An endoscopic three-dimensional axis distance decoupling En-DCR augmented reality preoperative quantitative planning method and system

Abstract

The application provides an endoscopic three-dimensional axis distance decoupling En-DCR augmented reality preoperative quantitative planning method and system, and is applied to the technical field of medical data processing. Based on the head thin layer CT of a target user, three-dimensional reconstruction of a lacrimal sac is completed, a double anatomical anchor point is calibrated to build a RAS reference coordinate system, a virtual endoscopic optical main shaft and an orthogonal imaging plane are constructed relying on the anchor point connecting line, and an exclusive optical projection mathematical model is established. The OAP algorithm is used for dimension reduction mapping of the vertex of the lacrimal sac, and a visual silhouette conforming to the endoscopic distortion is generated through convex hull fitting; then the profile extreme value is extracted to backtrack three-dimensional coordinates, diagonal line errors are eliminated along the anatomical axis, and standardized osteotomy parameters are obtained. The projection barycenter is used to automatically distinguish the left and right sides of the nasal cavity, adapt to the corresponding endoscopic bottom drawing, and adjust the scale, and finally, individualized AR planning views are generated by fusing the distorted visual layer and the accurate quantitative parameters, so that the endoscopic adaptive contour and the safe osteotomy index are synchronously output through the visual and data dual-track decoupling architecture, and preoperative accurate quantitative planning of En-DCR is realized.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to an En-DCR augmented reality preoperative quantitative planning method and system with endoscopic three-dimensional axis decoupling. Background Technology

[0002] Currently, preoperative planning and intraoperative guidance for En-DCR surgery mainly include three existing routine approaches, each with its own inherent technical limitations that cannot be addressed simultaneously: Clinically, fixed anatomical mean values ​​(e.g., 8mm upwards and forwards from the axilla of the middle turbinate) are commonly used to blindly estimate the osteotomy window, without considering individual variations in the height and anterior-posterior anatomy of the lacrimal sac. For low-lying lacrimal sacs, excessive upward osteotomy is prone to occur, damaging the orbital and skull base tissues; for high-lying dilatational lacrimal sacs, insufficient window opening and residual blind ends of the lacrimal sac significantly increase the risk of postoperative recurrence. The accuracy of the planning relies entirely on the physician's personal experience, resulting in extremely low standardization.

[0003] Current simplified AR methods directly project 3D CT models onto a 0° endoscopic view, without considering the anterior compression distortion caused by the natural oblique intersection of the endoscope and the nasal cavity sidewall. This distortion compresses the anatomical length within the 2D image. If the bone resection span is directly calculated based on the distorted projection pixels, the obtained value is far smaller than the actual anatomical size of the bone wall, easily leading to insufficient osteotomy area during surgery. If the distortion is eliminated to generate a non-fluoroscopic orthographic projection image, the image shape is completely misaligned with the intraoperative endoscopic field of view, rendering AR visualization guidance clinically useless, and visual reconstruction and accurate distance measurement cannot be achieved simultaneously.

[0004] Traditional 3D navigation uses the Euclidean 3D straight-line distance between two points to calculate osteotomy reference values, creating a "diagonal calculation trap." The 3D diagonal line simultaneously superimposes vertical and anteroposterior depth components, rather than a single anatomical axis in the coronal / horizontal direction of the mucosa and bone wall. The calculated values ​​are significantly inflated and cannot represent the true safe osteotomy boundary, directly misleading the judgment of the osteotomy range.

[0005] Meanwhile, all existing AR / navigation systems adopt a single-track architecture that uses visual simulation and parameter measurement to share a computing link, resulting in the superposition of perspective distortion error and three-dimensional diagonal error; virtual endoscope modeling simply connects two points to generate the light path, without matching the 0° endoscope oblique imaging logic, resulting in poor alignment accuracy of the projected contour; and it relies on manual differentiation of the left and right nasal cavities and manual switching of the endoscope base map and scale, which easily leads to side confusion and reverse labeling problems, resulting in low automation and cumbersome preoperative planning process.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0008] According to one aspect of this application, an En-DCR augmented reality preoperative quantitative planning method with endoscopic three-dimensional axis-distance decoupling is provided, comprising: acquiring simulated thin-slice CT images of the target user's skull, segmenting and reconstructing a three-dimensional model of the lacrimal sac and calibrating the anatomical anchor points of the middle turbinate axilla and the anterior nasal cavity, and constructing reference data of the RAS anatomical coordinate system; constructing a virtual endoscopic optical axis by connecting the anterior nasal cavity and the middle turbinate axilla, establishing a virtual imaging plane orthogonal to the optical axis, and building an endoscopic optical projection mathematical model; performing dimensionality reduction mapping of the three-dimensional vertices of the lacrimal sac based on the optical axis orthogonal projection algorithm of the endoscopic optical projection mathematical model, generating a two-dimensional anatomical silhouette that preserves endoscopic perspective distortion through convex hull fitting, and completing the rendering of the visual guidance layer; The extreme pixel values ​​of the projected contour are extracted and their corresponding 3D RAS coordinates are traced. The physical parameters for osteotomy safety without diagonal error are calculated by decoupling along the anatomical axis components, and a quantitative measurement data layer is constructed. The spatial centroid of the projection point set is calculated to determine the nasal cavity side, and the corresponding endoscopic baseline map is adaptively matched and the quantitative scale arrangement is adjusted by mirroring. The anatomical silhouette with perspective distortion and the anatomical axis distance parameters without distortion are fused to generate an individualized preoperative quantitative planning view for En-DCR surgery AR. Relying on multi-module algorithms such as 3D image segmentation, orthogonal projection imaging, anatomical axis distance axis component calculation, and automatic side recognition, the visual simulation and physical measurement calculation are separated, and the guiding contour and precise osteotomy quantitative indicators that fit the endoscopic vision are output simultaneously.

[0009] Another aspect of this application is an endoscopic three-dimensional axis-distance decoupling En-DCR augmented reality preoperative quantitative planning system, used to execute executable instructions to perform the above-described endoscopic three-dimensional axis-distance decoupling En-DCR augmented reality preoperative quantitative planning method.

[0010] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described En-DCR augmented reality preoperative quantitative planning method for endoscopic three-dimensional axisymmetric decoupling via executing the executable instructions.

[0011] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a second processor, implements the above-described En-DCR augmented reality preoperative quantitative planning method for endoscopic three-dimensional axisymmetric decoupling.

[0012] The beneficial technical effects of this application are as follows: First, a thin-slice CT scan of the patient's skull is acquired. A personalized three-dimensional model of the lacrimal sac is reconstructed using threshold segmentation and region growing algorithms. The anterior nasal cavity and the axilla of the middle turbinate are calibrated as dual anatomical anchor points, and a standardized RAS anatomical coordinate system is constructed. Using the line connecting the anterior nasal cavity to the axilla of the middle turbinate as the virtual endoscopic optical axis, an orthogonal virtual photosensitive plane is constructed. The three-dimensional vertices of the lacrimal sac are reduced in dimension using the OAP optical axis orthogonal projection algorithm. A two-dimensional visual silhouette, preserving the endoscopic anterior-contraction perspective distortion, is generated through convex hull fitting. The algorithm traces back to the three-dimensional coordinates corresponding to the extreme values ​​of the silhouette, performing calculations only along the single-axis component of the RAS, abandoning three-dimensional oblique line distance measurement, to obtain error-free osteotomy physical parameters A, B, and C. Then, the nasal cavity lateralization is automatically determined through the projection centroid, adaptively matching the corresponding endoscopic base map and mirroring the scale. Finally, the distortion guidance layer and the precise quantitative scale are fused to generate an AR planning view. The overall system adopts a decoupled dual-track architecture of visual simulation and physical measurement, with two independent parallel computational links, synchronously outputting an operational contour that conforms to the endoscopic field of view and personalized osteotomy safety scale.

[0013] This application aims to resolve the contradiction between the high fidelity of traditional AR visual measurements and precise measurement. The visual layer replicates the true deformation of the endoscope, facilitating intuitive positioning for doctors, while the quantification layer eliminates errors from both fluoroscopic and diagonal measurements, preventing misjudgments of the osteotomy range. It breaks away from the fixed 8mm cut standard, generating a personalized osteotomy limit based on the individual anatomy of the patient's lacrimal sac. This reduces damage to the orbital floor and skull base caused by excessive osteotomy, while also avoiding postoperative recurrence due to insufficient exposure, thus reducing surgical complications. It achieves automatic nasal cavity side recognition and adaptive matching of the base map and scale, with a high degree of automation throughout the entire process, simplifying preoperative planning and significantly improving clinical efficiency.

[0014] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0015] Figure 1 This document illustrates a flowchart of an En-DCR augmented reality preoperative quantitative planning method for endoscopic three-dimensional axis decoupling provided in an embodiment of this application. Figure 2 This image shows a schematic diagram of a coronal thin-section CT tomographic image of a patient's nasolacrimal duct region. Figure 3 A schematic diagram of a nasal cavity and sinus anatomy model reconstructed using a complete set of thin-section CT tomographic datasets of the skull is shown. Figure 4 A schematic diagram of surgical anchor point calibration for a three-dimensional reconstruction model of the nasal dacryocystorhinostomy cavity is shown. Figure 5 A schematic diagram of orthogonal projection of an endoscope optical axis to match the real scene is shown. Detailed Implementation

[0016] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0017] In one implementation, Figure 1 A schematic flowchart of an endoscopic three-dimensional axis-decoupled En-DCR augmented reality preoperative quantitative planning method according to an embodiment of this application is shown.

[0018] S101: Obtain simulated thin-slice CT images of the target user's skull, segment and reconstruct a three-dimensional model of the lacrimal sac, and calibrate the anatomical anchor points of the axillary region of the middle turbinate and the center of the anterior nasal cavity to construct the baseline data of the RAS anatomical coordinate system.

[0019] In one implementation, the acquisition layer outputs a sequence of continuous thin-slice CT images of the patient's skull with a slice thickness of 0.6 mm. The total number of image slices is no less than 300, completely covering the skull base, bilateral nasal cavities, lacrimal ducts, and infraorbital anatomical regions. The grayscale value range of all images is uniformly set to -1000 HU to 3000 HU. This batch of raw image data streams automatically flows into the image preprocessing stage. This stage consistently performs two core processing actions: irrelevant slice removal and multi-tissue thresholding + region growth composite grayscale segmentation. These two actions are executed sequentially, with the output of the previous step serving as the sole input for the next step. An example of the irrelevant slice removal execution rule is as follows: a longitudinal anatomical coordinate judgment threshold is set. Slices with a Z-axis coordinate greater than 20 mm above the upper limit of the AX anchor point and less than 15 mm below the lower limit of the anterior nasal cavity are directly judged as irrelevant slices, and the system deletes such images in batches.

[0020] Taking a patient with a unilateral low-lying dacryocystorhinos sac as an example, after removing 362 original CT slices, 217 effective tomographic images remained. All slices from the top of the skull and the bottom of the maxillary sinus without dacryocystorhinos anatomical information were removed, significantly reducing the computational load for subsequent reconstruction. Examples of the corresponding threshold values ​​for tissue grayscale segmentation are as follows: nasal cavity air grayscale range -900HU to -500HU, nasal cavity mucosa soft tissue grayscale range 20HU to 90HU, dacryocystic tissue grayscale range 10HU to 75HU, and nasal bone and maxillary bone grayscale range 180HU to 350HU. The system first coarsely divides different tissue pixel regions using fixed grayscale threshold ranges, then uses a 4-neighborhood region growth algorithm to expand the connected components of the dacryocystorhinos sac pixels. The algorithm's growth iterations are fixed at 4 times, filling only small pixel holes at the edge of the dacryocystorhinos sac without spreading outwards to surrounding nasal turbinate bone pixels. This accurately separates the target pixels of the dacryocystorhinos sac from irrelevant tissue pixels, ultimately outputting a subset of effective tomographic images containing only the anatomical structures of the dacryocystorhinos sac and middle turbinate. After preprocessing and filtering redundant anatomical information, the number of vertex generation per model in the subsequent 3D reconstruction stage can be reduced by 37%, while avoiding interference from extraneous tissues in the orbit and skull base that could affect the accuracy of the lacrimal sac contour restoration.

[0021] The effective tomographic image subset, after grayscale segmentation and slice selection, is automatically fed into the 3D model reconstruction stage. This stage uses a 3D volumetric interpolation model for computation. The model consists of three layers: a 2D tomographic contour extraction layer, a thickness direction interpolation layer, and a solid smoothing layer. Data is transmitted sequentially between these three layers without skipping between them. An example of the 2D tomographic contour extraction rules is as follows: Within a single effective tomographic image, all connected pixels conforming to the lacrimal sac grayscale range are traversed. An 8-neighborhood boundary tracing algorithm is used to extract the closed 2D contour line. The contour coordinates retain the original CT floating-point pixel precision without integer rounding.

[0022] The interpolation layer employs a linear 3D interpolation algorithm, with the interpolation step size perfectly matching the 0.6mm slice thickness of the CT slices. The lacrimal sac contour extracted from two adjacent slices is interpolated and filled point-by-point along the Z-axis, stitching together to generate a continuous closed 3D solid mesh. The solid smoothing layer has a fixed number of iterations of 6, with a smoothing weight coefficient set to 0.15, eliminating only the jagged mesh noise generated by slice stitching without altering the original coordinates of the overall lacrimal sac boundary. During the computation process, all 3D spatial vertices of the lacrimal sac are simultaneously acquired. The average number of vertices after reconstruction for a single patient is approximately 12,600. All vertices fully inherit the X / Y / Z 3D spatial floating-point coordinates output by the original CT equipment without coordinate simplification or offset compression. After the entire reconstruction process is completed, a 3D digital solid model is generated that fits the individualized anatomical shape of the lacrimal sac in terms of length and height, completely encompassing the bulging end and neck of the lacrimal sac, providing a complete 3D visualization carrier for subsequent dual anatomical anchor point spatial picking.

