Method and system for training anterior lateral free flap surgery based on mixed reality

Abstract

The invention relates to the technical field of surgical operation training, and particularly discloses a training method and system for an anterior-lateral free flap operation based on mixed reality. The method comprises the steps that by obtaining a preoperative image and a body surface grid, tissue layering labeling and individualized deformable registration are carried out, and projection stability parameters are generated; carrying out perforation discovery and ultrasonic fusion projection, carrying out layered approach planning, and generating instrument navigation information; and then microscopic anastomosis visual touch closed-loop training and complication treatment evaluation are realized, and finally a trainee ability portrait is generated and a self-adaptive course engine is driven. According to the invention, through multi-modal data fusion and real-time feedback, high-fidelity simulation and personalized training of key operation steps are realized, and the training precision and efficiency are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of surgical training technology, and in particular to a surgical training method and system for anterolateral thigh free flap based on mixed reality. Background Technology

[0002] The anterolateral thigh free flap is commonly used for head and neck reconstruction and complex wound repair due to its stable perforator source and wide range of skin island quality adaptation. In the early training phase, residents and young surgeons rely heavily on classroom lectures, surgical videos, and a limited number of clinical observations to accumulate experience. This results in a fragmented training chain in areas such as understanding the three-dimensional course of perforators, surface marking and localization, selection of layered approach pathways, and mastery of the rhythm of microsurgical anastomosis. Existing teaching models are mostly based on rigid registration and static template overlay, making it difficult to establish a point-to-point correspondence between preoperative images and the actual body surface. Changes in soft tissue posture and displacement triggered by local tension cannot be suppressed, leading to drift and jumps in the superimposed image on the head-mounted display, and insufficient stability of the surface projection. During ultrasound exploration, the probe posture and surface grid coordinates are often weakly coupled, making it difficult to synchronize the acoustic characteristics of the perforator with the three-dimensional path of the preoperative image. Exploration hit statistics and duration records are scattered across multiple devices, making it difficult to integrate the data into subsequent training scenarios. Layered approach practices typically rely on experience to establish "safe tunnels." Distance indicators and directional guidance between instrument tips and target structures lack a unified metric; the safety boundary radius cannot adaptively adjust to the organization's risk level and the current layer; and boundary violation annotations lack both temporal and spatial indexes. In microscopic anastomosis training, the overlay of microscopic views and the coordination of the tactile module are insufficient. Needle spacing and width recordings, patency assessments, and leakage-related determinations are sourced from multiple heterogeneous sources, lacking a common link with preceding approach and instrument navigation data, making it difficult to form operational rhythm indicators that run throughout the entire training process.

[0003] While mixed reality has been explored in medical training, common approaches focus on visualization or localized scoring of single aspects, lacking an end-to-end data framework covering "preoperative imaging—surface mesh—ultrasound exploration—access navigation—microscopic anastomosis—debriefing coaching." Particularly at the individual level, there is a lack of deformable correspondence between preoperative images and surface meshes; deformation descriptions are not co-produced with quality gating parameters; and there is a lack of unified parameter sets for surface projection triggering and freezing conditions. This results in subsequent perforator projection, ultrasound fusion, instrument navigation, and microscopic evaluation operating independently. The overlaid content in the training scenario is out of sync with real-world surface movements, making it difficult for data to flow stably between modules. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a surgical training method for the anterolateral thigh free flap based on mixed reality, comprising:

[0005] Preoperative images and body surface meshes are acquired, and the skin layer, superficial fascia layer and intermuscular channel are partitioned and hierarchically divided. Local correspondence constraints and deformable mappings are established between the control area and the image subdomain. Then, error estimation and stability are performed, individual deformable registration and steady-state projection initialization are performed, and projection stability parameters are generated.

[0006] Based on the projection stability parameters, perforator detection and ultrasound fusion projection are performed to generate projection correction values.

[0007] Based on the projection correction, a layered approach and instrument safety tunnel are constructed, and a distance sequence between the instrument and the target is generated.

