiment of the present invention is shown, the method comprising steps 102 to 112.

[0040] In one embodiment, the output includes a local cementoenamel boundary curve, a local alveolar ridge curve, and a apical point on at least one tooth surface. The detection results can be provided to a subsequent measurement module or used independently for display and quality control.

[0041] In step 102, three-dimensional volumetric data of the target tooth is acquired, and the long axis of the tooth is determined. The three-dimensional volumetric data is obtained from an oral cone-beam computed tomography (CBCT) scan with a voxel resolution of 0.25 mm. A three-dimensional region of interest (ROI) containing the target tooth is extracted based on the tooth position, with an ROI size of, for example, 256 × 128 × 128. The above dimensions are for illustrative purposes and can be adjusted according to the field of view.

[0042] The tooth long axis is used to constrain the axis that the slice plane must pass through. Multiple cross-sections are extracted from the tooth region along the initial principal axis direction. The center position of the tooth region within each cross-section is calculated. The sequence of center points from the crown to the mid-root is fitted with a straight line using least squares fitting to obtain the tooth long axis direction. Alternatively, the direction can be obtained from the direction returned by the segmentation interface. As long as all slices pass through the same tooth long axis, cross-sections with different angles will have a common crown-root reference.

[0043] The input is a single-channel grayscale, without adding enamel channels. Alveolar ridges and cementoenamel boundaries are determined on the section thermogram.

[0044] In step 104, multiple two-dimensional slices are cut along the tooth long axis at a set angle, covering at least two of the mesial, distal, buccal, and lingual directions.

[0045] Figure 2 This diagram illustrates a set of cross-sections along the long axis of the tooth at different angles, according to an embodiment of the present invention. In the diagram, θ represents the angle between adjacent cross-sections. The mesiodistal reference direction is taken as the zero-degree angle. The mesiodistal reference direction can be the tangent to the dental arch curve, or it can be the direction of proximal contact; this method does not impose any particular limitation on this. The angle is taken at set intervals. For example, an interval of 30 degrees yields 6 cross-sections, 15 degrees yields 12 cross-sections, or 45 degrees yields 4 cross-sections. The interval is selected according to the measurement requirements.

[0046] For each included angle, a plane passing through the tooth's long axis is generated. The normal to the plane is the unit vector obtained by the cross product of the tooth's long axis and the included angle direction. The two basis vectors in the plane are resampled along the tooth's long axis and the cross product direction, respectively, according to the voxel resolution. Interpolation can be performed using trilinear interpolation or cubic convolution. The area is cropped to a fixed field of view, such as 256×128, with the coronal direction facing upwards and the root direction facing downwards, to obtain a two-dimensional slice.

[0047] When cropping the region of interest, if the circumscribed rectangle is smaller than the fixed field of view, zeros are added around it. If it is larger than the fixed field of view, it is cropped at the center. The zero-padded area has a low response on the heatmap and is less likely to enter the two largest connected regions. In this way, teeth of different sizes enter the same image size in the model.

[0048] The crown-root orientation is determined by judging the projections of the crown center and the apical side onto the long axis of the tooth. Therefore, the same set of slice coordinates can be used for the maxilla and mandible. The angular interval does not need to be fixed. The interproximal surfaces of anterior teeth change gradually, so the interval can be wider. The buccal and lingual surfaces of molars and the area near the root bifurcation change rapidly, so the interval can be closer. The same tooth can also use a larger interval on the interproximal surfaces and a smaller interval on the buccal and lingual surfaces. The long axis of the tooth divides each slice into two half-slices, and each half-slice corresponds to one side of the tooth surface of the corresponding slice.

[0049] In step 106, the cementoenamel boundary point is determined based on the extreme values ​​of the cementoenamel boundary thermogram of each two-dimensional slice.

[0050] Figure 3 This diagram illustrates the correspondence between the extreme values ​​of a single-slice heatmap and the left and right sides of the midline, according to an embodiment of the present invention. The heatmap prediction model can employ a two-dimensional network of encoder and decoder, with a single-channel grayscale slice as input and a single-channel heatmap as output. The loss can be mean squared error or weighted mean squared error. During training, on each long-axis slice, the dentist marks two points where the slice intersects the cementoenamel junction, typically one point on each side of the slice. A two-dimensional Gaussian kernel is placed at these two points to generate a supervised heatmap. For example, using a Gaussian heatmap, the marked points are denoted as (…). , ), voxel resolution of Δ, spatial scale Approximately 2mm, but can also be selected within the range of 1.5-3mm. The monitoring heatmap is as follows: ,in .