[0023] After the complete 3D reconstruction model of the nasal cavity and dacryocystorhinostomy sac is generated, the full-space 3D mesh data is automatically transferred to the anatomical anchor point calibration stage. The stage has two types of fixed anatomical positioning rules built in, corresponding to the AX middle turbinate axillary anchor point and the Nostril anchor point at the center of the anterior nasal cavity, respectively. The two types of points are picked up independently and their coordinates are stored separately, and the data of the left and right nasal cavities are completely isolated. An example of the AX anchor point picking criteria is as follows: the bony junction between the anterior end of the middle turbinate and the lateral nasal wall is located. The threshold of the angle between the normal vectors of the bony mesh surface at this position is 72°. The bony intersection point with an angle reaching this value is the AX reference point. Only one set of AX coordinates is marked for each nasal cavity. The coordinate format retains 4 decimal places of floating-point precision, such as the original AX spatial coordinates of the right nasal cavity (12.4752, 26.8137, 18.0925).

[0024] The following is an example of the criteria for determining the center point of the anterior nasal cavity: Extract all boundary vertices of the bony opening of one anterior nasal cavity, and calculate the geometric center using the centroid calculation formula: , , In the formula, n represents the total number of vertices of the anterior nasal nostril boundary. , , The original three-dimensional coordinates of a single boundary vertex.

[0025] Taking the left nasal cavity as an example, there are a total of 84 vertices on the boundary of the anterior nasal cavity. Substituting these into the formula, the coordinates of the center point are calculated as (-11.9263, 24.7011, 6.1478). Point verification rules: After picking, the system automatically compares the point with the anatomical position threshold. If the shortest three-dimensional straight line distance between the AX anchor point and the lacrimal sac is less than 3mm, or the bony boundary of the center point of the anterior nasal cavity exceeds the opening range, it is determined that the picking deviates from the anatomical standard, and a pop-up window prompts the operator to reselect the point. The two sets of anchor points for the left and right nasal cavities are stored in independent coordinate arrays. The data between the arrays do not call or overwrite each other, avoiding the problem of confusion between the left and right anatomical coordinates. After the point verification is passed, the double anchor points are bound to the complete 3D mesh data of the nasal cavity and transferred downwards.

[0026] A complete 3D mesh dataset of the entire nasal cavity, precisely labeled with dual anatomical anchor points, was input into the RAS anatomical coordinate system baseline construction phase. The coordinate system transformation model consists of two layers: an origin translation layer and a three-axis component mapping layer. The original 3D coordinates of the AX anchor points are used as the global coordinate transformation origin, and a unified coordinate transformation across the entire space is performed. The origin translation transformation calculation formula is as follows: ; ; In the formula, X, Y, and Z are the original CT coordinates of any three-dimensional entity. , , The original three-dimensional coordinates of the AX anchor point of the corresponding lateral nasal cavity are converted. The R axis represents the left and right orientation, the A axis represents the front and back orientation, and the S axis represents the up and down orientation. All conversion results retain 4 decimal places of floating-point value without precision truncation.

[0027] The conversion covers two types of core spatial data: the first is the set of all three-dimensional vertices of the lacrimal sac entity, and the second is the three-dimensional coordinates of the Nostril center point. Substituting the original coordinates (14.2615, 29.3362, 21.7548) and the right AX coordinates (12.4752, 26.8137, 18.0925) of the right lacrimal sac vertex into the formula, the RAS coordinates of that vertex can be calculated as (1.7863, 2.5225, 3.6623). After all coordinate conversions are completed, the system packages and generates a standardized RAS anatomical coordinate system benchmark dataset. This dataset contains three structured storage fields: the first field is the unilateral nasal cavity identifier (left / right), the second field is the RAS three-axis benchmark coordinates of the AX anchor point, the third field is the RAS three-axis coordinates of the anterior nasal center point, and the fourth field is an array of RAS coordinates for all lacrimal sac vertices. This dataset serves as a unified underlying 3D spatial benchmark and is simultaneously distributed to three subsequent computational modules: virtual endoscopy modeling, OAP orthogonal projection imaging, and anatomical axis distance and axis component calculation. The entire process of image preprocessing, 3D reconstruction, anchor point calibration, and RAS coordinate conversion runs sequentially, forming a complete 3D spatial initialization full processing chain.

[0028] S102: A virtual endoscopic optical axis is constructed by connecting the anterior nasal cavity, the middle turbinate, and the axillary region. A virtual imaging plane orthogonal to the optical axis is established, and an endoscopic optical projection mathematical model is built.

[0029] In one implementation, the Nostril point, AX anchor point, 3D direction vector, orthogonal photosensitive plane, vector cross product, key entity variables of the virtual camera coordinate system, semantic rules for optical path construction, and modeling technical parameters are classified and integrated according to the endoscopic camera modeling target, anatomical entity, and optical operation dimensions to generate basic unit information for virtual endoscope optical path construction. After completing the calibration of the dual anatomical anchor points and converting them into a standardized RAS coordinate system benchmark dataset, the two sets of independent RAS 3D coordinate data of the AX anchor point and the Nostril anchor point of the anterior nasal cavity stored in the dataset will be completely integrated into the optical path basic unit integration stage. This stage strictly follows the hierarchical collection execution rules, and performs two fixed processing actions, anatomical entity classification and optical operation element collection, simultaneously and sequentially. The two processing actions have a sequential execution order; anatomical entity collection must be completed first, followed by optical element collection, and the processing order cannot be reversed.

[0030] The modeling is uniformly divided and categorized according to three fixed dimensions: endoscopic camera modeling targets, anatomical entities, and optical operations. Each of the three dimensions has its own independent scope, preventing cross-dimensional mixing of elements. The endoscopic camera modeling dimension uniformly includes all semantic rules for optical path construction and technical parameters for endoscope adaptation modeling; the anatomical entity dimension only includes two types of anatomical point entities: AX anchor points and Nostril points; the optical operation dimension uniformly includes four types of core optical variables: three-dimensional direction vectors, orthogonal photosensitive planes, vector cross product operation logic, and virtual camera coordinate system transformation. The scope of the three dimensions is strictly distinguished, and all optical path-related entities, variables, parameters, and rules are included in the collection scope without omission.

[0031] During the modeling process, all core spatial variables were broken down with the unified modeling goal of matching clinical 0° rigid endoscopy observation and adapting to the En-DCR nasal cavity anatomy scene. The collected materials fully covered two types of anatomical positioning points, three-dimensional direction vectors of optical paths, orthogonal photosensitive imaging planes, vector cross product operation rules, virtual camera coordinate system transformation logic, and various independent operational elements. Each type of element was established with a corresponding data entry, and the element's dimension and basic definition were simultaneously labeled within the entry to ensure that each optical path material had a unique identifier. The integration process strictly followed the dimensional classification rules to uniformly classify and organize the scattered entities, parameters, and optical path semantic rules. All elements under the same dimension were merged and included into independent material subsets. Subsets of different dimensions were isolated from each other, and independent reading identifiers were set between subsets. After the collection was completed, standardized virtual optical path construction basic unit information was generated.

[0032] After being categorized and aggregated, the modeling materials are strictly partitioned and stored according to dimensions. This clearly distinguishes between two independent material pools: anatomical entity content and optical computation content. When retrieving optical path materials, only the corresponding dimensional subset is read, completely avoiding issues such as variable loss, confusion of computation rules, and omission of modeling parameters caused by cross-dimensional retrieval. This basic unit information has a fixed data structure and can be directly used in the next optical path workflow orchestration stage. It is used to arrange optical path content, clarify the filling position of dual anchor point coordinates, solidify the standard expression of principal axis normalization and cross product, and divide the execution order of orthogonal plane construction. This provides compliant and complete pre-construction basic materials for the construction of the entire set of endoscopic optical projection models, while fully matching the dual business requirements of OAP algorithm pre-imaging and real rigid endoscope optical path reconstruction.

[0033] Based on the effectiveness requirements of endoscopic optical simulation, the content layout of the basic optical path units is designed, clarifying the filling positions of dual anchor point coordinate variables, the core expressions of principal axis normalization and cross product, and the proportion of orthogonal plane construction steps, generating complete process information for building the endoscopic optical projection model. After obtaining the basic unit information for building the virtual endoscopic optical path through the classification and integration of endoscopic camera modeling targets, anatomical entities, and optical three-dimensional dimensions, all collected materials flow into the projection model process arrangement stage. This stage executes two processing actions in a fixed order: variable arrangement planning and core operation rule definition. These two operations have a fixed order; the full variable arrangement planning must be completed first, and then the standard expression of optical operations must be unified and solidified. The execution order cannot be changed.

[0034] The core constraint for variable arrangement planning is that anchor point coordinates, optical path vectors, and imaging planes must follow a reasonable order of computation, and the disordered logic of placing imaging plane computation before optical path principal axis computation is not allowed. A unified filling position specification is set for dual anchor point coordinates, and the complete computation chain is divided into three execution stages: the first stage only loads anatomical anchor point related data, the second stage performs optical path vector related computations, and the third stage completes the construction of the orthogonal photosensitive plane. The coordinates of the Nostril anterior nasal nostril and the AX middle nasal turbinate axillary anchor points are uniformly set as the two types of variables read first in the chain. The data for these two types of anatomical points are only entered in the first layer of the chain and are not scattered after the vector and plane related computation steps.

[0035] For the two core optical operations, principal axis normalization and vector cross product, a unified and fixed textual description paradigm is adopted. The semantic rules for constructing the optical path are used throughout the process, with only the sentence description structure standardized, without adding new operational expressions. Simultaneously, the construction of the orthogonal photosensitive plane is defined as a fixed execution phase within the entire optical path modeling process. Orthogonal plane-related operations are only initiated after all principal axis-related operations are completed, without premature intervention in plane solving operations. After completing the design of the double anchor point filling positions, the standardization of core operation descriptions, and the division of the orthogonal plane execution order, a complete workflow for building the endoscope optical projection model is generated. This standardized workflow layout locks in the fixed execution sequence of the entire virtual endoscope optical path modeling, avoiding the problem of reversed principal axis operations and imaging plane construction. This ensures that all pre-processing materials delivered to the downstream OAP orthogonal dimensionality reduction step are compliant in timing and that all pre-processing conditions are met, preventing projection distortion caused by missing optical path parameters or reversed operational logic.

[0036] In line with the requirements for adapting to 0° rigid endoscopes and matching the En-DCR nasal anatomy scenario, the complete process information will clearly define a fixed execution sequence as follows: First stage, complete the reading and loading of two sets of coordinates, Nostril point and AX point; Second stage, based on the two sets of anchor point data, complete the construction of the three-dimensional direction vector of the optical path and related processing of principal axis normalization; Third stage, based on the normalized principal axis, perform vector cross product operation, and complete the construction of orthogonal photosensitive plane based on the axial reference obtained by cross product.

[0037] Combining the pre-imaging requirements of the OAP algorithm and the optical path reconstruction requirements of real rigid endoscopes, a step-by-step operation rule of anchor point connection - principal axis vectorization - orthogonal plane construction is set to match the clinical 0° endoscope observation logic, ensuring that the optical path model is adapted to the En-DCR nasal cavity anatomy scene. This step sets two core processing actions that are connected sequentially: the first is the solidification of the step-by-step operation logic, and the second is the clinical scene adaptation verification. These two processing actions must be executed in sequence. The scene adaptation verification is carried out only after the logic solidification is completed. The processing order cannot be reversed. The step-by-step operation logic solidification stage is based on the previously arranged anchor point filling order, optical operation standard expression, and plane construction sequence, locking three non-interchangeable continuous operation steps. The three steps are, in order, anchor point connection to generate the optical path principal axis, principal axis normalization and vectorization, and orthogonal photosensitive plane construction. The three steps have a fixed pre- and post-processing dependency relationship. The operation of the next step can only start after the previous step has been completely executed.

[0038] The first step involves connecting anchor points to generate the optical path axis. The calculation data is taken from two sets of coordinates that have been calibrated and converted to the RAS coordinate system: the Nostril anterior nasal point and the AX middle nasal turbinate axillary point. The line connecting these two points serves as the basic spatial path for endoscopic observation. Taking a unilateral nasal cavity sample as an example, the RAS three-axis coordinates corresponding to the Nostril point and the RAS three-axis coordinates corresponding to the AX point are all uniformly stored. Only the two sets of coordinates from the same side are used for the connection calculation, and the left and right nasal cavity point data are not mixed to generate the optical path.

[0039] The second step is principal axis normalization and vectorization. This involves standardizing the original spatial vector based on the line connecting two points, normalizing only the original optical path vector while preserving its original spatial orientation and not altering the actual observation direction of the endoscope's approach. This entire vectorization operation serves only the subsequent solution of the planar boundary vectors and does not modify the original spatial coordinates of the anchor points. The third step is orthogonal photosensitive plane construction. Using the normalized principal axis of the optical path as a reference, two sets of horizontal and vertical boundary axes are obtained through vector cross product. Based on these two sets of boundary axes, a virtual imaging plane perpendicular to the principal axis of the optical path is constructed. After the plane is generated, it completely replicates the spatial morphology of the 0° rigid endoscope's front-end CMOS photosensitive surface.

[0040] In the clinical scenario adaptation and verification phase, the En-DCR nasal cavity anatomical observation requirements and the OAP orthogonal projection algorithm pre-input requirements were used as dual verification benchmarks to check the adaptability of the three optical path operation steps segment by segment. The first verification criterion was whether the optical path principal axis generation logic conformed to the natural nasal cavity approach angle; the second verification criterion was whether the spatial orientation of the orthogonal photosensitive plane matched the imaging characteristics of the intraoperative endoscope obliquely intersecting the nasal cavity sidewall; and the third verification criterion was whether the output material of the entire optical path operation could be directly input into the OAP orthogonal dimensionality reduction process for vertex mapping.

[0041] For the optical path process corresponding to the low lacrimal unilateral nasal cavity, an adaptation and verification is performed. If there is a significant deviation between the virtual photosensitive plane generated after the execution of the three steps and the angle between the virtual photosensitive plane and the nasal cavity sidewall and the clinical 0° rigid endoscopy observation angle, the process is returned to the projection model process arrangement stage to readjust the execution ratio of the orthogonal plane construction step in the overall process. After generating the complete modeling process information again, it is sent back to this stage for verification until there is no significant deviation between the optical path perspective and the actual endoscopic field of view.

[0042] After all verification items pass the review, the process outputs complete optical path step-by-step calculation rules adapted to the En-DCR surgical scenario. These rules fully constrain the execution order of the three optical path operations, the data sources allowed for each operation, and the types of intermediate materials produced by each operation. This ensures a one-to-one match between the 3D space constructed by the virtual endoscope and the patient's individualized nasal cavity anatomy, eliminating angular deviations between the virtual optical path perspective and the actual intraoperative endoscopic view. The entire set of solidified step-by-step calculation rules will serve as a fixed pre-input carrier for subsequent OAP orthogonal dimensionality reduction, providing compliant optical model support that conforms to the true optical characteristics of the endoscope for matrix dot product orthogonal mapping of all 3D vertices of the lacrimal sac. This ensures that the 3D-to-2D projection process can completely retain the necessary anterior compression perspective distortion required for surgery.