[0008] Based on the distance sequence between the instrument and the target, microscopic anastomosis visual-touch closed loop and virtual perfusion are performed to generate microscopic manipulation rhythm indicators;

[0009] Based on the micromanipulation rhythm index, complication events are injected and strategy tree evaluation is performed to generate a treatment strategy performance matrix and a review timeline.

[0010] The system acquires a performance matrix of handling strategies and a review timeline. Based on event type and handling process, it aggregates the sequence of capability elements into a set of capability profile items, performs scene slicing and reorganization to bind level structure, scenario parameters and evaluation items, then generates recipes and writes them back, and uses an intelligent coaching and adaptive curriculum engine to generate data configuration.

[0011] Furthermore, the process of generating projection stabilization parameters includes:

[0012] Preoperative images and body surface meshes were acquired. Several traceable markers were set on the body surface and corresponding anatomical reference points were simultaneously marked in the preoperative images to form an initial correspondence set. Local correspondence constraints were established in order from superficial to deep. The skin layer, superficial fascia layer and intermuscular channels were partitioned and hierarchically divided. Local correspondence constraints and deformable mappings were established between the control area and the image subdomain. Data preparation was carried out for individual deformable registration and steady-state projection initialization to obtain the deformation field.

[0013] Furthermore, the process of generating projection stabilization parameters includes:

[0014] The location set of the representative areas covered by the deformation field in the sampling zone, including the raised area of ​​the skin layer, the flat area of ​​the superficial fascia, and the adjacent intermuscular communication area, is compared with the reference set. The errors corresponding to different levels are classified and summarized, and a short time window in time sequence is established to aggregate the degree of fluctuation. Adhesion error, interlayer error, strain error and fluctuation items are generated. Error estimation of individual deformable registration and steady-state projection initialization is performed to obtain the alignment quality index.

[0015] Furthermore, the process of generating projection stabilization parameters includes:

[0016] The alignment quality index is controlled by grouping the control areas based on the bonding error and interlayer error. A time-series buffer window is established in each group and the entry values ​​are read to form a time-space joint reference sequence. The drift compensation amount is calculated for the fluctuation entries and the projection update interval is set. The buffer window, drift compensation amount and projection update interval are established in the group. The stability of individual deformable registration and steady-state projection initialization is calculated to generate projection stability parameters.

[0017] Furthermore, preoperative imaging and surface mesh include:

[0018] Preoperative imaging specifically includes medical image slice data covering the anterolateral thigh region, as well as coordinate information of reference markers distributed on the body surface.

[0019] The surface mesh is specifically composed of a high-precision continuous curved surface mesh generated by optical or laser surface scanning equipment, and its vertex data includes spatial coordinates, normal vectors and local curvature attributes.

[0020] Furthermore, the process of generating the data configuration includes:

[0021] The system acquires a performance matrix of handling strategies and a review timeline. Using event type as the hierarchical key and handling process as the aggregation key, it aggregates the sequence of capability elements into a set of capability profile items and projects them onto a multi-dimensional coordinate domain. Based on event type and handling process, the system aggregates the sequence of capability elements into a set of capability profile items, constructs capability profiles for intelligent coaching and adaptive curriculum engines, and obtains a capability radar chart.

[0022] Furthermore, the process of generating the data configuration includes:

[0023] Based on the capability radar chart, the scene is sliced ​​and reorganized according to the segments and nodes corresponding to the weak elements. When the capability element item is the instrument path control, the same level and similar channel curvature segments are selected to generate continuous sub-levels. When it is the timeliness of complication treatment, the adjacent intensity level events are replayed and injected to form time-limited sub-levels. The level structure, scenario parameters and evaluation items are bound together to generate recipes for intelligent coaching and adaptive curriculum engines, resulting in individual training recipes and retraining parameters.

[0024] Furthermore, the process of generating the data configuration includes:

[0025] The retraining parameters are merged according to the projection, registration and acquisition configurations. The projection update interval, the target interval of the projection stability parameters and the confidence threshold of the cross-branch projection set are mapped to the initialization key value of the body surface projection module. The sampling density of the deformation field initialization point set and the alignment quality index threshold are written into the individual deformable registration initialization node and the quality gate node, and written into the regional density table and the convergence allowable range. The data configuration is generated by writing back the intelligent coach and adaptive curriculum engine.

