Three-dimensional reconstruction method for root canal system

Abstract

The invention relates to a three-dimensional reconstruction method for a root canal system, which comprises the following steps: based on a dental pulp cavity growth path backtracking mechanism, identifying an unclosed growth boundary between a dental pulp cavity and the root canal system and a directional flow trace of an inner cavity structure of the dental pulp cavity, and extracting a structure starting region of the root canal system; according to a physiological mechanics compliance principle, combining dentin development tension and anatomical configuration, constructing a direction barrier and impedance distribution of a root canal path in a space, and generating a three-dimensional channel trajectory conforming to a growth trend; performing closed stress group space gradient analysis of micro-density aggregation on the tooth root area, judging an expansion limit of a root canal path, and identifying a natural termination position of a channel; connection rule extraction is carried out through a closed-channel-closed structure generality mode, branch structures are screened, and false positive connection is inhibited; and back-projecting a track behavior of an iatrogenic instrument entering path in the root canal therapy process into the three-dimensional reconstruction model, performing path consistency verification on the generated model, and triggering structure reconstruction adjustment if spatial deviation exists.

Description

Technical Field

[0001] This invention relates to three-dimensional reconstruction of root canals, specifically a method for three-dimensional reconstruction of root canal systems. Background Technology

[0002] Current methods for 3D reconstruction of root canal systems, such as Chinese patent CN118135091A, "A Method for 3D Reconstruction of Root Canal Posts and Cores Based on Oral CBCT Data," still have shortcomings and drawbacks. These mainly focus on the anatomical and physiological adaptability of the model construction, the integration degree of the physician behavior feedback mechanism, the precision of the path structure evaluation mechanism, and the ability to control false positive connections. This patent mainly uses a deep learning model (ResUNet) to perform coarse and fine segmentation processing on CBCT images, achieving a certain degree of automation in root canal model recognition. It also optimizes segmentation accuracy through double boundary loss. Its innovations are mainly reflected in the training strategy of the image segmentation model, the setting of the loss function, and the simplification of the image processing flow. However, its solution is mainly limited to the image segmentation level and lacks the understanding and guidance of the actual physiological structural characteristics of the root canal. For example, no modeling basis for pulp cavity growth mechanism, dentin development tension, impedance distribution, or spatial direction barriers is introduced in the root canal path construction process. This leads to the reconstructed channel path being smooth and complete in morphology, but prone to non-physiological errors in structural orientation, branch generation, and termination determination.

[0003] Secondly, the scheme does not have a dynamic consistency verification mechanism with clinical operation pathways, nor does it consider reflecting the doctor's intraoperative trajectory behavior in model evaluation and reconstruction adjustment. Therefore, it cannot identify and correct path deviations caused by image errors or insufficient algorithm generalization ability in real time. Furthermore, the model accuracy evaluation system of this patent is relatively coarse, limited to boundary matching or coordinate overlap, and does not construct a path-level multi-factor credibility scoring model. It lacks a mechanism to comprehensively consider multi-dimensional parameters such as starting point identification, path physiological rationality, termination clarity, and consistency with doctor behavior. Therefore, it lacks fine control over the clinical acceptability of the reconstruction results. Further, the scheme does not design a path heatmap visualization mechanism, so doctors cannot intuitively grasp the channel quality distribution during the preoperative navigation stage, nor can they formulate zoning strategies based on credibility levels. In terms of false positive connection control, the scheme's strategy is mainly based on improving the model's segmentation accuracy to indirectly reduce non-target connections. However, it lacks structural suppression measures for the risk of misconnection in areas with blurred image density or complex curved root canal segments. For example, it does not introduce directional barrier crossing criteria, does not set closed-channel-closed structure rule recognition, and does not construct a low-potential impedance channel recognition and envelope layer limitation mechanism. Therefore, the model is prone to false positive generalization in multi-branch structure recognition.

[0004] Furthermore, this method does not handle grayscale gradient analysis of the channel termination region, making it difficult to identify whether there are unvisualized hidden channels at the termination boundary. Especially in cases of low contrast in CBCT or anatomical variations, the model cannot distinguish between closed termination and incompletely explored structures, which may lead to the potential risk of residual pulp or missed root canals in root canal treatment. Summary of the Invention

[0005] The purpose of this invention is to provide a three-dimensional reconstruction method for root canal systems, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.

[0006] The present invention addresses the aforementioned technical problems by employing the following technical solution: a three-dimensional reconstruction method for the root canal system, comprising: based on the pulp cavity growth path tracing mechanism, identifying the unclosed growth boundary between the pulp cavity and the root canal system, as well as the directional flow traces of the pulp cavity structure, and extracting the structural initiation region of the root canal system; then, based on the extracted structural initiation region, and according to the principle of physiological and mechanical compliance, combined with dentin developmental tension and anatomical configuration, constructing the directional barrier and impedance distribution of the root canal path in space, and generating a three-dimensional channel trajectory that conforms to the growth trend;

[0007] Spatial gradient analysis of micro-density clustered closed stress groups is performed on the root region involved along the three-dimensional channel trajectory to determine the expansion limit of the root canal path and identify the natural termination position of the channel. Based on the expansion limit and termination position, connection rules are extracted in multiple branch paths through the common structural pattern of closed-channel-closed, and branch structures with physiological rationality are screened to suppress false positive connections.

[0008] The trajectory behavior of iatrogenic instruments entering the path during root canal treatment is back-projected into the three-dimensional reconstruction model. The path consistency of the generated model is verified. If there is a spatial deviation, structural reconstruction adjustment is triggered.

[0009] Furthermore, the identification of the directional flow traces includes calculating the spatial anisotropy index in the image based on the longitudinal fiber arrangement direction of the residual tissue in the medullary cavity to determine the extension direction of the initial channel; the extracted structural starting region is marked as the initial channel seed point in three-dimensional space and serves as a restrictive starting point in the path generation step.

[0010] Furthermore, during the construction of the directional barrier and impedance distribution, an impedance tensor field based on the dentin micro-density distribution is introduced to dynamically adjust the growth priority of the channel trajectory in complex curved regions; during the generation of the three-dimensional channel trajectory, continuous low-potential channels in the impedance field are identified, and a dynamic envelope layer that restricts path deviation is constructed.

[0011] Furthermore, the spatial gradient analysis is based on a joint evaluation of the density change rate of the closed stress cluster, the consistency of the gradient direction, and the boundary ambiguity. In the process of identifying the multiple branch paths, the structural rationality score is calculated based on the local curvature of each branch, the deviation angle of the main channel, and the consistency of the growth direction, and only branch structures with scores higher than a preset threshold are retained.

[0012] Furthermore, in the process of suppressing false positive connections, if a branch path crosses two or more directional barriers at the same time, it is marked as a non-physiological path and removed; the acquisition of iatrogenic instrument trajectory behavior includes the time-series position information of the instruments during the operation, and the interpolation algorithm is used to map it into a three-dimensional reconstruction model to achieve path fitting and comparison.