[0051] The cementoenamel boundary is marked as a point, and the supervised target is a point-like thermal field. The variance of the two-dimensional Gaussian kernel is linked to the voxel resolution. At a resolution of 0.25 mm and a scale of 2 mm, the Gaussian kernel covers a radius of approximately 8 pixels.

[0052] During inference, the model outputs a heatmap, which is Gaussian smoothed and local maxima are taken. Connected domains are sorted by area or integral, and the centroids of the two top-ranked connected domains are selected as a pair of cementoenamel boundaries on the slice. The centroids of connected domains are weighted according to the thermal response. Heatmap smoothing can use a small-window Gaussian, with the window size on the same order as the supervised Gaussian. Before sorting connected domains by area, domains with excessively small areas are removed to avoid individual voxel responses being included in the centroid calculation. If only one valid extremum is detected, the slice is discarded or used only on one side of the tooth surface. If more than three are detected, only the two largest areas are retained, or one maximum extremum is taken from each of the left and right half-planes. Using the projection of the tooth's long axis in the slice as the midline, the left and right extrema are mapped to the tooth surface points on both sides of the slice. When the slice is close to the buccal-lingual direction, the two sides correspond to the buccal and lingual sides, or the labial and palatal sides. When the slice is close to the mesiodistal direction, the two sides correspond to the mesial and distal sides. The correspondence between the included angle and the tooth surface is determined by the zero-degree reference at the time of slice generation.

[0053] The left and right half-plane constraints can be written as retaining only the left maximum extreme value on the left side of the midline and only the right maximum extreme value on the right side of the midline. If there is no extreme value exceeding the response threshold on a certain side, then that side is marked as missing. The missing half-slice does not enter the local curve of the tooth surface, or leaves a break on the curve.

[0054] The two output channels can correspond to the cementoenamel junction and alveolar ridge respectively, or two models can be trained. In the channel scheme, two types of points are obtained simultaneously in one forward pass, and the slice is resampled only once. In the separate model scheme, the samples of the two tasks can be augmented separately. Both implementations output the extreme values ​​of the heatmap.

[0055] In step 108, the alveolar ridge point is determined based on the extreme values ​​of the alveolar ridge thermogram of each two-dimensional slice.

[0056] Even where the tooth and alveolar bone are no longer adjacent, the alveolar ridge point is still determined based on the extreme values ​​of the thermal map. Steps 108 and 106 use the same set of slices, and the correspondence rules for the left and right half-slices are the same as those for the cementoenamel boundary. The alveolar ridge thermal map can share the same set of slices as the cementoenamel boundary thermal map. It can be provided by two output channels of the same model or by two separate models. After smoothing, the extreme values ​​are extracted to obtain the alveolar ridge point on that slice. Each half-slice yields one alveolar ridge point. When there is no adjacent bone at the bone fissure or bone window, the half-slice still outputs one ridge point.

[0057] Figure 6 The diagram shows a comparison of the alveolar ridge at the location of bone fissure according to one embodiment of the present invention. On the left, points are taken at the junction of the tooth mask and the alveolar bone mask, and there is no ridge at the fissure. On the right, points are taken at the section heat map, and there is still a ridge at the fissure.

[0058] In step 110, the apical point is determined based on the extreme values ​​of the three-dimensional apical thermal map. The input is the volume data of the single tooth region of interest, and the enamel channel is not input. For a single root, the centroid of the extreme value with the largest integral is taken. For multiple roots, the extreme values ​​that all exceed the response threshold are taken as the set of apical points.

[0059] Figure 5 A schematic diagram illustrating the extreme values ​​of a three-dimensional root tip thermal map according to an embodiment of the present invention is shown. The three-dimensional thermal map prediction model outputs a single-channel thermal volume, and during training, points are marked at the root tip. A three-dimensional Gaussian kernel is placed, with one peak for a single root and one peak for each root tip of multiple roots. During inference, after smoothing, connected extrema are selected. For a single root, the centroid of the extremum with the largest area or integral is taken as the root tip. For multiple roots, all extrema that exceed the response threshold are taken as the set of root tips.