[0026] Furthermore, the process of generating projection stability parameters also includes:

[0027] Data preparation, error estimation, and stability determination for individual deformable registration and steady-state projection initialization are performed sequentially.

[0028] Furthermore, a mixed reality-based anterolateral thigh free flap surgical training system, applied to any of the methods described above, includes:

[0029] The registration initialization module is used to acquire preoperative images and body surface mesh, perform partitioning and hierarchical division of the skin layer, superficial fascia layer and intermuscular channel, and establish local correspondence constraints and deformable mapping between the control area and the image subdomain to generate a deformation field.

[0030] The error estimation module, connected to the registration initialization module, is used to compare the mapped position set with the reference set within the sampling band, generate fitting error, interlayer error, strain error and fluctuation items, and output alignment quality index.

[0031] The stability calculation module, connected to the error estimation module, is used to group the control area based on the bonding error and inter-layer error, and to establish a buffer window, drift compensation amount and projection update interval within the group to generate projection stability parameters.

[0032] The surface projection module, connected to the stability calculation module, is used to generate the perforator skeleton set and perform deformation field mapping under threshold total table gating. It uses drift compensation and projection update interval to implement timing gating and freeze control, and outputs the perforator projection set and the confidence map of the perforation point.

[0033] The ultrasound exploration module, connected to the body surface projection module, is used to delineate priority exploration sectors based on the confidence map of the penetration point, perform frame sector binding, constraint matching and course consistency check, and output the measured penetrating path and hit and duration indicators.

[0034] The correction calculation module, connected to the ultrasonic exploration module, is used to compare the measured penetrating path with the penetrating projection set in the priority exploration sector as the unit, generate an offset entry set through window smoothing, and output the projection correction amount.

[0035] The path constraint module, connected to the correction calculation module, is used to select target anchor points within the candidate entry zone and generate surface fitting entries, strip buffer entries, and tubular constraint entries, outputting a hierarchical path constraint set;

[0036] The safety navigation module, connected to the path constraint module, is used to calculate the dynamic safety radius curve and hazard zone coloring map based on the hierarchical path constraint set, and bind safety radius entries and prohibited entries to perform instrument navigation, outputting the distance sequence between the instrument and the target and boundary crossing event annotations.

[0037] The key innovations of this invention include:

[0038] (1) Focusing on the core links of individual deformable registration and steady-state projection initialization, the deformation field, alignment quality index and projection stability parameters are output together to establish a unified parameter group and unified triggering conditions that can be directly called by subsequent modules.

[0039] (2) Based on the discovery of perforators and ultrasound fusion projection based on the projection stability parameters, a set of perforator projections and a confidence map of the exit point are generated in the same body surface coordinates. The body surface projection is corrected online by the projection correction amount to complete the collaborative superposition of the image domain and the sound image domain.

[0040] (3) For the generation and risk assessment of target anchor points for layered access routes and instrument safety tunnels, a layered path constraint set and dynamic safety radius curve are formed, and connected with the distance sequence between instruments and targets, so that the path, boundary and instrument tracking have the same source identification and time sequence index.

[0041] (4) Construct a unified recording and evaluation link for microscopic anastomosis, visual-touch closed loop and virtual perfusion, and generate needle distance and needle width deviation, smoothness score and leakage risk prediction on the same data line, so that the quantitative indicators at the microscopic level and the entry navigation data can be traced.

[0042] (5) Through the event setting, intervention record and performance analysis of complication event injection and strategy tree evaluation, output the treatment strategy performance matrix and review timeline to realize the structured archiving of scenario triggering and treatment path.

[0043] (6) Design a write-back mechanism for intelligent coach and adaptive curriculum engine, and write back the individual training recipe and retraining parameters to the individual deformable registration and steady-state projection initialization, so that the deformation field, alignment quality index and projection stability parameters keep updated with the training process, and complete the closed-loop linkage of data, parameters and curriculum.