[0013] Furthermore, during the three-dimensional modeling of the root canal structure, the system introduces a dynamic behavioral consistency analysis mechanism between the surgeon's operating path and the reconstruction path. A comprehensive path consistency scoring index is constructed based on the difference in the directional angle between the two in three-dimensional space and the overlapping volume areas. This scoring index is not only used to evaluate the effectiveness of the model structure but also serves as the basis for the system to determine whether local reconstruction adjustments are triggered. In addition, in the final generated three-dimensional model of the root canal channel, each path segment is assigned a multi-factor joint confidence label. This label comprehensively evaluates the path generation quality using a nonlinear multivariate integral formula, considering the following four core variables:

[0014] Reliability of starting point identification;

[0015] The physiological rationale for path generation;

[0016] Structural clarity of the channel termination area;

[0017] Consistency between physician behavior pathways and modeling pathways;

[0018] An integral model is used as the path confidence scoring function:

[0019]

[0020] in:

[0021] C i The overall confidence score for channel segment i; Δt = t1 - t0 is the duration of the operation window within the evaluation period; θ i (t) represents the spatial angle difference between the operation path and the modeling path at time t; ρ i (t) represents the spatial offset distance of the starting point relative to the anatomical reference center; κ i (t) represents the rate of change of curvature of a local segment of the path, reflecting its physiological curvature consistency; χ i(t) represents the structural density gradient fluctuation of the termination region; ω(t) represents the system sampling confidence weighting function during the operation; ψ(·), ξ(·), η(·), and φ(·) are the included angle difference function, starting point accuracy function, physiological rationality function, and termination clarity function, respectively, which are nonlinear transformation functions defined in this invention and satisfy the monotonically decreasing characteristic; α, β, γ, and δ are the weight coefficients of the scoring factors of each dimension, satisfying α+β+γ+δ=1; this scoring function integrates spatial behavior deviation, physiological structure consistency, and doctor's operational stability through time integration, and can dynamically characterize the credibility of each path between actual operation and model;

[0022] Its integral weight ω(t) enables the system to emphasize the highly stable segments of the doctor's operation process while weakening short-term jitter or non-target operation segments, thereby making the scoring results more stable and reliable. Through the continuous feedback of this function, the system can trigger path adjustment under the following two conditions:

[0023] Multiple score segments consecutively fell below the set lower limit;

[0024] Doctors manually mark areas of path deviation;

[0025] In the process of three-dimensional reconstruction of root canals, to effectively quantify the local quality and clinical reliability of the access segment, this invention designs a dynamic integral path confidence function to comprehensively evaluate the geometric accuracy, physiological rationality, and behavioral consistency of the access segment. The design of this function is based on the following four mutually coupled technical observation points:

[0026] 1. Identify whether the initial point of the path deviates from the anatomical entry feature location;

[0027] 2. Does the path itself exhibit a physiological curve direction consistent with the direction of dentin tension?

[0028] 3. Does the terminal area of ​​the passage exhibit clear closure and expansion limits?

[0029] 4. Whether the doctor's surgical path actually traverses and conforms to the spatial orientation of the reconstructed structure;

[0030] Considering that the above four factors have different degrees of importance in different cases and that their dynamic behavior has a time continuity characteristic, this invention introduces a weighted integral function to score the path.

[0031] First, the analysis period for channel segment i is set to [t0, t1], with a total duration of Δt = t1 - t0. During this period, the system continuously collects temporal difference data between the doctor's instrument movement path and the reconstruction channel path. To quantify the consistency between the doctor's path and the reconstruction path, a time function θ is introduced. i(t) represents the angle difference index of this path segment at time t. The larger the value, the more serious the deviation between the doctor's operation trajectory and the system model. In order to convert this error into a score, this invention defines a monotonically decreasing function ψ(θ). i (t)), this function maps the angle deviation to the score value, and its value range is in the interval [0,1]. The closer it is to 1, the better the consistency.

[0032] Similarly, let ρ i (t) represents the spatial drift of the starting point at time t, indicating the distance of the initial channel identification position relative to the anatomical center of the pulp cavity; the corresponding function ξ(ρ) is defined. i (t) serves as the confidence scoring function for starting point identification, assigning higher scores to smaller initial drift values; the physiological rationality of the channel path is determined by the local curvature change rate function κ. i (t) represents; to quantify whether this morphology conforms to the direction of the dentin development tension curve, the function η(κ) is introduced. i (t) represents its compliance score, which gives a higher score for smooth, continuous paths;

[0033] The structural clarity of the channel end region is represented by the variable χ. i (t) describes the degree of discontinuity of the density gradient in the termination region; the more blurred the termination, the larger this value; corresponding to the function φ(χ). i (t) is converted into a clarity score; considering that behavioral stability during the operation has a moderating effect on the score, a time-weighted function ω(t) is introduced as a stability function of the doctor's operation behavior in different time periods; finally, the overall confidence score function of channel segment i is defined as C. i :

[0034] Wherein, α, β, γ, and δ are the weight factors for each scoring item, satisfying the constraint α+β+γ+δ=1, which can be customized according to the doctor's operating style or the type of lesion structure.

[0035] The advantages of this formula are: it introduces doctor's operational feedback while maintaining the precision of physiological structure control, and eliminates the influence of local perturbations through integration, so that the scoring results have both structural and behavioral robustness; its nonlinear mapping function (such as ψ, ξ, η, φ) can be a piecewise continuous function based on empirical curve fitting or an adaptive function approximator trained (such as spline surface or hyperbolic fitting) in the specific implementation, without relying on classical analytical functions, thus ensuring the novelty and specificity of the scoring function.

[0036] Furthermore, the angle difference is calculated based on the directional jump point of the doctor's path and the reconstruction path at the change of channel curvature; the overlapping volume ratio is achieved by setting a segmented volume superposition strategy for local comparison regions at the beginning, middle and end of the channel respectively.

[0037] Furthermore, a confidence heatmap is generated for each root canal segment and mapped onto the surface of the three-dimensional reconstruction model to visualize the channel quality during preoperative virtual navigation; the score is based on the angular dispersion between the channel path and the direction of the dentin microstructure, and a higher score is obtained if the angular change is smooth.

[0038] Furthermore, a contrast gradient is introduced at the end of the channel. The value of the contrast gradient reflects the grayscale transition degree of the termination boundary and is used to determine whether there are unidentified hidden channels at the termination boundary.

[0039] Furthermore, the reliability of the starting point is determined by combining the degree of spatial overlap with the actual entrance to the pulp cavity and the clarity of the image edges, with the weights of the two adjusted according to the case type.