[0043] S103 uses an orthogonal projection algorithm based on the optical axis of the endoscopic optical projection mathematical model to reduce the dimension of the three-dimensional vertices of the lacrimal sac. After convex hull fitting, a two-dimensional anatomical silhouette that preserves the endoscopic perspective distortion is generated, and the visual guidance layer rendering is completed.

[0044] In one implementation, based on the mathematical model of endoscopic optical projection, orthogonal dimensionality reduction operation (OAP), preservation of endoscopic anterior shrinkage perspective distortion, and extraction of two-dimensional anatomical contours, the system classifies and integrates the three-dimensional vertex set of the lacrimal sac, the virtual orthogonal photosensitive plane, matrix dot product mapping, convex hull contour extraction, and key entities and operational rules for endoscopic anterior shrinkage distortion. This generates basic unit information for three-dimensional vertex orthogonal dimensionality reduction, contour fitting, distortion preservation, and visual silhouette output. After completing the output of the optical path flow information for anchor point connection, principal axis vectorization, and orthogonal plane construction, the complete three-dimensional vertex dataset of the lacrimal sac in the RAS coordinate system is transmitted to the basic unit collection stage of the visual layer. This stage is a pre-processing unit of the spatial projection and visual generation module. The upstream data only comes from the optical path model and unilateral nasal cavity three-dimensional vertex data output from the virtual endoscope modeling stage, and the anatomical data of the left and right nasal cavities are collected and processed separately. This stage consistently performs two types of processing actions: optical algorithm element classification and anatomical entity element classification. A unified classification standard is set based on the mathematical model of endoscopic optical projection, and two independent storage partitions are divided. The materials in the two partitions are not mixed.

[0045] Before performing visual layer generation operations, four basic prerequisites must be met simultaneously. These four conditions are all prerequisites for the aforementioned operations: relying on the endoscopic optical projection mathematical model, performing OAP orthogonal dimensionality reduction operations, completely preserving endoscopic anterior shrinkage perspective distortion, and extracting two-dimensional anatomical contours. To match these four prerequisites, the system is divided into five core entities and corresponding operational rules. The entities are the three-dimensional vertex set of the lacrimal sac, the virtual orthogonal photosensitive plane, matrix dot product mapping, convex hull contour extraction, and endoscopic anterior shrinkage distortion constraints.

[0046] The system categorizes five entities according to two predetermined dimensions: optical modeling and graphical dimensionality reduction. The optical modeling dimension stores the virtual orthogonal photosensitive plane and endoscope anterior shrinkage distortion constraints. The graphical dimensionality reduction dimension stores the 3D vertex set of the lacrimal sac, matrix dot product mapping, and convex hull contour extraction logic. The collection phase uniformly performs a material normalization operation, which consists of two basic operations: removing duplicate parameters and supplementing entity computational relationships. After normalization, standardized basic unit information is generated, which carries four types of computational materials: 3D vertex orthogonal dimensionality reduction, contour fitting, distortion preservation, and visual silhouette.

[0047] The collected materials are stored in partitions according to two main dimensions. During computation, only the content of the corresponding partition is retrieved to avoid computational conflicts caused by the mixing of multiple types of parameters. Taking the RAS coordinates (1.7863, 2.5225, 3.6623) of a unilateral nasal cavity and lacrimal sac as an example, this vertex entity is classified into the graphics dimensionality reduction partition, while the corresponding photosensitive plane and distortion constraints are classified into the optical modeling partition. Unilateral computation only calls the corresponding materials on the same side. The basic unit information output after integrating all entities and classification rules is internally divided into four material packages corresponding to four computations, which are matched with three-dimensional vertex orthogonal dimensionality reduction mapping, two-dimensional contour convex hull fitting, physiological perspective distortion preservation, and two-dimensional erythema silhouette rendering. The unit data is unidirectionally transmitted to the downstream visual guidance layer layered planning stage.

[0048] Based on the overall architecture requirement of vision-data dual-track decoupling, the visual guidance layer generation task is designed with layered decoupling. This involves dividing the work into modules for 3D vertex orthogonal dimensionality reduction mapping, 2D contour convex hull fitting, complete preservation of physiological perspective distortion, and 2D erythema silhouette rendering output. The data flow relationship between CT 3D vertex input, matrix operations, contour extraction, and AR visual layer output is clearly defined, forming task execution information for the coordinated operation of projection mapping, contour fitting, distortion preservation, and layer rendering. After completing the classification and collection of materials such as 3D vertices, virtual photosensitive planes, and distortion constraints to form standardized basic unit information, all unit data flows into the layered planning stage of the visual guidance layer. This stage relies on the overall vision-data dual-track decoupling architecture as the top-level design constraint, and corely executes two types of standardized processing actions: complete imaging operation chain decomposition and full-process data flow relationship definition. These two actions are carried out simultaneously. First, independent operation units are decomposed, then the data input and output objects of each operation segment are uniformly labeled, ultimately generating continuously executable layered task execution information.

[0049] The entire visual guidance layer generation process only handles the generation of 2D images for endoscopic matching, without participating in the calculation of physical osteotomy parameters. It is computationally isolated from the subsequent anatomical axis calculation module, aligning with the overall architectural design concept of separating visual simulation and physical measurement. The system decomposes the complete OAP imaging process into four independent computational units with a fixed and non-replaceable sequence. These four units are, in order, 3D vertex orthogonal dimensionality reduction mapping, 2D contour convex hull fitting, complete preservation of physiological perspective distortion, and 2D erythema silhouette rendering output. The four units have a fixed upstream and downstream data transmission relationship, with the output of the previous unit directly serving as the sole input material for the next unit, and there is no direct cross-segment data retrieval.

[0050] For each computational unit, a fixed data flow path is defined. An example of single-sided nasal cavity sample flow is as follows: For the 3D vertex orthogonal dimensionality reduction mapping unit: the input material is the complete set of RAS vertices of the lacrimal sac in the single nasal cavity after 3D reconstruction, using single vertex coordinates (1.7863, 2.5225, 3.6623) as example input data. The computational carrier is the matrix dot product mapping rule corresponding to the virtual orthogonal photosensitive plane. The unit outputs a complete set of 2D projection points, which serves as the dedicated input for the next convex hull fitting. For the 2D contour convex hull fitting unit: the input material is the complete set of 2D projection points output in the previous step. The computation uses convex hull extraction logic, traversing all projection pixels to extract the closed outer boundary. The unit outputs a continuous, unbroken 2D closed contour of the lacrimal sac.

[0051] For the unit that fully preserves physiological perspective distortion: The input material is a 2D closed contour obtained by convex hull fitting. The entire calculation uses the perspective distortion characteristics generated by the 0° rigid endoscope obliquely crossing the nasal cavity, without contour stretching or compression correction. The unit output is a complete anatomical contour carrying the original perspective distortion. For the unit that renders 2D erythema silhouette: The input material is an anatomical contour carrying physiological distortion. The calculation completes the contour filling of the standardized erythema layer. The unit output can be directly superimposed on the AR visual layer of the endoscopic base image.

[0052] The four computational units are interconnected and execute synchronously. All data flow links are uniformly and standardizedly defined. The set of lacrimal sac vertices produced by the original CT 3D reconstruction is the initial input of the entire imaging process. Matrix dot product mapping serves as the intermediate data conversion carrier from 3D to 2D. The convex hull contour is the pre-processing material for the distortion retention stage. The final rendered erythema silhouette layer serves as the visual guide layer and is then transported to the next virtual-real fusion process. After the layered planning stage completes the decomposition and flow calibration of all links, a complete set of task execution information is generated. This information fully records the division method of the four computational units, the corresponding input and output material types for each segment, and the data transfer order between each unit. This enables the coordinated and orderly operation of four operations: 3D vertex orthogonal dimensionality reduction, 2D contour convex hull fitting, physiological perspective distortion retention, and erythema silhouette rendering. The entire layered design fully matches the OAP orthogonal projection algorithm's operating logic and forms a parallel, non-interfering computational system with the anatomical axis distance calculation module.

[0053] Combining the requirements for visual matching accuracy of En-DCR endoscope, intuitive AR guidance, and 3D-to-2D mapping accuracy, based on the oblique imaging characteristics of 0° rigid endoscope, the following criteria are established: 3D vertex matrix dot product dimensionality reduction standard, convex hull edge extraction rules, and constraints for complete retention of anterior projection distortion. This determines the orthogonal projection operation method, contour generation judgment conditions, and the final output format information of the visual guidance layer. After generating complete layered task execution information in the visual guidance layer layered planning stage, the entire set of layered operation sequence and data flow related data are sent to the imaging rule solidification stage. This stage consistently executes two types of serial processing actions: operation standard setting and output format locking. These two actions have a fixed execution order; the bottom-level operation execution standard of the entire chain must be unified first, and then the final output presentation format of the visual layer must be locked. The execution order of the two types of actions cannot be reversed.

[0054] The entire imaging rule setting process was based on three key operational requirements specific to clinical surgical scenarios. These requirements were: achieving satisfactory visual matching with the En-DCR endoscope, providing intuitive AR intraoperative guidance, and ensuring accurate mapping and restoration of 3D anatomical structures to 2D images. All three requirements served as the basis for formulating all imaging rules, and no standard could be omitted or relaxed. All imaging rules were set to perfectly match the actual imaging characteristics of a 0° rigid endoscope obliquely adhering to the nasal cavity wall during surgery, fully matching the objective imaging state where the endoscope and nasal cavity naturally form an oblique optical path during the procedure.

[0055] The process sequentially completes the unified definition of three types of core imaging standards: the first type is the dimensionality reduction standard of three-dimensional vertex matrix dot product, the second type is the two-dimensional convex hull edge extraction rule, and the third type is the hard constraint of complete retention of foreshortening perspective distortion. All three types of standards correspond to the entire OAP orthogonal projection calculation chain, and simultaneously determine the three types of supporting output information: the overall calculation execution path of orthogonal projection, the effective anatomical contour determination conditions, and the final standardized display format of AR visual layers.

[0056] The standard constraint ensures that all 3D vertices of the lacrimal sac in the RAS coordinate system participate in matrix mapping operations. Batch dimensionality reduction is performed independently only on one side of the nasal cavity, without mixing in the vertex dataset of the other side's nasal cavity. Taking the RAS coordinates (1.7863, 2.5225, 3.6623) of a single lacrimal sac vertex on one side of the nasal cavity as an example, this vertex completes a one-to-one matrix matching mapping with the virtual orthogonal photosensitive plane on the same side. All vertices synchronously perform dot product operations using the same set of mapping matrices adapted to the 0° endoscope optical path. Mapping parameters are not adjusted individually for each vertex, ensuring a uniform scaling ratio for the entire nasal cavity projection and preventing local anatomical contour stretching distortion. The entire set of matrix dot product operations can only be initiated after the virtual endoscope is built and the orthogonal photosensitive plane is generated. This serves as the sole execution path for 3D to 2D conversion, and other projection conversion methods are not allowed.

[0057] The rules stipulate that convex hull operations are performed only on the complete 2D discrete point set generated after all vertices have been projected. Scattered noise pixels within the point set are filtered out, and only the outermost continuous closed boundary that encloses all projected points is retained as the anatomical contour. The contour is considered valid if the closed boundary of the convex hull completely covers all projected points of the lacrimal sac. If the point set is missing or the boundary has breaks, the contour is deemed invalid, and the mapping operation must be re-executed in the 3D vertex dimensionality reduction step to avoid generating incomplete anatomical silhouettes that may affect the intraoperative visual guidance effect.

[0058] The constraints are non-adjustable hard rules. The entire dimensionality reduction and contour extraction process does not perform perspective distortion correction, fully preserving the natural anterior compression features generated by the 0° endoscopic oblique crossing of the nasal cavity, without stretching or compressing the two-dimensional convex hull contour. This constraint corresponds one-to-one with the clinical endoscopic imaging effect, ensuring that the anatomical morphology of the generated erythema silhouette on the screen completely overlaps with the visual morphology of the intraoperative endoscopic image, eliminating the problem of morphological deviation between the visual guidance layer and the actual surgical image.

[0059] After unifying the three core imaging standards and clarifying the corresponding execution paths and judgment conditions, the final output format information of the visual guidance layer was locked, and the basic style of the AR visual layer was standardized: the layer uses a closed contour of a convex hull filled with erythropomorphic spots as the anatomical marker carrier, and the layer size, viewing angle and imaging range of the virtual orthogonal photosensitive plane are completely aligned. The output layer can be directly sent to the downstream lateral adaptive matching stage without secondary cropping or deformation adjustment. After integrating and outputting all standardized imaging rule information, the execution logic of the entire process of 3D vertex dimensionality reduction, contour extraction, distortion preservation and layer rendering was fully standardized, providing a unified and compliant 2D visual material input for subsequent nasal cavity lateral determination and endoscopic base map matching processes, ensuring the uniformity of the visual generation branch operation standards and the standardization of output materials in the entire visual-data dual-track operation chain.

[0060] S104 extracts extreme pixels of the projected contour and traces the corresponding three-dimensional RAS coordinates. It decouples and calculates the osteotomy safety physical parameters without diagonal error along the anatomical axis components, and constructs a quantitative measurement data layer.

[0061] In one implementation, various anatomical measurement data, including extreme pixel values ​​of 2D projected contours, 3D RAS coordinates, anatomical single-axis components, and osteotomy limit indicators, are standardized and organized. This is matched with RAS axis distance decoupling calculation specifications to construct a quantization layer preprocessing workflow for extreme value tracing and axis difference calculation. After the 2D erythema silhouette layer and all lacrimal sac RAS vertex datasets retained from the 3D reconstruction stage are rendered and output by the visual guidance layer, the entire set of image projection-related measurement materials automatically enters the measurement data standardization stage. This unit belongs to the system's absolute physical axis distance calculation module and is a preprocessing unit for the quantization measurement data layer. It consistently receives the 2D anatomical silhouette pixel information output by the upstream visual guidance layer. The standardized materials after standardization are unidirectionally transported downwards to the multi-dimensional coordinate analysis framework for in-depth calculation, with no reverse data backflow throughout the entire process. This section sets two types of non-replaceable core processing actions. The first type is the collection of multiple types of anatomical measurement data, and the second type is the adaptation of RAS axis distance decoupling calculation specifications. The two types of actions are executed in sequence. First, all measurement materials are collected in a unified manner, and then the data is sorted and organized according to the established calculation standards.