[0044] The following are its main beneficial effects:

[0045] (1) This invention focuses on the homologous modeling of preoperative images and body surface meshes, constructs individual deformable registration and steady-state projection initialization, and produces three types of basic data: deformation field, alignment quality index and projection stability parameters. Through operations such as error estimation, stability calculation, drift compensation and threshold gating, the mixed reality superimposed on the soft tissue posture change and local tension scene maintains continuous mapping and orderly triggering, thereby providing unified coordinates and unified triggering conditions for the subsequent training process.

[0046] (2) Based on this unified baseline, the perforator detection and ultrasound fusion projection directly call the projection stability parameters to complete the surface projection and exploration registration, generate the perforator projection set, the exit point confidence map and the projection correction value, and realize the synchronous presentation and consistent recording of the image domain path and sonographic features in the same surface coordinates.

[0047] (3) The layered approach and instrument safety tunnel link generates a layered path constraint set and a dynamic safety radius curve under the constraints of the through-bracket projection set and projection correction amount. Together with the instrument and target distance sequence, it forms a spatiotemporally searchable navigation record, which solves the coordination problem between path anchoring, risk boundary and instrument approach prompt.

[0048] (4) Microscopic anastomosis, visual-touch closed-loop and virtual perfusion process: Based on the above navigation records, microscopic operation records and morphological evaluations are performed to obtain needle distance and needle width deviation, patency score and leakage risk prediction, thus completing the homogeneous connection from the entry route navigation data to the microscopic evaluation data.

[0049] (5) Complication event injection and strategy tree evaluation: Under the constraints of patency score and leakage risk prediction, event setting and intervention recording are carried out, and the treatment path, time and effect data are output. Through performance analysis, a treatment strategy performance matrix and review timeline are generated to realize a unified link of scenario triggering, intervention process and result archiving.

[0050] (6) The intelligent coach and adaptive course engine generates individual training recipes and retraining parameters based on the performance matrix of the handling strategy and the review timeline, and writes them back to the individual deformable registration and steady-state projection initialization, so that the deformation field, alignment quality index and projection stability parameters are updated from the same source as the student profile, forming a closed-loop data channel from initialization to write back. Attached Figure Description

[0051] Figure 1 A flowchart illustrating a mixed reality-based surgical training method for anterior lateral thigh free flap, as provided in this application embodiment;

[0052] Figure 2 This is a structural block diagram of a mixed reality-based anterolateral thigh free flap surgical training system provided in an embodiment of this application. Detailed Implementation

[0053] Example 1: Refer to Figure 1 This is a flowchart illustrating a mixed reality-based surgical training method for anterior lateral thigh free flap, as provided in an embodiment of the present invention. The flowchart may include at least steps S100-S600:

[0054] S100: Acquire preoperative images and body surface mesh, perform zonal labeling and hierarchical division of the skin layer, superficial fascia layer and intermuscular channel, and establish local correspondence constraints and deformable mapping between the control area and image subdomain. Then, perform error estimation and stability calculation, perform individual deformable registration and steady-state projection initialization, and generate projection stability parameters.

[0055] S200. Based on the projection stability parameters, the perforator skeleton set is generated and the deformation field is mapped under the threshold total table gating. The timing gating and freeze control are implemented with the drift compensation amount and the projection update interval. Then, the ultrasonic exploration registration and correction amount calculation are performed. The perforator is discovered and the ultrasonic fusion projection is performed to generate the projection correction amount.

[0056] S300: Based on the projection correction, the starting point, ending point and transfer point are selected as target anchor points within the candidate access zone, and surface fitting entries, strip buffer entries and tubular constraint entries are generated. Then, risk assessment and instrument navigation are performed, layered access and instrument safety tunnel are constructed, and instrument and target distance sequence is generated.

[0057] S400: Based on the distance sequence between the instrument and the target, the needle point set is extracted from the microscopic field frame sequence and a suture line segment set is generated. The geometric quantification is performed in combination with the change of tactile resistance. Then, morphological evaluation and rhythm analysis are performed to perform microscopic anastomosis visual-tactile closed loop and virtual perfusion, and microscopic operation rhythm index is generated.