[0040] This invention constructs a dynamic, high-fidelity, and clinically executable three-dimensional reconstruction method for root canal systems by integrating multi-scale structural analysis, physiological consistency modeling, and physician behavioral path feedback mechanisms. This significantly improves the overall performance of the reconstruction model in terms of path accuracy, clinical usability, and preoperative decision support. Its beneficial effects are mainly reflected in the following aspects:

[0041] First, by analyzing the spatial anisotropy of directional flow traces, this method can accurately pinpoint the starting direction of potential natural channels in the pulp cavity structure. Using this starting point as a limiting seed point, it guides the initial trend of channel generation paths, thus avoiding deviation from the physiological entry area. Second, during path generation, a dentin density-driven impedance tensor field is introduced to dynamically adjust the path growth priority based on the degree of anatomical curvature. Combined with the identification mechanism of continuous low-potential channels, it effectively avoids path jumps or structural punctures in complex bifurcation or curvature areas. Regarding spatial expansion judgment, this invention proposes a density gradient combined with fuzzy boundary evaluation mechanism for closed stress clusters, using the stable boundary of stress distribution as the natural termination limit of the path, effectively reducing structural misidentification caused by imaging artifacts. For multi-branch cases, the system introduces a structural rationality scoring model, comprehensively considering the local curvature, deviation angle, and consistency of the main channel for each branch, retaining only channels with scores higher than a threshold, achieving dual screening of structural rationality and physiological trend. To address the risk of false positive connections, the method sets a directional barrier crossing criterion, automatically identifying and eliminating non-physiological channels that cross multiple directional barriers, significantly improving the model's structural accuracy.

[0042] Simultaneously, by collecting the surgeon's intraoperative instrument path and mapping it to a 3D model, the system compares path consistency scores with trajectory behavior. If consecutive low consistency scores or manually marked deviation areas are found, a local reconstruction mechanism is automatically triggered to construct a realistic anatomical model consistent with the surgical behavior. Furthermore, the model embeds confidence labels in each channel segment, considering four dimensions: starting point overlap, path physiological curvature rationality, termination clarity, and behavioral trajectory consistency. An integral function is used for nonlinear joint scoring to ensure the system has a quantitative expression capability for path quality. This score can be further used to generate a confidence heatmap and map it onto the surface of the 3D model, providing intuitive visual guidance for reliable channel areas in preoperative virtual navigation. In addition, for termination area identification, this invention introduces a closed contrast gradient index, using the continuity of grayscale transition to evaluate the closure of the endpoint and identify potential hidden branch channel risks. For starting point judgment, the system uses a dual index of spatial overlap and edge clarity for joint scoring, combined with case type to set weighting factors, to improve the stability and adaptability of seed point identification. Attached Figure Description

[0043] Figure 1 This is a simplified flowchart of the root canal three-dimensional reconstruction and path correction of the present invention.

[0044] Figure 2 This is the main flowchart for the three-dimensional reconstruction and behavioral verification of root canals in this invention.

[0045] Figure 3 This is a diagram showing the integrated relationship between dynamic behavior consistency and path confidence in this invention.

[0046] Figure 4 This is a flowchart illustrating the implementation of three-dimensional main branch identification and consistency verification of root canals in Embodiment 1 of the present invention.

[0047] Figure 5 This is a flowchart illustrating the implementation of root canal three-dimensional channel consistency scoring and visualization in Embodiment 2 of the present invention. Detailed Implementation

[0048] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0049] Combined with appendix Figure 1This invention proposes a method for three-dimensional reconstruction of the root canal system. Based on the pulp cavity growth path tracing mechanism, a reverse structural tracing model for identifying the continuous structure of the pulp cavity and root canal is constructed. The method extracts the boundary region of the pulp cavity from CBCT image data and, combined with the directional growth characteristics of the tissues within the pulp cavity during development, focuses on identifying unclosed channels or low-density extensions from the bottom of the pulp cavity towards the root apex. These channels appear as blurred boundaries or low-contrast initial directions in the image space, thus being extracted as the starting region of the root canal system structure. Then, using this starting region as the seed point for path generation, and based on the principle of physiological and mechanical compliance, combined with the tissue force field direction and anatomical constraints exhibited by dentin during root development, the method determines the location of the reconstructed path. A spatial impedance distribution model is established. This model constructs a set of directional barrier regions by fitting the micro-density gradient and tension response direction in the dentin region. This shields spatial segments that lack physiological accessibility while opening growth paths consistent with the direction of tooth development. Based on the starting point, a three-dimensional channel trajectory that conforms to anatomical morphology and physiological tension curvature is generated through a path evolution algorithm. This trajectory achieves natural curvature and gradual narrowing from the pulp chamber to the root apex in space, while preserving common structural changes in the actual root canal system such as bifurcation and bending. The generated channel model not only improves the anatomical accuracy of the three-dimensional structural reconstruction but also has physiological acceptability that is highly consistent with the intraoperative operation path, providing a structural basis for subsequent path scoring, confidence judgment, and closed-loop correction.

[0050] Based on the obtained continuous three-dimensional channel trajectory, a refined micro-density aggregation analysis was further performed on the root region traversed by the trajectory. By extracting voxel regions adjacent to the path in the three-dimensional CBCT image and calculating parameters such as density continuity, local density change rate, and density gradient direction difference in these regions, a spatial gradient tensor field model was constructed to identify high-density contraction regions in the image. These regions often exhibit abrupt changes in image density or tissue hardening aggregation, typically representing dentin terminal structures or incompletely developed root apex regions. Such high-gradient aggregation regions were defined as closed stress clusters, indicating that the structural space had reached the limit boundary of natural tissue morphology termination or anatomical growth. By performing a closure trend analysis on the neighborhood of the path end using a spatial gradient function, when the gradient direction shows a convergence state and density continuity is broken, this location was determined to be the path expansion limit, thus identifying it as a root canal. The system identifies the natural termination point of the channel. After locating the termination point, based on the multi-branching trend throughout the path, it detects local density disturbances in the area adjacent to the main path to identify potential channels exhibiting a three-segment structure pattern of closure-channel-closure. This common structure is widely present in complex root canal systems, such as second root canals, lateral root canals, or apical bifurcation structures. In images, it typically appears as a closed area at the front and back with a low-density, thin channel in the middle. The system compares the path morphology by constructing a pattern recognition rule base, calculates the structural continuity, connection angle, radius contraction ratio, and angle trend with the main path for each detected branch structure, and comprehensively sets a physiological rationality threshold to screen out true branches with developmental potential, while eliminating false positive connection structures caused by noise, image artifacts, or irregular cementum. This enables the identification, accurate access, and path quality control of complex branches in the root canal system.

[0051] This system introduces iatrogenic instrument manipulation behavior as external input feedback. The actual entry paths of instruments used by the physician during root canal treatment (such as preparatory files, measuring files, and root canal locator probes) within the root canal are collected in real-time or recorded. Spatial trajectory data of the instruments is acquired through intraoperative navigation equipment, a 3D motion capture system, or an image fusion positioning system. This trajectory data includes the starting position, spatial direction, entry depth, curvature of the manipulation trajectory, and time-series characteristics of the instrument entry channel. This path is uniformly converted into a spatial behavior vector set and mapped onto the coordinate system of the reconstructed model, thereby achieving a behavioral consistency evaluation between the physician's actual manipulation path and the system-constructed path. The system calculates the spatial overlap between the behavioral trajectory and the model path, including the path's... The system considers multiple dimensions, including the angle difference between the two sides, the percentage of overlapping volume areas, curvature similarity, and entry depth deviation, and comprehensively constructs a consistency scoring index. When the score is lower than a set threshold, or when there are areas of spatial offset (such as the actual operation path penetrating areas not built in the model or significantly deviating from the model path), the system judges that there is a risk of structural misconstruction or path misidentification in the model segment, triggering a structural reconstruction and adjustment mechanism. The adjustment mechanism reverts to the path generation process locally based on the distribution characteristics of the deviation area, re-evaluates its starting point extraction, impedance distribution, and endpoint boundary, and combines the doctor's actual operation trajectory as an auxiliary path guidance constraint to reconstruct the three-dimensional channel segment of the area, ensuring that the final model, based on the consistency of the anatomical structure, constructs a closed-loop correction process of structure-behavior fusion.