[0062] The system collects calculation data comprising four types of fixed objects, which serve as the foundational data source for the entire set of osteotomy parameter calculations. These objects include: extreme pixel points of the two-dimensional projected contour, three-dimensional RAS coordinates of all target points corresponding to the contour, independent anatomical axis component values ​​obtained from decomposition, and clinical osteotomy limit indicators of categories A, B, and C. Data collection follows a unilateral nasal cavity isolation rule; silhouette pixels, lacrimal sac vertices, axis component values, and osteotomy indicator values ​​corresponding to one nasal cavity are packaged and stored separately, preventing the mixing of anatomical data from the contralateral nasal cavity for combined calculations. The following is an example of unilateral nasal cavity data collection. The projection contour recognition of the right lacrimal sac yielded three sets of points: top extreme pixels, anterior extreme pixels, and bottom extreme pixels. Each set of pixels is associated with a set of original RAS 3D coordinates. The coordinates corresponding to the top extreme values ​​are (1.7863, 2.5225, 3.6623), the coordinates corresponding to the anterior extreme values ​​are (1.2241, 4.1536, 2.9705), and the coordinates corresponding to the bottom extreme values ​​are (1.5028, 2.1104, 0.5267). The three sets of coordinates can be decomposed into independent components of the R, A, and S axes, which will be used to solve the osteotomy limit indices of A, B, and C.

[0063] All collected anatomical measurement data were uniformly matched with the RAS axis-distance decoupling calculation specification for standardization and organization. The standardization operation revolved around two core objectives: pixel coordinate tracing and single-axis difference calculation, building a complete pre-processing workflow. Each execution logic of the workflow is linked with the 3D space initialization, virtual endoscope modeling, and OAP projection stages. Pixel tracing execution rules: Each 2D contour extreme pixel is bound to a unique original 3D RAS coordinate index. The system uses the index to complete the reverse correspondence between 2D pixels and 3D anatomical points, preventing pixel-coordinate mismatches. Taking the top extreme pixel as an example, after reading the pixel's built-in index, the matching RAS coordinates of the highest point of the lacrimal sac can be directly retrieved as the original data for calculating parameter A. Axis difference calculation pre-constraints: After completing coordinate tracing, only the components corresponding to a single anatomical axis are extracted for difference calculation, abandoning the 3D space oblique line distance calculation mode. Only the S-axis component is used to solve for the upper and lower osteotomy heights, and only the A-axis component is used to solve for the anterior and posterior incision widths. This corresponds to three sets of predetermined difference calculation expressions built into the system: Parameter A: Parameter B: Parameter C: In the formula, C is the total height of the lacrimal sac; The S-axis component represents the highest point of the lacrimal sac. The A-axis component represents the anterior point of the lacrimal sac. The S-axis component represents the lowest point of the lacrimal sac. , These are the S-axis and A-axis components of the AX anchor point, respectively; The value of the S-axis (superior axis) component of the lowest point of the lacrimal sac in the RAS anatomical coordinate system.

[0064] Using the right nasal cavity values ​​as an example, the RAS coordinates of the AX anchor point on this side are (0,0,0), and the S-axis component of the highest point of the lacrimal sac is 3.6623. Substituting these values ​​into the expression yields the basic calculated value of parameter A for this case. Throughout the process, only the values ​​of a single axis are extracted for absolute difference calculations, avoiding the measurement error caused by the three-dimensional diagonal.

[0065] The entire extreme value tracing and axis difference calculation preprocessing chain has a fixed execution order. First, all contour extreme value pixels are bound to the 3D RAS coordinate index. Then, the coordinates of each group are split into three-axis components. Finally, the groups are classified according to the axis components corresponding to the osteotomy index. The completed set of calculation materials has a pixel-coordinate-axis component-osteotomy index correlation mapping relationship, which is directly supplied to the subsequent multi-dimensional coordinate analysis framework. This ensures that the subsequent calculation process does not require secondary processing of the original projection and 3D anatomical data, and uniformly eliminates the impact of perspective distortion and diagonal error on the osteotomy parameter calculation.

[0066] Based on a multi-dimensional analysis framework that utilizes projected pixel indexing, 3D spatial coordinates, and independent anatomical axis calculations, and integrates 2D silhouette point data with original CT 3D model data, a linked verification mechanism for pixel tracing, axis component decomposition, and diagonal error shielding is established. After the standardized anatomical measurement material package is processed in the measurement data normalization stage, the entire set of integrated data, including extreme pixel values, 3D coordinates, axis components, and osteotomy indicators, automatically flows into the multi-dimensional coordinate analysis stage. This stage is the intermediate processing unit of the absolute physical axis distance calculation module. The upstream and downstream data flows have unidirectional transmission characteristics, receiving only the complete set of anatomical measurement materials output by the preceding measurement normalization unit as the sole input source. After completing multi-dimensional fusion verification, coordinate binding and association intermediate data are generated and directed downwards to the parameter binding stage for osteotomy indicator conversion. The entire process does not involve any data back-to-back interaction.

[0067] This stage sequentially executes two core processing actions: building a multi-dimensional computational framework and fusing multi-source anatomical data. These two actions have an irreversible execution order; the framework for the three parallel analysis dimensions must be built first, followed by the fusion of 2D silhouettes and 3D models. The system constructs three independent, parallel analysis dimensions, each corresponding to a different type of original anatomical data source. Each dimension carries its own dedicated data reading and temporary storage area, with no data overlap between the three layers: The first layer is the silhouette pixel index dimension, specifically storing all extreme pixel positions and unique pixel index numbers of the 2D projection contour output by OAP computation; the second layer is the original 3D spatial coordinate dimension, used to store the RAS three-axis original coordinates corresponding to each extreme pixel; the third layer is the original CT reconstructed 3D model complete anatomical data dimension, completely preserving the complete set of vertices of the lacrimal sac on one side of the nasal cavity, the AX anchor point, and the Nostril anchor point reference coordinates.

[0068] The multi-source anatomical data fusion operation uses three parallel analysis dimensions as its carrier. It simultaneously loads two independent data sources: two-dimensional projection silhouette point information and original complete three-dimensional anatomical mesh data from CT. The two types of data establish a one-to-one mapping relationship based on pixel indexes, forming an integrated linkage verification rule for pixel positioning target points, coordinate backtracking to the original space, separate decomposition of axial components, and shielding of oblique line ranging errors. The entire set of linkage verification rules runs through the three analysis dimensions simultaneously. If any dimension of data is missing, the fusion operation will be terminated.

[0069] The following is an example of the step-by-step execution logic of the linkage verification rule. For the pixel positioning target point step, the extreme pixel index of the top of the lacrimal sac in a single location within the silhouette pixel index dimension is read, and the extreme coordinates of the top of the nasal cavity on one side (1.7863, 2.5225) corresponding to the example pixel are matched. Only the three-dimensional coordinates of the same group are retrieved through the index, so there will be no mismatch between pixels and coordinates.

[0070] For the coordinate backtracking to the original space step, the system jumps to the original 3D spatial coordinate dimension via an index, extracting the complete original values ​​of the R, A, and S axes of the target point without calling the coordinates of other irrelevant extreme points. For the step of separately decomposing axis components, the S-axis and A-axis components are separated from the complete 3D coordinates, retaining only the values ​​of these single axes for subsequent difference calculations, completely avoiding the artificial height error caused by oblique straight-line distance measurement in 3D space. For the step of shielding oblique distance measurement error, the system does not call the Euclidean distance calculation logic between two points in 3D space throughout the entire process, relying only on the separated independent anatomical axis components to participate in the subsequent calculation of the three osteotomy parameters A, B, and C, eliminating diagonal measurement deviations from the data source.

[0071] The two-dimensional silhouette of the right nasal cavity identified three sets of extreme pixels: top, front, and bottom. Each set of pixels corresponds to an independent index number. During the fusion process, the three-layer analysis dimensions simultaneously load all pixel indices on this side, match the three-dimensional RAS coordinates, and complete the CT reconstruction model of the dacryocystorhinostomy sac. Each set of extreme pixels performs a complete set of linkage verification processes separately. After all points are verified, the pixel index, corresponding three-axis coordinates, and split single-axis component association dataset are packaged together and used as the verified coordinate association data to flow down to the parameter binding stage.

[0072] To avoid 3D diagonal errors and obtain accurate bone resection safety scales, this approach integrates contour extreme points, RAS vertical / anterior / posterior axis coordinates, and three types of osteotomy parameters (A, B, and C) to construct a binding relationship between coordinate backtracking and single-axis difference calculation. After completing the integrated verification process of pixel positioning, coordinate backtracking, axis component splitting, and oblique line distance measurement shielding, an intermediate coordinate association dataset is generated. This dataset is then sent to the parameter association binding stage. This stage is the core transitional calculation unit within the anatomical axis distance calculation module. Upstream, it only receives the verified coordinate association data output from the multi-dimensional coordinate analysis stage. The processed binding parameter set is unidirectionally sent to the quantitative calculation data layer generation process, with no reverse data backflow operation throughout. The stage executes two processing actions in a fixed order: target anchoring and osteotomy parameter association binding. These two actions have a fixed execution order; the overall calculation target must be locked first, and then the binding and matching of points and difference calculation logic must be completed. The processing order cannot be reversed.

[0073] All data matching work in the entire process revolves around the core computational goal of avoiding three-dimensional diagonal errors and extracting accurate bone resection safety scales. All data integration and logical binding operations serve this goal, eliminating the numerical inaccuracies caused by traditional three-dimensional linear distance measurement from the computational level. The system uniformly collects four types of basic anatomical data for association and binding: the contour extreme points obtained from two-dimensional projection recognition, the upper and lower components of the S-axis of the target point in the RAS coordinate system, the anterior and posterior components of the A-axis of the target point in the RAS coordinate system, and the three clinical osteotomy limit indicators A, B, and C. Fixed binding mappings are established for each of the four types of data to prevent data cross-matching.

[0074] The system establishes a one-to-one binding relationship between the coordinate backtracking operation of each contour extreme point and the single-axis difference calculation. Each extreme point is matched with only one set of dedicated axis difference calculation logic, ensuring that the data source for each osteotomy parameter is unique and the measurement dimension is pure. This avoids the chaotic situation of a single point reusing multiple sets of axis components or multiple target points sharing the same difference formula. The three types of osteotomy indicators correspond to dedicated matching rules and standard calculation formulas: Parameter A (upward osteotomy limit) binding rule: Only match the S-axis component of the extreme point at the top of the lacrimal sac and the AX anchor point. Calculation formula: .

[0075] Parameter B (forward cutting limit) binding rule: Only match the A-axis component of the anterior extreme point of the lacrimal sac and the AX anchor point. Calculation formula: .

[0076] Parameter C (total lacrimal sac height) binding rule: Only match the two sets of S-axis components: the extreme points at the top and bottom of the lacrimal sac. The formula is: .

[0077] The following is an example of unilateral nasal cavity binding values: the RAS coordinates of the AX anchor point in the right nasal cavity are (0,0,0); the RAS coordinates of the extreme point at the top of the lacrimal sac are (1.7863,2.5225,3.6623), and this point is separately bound to the S-axis difference calculation logic corresponding to parameters A and C; the RAS coordinates of the extreme point at the front of the lacrimal sac are (1.2241,4.1536,2.9705), and this point is separately bound to the A-axis difference calculation logic corresponding to parameter B; the RAS coordinates of the extreme point at the bottom of the lacrimal sac are (1.5028,2.1104,0.5267), which is only paired with the top point to participate in the calculation of parameter C. The calculation logic bound to each point is not mixed.

[0078] The binding execution rules between points and computational logic are standardized. A single contour extreme pixel indexes only one set of 3D RAS coordinates; a set of 3D coordinates is split into only one set of independent A and S-axis components; and each set of axis components corresponds to only one set of osteotomy parameter difference calculations. The entire process is independent of 3D spatial diagonal distance calculations, completely cutting off interference paths from perspective pixels and 3D diagonals to the physical scale. After all extreme points are bound one-to-one, a complete set of A, B, and C osteotomy measurement data with binding identifiers is integrated and output. The data clearly records the original contour point, corresponding anatomical axis component, and corresponding standard deviation calculation formula for each parameter. This data can be directly used to build a standardized quantitative measurement data layer without re-matching points and computational logic.

[0079] By employing triple verification—projection extreme pixel recognition, 3D coordinate reverse tracing, and independent calculation of anatomical axis components—the spatial oblique line distance measurement method is abandoned, generating precise osteotomy physical parameters and completing the construction of a quantitative measurement data layer. The associated datasets, bound together by coordinate backtracking and single-axis difference calculations, flow to the quantization layer to generate the final processing stage. This stage belongs to the absolute physical axis distance calculation module and is the final output unit of the entire quantization measurement chain. The only upstream input is the coordinate-parameter mapping associated data output from the parameter association binding stage. After standardized parameter packaging, it is unidirectionally supplied to the AR virtual-real fusion module for reading and calling, with no reverse data return logic. The stage consistently executes two sequential processing actions: multi-verification review and standardized parameter output. The unified parameter encapsulation and output must be performed only after the complete triple verification process is finished; the execution order of these two actions cannot be changed.

[0080] This step incorporates a three-layer serial independent verification process. The three layers of verification are executed sequentially and none can be omitted. The entire verification mechanism completely avoids the numerical inflation defects caused by traditional three-dimensional Euclidean oblique line ranging. The three layers of verification have clear division of labor: the first layer completes the identification of projection extreme value pixels, the second layer completes the reverse tracing of the pixel's corresponding three-dimensional coordinates, and the third layer completes the independent difference calculation of the dissected single-axis components. The three layers of verification form a closed loop constraint. Only when the verification result of the previous layer is compliant can the next layer of calculation steps be entered.

[0081] Layer 1: Projection Extreme Value Pixel Recognition and Verification. The system iterates through the complete 2D erythema silhouette generated by the preceding convex hull fitting, automatically filtering out three types of boundary extreme value pixels: the top, front, and bottom of the lacrimal sac. Only valid contour extreme values ​​that completely enclose the anatomical region of the lacrimal sac are included in subsequent calculations, while scattered noise pixels at the contour edges are filtered out. Taking the silhouette of the right nasal cavity on one side as an example, the program selects three sets of valid extreme value points, corresponding to the top, front, and bottom anatomical positions, respectively. Each set of points generates an independent pixel index number. If any type of extreme value point is missing, the current silhouette material is determined to be non-compliant, and the data flow reverts to the visually guided layered planning stage to regenerate the 2D projection contour.