[0060] The field of view in a 3D periapical thermal map is centered on the tooth region, with an extra margin on the apical side to prevent the true root apex from falling outside the region of interest. For multi-rooted teeth, multiple peaks are spatially separated; non-maximum suppression can be implemented based on 3D distance, merging peaks less than 1.5 mm apart, and sorting the remaining peaks by integral. For single-rooted teeth, only the first peak is taken; for multi-rooted teeth, all peaks exceeding a threshold are taken. The threshold can be a certain percentage of the maximum peak integral, such as 30%, used to remove background responses.

[0061] The role of tooth position numbering is limited to determining whether the tooth is maxillary or mandibular to unify the crown-root direction; and to provide optional constraints for the expected number of peaks. The expected number of peaks does not replace the image. When the measured number of peaks is inconsistent with the expected number, the image shall prevail.

[0062] In step 112, the two-dimensional points are mapped back to three dimensions to form a local curve of at least one tooth surface, and the local cementoenamel boundary curve, local alveolar ridge curve, and apex point are output.

[0063] Figure 4 A schematic diagram of a local curve of the tooth surface formed after two-dimensional point back mapping according to an embodiment of the present invention is shown. The mapping uses the origin of the slice plane and two basis vectors, and the back mapping uses the origin o of the slice plane and basis vectors u and v. The pixel coordinates (p, q) correspond to three-dimensional points. u and v have been calibrated according to voxel resolution when the slice is generated, so the points after back mapping have millimeter units.

[0064] Arrange the cementoenamel junction points of each slice along the same tooth surface direction in a circumferential manner, and connect them with broken lines or splines to form a local cementoenamel junction curve for that tooth surface. Alveolar ridge points are also connected to form a local alveolar ridge curve. Local curves are allowed to be broken, and are not required to be closed into full-circumference curves. The thermal response of each point can be output simultaneously, and low response points may not be included in the curve.

[0065] If the axial coordinate deviation of the cementoenamel junction on adjacent angled sections exceeds a deviation threshold, such as 2 mm, then that point is downweighted or removed. The cementoenamel junctions are then projected onto the tooth's long axis to obtain an axial coordinate sequence. =( -o)·e. For circumferentially adjacent points, if the axial difference exceeds the threshold τ, the point with the higher response is retained, or both ends are downweighted. The alveolar ridge is treated similarly. The axial coordinates of the ridge at the bone fissure location will be significantly biased towards the root. The root displacement caused by the fissure will appear continuously at multiple adjacent angles. Artifacts and burrs often only appear on one or two slices. Output at least one local cementoenamel boundary curve, local alveolar ridge curve, and apex point of the tooth surface. The response value of each point on the thermogram can be output simultaneously as a confidence level. These curves and points can be provided to the subsequent measurement module.

[0066] The alveolar ridge and cementoenamel junction on the slice use the same midline left-right rule. Therefore, a pair of points on the same half-slice corresponds to the same tooth surface. After mapping the detection points back to 3D, the slice number and the left-right markings of the half-slice can be retained. The field of view of the 3D thermal map of the periapex and the 2D thermal map of the slice need to be roughly aligned. The root side of the slice should cover the vicinity of the periapex, and the coronal side of the 3D region of interest should cover the vicinity of the cementoenamel junction, so that the three types of structures output can be placed into the same coordinate system by subsequent modules. An example of a tooth with bone cracking is given below.

[0067] The target tooth is the left mandibular first premolar with buccal bone fissure. On a section with an angle of approximately 90 degrees across the tooth's long axis, the adjacent bone is not visible on the outer side of the tooth outline in the buccal half-section. Our method still yields an extreme value on the alveolar ridge thermogram of this half-section. After mapping back to 3D, this point is located near the coronal residual bone ridge in the fissure area. The lingual half-section boundary exists, and the thermal extreme value is close to the boundary location. Therefore, this tooth simultaneously possesses a lingual ridge point and a ridge point in the buccal fissure area.