[0052] Combined with appendix Figure 2In the extraction of the structural initiation region and path generation, local image structure analysis is performed on the bottom region of the pulp cavity, focusing on identifying the fiber arrangement direction with longitudinal structural features in the residual tissue of the pulp cavity. This arrangement direction usually originates from the physiological direction of the pulp tissue extending along the root during development. By extracting spatial orientation features from the voxel set in CBCT images, a local anisotropic response tensor field is constructed. The anisotropy index is used to measure whether there is a stable single dominant direction in a local region. The higher the value, the more obvious the directional tendency of the tissue structure in that region. The system searches for the maximum anisotropy point and its dominant direction vector in the bottom region of the pulp cavity. This directional result is used as a flow trace representing the potential channel direction. By combining the continuity of image grayscale with the regularity of tissue texture, this region is extracted as the structural starting region. After extraction, the system calibrates this region in three-dimensional space, generating an initial channel seed point containing the starting coordinates, direction vector, and physiological confidence label. The seed point not only serves as the starting input of the path generation algorithm, but also serves as a spatial constraint point in the subsequent path evolution process. That is, the system must not bypass or deviate from the directional anchoring range defined by this point during the three-dimensional path growth, thereby ensuring that the generated path has high anatomical accuracy and directional stability in the initial stage, preventing path drift caused by image blurring or unclear boundaries, and providing dual protection for the subsequent three-dimensional channel reconstruction in terms of biological developmental logic and spatial stability.

[0053] In constructing the three-dimensional channel path, to achieve spatial coordination and physiological rationality between the channel trajectory and the actual anatomical tissue, a directional barrier construction mechanism based on the micro-density distribution characteristics of dentin is adopted. In this mechanism, voxel-level density analysis is performed on the root region in CBCT images to extract the micro-density gradient changes of dentin at different spatial locations. Based on this, a continuous three-dimensional impedance tensor field is constructed. This impedance tensor field converts density increments in different directions into growth impedance values, forming a set of directional reference systems for channel growth path regulation. During the system path growth process, the selection of the growth direction at each step is regulated by the local tension distribution of this tensor field, especially in cases of complex root curvature, bifurcation, or abrupt curvature changes. The system identifies continuous low-resistance path segments within the impedance field—regions exhibiting slow density gradient changes and minimal resistance—and defines them as low-potential channels, prioritizing them as the primary extension direction for path evolution. Simultaneously, based on the aforementioned impedance tensor field, a dynamic path offset constraint mechanism is constructed. During channel growth, a spatial envelope layer tangent to the current path is generated in real-time. The extent of this envelope layer is determined by the local impedance distribution, and the channel path must not deviate from the boundary of this dynamic envelope. This effectively restricts path drift, jumps, and non-physiological bends in space, ensuring the entire channel trajectory remains within a constraint trajectory consistent with the dentin tissue growth trend. Furthermore, the system adjusts growth priority and direction selection strategies based on changes in the impedance field.

[0054] In the process of extending the boundary and screening branch structures of the channel path, a spatial gradient field is constructed in the 3D CBCT image space to analyze the density changes of the root tissue. The system uses closed stress clusters as the core of analysis, and obtains the velocity characteristics of density increase or decrease in continuous space by multi-scale sampling of the density change rate of the voxels in the vicinity of the channel end. Simultaneously, it performs consistency analysis on the gradient direction of density change to identify whether there is a gradient vector field converging towards the center point. If gradients in multiple directions tend to converge, it indicates that there is a stress closure trend in the region. Furthermore, the system introduces boundary ambiguity as a measure of the strength of the boundary signal, fitting the transition width of image grayscale at the boundary. If the boundary region exhibits a long-distance ambiguity transition and a low grayscale variance, it can also be determined that the region is a closed termination structure. By jointly evaluating the above three dimensions—density change rate, gradient direction consistency, and boundary ambiguity—the system forms a multi-factor termination reliability model. The system first determines the natural termination point of the main channel. After determining the termination position, the system enters the branch path structure identification stage. This stage involves searching for lateral or multi-channel paths in the vicinity of the main channel and calculating the structural rationality score for each candidate branch structure. This score comprehensively considers the following three indicators: first, the curvature level of the local branch path; if the curvature is too high, it is a misidentified structure; second, the deviation angle between the branch path and the main channel; if the deviation angle is too large, it does not conform to the actual branch morphology; and third, the consistency between the branch extension direction and the overall growth direction, i.e., whether the branch is still within the envelope of the root tension distribution axis. The three indicators are assigned weights and then weighted and integrated to form a structural rationality score. Finally, the system sets a preset scoring threshold and only retains branch structures with scores higher than the threshold in the final three-dimensional model. Other paths that do not meet the structural physiological logic are eliminated, thereby achieving reliable construction of complex branch systems and suppressing false positive identification.

[0055] In the process of constructing branch structures and evaluating path quality, to effectively suppress false positive connection paths caused by image artifacts, noise interference, or algorithm misjudgment, a branch rejection rule based on directional barrier crossing behavior was established. After the system completes the initial path generation, the interaction between each candidate branch path and the directional barrier region generated by the impedance tensor in space is analyzed. The directional barrier is a path restriction layer established by the system based on dentin density distribution and anatomical morphology, used to prevent paths that do not conform to physiological orientation from crossing. If a branch path crosses two or more directional barrier regions in its complete pathway, the system determines that the path has non-physiological characteristics such as significantly inconsistent with the direction of tooth tension, excessive spatial offset, or crossing of high impedance regions, and thus marks it as a non-physiological path and rejects it, preventing it from entering the final model output, ensuring that the 3D reconstructed structure conforms to the biological composition logic. On this basis, to further improve the modeling results and actual intraoperative... To ensure operational consistency, the system incorporates iatrogenic instrument trajectory behavior as an external verification reference. Specifically, it works in conjunction with intraoperative navigation, 3D positioning, or optical tracking systems to collect real-time temporal position information of the root canal treatment instruments used by the surgeon during the procedure. This information includes the instrument's 3D coordinates, direction of movement, and entry depth at different time points. The system converts this information into a time-series path vector and uses an interpolation algorithm to smoothly reconstruct discrete trajectory points, forming an operational path curve with continuous spatial definition. Subsequently, this trajectory behavior is back-projected onto the generated 3D root canal model and spatially matched and fitted with the system's reconstructed path. By calculating indicators such as the spatial angle, volume overlap rate, and entry point deviation between the instrument trajectory and the model path, the system comprehensively evaluates the consistency level between the actual operation and the model structure. This information is used for subsequent confidence correction, path adjustment, or structural reconstruction trigger judgments, achieving a two-way closed-loop verification mechanism between structural modeling and clinical behavior.