[0058] S500, based on micromanipulation rhythm indicators, determines the type of complication according to leakage risk items and low value range of patency score, and adds out-of-bounds gating items and rhythm sensitive markers. Then, intervention records and performance analysis are carried out, complication events are injected and strategy tree evaluation is performed, and a treatment strategy performance matrix and review timeline are generated.

[0059] S600 acquires the performance matrix of the handling strategy and the review timeline. Based on the event type and handling process, it aggregates the sequence of capability elements into a set of capability profile items, and performs scene slicing and reorganization to bind the level structure, scenario parameters and evaluation items. Then, it performs recipe generation and write-back processing, performs intelligent coaching and adaptive curriculum engine, and generates data configuration.

[0060] Step S100 includes at least steps S110-S130:

[0061] S110. Acquire preoperative images and body surface mesh, prepare data for individual deformable registration and steady-state projection initialization, and obtain the deformation field;

[0062] Preoperative images specifically include slice data of medical images covering the anterolateral thigh region (such as CT, MRI, or ultrasound sequences), as well as coordinate information of reference markers distributed on the body surface. The body surface mesh is composed of a high-precision continuous curved surface mesh generated by optical or laser surface scanning equipment. Its vertex data includes spatial coordinates, normal vectors, and local curvature attributes. Noise points have been removed and topological holes have been repaired through preprocessing to ensure geometric continuity and differential properties in areas with abrupt curvature changes such as anatomical protrusions and depressions. Both types of data have undergone coordinate system unification and de-identification processing. An initial correspondence set is formed by pre-set traceable markers on the body surface and corresponding anatomical reference points marked in the images. This provides a multimodal fusion basis for subsequent establishment of hierarchical correspondence constraints and deformable mappings from shallow to deep.

[0063] Specifically, this step receives preoperative images and a surface mesh from the training subject as input. The preoperative images include layered slice data covering the anterolateral thigh region and corresponding surface reference markers. The surface mesh consists of a continuous curved surface mesh generated by a surface scanning device. Specifically, the preoperative images are imported, formatted, and de-marked to establish a correspondence between image coordinates and surface coordinates. To ensure stable operation of subsequent individual deformable registration and steady-state projection initialization, further, the preoperative images undergo zonal labeling and hierarchical division, structurally marking the skin layer, superficial fascia layer, and intermuscular channels within the image domain; noise point removal and hole repair are performed on the surface mesh to maintain continuity at points of abrupt curvature changes. Understandably, to establish a deformable correspondence between the preoperative images and the surface mesh, several traceable markers are set on the surface, and corresponding anatomical reference points are simultaneously marked in the preoperative images, forming an initial correspondence set through mutual constraints. Subsequently, by employing a layer-by-layer dense sampling method, the distribution density of the initial correspondence set was refined, extending the correspondence information between the skin layer and the superficial fascia layer to the entire anterolateral thigh region. Based on the above preparations, the data pipeline required for individual deformable registration and steady-state projection initialization was constructed: firstly, control patches for local fitting were defined on the body surface mesh, and simultaneously, image subdomains matching the control patches were established in the preoperative images; then, local correspondence constraints were established between each pair of control patches and image subdomains in a progressively deeper order, forming a continuous deformable mapping globally. Through iterative and local refinement, the deformable mapping maintained a smooth connection at the boundaries of the control patches, and local buffer transitions were introduced in areas of abrupt changes in body surface curvature. Finally, the deformation field is output, which describes the point-by-point mapping relationship from the preoperative image space to the body surface grid space. It is stored together with the running parameters of individual deformable registration and steady-state projection initialization for subsequent steps. This output will serve as the unified input for subsequent steps of the present invention. The deformation field is directly provided to the error estimation process as a data source and as the initial reference for the body surface projection transformation in subsequent perforator detection and ultrasound fusion projection.