[0068] In the thin bone plate embodiment, the labial bone plate of the maxillary anterior teeth is close to one voxel thick, and the junction point jumps up and down about 2 mm along the labial side. The axial coordinate difference of the extreme values ​​of the thermal map on adjacent angle slices is less than 1 mm. After the consistency constraint of adjacent slices, it enters the local curve, and the local alveolar ridge curve on the labial side is continuous.

[0069] In the artifact example, the adjacent tooth metal crown creates a strip artifact on the distal section of the target tooth. The cementoenamel junction thermal map shows a third weak response on the artifact strip. After area sorting, this weak response is discarded, and only the two extreme values ​​with the largest areas on both sides of the midline are retained. The local curve of the distal surface does not pass through the artifact strip.

[0070] In the single-rooted tooth apex embodiment, the three-dimensional thermal map of the mandibular canine has only one main peak, and the centroid of this peak is taken as the apex point. In the multi-rooted tooth embodiment, the maxillary first molar has three peaks exceeding the threshold, and the output is a set of apex points, without labeling the three peaks as mesobuccal root, distal buccal root, and palatal root respectively.

[0071] In the bone fenestration example, there may be no bone in the middle section of the fenestration area, but there is still bone wall in the coronal and apical regions. The slice heat map may show a weaker ridge response in the middle section of the fenestration and a stronger response at the coronal boundary of the fenestration. The consistency of adjacent slices will preserve the coronal boundary point, and the output local alveolar ridge curve will be broken in the fenestration area.

[0072] In the subosseous pocket example, the crest is significantly apical displaced relative to the cementoenamel junction, but the tooth and bone remain adjacent, and the ridge point is also given at this location by thermography.

[0073] If no valid extreme values ​​can be obtained from all slices of a tooth, the tooth is marked as a failed test, and no curve is output. Failed teeth can be manually sampled. In this method, the tooth region is mainly used to determine the tooth's long axis, the field of view for cutting slices, and to determine the midline. The mesial and distal reference directions have two sources in this method: the arch curve tangent, suitable for rotated and crowded teeth; and the proximal contact direction, suitable for cases where the arch curve is unstable but the proximal contact points are clear. The contact direction can be approximated by the projection of the line connecting the center of the target tooth to the mesial and distal adjacent teeth onto the horizontal plane. Both sources are used to provide a 0-degree reference for the slice angle.

[0074] When outputting to subsequent measurements, the system can simultaneously provide the 3D coordinates of each point, its corresponding tooth surface, the included angle, the thermal response, and whether it has undergone consistency constraints. The measurement module can weight points according to their response or discard low-response points. When visualizing the test results, local curves can be overlaid on excessively long axis slices or 3D surfaces. Colors can change according to the thermal response, and low-response segments are represented by dashed lines or breaks. The visualization serves quality control and does not constitute disease diagnosis.

[0075] During training, dental specialists annotate the cementoenamel boundary and alveolar ridge on the long axis slices, one point on each side of each slice. The annotation results are reviewed and used as the gold standard. Volume data and slices can be enhanced synchronously, including small-angle rotation, translation, and grayscale jitter. Images are interpolated continuously, and each original sample can generate several enhanced samples. During inference, the heatmap is first smoothed with a small window Gaussian, then local maxima are taken. Connected components are sorted by area or integral. Connected components with too small an area are first removed, and the two top-ranked connected components are selected. The centroid is calculated by weighting the thermal response. With the tooth long axis projection as the midline, the left extreme value corresponds to one side of the tooth surface, and the right extreme value corresponds to the other side of the tooth surface.

[0076] The 3D periapical thermal mapping prediction model can also employ an encoder-decoder structure, with single-tooth region of interest volume data as input and single-channel thermal volume as output. During training, a 3D Gaussian kernel is placed at the periapical annotation point. During inference, after smoothing, the connectivity extrema are taken. The distance threshold for non-maximum suppression can be set to 1.5 mm, and the response threshold can be set to 30% of the maximum peak integral. After the slice plane is uniquely determined by the tooth's long axis and the included angle, the resampling mesh takes the center of the tooth region's projection on this plane as its origin, with a mesh step size equal to the voxel resolution. The coronal plane is upward and the apical plane is downward. The same slice coordinate system is used for maxillary and mandibular teeth, and the anatomically downward direction is corrected according to the tooth position.