[0056] Combined with appendix Figure 3 This study introduces a dynamic behavioral consistency analysis mechanism based on the doctor's operational path and the system's reconstructed path. By comparing the spatial angle differences and volume overlap areas between the doctor's instrument path during surgery and the system-generated 3D root canal path in real time, a comprehensive path consistency score is constructed. This score not only assesses whether the model structure is consistent with the actual operational behavior but also serves as an important basis for determining whether the system needs to trigger local structural reconstruction adjustments. Furthermore, to achieve model structural transparency and traceability, each path segment in the final generated 3D root canal channel model is assigned a multi-factor joint confidence label. This label is calculated through a dynamic nonlinear integral function, comprehensively considering the following four core factors: the credibility of the starting point identification, the physiological rationality of the path generation, the structural clarity of the channel termination region, and the spatial consistency between the doctor's instrument path and the system's modeled path.

[0057] The path confidence integral scoring function is defined as follows:

[0058]

[0059] Among them, C i The overall confidence score of path segment i is represented by Δt = t1 - t0, which represents the length of the evaluation time window for that path segment; θ i (t) represents the spatial angle difference between the doctor's operational path and the modeled path at time t, used to reflect behavioral consistency; ρ i (t) represents the spatial offset distance of the starting point of the path segment relative to the anatomical reference center at time t, used to evaluate the accuracy of starting point identification; κ i (t) represents the rate of change of local curvature along the path, reflecting whether it conforms to the natural physiological curve characteristics of the dentin tension direction; χ i ω(t) represents the density gradient fluctuation value of the termination region, used to evaluate structural closure and boundary clarity. The functions ψ(·), ξ(·), η(·), and φ(·) correspond to the included angle difference scoring function, the starting point accuracy scoring function, the physiological curvature scoring function, and the termination clarity scoring function, respectively. All are monotonically decreasing nonlinear functions with an output range of [0,1], and are sensitive to errors and enhance stability output within reasonable intervals. The weighting coefficients α, β, γ, δ ∈ [0,1], satisfying the constraint α + β + γ + δ = 1, and can be configured according to specific clinical operational preferences and tooth structure types. ω(t) is a time-weighted function reflecting the system's judgment of the stability of the operation at different time points. This function assigns higher weight to stable advancement segments in the operation path, while reducing the scoring impact on segments with frequent shaking or mis-entry into the path, thus ensuring priority emphasis on high-confidence behavioral regions during integration.

[0060] This integral scoring model focuses on the accuracy of anatomical structure recognition and the synergy of behavioral paths. In application, the system continuously collects spatial trajectory data of the doctor's instruments during operation and matches it with the reconstruction model. The score value obtained by the above integral function is calculated in real time as dynamic confidence feedback. If the system finds that the score Ci is continuously lower than the set threshold in multiple consecutive path segments, or if the doctor actively marks the area with deviation in the interface, the system will trigger a local reconstruction adjustment mechanism to re-estimate the structure in the path area and perform corrective modeling.

[0061] To achieve high-precision consistency assessment between the doctor's operative path and the system's reconstructed path, a difference quantification mechanism targeting key geometric feature points of the channel is proposed. The calculation of the angle difference not only considers the directional deviation of the path in the overall space but also focuses on directional jump points where the channel curvature changes. Specifically, the system first identifies key locations in the channel path where there are significant geometric changes, i.e., locations corresponding to abrupt changes in the first or second derivative of the path, typically appearing in areas such as bends, bifurcations, and corners of the root canal. These points are defined as directional jump points. Then, the system extracts the tangential direction vectors of the doctor's operative path and the model's reconstructed path at these jump points, calculates the spatial angle between these vectors to obtain the local angle difference, and weights and averages the angle differences at all jump points to form the overall structural behavior angle deviation of the channel. The difference index can more sensitively reflect the consistency of the path at key bends, effectively avoiding false matching problems caused by similar overall trends but deviations in details. In terms of calculating the overlap volume ratio, a segmented volume superposition strategy is introduced, that is, each segment of the root canal path is divided into three structural functional segments: the starting segment, the middle segment, and the ending segment. Local comparison areas of fixed length or proportion are set for each segment. In each comparison area, the voxel block of the system-reconstructed path and the equivalent channel voxel block formed by the doctor's operation path are extracted. The ratio of their spatial overlap volume to the combined volume is calculated to obtain the segment overlap rate. Finally, the volume overlap index of the three regions is integrated through a weighted aggregation strategy to form the overall overlap volume ratio. This takes into account the anatomical differences and operational sensitivity of the root canal structure in different segments, and especially strengthens the local attention to the accuracy of the starting positioning and the closure of the ending.

[0062] For each root canal segment, a corresponding confidence heatmap is generated, and this heatmap information is mapped onto the surface mesh of the 3D reconstructed model in a fitted manner. This allows users to directly identify the modeling reliability of the path segment through color gradients in the navigation interface. Specifically, the heatmap generation is based on the calculation of the angular dispersion between the channel path direction and the dentin microstructure arrangement direction. The system first extracts the tangential direction vector of the path at fixed intervals on the reconstructed channel path and registers it with the pre-identified dentin fiber direction within the tooth body region where the path is located. Then, the angle between the two is calculated at each corresponding position, and the fluctuation range of the angle sequence is evaluated using a sliding window method, i.e., the local standard deviation or range of the angle change. If the... A relatively stable change within a continuous path segment indicates that the path segment better conforms to physiological structures, has higher modeling confidence, and therefore a higher score. Conversely, if the angle changes frequently and abruptly, it indicates that the path traverses areas of abnormal impedance or areas with unclear anatomical structures, and the corresponding score will decrease. Finally, the system spatially converts the confidence score into a color map, generating a heat map from high confidence (e.g., green) to low confidence (e.g., red), and attaches it to the channel surface of the 3D reconstruction model using surface texture rendering technology. During the doctor's preoperative review or path planning process, this enables real-time perception and dynamic guidance of the channel quality status, thereby improving the safety of navigation path selection and the pertinence of strategy formulation.

[0063] The system introduces the channel termination contrast gradient index as a key parameter for structural termination judgment. This parameter is obtained by analyzing the gray-level distribution variation characteristics of the channel termination region in 3D image data. A multi-layered spherical sampling region is established near the termination point of each generated root canal path, and a gray-level value sequence during the transition from the channel interior to the external tissue is extracted within this region. The spatial gradient of this gray-level sequence is then calculated, which is the ratio of the gray-level difference between any two adjacent sampling points to their spatial distance, thus constructing a radial contrast curve at the channel termination. The system further performs statistical analysis on this curve. If the gray-level transition in the channel termination region is abrupt and exhibits a high-amplitude gradient peak, it is determined to be clearly closed and without any extended structures. If the gray-level transition is slow, the contrast gradient value is small, or even tends to be flat, there is a potential for unidentified hidden channels or branch extensions. Based on this, the system outputs a "termination boundary unclear" label and, combined with the surrounding impedance tensor field and anatomical structure information, re-triggers the deep structure recognition module to expand the scanning area to find potential missed pathways. This method captures weak structural clues through the contrast gradient variation pattern at the channel termination, significantly improving the detection rate of hidden branches.