[0064] S120. Estimate the error of individual deformable registration and steady-state projection initialization from the deformation field to obtain the alignment quality index;

[0065] After obtaining the deformation field, this step uses the deformation field as input to estimate the spatial consistency and hierarchical continuity of the deformation correspondence, outputting alignment quality indicators. Specifically, representative sampling bands are selected on the body surface mesh based on the distribution of control areas. These sampling bands cover the raised areas of the skin layer, the flat areas of the superficial fascia, and the adjacent intermuscular communication areas, ensuring that the error estimates are comparable under different geometric and tissue environments. Subsequently, the deformation field is applied point-by-point to the structured markers of the preoperative image, generating a set of mapped positions, which is then compared with the corresponding reference set on the body surface mesh. During the comparison, the corresponding errors at different levels are classified and summarized to form fitting error entries describing the degree of surface adhesion, interlayer error entries describing interlayer continuity, and strain error entries describing local strain changes. Furthermore, for areas prone to drift, short-time windows are established in temporal order, and the fluctuation degree of the mapped position set within the short-time window is aggregated to obtain stability-related fluctuation entries. All of the above entries are merged with a unified identifier to form a set of alignment quality indicators. To ensure direct compatibility between the alignment quality index and subsequent steady-state projection, this step standardizes the description of the item set, clarifying the source, calculation order, and corresponding control area of ​​each item, and recording the hierarchical relationship and spatial coverage between items. Finally, the alignment quality index is output. This output is directly passed to the steady-state projection process in the data stream to guide the calculation of parameters such as projection update interval, drift compensation, and buffer window. Simultaneously, the alignment quality index and deformation field are deposited together in the data storage unit, providing a quality gating reference for the perforator detection and ultrasound fusion projection processes during surface projection, ensuring that subsequent surface projection operations are aware of the current individual's deformable registration and steady-state projection initialization state.

[0066] S130. The stability of individual deformable registration and steady-state projection initialization of the alignment quality index is obtained, and projection stability parameters are generated.

[0067] This step takes the alignment quality index as input and, under the operating environment of individual deformable registration and steady-state projection initialization, completes the systematic acquisition of steady-state projection-related parameters and outputs projection stability parameters. Specifically, firstly, based on the fitting error entries and inter-layer error entries in the alignment quality index, local control areas are grouped. Control areas with similar fitting degrees and inter-layer continuity are grouped into the same group, and a time-series buffer window is established within each group. Within the buffer window, the entry values ​​corresponding to the alignment quality index are read sequentially and correlated with the mapping results of the deformation field at the same time position to form a time-space joint reference sequence. Further, for the fluctuation entries in the reference sequence, the drift compensation amount is obtained by combining the control area and layer from which they originate, and a projection update interval coordinated with the fluctuation degree is set within the group. For control areas with boundary transitions, a buffer zone is established, and a buffer window scale adapted to the group is configured in the buffer zone so that the buffer zone can absorb the changes of adjacent groups within the projection update interval. Subsequently, a threshold set for steady-state projection is generated within each group. This threshold set constrains the triggering and freezing conditions for surface projection. The threshold sets of each group are then combined to form a total threshold table covering the entire control area. After the total threshold table is determined, the drift compensation, projection update interval, and buffer window for each group are summarized to generate projection stability parameters. To ensure direct connection with subsequent steps, this step, while generating the projection stability parameters, marks the correspondence between the parameters and control areas, levels, and buffer sidebands, and provides the reading order and triggering conditions for surface projection, enabling seamless invocation of perforator detection and ultrasound fusion projection. Finally, the projection stability parameters are output. The projection stability parameters are passed backward at the data level. In the surface projection stage, the projection stability parameters are invoked in the reading order to execute surface projection. In the perforator detection registration stage, the projection update interval and the total threshold table are referenced for gating during probe attitude alignment, and the drift compensation is combined with the drift compensation during projection correction calculation to update the surface projection. Simultaneously, the layered approach and instrument safety tunnel can read and control the cache windows and buffer sideband annotations related to the control area to maintain consistency with steady-state projection when generating the layered path constraint set. Microscopic anastomosis tactile closure and virtual perfusion can reuse the freezing conditions consistent with the buffer sidebands within the micro-segment. Complication event injection, strategy tree evaluation, intelligent coaching, and adaptive curriculum engine can reference the records of projection update intervals and drift compensation amounts on the review timeline to reproduce the operational status of individual deformable registration and steady-state projection initialization during training. At this point, the projection stability parameters, as a unified baseline for subsequent processes in this invention, are output and archived along with the deformation field and alignment quality indicators, forming stable references for penetrating bronchus detection and ultrasound fusion projection, layered approach and instrument safety tunnel, and microscopic anastomosis tactile closure and virtual perfusion.