[0077] In one implementation, the cementoenamel junction and alveolar ridge are output separately, yielding two types of points in a single forward pass. In another implementation, two models are trained for the two tasks respectively, with an independent 3D model used for the apex. The three types of outputs are aligned in the same tooth region coordinate system. If the axial coordinate deviation of the same tooth surface point on adjacent angle slices exceeds 2mm, the point with the higher response is retained, or both ends are downweighted. When connecting local curves, circumferentially adjacent points are connected with polylines or splines, and points with excessive axial jumps are disconnected. The disconnected curve segment is still used as the local curve output for that tooth surface. During quality control, the 3D coordinates, tooth surface, angle, thermal response, and whether consistency constraints have been applied to each point can be output. Low-response points may not be included in the curve. If no pair of valid extrema can be obtained for any of the slices, the tooth is marked as a failed test.

[0078] Metal artifacts, thin bone plates, and bone cracks are not mutually exclusive. Weak responses on artifact strips are discarded during area sorting. At thin bone plates, the axial difference between adjacent slices is included in the curve after consistency constraints. The heatmap at cracks still gives the ridge point. This section of the curve can be continuous or broken according to the response.

[0079] The contact direction of the proximal surfaces can be approximated by the projection of the line connecting the center of the target tooth with the mesial and distal adjacent teeth onto the horizontal plane. When the dental arch curve is unreliable due to continuous tooth loss, the contact direction can still provide a usable 0-degree reference for the slice. The included angle reference is only used to generate slices.

[0080] The spatial scale of the Gaussian kernel is set according to the voxel resolution. When the resolution is 0.25mm and the scale is 2mm, the Gaussian kernel covers a radius of about eight pixels. The scale can also be selected in the range of 1.5-3mm. If the window is too large, the two left and right extrema will be connected into a peak, and the left and right sides of the midline will fail to correspond. If the window is too small, the artifact peak will be treated as an extremum. In implementation, the smoothing window should be of the same order of magnitude as the supervised Gaussian scale.

[0081] When cutting the region of interest, leave more margin on the apical side to avoid the true root apex falling outside the region of interest. The root side of the slice should cover the vicinity of the root apex, and the coronal side of the three-dimensional region of interest should cover the vicinity of the cementoenamel junction. If the circumscribed rectangle is smaller than the fixed field of view, zeros should be added around it. If it is larger than the fixed field of view, it should be cut according to the center.

[0082] When connecting local curves, points on the same tooth surface are ordered circumferentially. The circumferential direction can be represented by the angle between the relative mesial and distal reference directions. Adjacent points are connected by broken lines or splines. If the thermal response of a certain point is lower than the response threshold, that point will not enter the curve, and the curve will break at that point. Local curves are not forced to close into full-circumference curves.

[0083] Tooth position number is used to determine whether it is maxillary or mandibular to unify the crown-root direction and provide optional constraints on the expected number of peaks. The expected number of peaks does not replace the image. When the measured number of peaks is inconsistent with the expected number, the image shall prevail. For a single tooth, the centroid of the extreme value with the largest integral is taken. For multiple teeth, all extreme values ​​that exceed the response threshold are output as the set of apical points. If no pair of valid extreme values ​​can be obtained from all slices of a tooth, the tooth is marked as a failed test and no curve is output. Failed teeth can be manually sampled. The test results can be provided to the subsequent measurement module or used alone for display and quality control.

[0084] After steps 102 to 112, the local cementoenamel boundary curve, local alveolar ridge curve, and apical point of multiple tooth surfaces are obtained. The angular interval can be adjusted according to the tooth position. The change is gradual on the proximal surfaces of anterior teeth, so the interval can be 45 degrees. The change is rapid on the buccal and lingual surfaces of molars and near the root bifurcation, so the interval can be 15 degrees. For the same tooth, a larger interval can be used on the proximal surfaces and a smaller interval on the buccal and lingual surfaces. The output can include the three-dimensional coordinates of each point, the tooth surface it belongs to, the angle it belongs to, the left and right markings of the half-slice, the thermal response, whether it has passed the consistency constraint, and the local cementoenamel boundary curve, local alveolar ridge curve, and apical point obtained by connecting these points. Low response points can be excluded from the curve, and teeth that fail to be detected are marked separately.