[0064] The system establishes a reliability evaluation mechanism for the starting point of each channel. This mechanism comprehensively considers two dimensions of indicators: First, the degree of spatial overlap between the extracted starting point and the actual entrance to the pulp cavity. During the preprocessing stage, the system uses image registration and anatomical template matching to locate the reference entrance region of the pulp cavity and constructs its volume boundary model in three-dimensional space. Subsequently, the spatial distance between the extracted starting point and this model is calculated and the voxel intersection is evaluated. The higher the overlap rate, the more accurate the spatial positioning. Second, the edge sharpness index of the image around the point in the cross section. The system calculates the local gradient magnitude and edge connectivity of the region using image gradient operators (such as 3D Sobel or higher-order edge enhancement). The reliability score is determined by the sharpness of the edges and the stability of the boundary changes. The reliability score of the final starting point is obtained by weighted fusion of the above two indicators. To adapt to the differences in imaging characteristics of different types of root canal cases, the system allows the adjustment of the weight coefficients of the two indicators based on the case label. For example, in cases of pulp cavity calcification or blurred boundaries, the weight of edge sharpness is reduced and the influence of spatial overlap is increased; while in cases with clear edges but abnormal spatial structure, the opposite configuration is applied. This kind of dynamic weight setting strategy can be manually specified by the doctor or allocated by the system based on the case training model, thereby achieving robust control over the localization of complex starting areas and providing stable and reliable structural starting point support for subsequent three-dimensional path generation.

[0065] Example 1:

[0066] Combined with appendix Figure 4 In this embodiment, a 45-year-old male patient presented with suspected multiple root canal branches in his right mandibular first molar on clinical imaging. After obtaining the raw data from the CBCT scan, the data was input into a three-dimensional reconstruction system. First, the system identified directional flow traces through image feature analysis of residual tissue in the pulp cavity. In the three-dimensional reconstructed voxel model, the pulp cavity region was selected as the ROI, and anisotropic tissue points with gray levels in the range of [1200, 1500] HU were extracted. The system applied structural tensor analysis to construct a feature vector field in this region and fitted the fiber arrangement trend through the distribution of principal axis vector directions. For example, in the slice layer numbered "Z=54", the longitudinal principal axis direction is (0.85, 0.12, 0.51), indicating that the fiber structure in this region is inclined distally, lingually, and inferiorly. The system used this direction as a guide to set the initial channel extension direction, and combined with the position coordinates (42.3, 61.5, 54.0), it was marked as the starting channel seed point in three-dimensional space and stored in the path generation module as a restrictive starting point.

[0067] Before generating the 3D path, the system enters the impedance tensor field construction step. First, the dentin density distribution is extracted from the reconstructed volume containing the entire root region and converted into a relative impedance index through voxel grayscale. For example, if the voxel density is 1350 HU, the corresponding impedance weight coefficient R = 1 - (HU - 1200) / 600 = 0.75. After constructing the full-volume impedance tensor field, in the channel trajectory numbered "path segment #1", the system recursively searches along the starting direction with a step size of 0.5 mm. At each position point, a directional gradient tensor is constructed. By judging the path with the minimum impedance in the direction of the tensor principal axis, the system identifies that the current step direction belongs to a continuous low-potential channel in the 3D impedance field. At the same time, the system constructs a dynamic envelope layer in this region, forming a resistance offset restriction band around the center point of the current path with a radial radius of 0.5 mm. If the path is to break through this envelope layer in subsequent growth, a directional correction weight will be applied to return to the low-resistance direction.

[0068] Near step 5 (45.0, 62.1, 57.2), the path orientation has an angle of 18.2° with the principal axis of the impedance tensor, which is a region with high tensor compliance. The system assigns a channel confidence weight of 0.92 to this point and continues to advance. At step 9, the system detects that the path direction is about to cross a region with significantly increased impedance (R rises to 0.88), and the angle between the principal axis of the tensor and the current direction rises to 47.6°. The system adjusts its direction to the nearby resistance depression (46.7, 63.2, 59.8) through envelope control, finally forming a continuous low-resistance channel from the seed point of the pulp chamber to the junction of the middle segment of the root canal, with a length of 7.2 mm and an average path curvature of 0.042 / mm. The system marks it as a first-order physiological compliance path and submits it to the subsequent extension boundary and termination analysis module.

[0069] After generating the spatial trajectory of the main channel, the system enters the stage of identifying root canal branch paths and analyzing false positives. First, it performs spatial gradient analysis on the neighborhood of the path endpoint. The system scans the density gradient change rate of each voxel point within a 3mm range of the endpoint, identifies point clusters with abrupt changes in local density between 1350 and 1520 HU, and applies a spherical expansion operation to these point clusters to construct a closed stress cluster envelope. Then, it analyzes the consistency of the gradient direction. For example, if the principal axis of the gradient direction of a stress cluster is (0.82, 0.03, 0.57) and the direction consistency score is 0.91, it indicates that the cluster has high structural orientation. Further, its confidence is corrected by boundary ambiguity calculation. If the slope of the grayscale boundary change is less than 15 HU / mm, the ambiguity score is less than 0.3, and the stress cluster is excluded. Finally, 6 closed stress clusters with a confidence score higher than 0.7 are selected as seed points for potential branch paths.

[0070] The system recursively branches from each of the aforementioned high-confidence points. Among the 11 generated branch paths, the system calculates three structural rationality indices for each path: local curvature κ, deviation angle θ from the main channel, and growth direction consistency ε. Path number 4 has a local curvature mean of 0.061 / mm, a main channel deviation angle of 26.5°, and a growth direction consistency score of 0.78. The joint structural rationality scoring function is applied.

[0071]

[0072] With weights set as λ1 = 0.3, λ2 = 0.4, and λ3 = 0.3, then S4 = 0.3 × (1 - 0.061 / 0.15) + 0.4 × (1 - 26.5 / 90) + 0.3 × 0.78 ≈ 0.84. The system sets the structural rationality threshold to 0.75, and ultimately retains the four branch structures with scores above 0.75 for subsequent merging.

[0073] To further suppress non-physiological connection paths, the system applies a directional barrier crossing check to retained branch paths. The location of the directional barrier layer is marked when constructing the impedance tensor field of the main path. For example, there is a tensile resistance region dominated by the normal vector (-0.5, 0.7, 0.3) in the Z=65 to Z=69 layers. If a path segment crosses more than two directional barrier bodies, it is marked as a non-physiological path. For example, if the second branch path crosses the barrier regions Z=68 and Z=72 consecutively during the tracking process, the system immediately marks it as an invalid path and removes it to avoid abnormal extension of the channel structure.