[0082] The second layer: 3D coordinate reverse tracing verification. Based on the preset index mapping relationship between pixels and RAS coordinates, reverse matching is performed on each set of extreme pixels obtained from the first layer. The complete R-axis, A-axis, and S-axis coordinates of the pixel in the original 3D space of the CT reconstruction are retrieved to ensure that each 2D extreme pixel is uniquely bound to a set of 3D anatomical coordinates, and there are no mismatches such as one-to-many or many-to-one. Using the sample values ​​of the right nasal cavity, the index of the top extreme pixel is matched with the original RAS coordinates (1.7863, 2.5225, 3.6623), the front extreme pixel is matched with (1.2241, 4.1536, 2.9705), and the bottom extreme pixel is matched with (1.5028, 2.1104, 0.5267). If there are pixels without matching 3D coordinates or multiple pixels sharing the same set of 3D coordinates, the verification is marked as abnormal and the process is terminated.

[0083] The third layer: Independent calculation and verification of anatomical axis components. After completing coordinate tracing, the system separates the A-axis and S-axis components of each group of three-dimensional points. Throughout the process, it does not call the logic for calculating the three-dimensional oblique distance between any two points, but only uses the absolute difference of the axis components to solve for the three types of osteotomy parameters. The standard formula is as follows: Parameter A = |Ptop[S] AXpos[S]|;Parameter B = |Pfront[A] AXpos[A]|;Parameter C = |Ptop[S] Pbottom[S]|. For example, the reference RAS coordinates of the AX anchor point on this side are (0,0,0), and Ptop[S] takes the value of 3.6623. Substituting it into the calculation formula of parameter A, only the single-axis value is extracted for difference, completely eliminating the measurement distortion caused by depth deviation; after all three layers of verification are passed, it is confirmed that there is no diagonal error in the three sets of osteotomy parameters, no misalignment in pixel-coordinate mapping, and no missing extreme points.

[0084] After all triple verifications pass, the system proceeds to standardized parameter output processing. It automatically collects five types of information: three categories of extreme pixel indices, matching complete 3D RAS coordinates, split single-axis components, three sets of osteotomy limit parameters (A, B, and C), and unilateral nasal cavity markers. These are then uniformly assembled and packaged to generate a complete quantitative measurement data layer. The packaged dataset includes a complete traceability link, clearly recording the original silhouette pixels, matching 3D target points, and used axis components for each osteotomy parameter. When the AR layer fusion module retrieves this data, it can directly read all distortion-free physical scale data without repeatedly performing pixel recognition, coordinate tracing, and axis difference calculations. This seamlessly integrates the vision-data dual-track decoupled architecture, providing precise quantitative underlying material for personalized En-DCR surgical AR planning views.

[0085] S105, calculate the spatial centroid of the projection point set to determine the nasal cavity side, adaptively match the corresponding endoscopic baseline map and mirror the quantitative scale arrangement.

[0086] In one implementation, by combining the X-axis distribution of the two-dimensional projection point set, the positive and negative thresholds for the centroid, the anatomical symmetry attributes of the left and right nasal cavities, and the classification rules for storing endoscopic base maps, the features of the projection centroid coordinates, nasal cavity lateral identification, and the arrangement direction of the quantization scale are extracted and distinguished to classify two anatomical application scenarios: the right nasal cavity and the left nasal cavity. After generating a complete two-dimensional projection point set through OAP orthogonal dimensionality reduction and convex hull fitting, all point set data automatically flows into the pre-stage of nasal cavity lateral scene classification. This stage belongs to the pre-discrimination subunit of the AR virtual-real fusion and interaction module. The only upstream input is the complete two-dimensional projection coordinate set output by the visual guidance layer, with no other irrelevant optical paths or measurement data mixed in. The scene classification result is unidirectionally transmitted to the base map adaptive matching process, and the entire discrimination process does not generate reverse data feedback.

[0087] The process incorporates four types of fixed basic judgment constraints, all matching the anatomical symmetry features of the nasal cavity and the system's base map library storage architecture. These four conditions are, in order: the distribution range of the X-axis global coordinates of the 2D projection point set, the positive and negative judgment thresholds for the projection centroid, the natural anatomical symmetry attributes of the left and right nasal cavities, and the lateral storage classification rules of the endoscopic baseline map. All four conditions participate simultaneously in feature extraction, and none can be omitted. The system simultaneously extracts three core discriminative features from the input projection point set: the overall centroid coordinates of the 2D point set, the nasal cavity lateral identification label, and the preset layout direction of the quantized scale. Based on the four judgment conditions, these three features are classified and distinguished, ultimately dividing the system into two independent anatomical application scenarios: the right nasal cavity and the left nasal cavity. After scenario division, a feature dataset containing centroid values, lateral labels, and scale layout references is packaged and passed down. If the input 2D projection point set has missing points or is empty, the system automatically pauses the scenario division operation and re-pulls the complete 2D silhouette projection data to the upstream visual guidance hierarchical planning stage. The discrimination process restarts only after a compliant point set is input.

[0088] The system calculates the centroid coordinates using all vertices of the 2D projection. The centroid calculation follows the aforementioned mean-based solution logic, only performing mean calculation on the projected X-axis coordinates as the basis for lateral discrimination. The centroid solution expression is as follows: In the formula, The x-axis barycenter coordinate of the entire two-dimensional projection is represented by n, where n is the total number of vertices in the two-dimensional projection of the lacrimal sac. This represents the X-axis coordinate value corresponding to a single projected vertex.

[0089] The following is an example of unilateral nasal cavity calculation. A total of 1286 vertices were extracted from the two-dimensional projection of the right nasal cavity and the average X-axis coordinate of all vertices was calculated. =-4.27; Mean value of the projection vertex of the left nasal cavity of the same specification =3.91, the system's preset center of gravity determination threshold is 0, using 0 as the boundary between the left and right nasal cavities. When When <0, it is automatically determined to be a right nasal cavity application scenario; when When the value is >0, it is automatically determined to be a left nasal cavity application scenario; if As the value approaches zero, the distribution of the system's marked points is abnormally centered. The centroid is then recalculated by retrieving the complete set of projection points of the lacrimal sac convex hull.

[0090] The 2D projection X-axis coordinate distribution range constraint ensures that the X-axis of the projection vertices of the unilateral nasal cavity has a fixed distribution range. All projection X-axis values ​​of the right nasal cavity are uniformly less than the boundary threshold 0, while all projection X-axis values ​​of the left nasal cavity are uniformly greater than the boundary threshold 0. The system traverses all vertex X-coordinates. If both positive and negative X-values ​​exist simultaneously in the same set of points, the projection contour is determined to be incomplete, the lateral segmentation is paused, and the data is sent back upstream to regenerate the silhouette. Using the example, the X-range of the right nasal cavity vertices is [-7.12, -0.45], and the X-range of the left nasal cavity vertices is [0.38, 6.94]. The interval boundaries and the centroid determination threshold form a double verification.

[0091] The centroid positive and negative threshold matching rule uses a fixed value of 0 as the unique dividing threshold, relying solely on... It distinguishes between positive and negative scenarios for left and right, does not set multi-segment grading thresholds, simplifies the side discrimination logic, and matches the natural left and right symmetrical anatomical structure of the nasal cavity, conforming to the En-DCR bilateral nasal cavity surgical anatomical benchmark.

[0092] The anatomical symmetry adaptation rule for the left and right nasal cavities is that the anatomical structures on the left and right sides of the nasal cavity are mirror symmetrical, and the corresponding quantitative scales are arranged in completely opposite directions. In the scene segmentation process, while marking the side identification, the corresponding scale reference arrangement features are simultaneously bound: when it is determined to be a right nasal cavity scene, the horizontal reference of the scale is on the left; when it is determined to be a left nasal cavity scene, the horizontal reference of the scale is on the right, providing a preliminary feature basis for the subsequent UI mirror flip.

[0093] Endoscopic base maps are stored separately according to different scenarios. The matching rule system base map library archives the results independently, with preset... , Two sets of dedicated endoscopic baseline base map files are synchronously written into the corresponding base map file index within the scene feature data. The downstream matching process can directly retrieve the corresponding base map based on the side identification without the need for a secondary image library search.

[0094] After completing the lateral discrimination, the output feature dataset contains three fixed structured fields: the first field: the floating-point value of the centroid of the two-dimensional projection X-axis (retaining two decimal places); the second field: the lateral identifier of the unilateral nasal cavity (right nasal cavity / left nasal cavity); and the third field: the identifier of the quantized scale baseline arrangement direction. The entire feature dataset corresponds only to a single lateral discrimination scenario and will not mix the discrimination information of the left and right nasal cavities at the same time. It fully supports the subsequent base map retrieval and scale mirror linkage of the entire adaptive processing flow, avoiding the problems of base map mismatch and reversed annotation scale caused by lateral discrimination confusion.

[0095] Combining the positive and negative projection centroid determination output, the endoscopic baseline image library indexing logic, and the ruler UI mirror flipping linkage mechanism, the adaptation execution rules for the two types of nasal cavity scenes are organized based on the lateral intelligent recognition fusion judgment logic, clarifying the base map files matched to the left and right nasal cavities and the ruler layout adjustment methods. After packaging and outputting three types of feature information—two-dimensional projection centroid coordinates, nasal cavity lateral identifiers, and ruler layout direction—the entire set of scene classification feature datasets flows into the adaptation rule organization stage. This stage belongs to the post-processing step of the lateral recognition subunit of the augmented reality fusion and interaction module. Upstream, it only receives the global projection X-axis centroid value and the original data of the single nasal cavity label output from the lateral scene segmentation stage, without mixing in irrelevant parameters such as optical path modeling and osteotomy calculation. After the rule organization is completed, it generates matching instructions for base map retrieval and ruler flipping linkage, which are then sent down to the layer fusion rendering process. The entire data flow is unidirectional, and there is no operation of rewriting the lateral judgment parameters in reverse.

[0096] The process incorporates two sets of fixed linkage logic: the endoscopic baseline image library index logic and the quantitative scale UI mirror flip linkage control mechanism. Both mechanisms rely on the positive and negative determination of the projection center of gravity as the trigger condition, and simultaneously generate exclusive adaptation execution rules that are not universal for the two types of anatomical scenarios: the right nasal cavity and the left nasal cavity. The rules for each type of scenario are independently encapsulated, and there will be no situation where the left and right base images and scale parameters are mixed.

[0097] The system pre-stores two independent sets of endoscopic baseline image files. The right nasal cavity matching baseline image is marked as... The left nasal cavity matching base map is marked as The base map index binding logic takes effect entirely based on the centroid positive / negative determination result: If the center of gravity of the projection <0, indicating a right nasal cavity condition, automatically bound. The base image index is directly read during rendering and used as the underlying background of the AR layer. If the center of gravity of the projection >0, indicating a left nasal cavity condition, automatically bound. Base map index, using the standard endoscopic image on the left as the visualization base; Fuzzy fallback rule for determining center of gravity: When the projected center of gravity... When the value approaches 0 infinitely, and the projection X-axis interval contains both positive and negative coordinates, making it impossible to clearly assign it to a single nasal cavity, the system terminates the generation of adaptation rules, clears the current cache of the entire set of projection points, and returns to the previous scene division stage to re-execute the calculation of the average centroid of the global vertex until a clear positive and negative centroid value is produced before the adaptation rules can be further organized.

[0098] The centroid calculation follows the same formula as the mean calculation: In the formula Here, n represents the centroid coordinates of the two-dimensional projection along the X-axis, and n represents the total number of vertices in the unilateral lacrimal sac projection. This represents the X-axis coordinate of a single projected vertex. Example calculation: The total number of projected vertices on the right side of the nasal cavity is n=1286, calculated by averaging... =-4.27, value less than 0, triggered. Base map index; Total number of vertices in the left nasal cavity projection on one side n=1312, mean =3.91, value greater than 0, retrieve Base map file; if the projection calculation for a certain patient yields... =-0.03, the value is infinitely close to the boundary threshold of 0, the judgment result is ambiguous, no base map index or scale flip rule is generated, and the projected points are re-collected and the centroid is recalculated upstream.

[0099] The ruler layout logic conforms to the mirror-symmetrical anatomical features of the human nasal cavity, and works synchronously with the base map index. Rules distinguish between two scenarios as follows: Right nasal cavity adaptation rules: Matching... In the initial mapping, the three sets of osteotomy quantification rulers (A, B, and C) are positioned with their overall horizontal reference on the left. The vertical annotation lines for parameter A and the horizontal annotation lines for parameter B retain their original generated orientation and do not require mirroring. Left nasal cavity fitting rules: Matching When creating the base map, the entire set of quantitative scales is horizontally mirrored and flipped, with the horizontal reference of the scales simultaneously placed to the right to adapt to the left nasal cavity anatomical mirror view, thus avoiding misalignment between the labeled direction and the actual anatomical position.

[0100] In the AR view of the right nasal cavity, the B-parameter (anterior cutting limit) scale extends to the left from the AX anchor point; after the left nasal cavity is mirrored, the same type of B-scale extends to the right from the AX anchor point, maintaining visual consistency with the natural symmetrical structure of both nasal cavities.

[0101] Read the floating-point value of the projection centroid and the single-sided scene label from the upstream output; substitute them into the centroid positive / negative determination criteria to distinguish between clear left / right nasal cavity conditions and blurred conditions; when the condition is clear, simultaneously bind the corresponding base map library index and generate a matching ruler mirror flip instruction, integrating them into a complete set of scene adaptation execution rules; when the condition is blurred, do not output any base map or ruler-related rules, send a data recalculation instruction to the upstream scene segmentation stage, re-collect the complete 2D projection point set, and solve again. After the effective adaptation rules are generated, three types of structured data—side label, base image file name, and ruler flip switch—are uniformly encapsulated and transmitted to the AR virtual-real fusion stage for overlaying and layout of silhouette layers and quantization parameters.