[0085] The output of the heatmap is a spatial thermal field. The cementoenamel junction and alveolar ridge take extreme values ​​on the two-dimensional slice, while the root apex takes extreme values ​​on the three-dimensional volume data. The tooth region is used to determine the long axis of the tooth, cut the slice field of view, and determine the midline.

[0086] When there is no adjacent bone at the bone fissure or fenestration, the half-slice still outputs a ridge point based on the extreme value of the alveolar ridge heatmap. At the thin bone plate location, the axial coordinates of adjacent angle slices are entered into the local curve after consistency constraints. The weak response caused by metal artifacts is discarded when sorting the area of ​​the connected domain. The local curve is allowed to be broken, and the broken curve segment is still output as the detection result of the tooth surface.

[0087] In one implementation, the mesial-distal reference direction is the tangent of the dental arch curve; in another implementation, the proximal contact direction is used. Both sources are only used to provide a 0-degree reference for the slice angle. The slice covers at least two of the mesial, distal, buccal, and lingual directions, and the set of apical points can be provided for use by subsequent modules.

[0088] According to another aspect of the present invention, a periodontal critical structure detection system is provided. Figure 7 A schematic diagram of a periodontal critical structure detection system according to an embodiment of the present invention is shown.

[0089] The periodontal key structure detection system may include a data acquisition module 1002, a slice generation module 1004, an enamel-cementum boundary detection module 1006, an alveolar ridge detection module 1008, a periapical detection module 1010, and an output module 1012.

[0090] The data acquisition module is used to acquire the three-dimensional volume data of the target tooth and determine the long axis of the tooth. The slice generation module is used to extract multiple two-dimensional slices along the long axis of the tooth at a set angle. The cementoenamel boundary detection module is used to determine the cementoenamel boundary point based on the extreme values ​​of the cementoenamel boundary thermogram of each two-dimensional slice. The alveolar ridge detection module is used to determine the alveolar ridge point based on the extreme values ​​of the alveolar ridge thermogram of each two-dimensional slice. The apex detection module is used to determine the apex point based on the extreme values ​​of the three-dimensional apex thermogram. The output module is used to output the local cementoenamel boundary curve, local alveolar ridge curve, and apex point of at least one tooth surface.

[0091] The slice generation module also includes an angle reference submodule and a resampling submodule. The cementoenamel boundary detection module may also include a thermal smoothing submodule and a left-right correspondence submodule. The output module may also include a back mapping submodule and a local curve connection submodule. The processing logic of each submodule corresponds to the above-described method implementation, and will not be repeated here.

[0092] The system can be deployed independently as a detection service, providing single-tooth regions of interest and tooth long axes upstream, and receiving points and curves downstream. The system can be deployed on a local imaging workstation, on a server that interfaces with the hospital's information system, or provide analysis through cloud services.

[0093] Figure 8 A schematic diagram of an electronic device according to an embodiment of the present invention is shown. The electronic device includes a memory 1300 and a processor 1200. The memory 1300 stores execution instructions, and the processor 1200 executes the execution instructions stored in the memory 1300 to implement a method for detecting key periodontal structures.

[0094] The device may include corresponding modules that perform one or more steps in the above process. Modules may be hardware modules specifically configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination thereof.

[0095] This hardware architecture can be implemented using a bus architecture. A bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints. The bus connects various circuits, including one or more processors, memory, and hardware modules. It can also connect other circuits such as peripherals, voltage regulators, and power management circuits.

[0096] A bus can be an industry-standard architecture bus, an external device interconnection bus, or an extended industry-standard architecture bus, etc. Buses can be classified as address buses, data buses, control buses, etc. For ease of illustration, only one connection line is used in the attached diagram, but this does not mean that there is only one bus or one type of bus.

[0097] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. The scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed in a different order than those shown or discussed.

[0098] The processor executes the various methods and processes described above. The method implementations in this invention can be implemented as software programs tangibly contained in a machine-readable medium, such as memory. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above can be performed. Alternatively, the processor can be configured in other suitable ways to execute the methods described above.