[0074] Based on the aforementioned path recognition, iatrogenic instrument trajectory behavior was further introduced for model verification during reconstruction. Intraoperatively, time-series coordinates were recorded using an ultrasonic root canal file in conjunction with a position tracking system. For example, 83 frames of instrument position information were collected from the 10th to the 12th minute, forming a path point sequence P(t). The system performed five B-spline interpolations on the path to generate a smooth trajectory curve, which was then compared with the reconstructed path in terms of volume overlap. Cylinders with a diameter of 1.5 mm were set at the beginning, middle, and end of the path for local comparison. The system detected that the overlap rate between the iatrogenic path and the modeled path was 92% at the beginning, 86% at the middle, and 54% at the end. The maximum angle difference occurred at layer Z=74, at 42.1°. The system calculated the consistency score of the entire path to be 0.68, which was lower than the system's set lower limit of 0.75. Since two consecutive segments were below the threshold, the system triggered reconstruction adjustment for that path area, reselected the main direction, and fitted a structural path with better physiological consistency.

[0075] Example 2:

[0076] Combined with appendix Figure 5In this embodiment, a 45-year-old male patient (patient number P045) was examined, and his mandibular first molar was suspected of having a complex root canal structure. After clinical acquisition of CBCT images, the system identified three suspected main channel regions and used the anisotropy index of residual pulp tissue (FA value around 0.72) to determine the starting direction of the main channel, setting the starting seed point as (x0, y0, z0) = (12.4, 24.1, 5.8) mm. During the path generation stage, based on the intrinsic density tensor field distribution, a high-resistance bending region was found in the right middle root, with its equipotential channel direction offset by 17°. The system adjusted the growth priority to the left middle branch and simultaneously initiated dynamic envelope control of path offset to limit the path deviation to no more than ±2 mm.

[0077] The system then proceeds to the confidence scoring phase for channel segment i. Within the timeframe [t0, t1] = [0s, 10s], the system performs consistency analysis sampling between the doctor's operational path and the model path, with a sampling frequency of 1Hz, collecting a total of 10 sets of data. The operational data is now set as follows:

[0078] Spatial angle difference θ i (t):[5°,8°,7°,10°,6°,4°,5°,12°,6°,8°]

[0079] Initial drift ρ i (t):[0.5,0.3,0.6,0.4,0.5,0.2,0.3,0.6,0.4,0.5]mm

[0080] Rate of change of curvature κ i (t):[0.15,0.2,0.18,0.16,0.12,0.1,0.14,0.19,0.13,0.17]

[0081] Termination density gradient χ i (t):[0.4,0.45,0.42,0.38,0.4,0.43,0.41,0.48,0.39,0.4]

[0082] The nonlinear scoring functions are defined as follows (all are monotonically decreasing):

[0083] ψ(θ)=e -0.05θ

[0084] ξ(ρ)=e -1.2ρ

[0085] η(κ)=1-0.8

[0086] φ(χ)=e -1.5χ

[0087] Let the weighting coefficients be α = 0.3, β = 0.2, γ = 0.3, δ = 0.2, and the operational stability weighting function be ω(t) = 1 (i.e., equal weight at each time point), then:

[0088]

[0089] The calculation for the first time point is as follows:

[0090] ψ(5°)=e -0.25 ≈0.7788

[0091] ξ(0.5)=e -0.6 ≈0.5488

[0092] η(0.15)=1-0.8×0.15=0.88

[0093] φ(0.4)=e -0.6 ≈0.5488

[0094] Overall score for point 1:

[0095] C1=0.3·0.7788+0.2·0.5488+0.3·0.88+0.2·0.5488

[0096] =0.2336 + 0.1098 + 0.264 + 0.1098 = 0.7172

[0097] Similarly, the system averages the results over 10 time points to obtain the final comprehensive confidence score C for segment i. i ≈0.701.

[0098] The system's preset threshold is 0.65, therefore this path segment is considered a high-confidence path and included in the final 3D model. If there are three consecutive C segments... i If the score is less than 0.6, the system will activate the local reconstruction module to readjust the seed point starting point, low-potential direction barrier, or behavior fitting function until the path consistency score returns to a stable range.

[0099] The 3D modeling of the root canal system has entered the stage of path consistency verification and channel confidence visualization. This step is based on the consistency score between the surgeon's intraoperative operative path and the system's reconstructed path, visually representing the quality of the path structure and providing navigational prompts for the confidence of different channel segments in 3D space. The calculation of the angle difference is particularly crucial. The system first identifies directional jump points at locations where the channel curvature changes significantly. These jump points generally occur in physiological transition areas of the channel, such as the curved section at the middle third of the root canal or the bifurcation between the mesial and distal roots. Taking the right middle root channel of patient P045 as an example, the system identified the first significant jump point in the main channel at z = 3.2 mm, where the reconstructed path normal vector is... The direction vector of the doctor's operation trajectory is The angle difference θ can be calculated using the vector dot product formula:

[0100]

[0101] The system samples four such transition points, located near the starting, middle, and ending segments. The included angle for each segment is calculated and averaged to form a path directionality consistency score for that segment. Simultaneously, to further quantify the degree of path spatial overlap, the system constructs local spherical Regions of Interest (ROIs) with a radius of 1.0 mm at the starting (0–2 mm), middle (2–6 mm), and ending (6–10 mm) segments, and calculates the overlap ratio between the voxel representations of the doctor's path and the model path for each segment. Let the volume intersection of the starting segment be V. inter,start =6.2mm 3 The union of V union,start =8.4mm 3 Then the overlap ratio is:

[0102]

[0103] The overlap ratios of the other two segments are 0.702 and 0.768, respectively, and the overall average overlap ratio is R. avg =0.736. The system's preset overlap ratio threshold is 0.7, therefore the path fit is considered good.

[0104] Based on this, the system constructs a confidence heatmap for the entire channel path, using color gradients to represent segments from high confidence (red) to low confidence (blue), and maps the score values ​​to the vertex attributes of the 3D reconstructed model surface, achieving real-time feedback during preoperative navigation. In the right mid-root channel segment, the standard deviation of the angle dispersion is 1.2° at the beginning, 3.8° in the middle, and 2.1° at the end. Therefore, according to the inverse dispersion function, the converted score values ​​are 0.92, 0.76, and 0.88, respectively. The final fused score is:

[0105]

[0106] This score reflects the channel's high path smoothness and microstructure orientation compliance in three-dimensional space. Therefore, the system labels it as a high-quality path and displays a green channel line in the navigation interface for preoperative path selection and intraoperative navigation. If the angular dispersion score of a certain segment is lower than 0.65, the system will alert that there is a potential modeling anomaly, requiring manual review.