[0102] Based on the verification of the mean of the projection global coordinates and the dual-layer verification rules for the positive and negative centroid discrimination, the nasal cavity lateral segmentation results are further discriminated to accurately distinguish between the imaging conditions of the right and left nasal cavities, avoiding lateral segmentation misjudgment. After completing the indexing of the endoscope baseline image library and the organization of scene-specific adaptation rules for the scale UI mirror flip, the complete scene information, including centroid values, lateral labels, base map index, and scale layout instructions, is transferred to the dual-layer verification and discrimination stage. This stage belongs to the verification process of the lateral recognition subunit of the augmented reality fusion and interaction module. The only upstream input is the full-scene structured data output from the scene adaptation rule organization stage. The reliable lateral conclusion after verification is unidirectionally sent to the AR layer fusion and rendering stage. If the two verification conclusions conflict, an abnormal instruction is generated and sent back to the projection centroid calculation step. The entire verification process does not involve intermediate data caching and reuse or cross-stage jump calculations.

[0103] The process incorporates two sets of independent, sequentially executed verification rules: a global projection coordinate mean verification rule and a centroid positive / negative discrimination rule. The two verifications have a fixed execution order: the global projection coordinate interval verification is performed first, followed by the centroid positive / negative discrimination verification. Only when the nasal cavity condition conclusions obtained from the two verifications are completely consistent can a reliable nasal cavity lateral discrimination conclusion with usability be output. If the left and right nasal cavity conclusions output by the two layers of verification conflict with each other, the current lateral discrimination identification process is marked as abnormal, the current discrimination link is terminated, a recalculation instruction is issued to the projection centroid calculation step, the entire set of two-dimensional projection vertex sets is reread, the X-axis centroid value is solved again, and the entire two-layer discrimination process is restarted.

[0104] First layer: Verification rule for the mean of the projection coordinates across the entire domain. This layer's verification relies on the distribution range of the X-coordinates of all 2D projection vertices. Using 0 as the boundary between the left and right nasal cavity coordinates, compliance is achieved if the X-coordinates of all projection vertices on one side of the nasal cavity fall within the same lateral interval. The centroid calculation follows the aforementioned mean formula: In the formula Represents the centroid coordinates of the two-dimensional projection along the X-axis, where n is the total number of vertices in a single-sided teardrop projection. This refers to the X-axis coordinate value of a single projected vertex. Example 1 (Right Nasal Cavity): The total number of projected vertices on one side is n=1286, and the X-coordinate distribution range of all vertices is [-7.12, -0.45]. All values ​​in this range are less than 0, so this layer's verification determines it as a right nasal cavity imaging condition. Example 2 (Left Nasal Cavity): The total number of projected vertices on one side is n=1312, and the X-coordinate distribution range of all vertices is [0.38, 6.94]. All values ​​in this range are greater than 0, so this layer's verification determines it as a left nasal cavity imaging condition. An exception is as follows: If the same set of projected points contains both X>0 and X<0 vertices, and the coordinate range crosses the boundary threshold of 0, this layer directly determines the coordinate distribution is non-compliant. The two layers' verifications naturally conflict, triggering a recalculation logic.

[0105] The second layer: the centroid sign discrimination rule, this layer's verification directly reads the value calculated using the mean formula. Floating-point values, with 0 as the dividing threshold for determining the working condition: when <0, indicating a right nasal cavity imaging condition; when >0, indicating a left nasal cavity imaging condition; if Approaching infinitely close to 0, the two-layer verification naturally conflict, directly marking an anomaly. Continuing from the above sample calculation results: the sample on the right calculates... =-4.27, less than 0, the second layer of verification judges the right side; the left sample is calculated. =3.91, greater than 0, the second layer of verification judges the left side; the two sets of samples have completely consistent two-layer verification conclusions, and output the reliable side identifier.

[0106] Read the complete set of 2D projected vertex X coordinate arrays output from the upstream, and the initial calculations. Specifically, for the centroid value, the first-layer global coordinate interval verification is performed, outputting the first-layer nasal cavity condition judgment result; the second-layer centroid positive / negative discrimination is performed, outputting the second-layer nasal cavity condition judgment result. The two layers of condition labels are compared: ① If the two judgment results are completely identical: verification passes, and the side identification label, corresponding base map index, and ruler mirroring command are packaged and transmitted to the downstream layer fusion stage; ② If the two judgment results are inconsistent / the centroid approaches 0 infinitely / the coordinate interval crosses the 0 boundary: the side identification is abnormal, the current centroid cache data is cleared, and the process reverts to the projection centroid calculation step, reloading the complete 2D projection point set for iterative calculation. Repeat the double-layer verification process until the conclusions of the two layers are consistent.

[0107] The process of intelligent lateralization identification and adaptive matching is streamlined by integrating and summarizing the nasal cavity lateralization type, corresponding endoscopic base image retrieval, ruler mirroring adjustment rules, and dual-layer verification results. After obtaining a reliable nasal cavity lateralization judgment conclusion based on the projection global coordinate range verification and the dual-layer verification process of centroid positive / negative discrimination, the entire set of process data, including centroid values, projection coordinate distribution, preliminary lateralization labels, and ruler pre-arrangement, is uniformly transferred to the information aggregation and finalization stage. This stage belongs to the final integration process of the lateralization identification subunit of the augmented reality fusion and interaction module. The only upstream input is the complete working condition data output by the dual-layer verification, with no external data such as optical path modeling or osteotomy calculation mixed in. The standardized data after integration is unidirectionally sent to the AR virtual-real fusion module as fixed pre-input material. The entire lateralization adaptive processing link is serial and irreversible, with no data rollback or recalculation operations.

[0108] This step is designed as a single integrated processing action, with the core being the collection, binding, and structured encapsulation of four types of core information. The four types of basic information that are fixed to be included are: nasal cavity lateral discrimination determination type, matching and retrieved endoscopic baseline map index, quantitative scale UI mirror adjustment execution rules, and double-layer verification of the original discrimination results. The four types of information are bound together one by one, and the four types of materials are bound as a set under a single nasal cavity condition and are not split and output separately.

[0109] Example of binding nasal cavity lateral type with endoscopic base map index: Taking the right nasal cavity sample as an example, double-layer validation output. =-4.27, global X-coordinate distribution interval [-7.12, -0.45], both layers of judgment results are uniformly marked as the right nasal cavity, and the baseline map index is simultaneously bound. Taking the left nasal cavity sample as an example, the projection centroid... =3.91, global X-coordinate distribution interval [0.38, 6.94], double-layer verification uniformly determines it as the left nasal cavity, matching the base map file identifier. .

[0110] The formula for calculating the center of gravity follows the original expression for the mean: In the formula Here, n represents the centroid coordinates of the two-dimensional projection along the X-axis, and n represents the total number of vertices in the unilateral lacrimal sac projection. This represents the X-axis coordinate value of a single projected vertex.

[0111] Example of ruler mirror adjustment rule binding: The binding logic is synchronously written to the summary data according to the side type: When the condition is determined to be the right nasal cavity, the horizontal reference of the quantization ruler remains in its original left-hand arrangement and does not perform horizontal mirror flipping. The A, B, and C parameter annotation lines maintain their original generated orientation. When the condition is determined to be the left nasal cavity, the entire set of quantization rulers performs a complete horizontal mirror flipping. The horizontal reference of the ruler is automatically switched to the right side of the screen to adapt to the nasal cavity mirror anatomy and avoid misalignment between the annotation orientation and the actual anatomy.

[0112] The summary of the dual-layer verification and discrimination result retention rules is fully preserved. The original judgment basis of the two-layer verification is: the first layer is the range of the entire coordinate interval of the projection X-axis and the positive and negative attributes of the interval; the second layer is the floating-point value of the centroid and the comparison relationship between the centroid and the 0 threshold. The two types of original discrimination data are fully archived and can be used for the traceability and discrimination logic of the downstream AR fusion process.

[0113] The information collection and sorting rules integration operation follows a fixed collection order: the first step is to enter the original discrimination values ​​and interval data of double-layer verification; the second step is to lock the final nasal cavity side type label; the third step is to match the corresponding endoscopic base map file index; and the fourth step is to bind the matching ruler mirror adjustment switch command. The four steps are completed in sequence and the collection order cannot be disrupted.

[0114] The complete set of data output structure integrates the side identification intelligent recognition process data into a fixed structured data package, which contains four ordered information blocks: verification raw data block, side identification block, base map index block, and ruler layout control block. Each information block is independently divided and read. When the AR virtual-real fusion module retrieves the layer rendering material, it can read single-type information as needed without fully parsing the entire data package.

[0115] The upstream and downstream link connection rules: This step is the last unit of the nasal cavity side adaptive processing link. The preceding process is OAP projection point set generation -- side scene division -- adaptation rule organization -- double-layer verification and discrimination. After the information is summarized in this step, it directly connects to the AR virtual and real fusion process. The entire side recognition, base map matching, and scale adjustment complete link is closed loop from beginning to end. The visual generation branch operation and the quantitative measurement branch operation are kept isolated on two tracks throughout the process and do not interfere with each other.

[0116] If the two-layer verification has resulted in conflicting conclusions or abnormal conditions where the center of gravity approaches zero, the system will not generate a complete set of standardized matching data. Instead, it will directly send a recalculation instruction to the upstream two-layer discrimination stage. Only after the recalculation yields a unified discrimination conclusion and the four types of information can be completely bound will the collection and integration operation in this stage be initiated. Standardized discrimination adaptation data that can be directly called will only be output when the two-layer verification conclusions are completely consistent and the four types of information can be completely matched one by one. This prevents abnormal discrimination data from flowing into the AR layer rendering stage, causing visualization deviations such as base map mismatch and scale reversal.

[0117] S106, integrates anatomical silhouettes with perspective distortion with distortion-free anatomical axis parameters to generate a personalized En-DCR surgical AR preoperative quantitative planning view.

[0118] In one implementation, by combining physiological perspective distortion silhouettes, RAS decoupling osteotomy parameters, lateral nasal cavity endoscopic base maps, and AR layer overlay hierarchical elements, the distorted visual layers, error-free quantized annotations, left and right baseline base maps, and planning view carrier features are extracted and differentiated to classify the output scenes into two categories: right nasal cavity AR planning view and left nasal cavity AR planning view. After completing lateral adaptive matching of the nasal cavity to obtain the endoscopic base map index, mirror scale parameters, two-dimensional erythema silhouette with anterior compression perspective distortion, and RAS axis component decoupling A, B, and C osteotomy parameters, the entire set of basic layer materials are uniformly flowed into the AR view scene division stage. This stage belongs to the first layer classification unit of the augmented reality fusion and interaction module. Upstream, it only receives the standardized feature dataset output by the lateral adaptive matching stage and does not access the original intermediate data of optical path modeling and 3D reconstruction. The stage operation only distinguishes the independent view of one nasal cavity, and the left and right nasal cavity layer materials are completely isolated and collected. A single data stream only carries the rendering material corresponding to the single-sided anatomy and does not mix the bilateral base maps and silhouettes for synchronous processing.

[0119] The process uniformly collects four core elements of fixed AR layers, which serve as the sole criterion for view division. These four elements are: anatomical erythema silhouettes with endoscopic physiological fluoroscopy distortion; osteotomy quantification parameters without diagonal errors obtained through RAS single-axis difference calculations; endoscopic baseline maps archived independently by left and right sides; and multi-layer image overlay constraint rules. The system simultaneously extracts the visualization feature information corresponding to the four elements: distorted visual layer morphology, error-free quantization annotation style, side-specific baseline map identifiers, and planning view carrier size specifications. Based on the natural left-right mirror anatomical orientation of the nasal cavity, it divides the output into two independent output scenes: right nasal cavity AR planning view and left nasal cavity AR planning view.

[0120] The entire scene segmentation process includes pre-processing material verification logic. If the corresponding endoscope base image material is missing from the current single-side computation link, the system directly pauses the view classification process and sends a resource retrieval instruction to the upstream side identification and base image matching modules, waiting for the complete image to be retrieved. or After the base map is loaded, the view scene division is re-executed; only when all four types of layer elements are complete, and the sides and base are visible. Figure 1 Only when a one-to-one correspondence is established is the single-side scene feature dataset packaged and then transferred to the virtual-real overlay rule processing stage.

[0121] Unilateral right nasal cavity data collection elements: a red lacrimal sac silhouette with oblique perspective distortion, three sets of quantitative parameters (A=5.4mm / B=7.4mm / C=8.1mm), and base map markings. Layer three-layer overlay rules; the system categorizes it as a right-side nasal cavity AR planning view. Elements collected for the left-side nasal cavity: mirrored flipped erythema silhouette, three sets of quantized parameters A=5.0mm / B=6.6mm / C=8.5mm, and base map markers. The same three-layer overlay rule; the system classifies it as the left nasal cavity AR planning view.

[0122] The scene feature dataset for right / left nasal cavity classification is fed into the virtual-real overlay rule organization stage. This stage uses the entire system's vision-data dual-track decoupled architecture as the underlying fusion logic, and simultaneously introduces the output results of three pre-processing steps as constraints for rule compilation: the visual layer fully preserves the anterior compression perspective distortion caused by the 0° rigid endoscope oblique intersection; the quantization layer completely abandons 3D Euclidean oblique line ranging and only uses full osteotomy quantization annotation and scale line segments; RAS single-axis difference calculation; and each side view is matched with a corresponding dedicated endoscopic baseline map. Based on the three types of pre-processing constraints, two sets of independent layer overlay execution rules are compiled for the right and left nasal cavities, clearly specifying the fixed layer arrangement order from bottom to top for the two types of views, and the placement style of the three sets of osteotomy parameter annotations (A, B, and C).

[0123] The layers are arranged in a fixed and uniform pattern from top to bottom (the only difference between left and right is the mirror image of the ruler; the layer order remains unchanged). First layer (bottom layer): Corresponding to the baseline image of the side endoscope (…). / The first layer (middle layer) replicates the real endoscopic image during the operation; the second layer (middle layer) is a silhouette of the dacryocystorhinosus with original perspective distortion, which covers the lesion area in accordance with the anatomical position of the base map; the third layer (top layer) is a quantitative label and scale line segment for safe osteotomy of A, B, and C.