[0099] A readable storage medium can be any means that can contain, store, communicate, propagate, or transmit programs for use by an instruction execution system, apparatus, or device. More specific examples include: an electrical connection having one or more wires, a portable computer disk drive, a random access memory, a read-only memory, an erasable and editable read-only memory, an optical fiber device, and a portable read-only memory.

[0100] Various parts of this invention can be implemented in hardware, software, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software stored in memory and executed by a suitable instruction execution system. If implemented in hardware, they can be implemented using discrete logic circuits, application-specific integrated circuits, programmable gate arrays, field-programmable gate arrays, etc., which are well known in the art.

[0101] Those skilled in the art will understand that all or part of the steps of the methods described above can be implemented by a program instructing related hardware. The program can be stored in a readable storage medium. When executed, the program includes one or a combination of the steps of the method implementation.

[0102] In the various embodiments of the present invention, each functional unit can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or in the form of a software functional module.

[0103] The present invention also provides an electronic device, comprising: a memory storing execution instructions; and a processor or other hardware module, wherein the processor or other hardware module executes the execution instructions stored in the memory, causing the processor or other hardware module to perform the above-described method.

[0104] The present invention also provides a readable storage medium storing executable instructions, which, when executed by a processor, are used to implement the above-described method.

[0105] In one implementation, the entire mouth process extracts regions of interest tooth by tooth, determining the long axis and generating multi-angle slices along the long axis. The cementoenamel junction and alveolar ridge are represented by extreme values ​​on the slice thermogram, while the apex is represented by extreme values ​​on the 3D thermogram. Two-dimensional points are mapped back to 3D and connected to form local curves along the tooth surfaces. Output fields include the 3D coordinates of each point, the corresponding tooth surface, the corresponding angle, the thermal response, the local curve, and the apical point. Failed tooth locations are individually marked. The angle interval, Gaussian scale, and axial deviation threshold can be adjusted according to imaging conditions and measurement needs.

[0106] This invention uses thermal map extreme values ​​to determine the cementoenamel junction and alveolar ridge on multi-angle slices along the long axis of the tooth, and uses thermal map extreme values ​​to determine the apex on three-dimensional volume data, and outputs local curves according to the tooth surface tissue.

[0107] The use of terms such as "one embodiment," "some embodiments," "example," or "some examples" indicates that a specific feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. Those skilled in the art should understand that the above embodiments are merely for clearly illustrating the present invention and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications can be made based on the above disclosure, and these changes or modifications are still within the scope of the present invention.

Claims

1. A method for detecting key periodontal structures, characterized in that, include: Obtain the three-dimensional volumetric data of the target tooth and determine its long axis; Multiple two-dimensional slices are cut along the tooth long axis at a set angle, and the multiple two-dimensional slices cover at least two of the mesial, distal, buccal and lingual directions; The cementoenamel boundary point is determined based on the extreme values ​​of the thermogram of the cementoenamel boundary in each two-dimensional slice; The alveolar ridge point is determined based on the extreme values ​​of the alveolar ridge thermogram of each two-dimensional slice. The alveolar ridge point is still determined based on the extreme values ​​of the thermogram at the position where the tooth and alveolar bone are no longer adjacent. Determining the root apex point based on the extreme values ​​of a three-dimensional root apex thermogram; and Output the local cementoenamel boundary curve, local alveolar ridge curve, and apex point of at least one tooth surface.

2. The method as described in claim 1, characterized in that, The step of cutting multiple two-dimensional slices along the tooth's long axis at a predetermined angle includes: Determine the mesial and distal reference directions, which are either the tangent direction of the dental arch curve or the contact direction of adjacent surfaces; Using the mesial-distal reference direction as the angular reference, a set of planes passing through the tooth long axis are generated at set intervals; and The two-dimensional slice is obtained by resampling the plane.

3. The method as described in claim 2, characterized in that, The voxel resolution for resampling is determined according to the image spatial resolution; The view is cut off to a fixed field of view centered on the circumscribed rectangle of the tooth region on the plane; and Orient the coronal side toward the first side of the slice image and the radix side toward the second side of the slice image.