[0068] The technical effects of this step can be summarized as follows: Under the premise of ensuring continuous correspondence between preoperative images and body surface mesh, a stable projection parameter is established through error estimation and stability calculation. This parameter can be directly called by body surface projection, perforator exploration registration, and projection correction calculation. This ensures that the reading, triggering, and updating of individual deformable registration and steady-state projection initialization in subsequent processes have consistent timing and constraints. In turn, this provides a stable baseline that can be continuously referenced for training stages such as layered approach and instrument safety tunnel, as well as microsurgical anastomosis visual-tactile closed loop and virtual perfusion.

[0069] Step S200 includes at least steps S210-S230:

[0070] S210. Based on the projection stability parameters, perform surface projection of perforator detection and ultrasound fusion projection to obtain the perforator projection set and the confidence map of the exit point.

[0071] After individual deformable registration and steady-state projection initialization are completed, the system reads projection stability parameters from the data storage unit as the gating input for this step. Simultaneously, it imports preoperative images and surface meshes and establishes time and coordinate markers consistent with the previous data archive, ensuring that the preoperative images, surface meshes, and projection stability parameters are in the same reference relationship. Specifically, the system performs layered analysis on the anterolateral thigh region of the preoperative images, sequentially completing the structured annotation of the skin layer, superficial fascia layer, and intermuscular channels. Under the constraints of the structured annotation, it traces the connectivity of continuous bright and dark channels with tubular morphology, forming a perforator candidate set. For entries with incomplete morphology or significant contradictions with existing anatomical courses, the system performs gating elimination based on the threshold table in the projection stability parameters, retaining entries that satisfy connectivity and course consistency. The retained entries are then processed into a skeletonized set of perforator skeletons. After completing the above analysis, the system calls the previously generated deformation field to map the point-by-point positions of the perforator skeleton set in the preoperative image space, placing the mapping results onto the corresponding positions on the body surface grid. The system uses drift compensation and projection update intervals in the projection stability parameters to implement temporal gating and freeze control on the mapping process, ensuring continuity during abrupt changes in body surface curvature or field of view switching. The system performs neighborhood smoothing and breakpoint suturing on the mapped trajectory to form a perforator projection set. Simultaneously, the system integrates the continuity of the perforator skeleton in the preoperative image, the stability during the body surface mapping process, and the matching degree with alignment quality indicators to generate perforation point position entries according to body surface coordinates, assigning a confidence scale to each entry to form a perforation point confidence map. After completing the above processing, the system writes the perforator projection set and the perforation point confidence map into a shared buffer, registering them with body surface coordinate indices and timestamps for direct reading during subsequent ultrasound examination and registration. The system also retains the call records of the projection stability parameters in this step for reference in subsequent correction calculations.

[0072] S220. Perform ultrasonic detection registration of the perforator discovery and ultrasonic fusion projection from the perforator projection set to obtain the measured perforator path and hit and duration indicators.