[0124] The following is an example of the arrangement of lateralization-specific annotations: Right nasal cavity view: Parameter A, the vertical scale is placed vertically along the right side of the lacrimal sac; Parameter B, the horizontal scale extends to the left from the AX anchor point. Left nasal cavity view: The entire set of quantization scales is horizontally mirrored; Parameter A, the vertical scale is placed vertically along the left side of the lacrimal sac; Parameter B, the horizontal scale extends to the right from the AX anchor point. The validation trigger rule is as follows: If the lateralization determination label of the current dataset does not match the identifier of the retrieved base map file (e.g., determining the left side but loading...), the lateralization determination label will be applied. The system directly terminates the generation of overlay rules, returns the entire set of scene feature data to the AR view scene segmentation stage, and re-executes the side and base map matching verification until the label and base map are correctly matched. Figure 1 After a one-to-one correspondence, the layer overlay rules are then formulated.

[0125] After completing the layer overlay and parameter annotation rules for each scene, they are all sent to the blending effect verification stage. This stage sets up two sets of parallel and independent verification criteria: a silhouette distortion fidelity verification standard and an anatomical axis distance error-free verification standard. Both standards must pass the verification before a compliant AR layer blending solution can be generated. If a single standard fails the verification, the current view rendering solution is deemed invalid.

[0126] The silhouette perspective distortion fidelity verification logic is as follows: the verification rule is to extract the silhouette outline of the second layer of erythema and compare it with the anatomical boundary of the bottom endoscopic base image. The silhouette deformation shape must be completely consistent with the anterior compression perspective characteristics generated by the 0° endoscope obliquely crossing the nasal cavity. The system is not allowed to automatically stretch or compress the silhouette to eliminate natural perspective distortion. If the silhouette generated by the system is a distortion-free orthographic projection outline, and the distortion deviation from the actual endoscopic image is too large, the verification is judged to fail, and the process returns to the virtual-real overlay rule step to re-lock the distortion retention constraint and adjust the layer rendering logic.

[0127] The logic for verifying the accuracy of anatomical axis distances is as follows: The verification rule is to read the top-level A, B, and C sets of labeled values ​​and verify that the values ​​are derived solely from the absolute difference of the RAS single axis. The use of three-dimensional spatial oblique line distances in parameter calculations is prohibited. For example, the AX coordinate of the right nasal cavity is (0,0,0), and the S-axis component of the top of the lacrimal sac is 3.6623. After single-axis difference calculation, A = 3.6623 mm. If the layer labeled values ​​are converted from three-dimensional oblique line distances, the verification is deemed invalid, and the original axis component data is retrieved from the quantization measurement layer to update the labeled parameters.

[0128] Two sets of review standards were used to simultaneously identify each set of views and output two standardized and compliant layer fusion schemes that were adapted to the right nasal cavity and the left nasal cavity, respectively. Only after all double verifications were completed and met the standards were the complete layer layout, annotation rules, and base map related data packaged and sent to the final summary stage of the entire process.

[0129] After the double-verification outputs a compliant layer fusion solution, the data flow enters the final stage of the entire process summary and completion. This unit is the final output unit at the end of the augmented reality fusion and interaction module, which uniformly collects and archives five types of essential core information. The five types of information are as follows: the final determination type of nasal cavity lateralization, the anatomical silhouette material of the lacrimal sac with physiological perspective distortion, the three sets of RAS decoupling osteotomy quantitative parameters (A, B, and C), the binding relationship of the endoscopic baseline map for view matching, and the final result of layer composite rendering that has passed the double verification.

[0130] The process integrates five types of information in a structured manner to generate a complete, personalized En-DCR surgical AR quantitative planning view, documenting the entire workflow from OAP projection, quantitative calculation, lateral identification, layer division, overlay arrangement, and effect verification. The entire set of visualized data can be directly output for preoperative clinical planning, seamlessly connecting the two decoupled computational branches of visual simulation and physical quantification. The four processes—scene division, overlay rule organization, dual verification of fusion effects, and summary of full-process information—are strictly sequentially linked, without skipping steps or recalculating, forming a closed-loop, complete personalized AR view generation and processing chain.

[0131] S107 relies on multi-module algorithms such as 3D image segmentation, orthogonal projection imaging, anatomical axis distance and axis component calculation, and automatic lateral identification to separate visual simulation and physical measurement calculation, and simultaneously output guide contours and precise osteotomy quantitative indicators that conform to endoscopic vision.

[0132] In one implementation, the entire algorithm system is built by integrating four types of serial front-end low-level computing units. These four units operate sequentially in a fixed order, with data flowing unidirectionally downwards between units. The complete output of one unit serves as the sole input to the next unit, eliminating any logic for cross-unit or hierarchical data retrieval. After all four types of units have completed their operations, the resulting structured data is aggregated and connected to the central control module, serving as the raw input for the central control process. The four independent low-level units have fixed roles, each outputting its own dedicated structured data carrier. The output content of each unit is as follows: 3D Image Segmentation Unit: The input is a thin-slice CT image sequence of the patient. After grayscale thresholding and region growth reconstruction, the output is a complete 3D vertex dataset of the individualized dacryocystorhinostomy sac of one nasal cavity. A single dataset contains approximately 12,600 sets of RAS three-axis floating-point coordinates. The converted coordinates of an example single vertex are (1.7863, 2.5225, 3.6623). The complete 3D spatial information of the anatomical boundaries of the dacryocystorhinostomy sac is preserved, providing the original anatomical benchmark for subsequent optical path modeling and axis distance calculation. Orthogonal projection imaging unit: Using the lacrimal sac vertex and dual anatomical anchor point RAS coordinates output by the three-dimensional image segmentation unit as input, it completes the entire set of calculations, including virtual endoscope optical axis construction, OAP matrix dot multiplication and dimensionality reduction, and convex hull contour fitting, and outputs a two-dimensional erythema anatomical silhouette that fully preserves the anterior compression fluoroscopic distortion caused by the 0° rigid endoscope oblique intersection. The silhouette shape is completely matched with the anatomical contour of the intraoperative endoscopic real-shot image. Anatomical axis distance and component calculation unit: Reads the extreme pixel values ​​of the two-dimensional projection and completes the reverse tracing of the three-dimensional coordinates. It extracts only the independent A and S axis components of the RAS coordinate system and performs absolute difference calculations, abandoning the three-dimensional space oblique distance calculation logic. It outputs three types of osteotomy safety parameters A, B, and C without diagonal measurement errors, along with the aforementioned fixed calculation expression: A=|Ptop[S] AXpos[S]|、B=|Pfront[A] AXpos[A]|、C=|Ptop[S] Pbottom[S]|. Taking the AX anchor point coordinates of the right nasal cavity (0,0,0) and the S-axis component of the top of the lacrimal sac (3.6623) as an example, the parameter A = 3.6623 mm is obtained by substituting into the formula. The parameter value only represents the actual osteotomy height of the superficial bone wall and is not affected by perspective distortion or three-dimensional depth difference. The nasal cavity lateralization automatic identification unit receives all two-dimensional projected X-axis coordinates generated by orthogonal projection and completes lateralization discrimination using the global vertex centroid mean formula. The centroid calculation formula is as follows: If the calculation yields <0 indicates the right nasal cavity, matching The endoscope base image and scale are arranged regularly with the left side facing the left. >0 indicates the left nasal cavity, matching The endoscopic base map and the horizontal mirror arrangement rules of the scale are used to simultaneously output three types of supporting data: nasal cavity side judgment label, baseline base map index, and quantitative scale adjustment instructions.

[0133] After all the data from the four types of front-end units are aggregated and flow into the central control module, the central control module completes task splitting based on two physically isolated independent computing links: a visual simulation link and a physical measurement link. The hardware computing resources and data storage partitions of the two links are completely separated, and the entire set of operations is executed in parallel and independently. There is no real-time data cross-reading and writing between the links, and no parameter cross-referencing, which completely avoids the superposition and amplification of visual perspective distortion and three-dimensional diagonal error during the operation. The visual simulation link only retrieves the two-dimensional distorted erythema silhouette output by the orthogonal projection imaging unit, without reading three-dimensional coordinates or axis component calculation data. The link's computational goal is only to replicate the real 0° rigid endoscopy intraoperative visual effect, fully preserving the natural perspective compression characteristics brought by the oblique optical path, and generating an anatomical guide contour that is completely aligned with the anatomical position and contour deformation of the endoscopic image. It is only used for intraoperative visual area marking and does not participate in any length or height numerical calculations. The physical measurement link only reads the original lacrimal sac vertex of the three-dimensional image segmentation unit and the independent A / S axis components after the anatomical axis distance calculation unit is split. It does not introduce two-dimensional projection pixels or perspective distortion-related materials throughout the process. All osteotomy safety scales are calculated solely based on the RAS single-axis difference, completely eliminating the calculation deviations caused by two-dimensional image deformation and three-dimensional depth difference. It is specifically designed to complete the individualized bone osteotomy safety numerical quantification and output pure physical dimension parameters.

[0134] After the parallel processing of the two isolated links is completed, the central control module synchronously outputs two independent clinically usable data carriers in both directions. The two types of data are generated and flow downwards synchronously, with no sequential output order: The first type of carrier is a distorted anatomical guide contour adapted to the endoscopic view, which is directly sent to the AR layer fusion stage as a layer overlay material to fit the endoscopic base map annotation of the dacryocystorhinostomy operation area; The second type of carrier is the A, B, and C osteotomy quantitative indicators that eliminate double errors, which are synchronously bound to the base map index and scale mirror arrangement rules output by the side automatic recognition unit and sent to the view rendering stage together to complete the numerical annotation overlay according to the scale orientation of the corresponding nasal cavity.

[0135] After visual contour materials and quantitative annotation materials are simultaneously delivered to the AR planning view generation process, the system completes the composite rendering of three layers: base map, erythema silhouette, and individualized osteotomy scale, ultimately producing the En-DCR individualized preoperative quantitative planning view. The entire system forms a complete serial processing link of "3D image segmentation - orthogonal projection imaging - anatomical axis calculation - automatic lateral recognition - overall control dual-track parallel operation - AR view fusion rendering". Relying on the dual-link physical isolation architecture, it achieves decoupling of visual display and physical measurement, solving the inherent defects of traditional single AR links where perspective distortion and diagonal error are superimposed, making it impossible to balance the realism of the image and the accuracy of the calculation.

[0136] like Figure 2 As shown, Figure 2The images used are the original coronal thin-section CT images of the nasolacrimal duct region of the patient, serving as the two-dimensional data source for the automatic target area segmentation and extraction steps of this protocol. The images include grayscale images of the bilateral maxillary sinuses, ethmoid sinuses, bony nasolacrimal duct, lacrimal sac fossa, nasal cavity and turbinates, and skull base bone. The white high-density shadow in the right lacrimal sac fossa is the contralateral bony marker, and the red-filled area on the left is the target lacrimal sac soft tissue, individually identified by the algorithm after threshold segmentation and region growing operations. Grayscale differences are used to distinguish the lacrimal sac soft tissue from the surrounding bone and nasal airways, accurately locating the two-dimensional pixel boundaries of the lacrimal sac and extracting the coordinates of all tomographic pixels of the lacrimal sac, providing the underlying pixel dataset for subsequent three-dimensional solid reconstruction. This visually demonstrates the adjacent anatomical relationships between the lacrimal sac, lacrimal sac fossa, and nasal airways at the two-dimensional level, completing the target area anatomical localization preprocessing.

[0137] like Figure 3 As shown, Figure 3 This is a schematic diagram of a nasal cavity and sinus anatomy model reconstructed using a complete set of thin-section CT tomographic datasets of the skull, providing a 3D digital model carrier for subsequent virtual endoscopic coordinate system construction and AR projection planning.

[0138] The model visualization rules are as follows: the craniofacial bones, nasal cavity and sinus air cells are rendered with a semi-transparent light gold body, which can penetrate and observe the deep anatomical adjacent relationships, avoiding the bony tissue from obscuring the lacrimal sac target area; the red three-dimensional entity in the figure is the individualized target lacrimal sac complete three-dimensional solid model obtained by grayscale threshold segmentation and region growth operation, which is displayed independently of the bone structure layer by layer, and fully preserves the original spatial morphology, long axis direction and geometric dimensions of the lacrimal sac.

[0139] The letter "S" at the top of the image represents the Superior axis of the RAS medical anatomical coordinate system, and the letter "P" in the middle represents the Posterior axis. The outer purple frame is the spatial boundary of the 3D reconstruction data. This is used to construct a standardized RAS anatomical 3D coordinate system and unify the 3D coordinate reference of all anatomical structures and markers.

[0140] This 3D model integrates three types of anatomical objects—craniofacial bones, nasal airways, and the lacrimal sac target area—into the same digital space. This provides core 3D foundational support for subsequent calibration of surgical anchor points in the middle turbinate axilla and anterior nasal aperture, construction of a virtual endoscopic optical axis, and realization of orthogonal projection conversion from 3D anatomy to the intraoperative endoscopic 2D field of view.

[0141] like Figure 4 As shown, Figure 4 This diagram illustrates the surgical anchor point calibration for the 3D reconstruction model of the nasal cavity and dacryocystorhinostomy sac. It is a semi-transparent 3D anatomical model of the nasal cavity and sinuses reconstructed based on a thin-slice CT dataset, continuing from the previous appendix. Figure 2 Two-dimensional fault segmentation, attachment Figure 3The overall 3D modeling results are visualized in the surgical anatomical anchor point extraction and spatial coordinate system construction stages. The craniofacial bones, nasal cavity, and sinus air cells are rendered with a semi-transparent light gold body, allowing for deep anatomical observation of their adjacent relationships. The red solid in the upper left corner is an individualized 3D model of the target lacrimal sac generated through pixel segmentation, fully preserving the original spatial morphology and geometric dimensions of the lacrimal sac. The pink outline around the image represents the spatial boundary of the 3D reconstructed data, using the standardized RAS medical anatomical coordinate system to unify the 3D coordinate references of all anatomical structures and marker points.

[0142] Two discrete red spherical markers are set within the model as key spatial reference anchors for En-DCR endoscopic minimally invasive surgery, as detailed below. (Middle turbinate axillary anchor point, marked with a small circle at the top of the image), anatomically located at the axillary root of the middle turbinate, serves as the origin of the three-dimensional coordinate system for the entire measurement system. It is the reference center for measuring quantitative parameters such as the intraoperative osteotomy range, lacrimal sac height, and mucosal incision distance. All three-dimensional spatial distances and axial offsets are decoupled and calculated with this anchor point as the zero point, eliminating measurement errors caused by the three-dimensional diagonal distance. Marker point. (Anterior nasal center anchor point, marked with a small circle at the bottom of the image), its anatomical location corresponds to the geometric center of the anterior nasal cavity, simulating the optical starting point for insertion of the rigid nasal endoscope during surgery; this point is then compared with the origin of the coordinate system. Vector connections are made to generate the virtual endoscope optical principal axis. This principal axis is used to construct a local orthogonal imaging plane, providing a geometric basis for the subsequent OAP optical axis orthogonal projection algorithm.