[0107] The system then enters a fine-tuning stage, determining the reliability of the channel's starting point and the closure boundary of its endpoint. This step is crucial for determining whether the reconstructed channel has a "false termination" or "hidden branch channel," and it is also used to verify the accuracy of the initial channel identification. During operation, the system first performs a spatial overlap analysis on the starting point region. Based on the segmented entrance structure from the pulp chamber CT image, the system calculates the center P of the reconstructed path's starting point and the anatomical entrance point. a =(x a ,y a ,z a The Euclidean distance between the reconstructed path and the starting point P. r =(x r ,y r ,z r = (2.2, 4.3, 5.1), and the CT segmentation entry point is P. a = (2.1, 4.2, 5.0), then the spatial coincidence error is:

[0108]

[0109] Simultaneously, in the image edge intensity map, the system calculated the edge sharpness of the region at this point to be 0.86 (normalized range 0–1) based on the Sobel operator. Combining the judgments from both dimensions, the system classified this case as "narrow medullary cavity type," with the starting point accuracy index primarily based on structural overlap, and weights set at w1 = 0.7 and w2 = 0.3. Therefore, the confidence level of the starting point S... start The calculation is as follows:

[0110] S start =w1·(1-ρ norm )+w2·Edge norm =0.7·(1-0.173)+0.3·0.86=0.7·0.827+0.258

[0111] ≈0.835

[0112] This value is higher than the system's recommended threshold of 0.75, indicating that the starting point has high reliability and meets the seed point setting conditions for subsequent path growth. The system then enters the structural analysis stage of the channel termination region, focusing on detecting abnormal grayscale transitions in the closure contrast gradient index to identify hidden branches or unclosed structures. Taking the right middle root termination segment image of patient P045 as an example, the system constructs a 1.5mm section of the terminal region... 3 A grayscale window is created, and the grayscale change vector G(x,y,z) is calculated along the end of the reconstruction path. The closed contrast gradient value is defined as:

[0113]

[0114] Where I iLet [112, 116, 119, 122, 124, 125, 125, 125, 126, 126] be the grayscale value of the continuous voxels in the path endpoint direction. In this example, sampling 10 consecutive points yields the grayscale sequence: [112, 116, 119, 122, 124, 125, 125, 125, 126, 126]. The mean grayscale difference is:

[0115]

[0116] The system presets a closed contrast gradient threshold of 2.5. If the value is below this threshold and the grayscale tends to be stable and flat, the termination region is considered to have high closure and a clear structure. If the value is above this threshold and fluctuates sharply, it indicates the presence of unidentified branch channels or sectional breaks. In this case, 1.56 represents a relatively smooth grayscale transition, and the system considers the termination boundary of the channel to be reliable, without triggering end extension or branch channel identification.

[0117] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional reconstruction method for root canal systems, characterized in that... include: Based on the pulp cavity growth path tracing mechanism, the unclosed growth boundary between the pulp cavity and the root canal system and the directional flow traces of the pulp cavity structure are identified, and the structural initiation region of the root canal system is extracted. Then, based on the extracted structural initiation region, according to the physiological and mechanical compliance principle, combined with dentin development tension and anatomical configuration, the directional barrier and impedance distribution of the root canal path in space are constructed to generate a three-dimensional channel trajectory that conforms to the growth trend. Spatial gradient analysis of micro-density clustered closed stress groups is performed on the root region involved along the three-dimensional channel trajectory to determine the expansion limit of the root canal path and identify the natural termination position of the channel. Based on the expansion limit and termination position, connection rules are extracted in multiple branch paths through the common structural pattern of closed-channel-closed, and branch structures with physiological rationality are screened to suppress false positive connections. The trajectory behavior of iatrogenic instruments entering the path during root canal treatment is back-projected into the three-dimensional reconstruction model. The path consistency of the generated model is verified. If there is a spatial deviation, structural reconstruction adjustment is triggered.

2. The three-dimensional reconstruction method for the root canal system according to claim 1, characterized in that... The identification of the directional flow traces includes calculating the spatial anisotropy index in the image based on the longitudinal fiber arrangement direction of the residual tissue in the medullary cavity to determine the extension direction of the initial channel; the extracted structural starting region is marked as the initial channel seed point in three-dimensional space and serves as a restrictive starting point in the path generation step.

3. The three-dimensional reconstruction method for the root canal system according to claim 1, characterized in that... In the process of constructing the directional barrier and impedance distribution, an impedance tensor field based on the dentin micro-density distribution is introduced to dynamically adjust the growth priority of the channel trajectory in complex curved regions. In the process of generating the three-dimensional channel trajectory, continuous low-potential channels in the impedance field are identified, and a dynamic envelope layer that restricts path deviation is constructed.

4. The three-dimensional reconstruction method for the root canal system according to claim 1, characterized in that... The spatial gradient analysis is based on a joint evaluation of the density change rate of the closed stress cluster, the consistency of the gradient direction, and the boundary ambiguity. In the process of identifying the multiple branch paths, the structural rationality score is calculated based on the local curvature of each branch, the deviation angle of the main channel, and the consistency of the growth direction, and only the branch structures with scores higher than a preset threshold are retained.

5. The three-dimensional reconstruction method for the root canal system according to claim 1, characterized in that... In the process of suppressing false positive connections, if a branch path crosses two or more directional barriers at the same time, it is marked as a non-physiological path and removed; the collection of iatrogenic instrument trajectory behavior includes the time sequence position information of the instruments during the operation, and the interpolation algorithm is used to map it into the three-dimensional reconstruction model to achieve path fitting and comparison.

6. The three-dimensional reconstruction method for the root canal system according to claim 1, characterized in that... During the trajectory behavior back-projection verification process, the model calculates the path consistency score based on the angle difference between the doctor's operation path and the reconstruction path and the overlap volume ratio; the triggering basis for the structural reconstruction adjustment includes multiple consecutive path consistency scores being lower than the threshold, or the response after the doctor manually marks the area with path deviation; each root canal channel in the generated three-dimensional reconstruction model is accompanied by a confidence label, which is obtained by jointly calculating the confidence of the starting point identification, the physiological rationality of the path generation, the clarity of the termination boundary, and the behavior trajectory consistency score.

7. The three-dimensional reconstruction method for the root canal system according to claim 6, characterized in that... The angle difference is calculated based on the directional jump point of the doctor's path and the reconstruction path at the change of channel curvature; the overlapping volume ratio is achieved by setting a segmented volume superposition strategy for local comparison areas at the beginning, middle and end of the channel respectively.

8. The three-dimensional reconstruction method for a root canal system according to claim 7, characterized in that... A confidence heatmap is generated for each root canal segment and mapped onto the surface of a three-dimensional reconstruction model to visualize the channel quality during preoperative virtual navigation. The score is based on the dispersion of the angle between the channel path and the direction of the dentin microstructure; a smooth angle change results in a high score.

9. The three-dimensional reconstruction method for the root canal system according to claim 6, characterized in that... A contrast gradient is introduced at the end of the channel. The value of the contrast gradient reflects the grayscale transition degree of the termination boundary and is used to determine whether there are unidentified hidden channels at the termination boundary.

10. The three-dimensional reconstruction method for a root canal system according to claim 9, characterized in that... The reliability of the starting point is determined by a combination of the degree of spatial overlap with the actual entrance to the pulp cavity and the clarity of the image edges, with the weights of the two adjusted according to the case type.

Images