4. The method as described in claim 1, characterized in that, The determination of the cementoenamel boundary point based on the extreme values ​​of the cementoenamel boundary thermogram of each two-dimensional slice includes: The thermal map prediction model outputs a thermal map of the cementum-enamel interface; After smoothing the thermal map of the cementoenamel interface, locally maximal connected regions were extracted; and The centroids of the two connected regions with the highest area ranking are determined as a pair of cementoenamel boundaries on the slice.

5. The method as described in claim 4, characterized in that, During training, the physician marks two points on the slide where the slide intersects with the cementoenamel junction, and places a two-dimensional Gaussian kernel at the two points to generate a supervised heatmap; The spatial scale of the Gaussian kernel is set according to voxel resolution; as well as The monitoring target is a point-like thermal field, rather than a enamel region segmentation mask.

6. The method as described in claim 4, characterized in that, Using the projection of the tooth's long axis in the slice as the midline, the left and right extreme values ​​are mapped to tooth surface points on both sides of the slice, respectively. as well as Based on the direction of the included angle of the slice, the tooth surface point is mapped to the buccal and lingual surfaces, or to the mesial and distal surfaces.

7. The method as described in claim 1, characterized in that, The determination of alveolar ridge points based on the extreme values ​​of the alveolar ridge thermal map of each two-dimensional slice includes: The heatmap prediction model outputs alveolar ridge heatmaps on the same set of two-dimensional slices; After smoothing the heatmap of the alveolar ridge, the extreme points are extracted; and The alveolar ridge thermal map and the cementoenamel interface thermal map can be output in separate channels, or output separately from the two models.

8. The method as described in claim 1, characterized in that, Also includes: Based on the origin and basis vectors of the slice plane, the two-dimensional cementoenamel boundary point and the two-dimensional alveolar ridge point are mapped to three-dimensional points; Three-dimensional points along the same tooth surface are ordered circumferentially and connected to form a local curve of that tooth surface; and The local curve is allowed to be broken, but is not forced to be closed into a full-circuit curve.

9. The method as described in claim 8, characterized in that, If the axial coordinate deviation of the cementoenamel junction point on adjacent angle slices exceeds the deviation threshold, then that point will be downweighted or removed.

10. The method as described in claim 1, characterized in that, The determination of the root apex point based on the extreme values ​​of the three-dimensional root apex thermogram includes: The root tip thermal map is output from the 3D thermal map prediction model; After smoothing the root tip thermogram, the connectivity extrema are extracted; Under the condition of a single root, the centroid of the maximum integral is determined as the apex; and Under multiple root conditions, the extreme values ​​exceeding the response threshold are output as the set of root apex points.

11. The method as described in claim 10, characterized in that, Determine whether the tooth is maxillary or mandibular based on its position number to unify the crown-root direction; and The expected number of roots given by the tooth position number is only an optional constraint and does not divide the set of apical points to a specific tooth surface.

12. The method as described in claim 1, characterized in that, The outputs of the cementoenamel boundary thermal map, alveolar ridge thermal map, and periapical thermal map are spatial thermal fields, not the center of the detection frame.

13. The method as described in claim 1, characterized in that, The local cementoenamel boundary curve, local alveolar ridge curve, and apical point are provided to the subsequent measurement module.

14. A system for detecting key periodontal structures, characterized in that, include: The data acquisition module is used to acquire the three-dimensional volume data of the target tooth and determine the long axis of the tooth; The slice generation module is used to extract multiple two-dimensional slices along the tooth long axis at a set angle, the multiple two-dimensional slices covering at least two of the mesial, distal, buccal and lingual directions; The cementoenamel boundary detection module is used to determine the cementoenamel boundary point based on the extreme values ​​of the cementoenamel boundary thermogram of each two-dimensional slice. The alveolar ridge detection module is used to determine the alveolar ridge point based on the extreme values ​​of the alveolar ridge thermogram of each two-dimensional slice. The alveolar ridge point is still determined based on the extreme values ​​of the thermogram at the position where the tooth and alveolar bone are no longer adjacent. The root tip detection module is used to determine the root tip point based on the extreme values ​​of the three-dimensional root tip thermogram; and The output module is used to output the local cementoenamel boundary curve, local alveolar ridge curve, and apex point of at least one tooth surface.

15. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 13.

16. A readable storage medium, characterized in that, The readable storage medium stores execution instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 13.

Images