[0143] like Figure 5 As shown, Figure 5 This is a schematic diagram of orthogonal projection of an endoscope's optical axis to match the real-world scene, with corresponding appendix. Figure 4 The axillary dacryocystorhinostomy reference point, calibrated in the 3D model, is spatially mapped onto the 2D endoscopic view. This serves as a unified reference point for the quantitative measurement of osteotomy and incision ranges, anchoring the coordinate correspondence between the 3D space and the 2D endoscopic view. The green dashed rectangle represents the effective imaging field of view boundary of the endoscope, defining the effective rendering area of ​​the OAP orthogonal projection algorithm. Only within this rectangle is the lacrimal sac contour distortion fitted, matching the natural anterior-posterior perspective law of the endoscopic lens. The red filled contour area is a 2D silhouette of the lacrimal sac generated by orthogonal dimensionality reduction projection along the optical axis of "anterior nasal cavity - middle turbinate axilla" from the aforementioned 3D lacrimal sac solid model. The core innovation is the preservation of endoscopic physiological perspective distortion; the silhouette tilt, length, and width distortion are completely consistent with the intraoperative visual effect of the naked eye endoscopically, intuitively marking the safe operating area of ​​the lacrimal sac's surface projection under the nasal mucosa.

[0144] In one embodiment, this application also provides an En-DCR augmented reality preoperative quantitative planning system with endoscopic three-dimensional axis decoupling, comprising: The three-dimensional image preprocessing module is used to acquire thin-slice CT images of the patient's skull, complete the segmentation of the lacrimal sac tissue and the reconstruction of the three-dimensional model, accurately mark the two anatomical anchor points of the middle turbinate axilla and the anterior nasal cavity, and build a standardized RAS anatomical coordinate system benchmark dataset to provide a unified three-dimensional spatial benchmark for endoscopic optical path modeling and anatomical parameter calculation. The virtual endoscope modeling module is used to construct the optical axis of endoscope observation by connecting two anatomical anchor points, build a virtual photosensitive plane orthogonal to the optical axis at the anchor point position, complete the construction of the camera coordinate system through three-dimensional vector cross product, and generate an optical projection mathematical model adapted to 0° rigid endoscope. The OAP visual rendering module is used to perform three-dimensional vertex orthogonal dimensionality reduction mapping based on the established endoscopic optical projection mathematical model, extract the projection contour using the convex hull algorithm, fully preserve the endoscopic anterior fluoroscopic distortion and render two-dimensional anatomical silhouettes, and generate an AR visual guidance layer that matches the operation. The anatomical axis distance calculation module is used to identify the extreme pixel of the silhouette outline and backtrack the corresponding RAS three-dimensional coordinates. It abandons the spatial oblique line distance measurement method, calculates the difference along the anatomical axis components separately, eliminates the diagonal measurement error, generates standardized osteotomy safety parameters, and constructs a quantitative measurement data layer. The lateral adaptive matching module is used to solve the centroid of the two-dimensional projection point set on the X-axis to determine the left and right nasal cavities, automatically retrieve the corresponding endoscopic baseline map, and synchronously mirror and flip the UI layout of the A, B, and C osteotomy scales to adapt to the anatomical symmetry display requirements of both sides of the nasal cavity. The AR virtual-real fusion module is used to overlay and fuse distorted anatomical visual silhouettes and error-free anatomical osteotomy quantitative parameters, match the corresponding nasal endoscope base map to complete layer composite rendering, and output an individualized En-DCR surgery AR preoperative quantitative planning view. The dual-track decoupled computing control module integrates four types of underlying algorithms: image segmentation, orthogonal projection, axis distance calculation, and lateral recognition. It separates the two computing links of endoscopic visual simulation and anatomical physical measurement, and outputs anatomical contours and precise osteotomy quantitative indicators that conform to the real endoscope in parallel.

[0145] The computer-readable storage medium provided in the above embodiments of this application and the En-DCR augmented reality preoperative quantitative planning method for endoscopic three-dimensional axis decoupling provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0146] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for evaluating the En-DCR augmented reality preoperative quantitative planning method, system, electronic device, and readable storage medium with endoscopic three-dimensional axisymmetric decoupling are basically similar to the embodiments of the En-DCR augmented reality preoperative quantitative planning method with endoscopic three-dimensional axisymmetric decoupling described above, and are therefore described relatively simply. Relevant parts can be referred to in the descriptions of the embodiments of the En-DCR augmented reality preoperative quantitative planning method with endoscopic three-dimensional axisymmetric decoupling described above.

Claims

1. A preoperative quantitative planning method for endoscopic three-dimensional axis-distance decoupling in En-DCR augmented reality, characterized in that, include: Acquire simulated thin-section CT images of the target user's skull, segment and reconstruct a three-dimensional model of the lacrimal sac, and calibrate the anatomical anchor points of the axillary part of the middle turbinate and the center of the anterior nasal cavity to construct the baseline data of the RAS anatomical coordinate system. The virtual optical axis of the endoscope is constructed by connecting the nostrils, the middle turbinate and the axilla, and a virtual imaging plane orthogonal to the optical axis is established to build a mathematical model of endoscopic optical projection. The optical axis orthogonal projection algorithm based on the mathematical model of endoscopic optical projection is used to reduce the dimension of the three-dimensional vertices of the lacrimal sac and generate a two-dimensional anatomical silhouette that preserves the endoscopic perspective distortion through convex hull fitting, thus completing the rendering of the visual guidance layer. Extract the extreme pixels of the projected contour and trace their corresponding three-dimensional RAS coordinates. Decouple and calculate the osteotomy safety physical parameters without diagonal error along the anatomical axis components to construct a quantitative measurement data layer. Calculate the spatial centroid of the projection point set to determine the nasal cavity side, adaptively match the corresponding endoscopic baseline map and mirror the quantitative scale arrangement; By fusing anatomical silhouettes with perspective distortion with anatomical axial distance parameters without distortion, a personalized preoperative quantitative planning view for En-DCR surgery using AR is generated. Relying on multi-module algorithms such as 3D image segmentation, orthogonal projection imaging, anatomical axis distance and axis component calculation, and automatic lateral identification, the system separates visual simulation and physical measurement calculation, and simultaneously outputs guiding contours and precise osteotomy quantitative indicators that conform to endoscopic vision.

2. The method as described in claim 1, characterized in that, A virtual endoscopic optical axis is constructed by connecting the anterior nostril, middle turbinate, and axillary region. A virtual imaging plane orthogonal to the optical axis is established, and a mathematical model of endoscopic optical projection is built, including: Based on the modeling target, anatomical entity, and optical operation dimensions of the endoscope camera, Nostril point, AX anchor point, three-dimensional direction vector, orthogonal photosensitive plane, vector cross product, key entity variables of virtual camera coordinate system, semantic rules of optical path construction, and modeling technical parameters are classified and integrated to generate basic unit information for virtual endoscope optical path construction. Based on the effectiveness requirements of endoscope optical simulation, the content layout of the optical path basic unit is designed, clarifying the filling position of dual anchor point coordinate variables, the core expression of principal axis normalization and cross product, the proportion of orthogonal plane construction steps, and generating complete process information for building the endoscope optical projection model. Combining the pre-imaging requirements of the OAP algorithm with the optical path restoration requirements of real rigid endoscopes, a step-by-step operation rule of anchor point connection - principal axis vectorization - orthogonal plane construction is set to match the clinical 0° endoscope observation logic, ensuring that the optical path model is adapted to the En-DCR nasal cavity anatomy scenario.

3. The method as described in claim 1, characterized in that, An orthogonal projection algorithm based on an endoscopic optical projection mathematical model is used to reduce the dimensionality of the three-dimensional vertices of the lacrimal sac. A convex hull fitting process is then used to generate a two-dimensional anatomical silhouette that preserves endoscopic perspective distortion. This process completes the rendering of the visual guidance layer, including: Based on the mathematical model of endoscopic optical projection, orthogonal dimensionality reduction operation of OAP, preservation of endoscopic anterior shrinkage perspective distortion, and extraction of two-dimensional anatomical contours, the three-dimensional vertex set of the lacrimal sac, virtual orthogonal photosensitive plane, matrix dot product mapping, convex hull contour extraction, key entities and operation rules of endoscopic anterior shrinkage distortion are classified and integrated to generate basic unit information for three-dimensional vertex orthogonal dimensionality reduction, contour fitting, distortion preservation, and visual silhouette output. Based on the overall architecture requirement of vision-data dual-track decoupling, the visual guidance layer generation task is designed in a layered and decoupled manner. The module division is divided into three-dimensional vertex orthogonal dimensionality reduction mapping, two-dimensional contour convex hull fitting, physiological perspective distortion preservation, and two-dimensional erythema silhouette rendering output. The data flow relationship of CT three-dimensional vertex input, matrix operation, contour extraction, and AR visual layer output is clarified, forming task execution information for the collaborative operation of projection mapping, contour fitting, distortion preservation, and layer rendering. Combining the requirements of En-DCR endoscope visual matching degree, AR guidance intuitiveness, and 3D to 2D mapping and restoration accuracy, based on the 0° rigid endoscope oblique imaging characteristics, we set the 3D vertex matrix dot product dimensionality reduction standard, convex hull edge extraction rules, and anterior compression perspective distortion integrity retention constraint, and determined the orthogonal projection operation method, contour generation judgment conditions, and final output form information of the visual guidance layer.

4. The method as described in claim 1, characterized in that, Extracting extreme pixels of the projected contour and tracing their corresponding 3D RAS coordinates, decoupling along the anatomical axis components to calculate the safe physical parameters of osteotomy without diagonal error, and constructing a quantitative measurement data layer, including: We organize and sort out various anatomical measurement data, including extreme pixel values ​​of two-dimensional projected contours, three-dimensional RAS coordinates, anatomical single-axis components, and osteotomy limit indicators. We match the RAS axis distance decoupling operation specifications and construct a quantitative layer preprocessing process for extreme value tracing and axis difference calculation. Based on a multi-dimensional analysis framework that relies on projected pixel index, three-dimensional spatial coordinates, and independent operation of anatomical three axes, and integrates two-dimensional silhouette points with original CT three-dimensional model data, a linkage verification mechanism for pixel tracing, axis component splitting, and diagonal error shielding is established. With the goal of avoiding three-dimensional diagonal errors and obtaining a safe scale for bone osteotomy, the contour extreme point position, RAS upper / lower / anterior axis coordinates, and three types of osteotomy parameter information (A, B, and C) are integrated to construct a binding relationship between coordinate backtracking and single-axis difference calculation. By employing triple verification through projection extreme pixel recognition, three-dimensional coordinate reverse tracing, and independent calculation of anatomical axis components, the spatial oblique line distance measurement method is abandoned, and precise osteotomy physical parameters are generated and a quantitative measurement data layer is constructed.

5. The method as described in claim 4, characterized in that, Calculate the centroid of the projection point set to determine the nasal cavity side, adaptively match the corresponding endoscopic baseline map and mirror and adjust the quantization scale arrangement, including: Combining the X-axis distribution of two-dimensional projection point set, the positive and negative determination threshold of the centroid, the anatomical symmetry attributes of the left and right nasal cavities, and the classification rules for storing endoscopic base maps, the features of projection centroid coordinates, nasal cavity side identification, and quantitative scale arrangement direction are extracted and distinguished to classify two anatomical application scenarios: right nasal cavity and left nasal cavity. Combining the positive and negative projection center of gravity determination output, the endoscope reference image library index logic, and the ruler UI mirror flip linkage mechanism, the adaptation execution rules corresponding to the two types of nasal cavity scenarios are organized based on the side-specific intelligent recognition fusion judgment logic, clarifying the base image files matched for the left and right nasal cavities and the ruler layout adjustment methods respectively; Based on the verification of the mean of the projection global coordinates and the double-layer verification rule of the centroid positive and negative discrimination, the nasal cavity lateral classification results are judged a second time to accurately distinguish the imaging conditions of the right nasal cavity and the imaging conditions of the left nasal cavity, thus avoiding the problem of lateral misjudgment. The entire process of intelligent lateral identification and adaptive matching was streamlined by integrating and summarizing the nasal cavity lateral identification type, corresponding endoscopic base image retrieval, ruler mirror adjustment rules, and double-layer verification results.

6. The method as described in claim 1, characterized in that, By fusing anatomical silhouettes with perspective distortion with distortion-free anatomical axis parameters, a personalized preoperative quantitative planning view for En-DCR surgery (AR) is generated, including: Combining physiological perspective distortion silhouette, RAS decoupled osteotomy parameters, nasal cavity lateral endoscopic base map, and AR layer overlay hierarchical elements, the distortion visual layer, error-free quantitative annotation, left and right reference base maps, and planning view carrier features are extracted and distinguished, and two types of output scenes are divided into right nasal cavity AR planning view and left nasal cavity AR planning view. Combining the visual layer to preserve endoscopic distortion, the quantization layer to remove diagonal errors, and the output of the side-matching base map, the virtual-real overlay rules corresponding to the two types of views are organized based on the visual-data dual-track decoupling and fusion logic, clarifying the layer overlay order and parameter annotation arrangement of different side views; Based on the dual verification rules of silhouette distortion fidelity verification and anatomical axis distance error-free verification, the layer fusion effect is reviewed and judged, and two standardized planning output conditions are distinguished: one adapted to the right nasal cavity and the other adapted to the left nasal cavity. The entire process of individualized En-DCR surgery AR quantitative planning view was completed by integrating and summarizing the nasal cavity lateralization, distorted anatomical silhouette, A, B, and C osteotomy quantitative parameters, base map matching relationship, and layer fusion results.

7. An En-DCR augmented reality preoperative quantitative planning system with endoscopic three-dimensional axis-distance decoupling, characterized in that, The system is used to execute executable instructions to perform the En-DCR augmented reality preoperative quantitative planning method for endoscopic three-dimensional axis decoupling as described in any one of claims 1 to 6.

8. An electronic device, characterized in that, include: First processor; And a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the En-DCR augmented reality preoperative quantitative planning method of any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the En-DCR augmented reality preoperative quantitative planning method for endoscopic three-dimensional axis decoupling as described in any one of claims 1 to 6.

Images