[0073] This step reads the perforator projection set and the exit point confidence map from the shared buffer, and simultaneously accesses the ultrasound image stream and probe pose information from the training site. The time sequence of the ultrasound image stream is synchronized with the time stamp of the perforator projection set using the projection update interval in the projection stabilization parameters. The probe pose and the body surface mesh are then surface-constrained to ensure that the ultrasound sectors and the body surface projections are consistent. Specifically, the system delineates several priority exploration sectors on the body surface mesh based on the exit point confidence map, and configures a matching sector-shaped scanning path for each priority exploration sector. During the scanning process, the system performs frame-sector binding between the ultrasound image frames and the probe pose within the same time period. Using the body surface trajectory of the corresponding sector in the perforator projection set as a priori, the tubular acoustic echoes within the image frames are bound and matched with the body surface trajectory frame by frame to obtain a candidate echo sequence. The system performs connectivity analysis and course consistency checks on candidate echo sequences, eliminating segments that deviate from the body surface trajectory by more than a threshold. Adjacent segments that meet the consistency criteria are spliced ​​together temporally and spatially to form a set of continuous echo segments covering the depth range of the sector. After splicing, the system projects the set of continuous echo segments onto a body surface reference to generate a measured perforator path. To output hit and duration metrics, the system records the duration from entering the sector to the first identification of the measured perforator path within each priority exploration sector, and counts the number of intersections between the measured perforator path and the corresponding body surface trajectory. These durations and intersection counts are aggregated in a unified format to form hit and duration metrics. When a continuous echo segment breaks within the same sector during the projection update interval, the system performs a one-time window translation based on the drift compensation in the projection stability parameters and then performs a re-verification. Segments that still do not meet the course consistency criteria after the re-verification are not included in the above metric statistics. After completing the above processing, the system writes the measured penetrating path, hit and duration indicators into the shared buffer and marks them according to the priority exploration sector so that they can be matched with the same name when calculating the correction amount; at the same time, it retains the synchronization and gating call records of the projection stabilization parameters in this step to realize the correspondence with subsequent updates.

[0074] S230. Calculate the correction amount for the measured perforation path by combining the perforation detection with the ultrasonic fusion projection, and generate the projection correction amount.

[0075] This step uses the measured penetrating path and hit / duration indices as input, and the penetrating projection set as a reference, to initiate the calculation process of the correction amount for the body surface projection. The system first pairs the measured penetrating paths with the penetrating projection set on the body surface mesh, using priority exploration sectors as units. If multiple measured segments exist within the same sector, the segment with the shorter duration and higher number of intersections in the hit / duration indices is determined as the main segment, and the other segments are registered as candidate segments. Within each sector, the system performs point-to-point comparison between the main segment and the corresponding body surface trajectory, records displacement entries in the tangential and normal directions, and performs window smoothing on the displacement entries based on the drift compensation amount in the projection stability parameters, obtaining a constrained offset entry set. For candidate segments, the system only retains the portion with high consistency with the main segment's movement after window smoothing and incorporates it into the offset entry set. After the set of offset entries is determined, the system converts them into adjustment entries for the surface mapping layer and fine-tuning entries for the shallow structure based on the spatial coverage of the entries within the sector. These two types of entries together form a candidate set for projection corrections. Subsequently, the system reads the quality entries related to the current sector from the alignment quality index, maps the quality entries to the candidate set one-to-one, deletes adjustment entries that conflict with the quality entries, and rearranges the application order of the remaining entries so that the adjustments are applied to the surface mapping layer first, and then to the shallow structure. The system applies the rearranged adjustment entries to the perforator projection set in batches, using the projection update interval in the projection stability parameters as the beat, to complete the batch update of the surface trajectory and the position of the exit point, thereby generating projection corrections. After each batch of applications is completed, the system writes an update marker to the shared buffer. The update marker includes the applied adjustment batch number and the corresponding sector, so that the layered approach and the instrument safety tunnel remain consistent with the surface projection state during subsequent readings. Furthermore, at the end of this step, the system performs a global stitching of the updated perforator projection set to ensure the continuity of cross-sector trajectories at the boundaries. The projection correction values ​​and the updated perforator projection set are archived together for direct reference when generating the layered path constraint set for the layered approach and instrument safety tunnel. This also provides an indirect reference for determining the range of the microscopic working area in the microsurgical anastomosis visual-tactile closed loop and virtual perfusion. Hit and duration indicators and update markers are registered in the review timeline for retrospective analysis of complication events, strategy tree evaluation, and for the intelligent coach and adaptive curriculum engine to ensure temporal consistency. At this point, the surface projection, ultrasound examination registration, and correction value calculation have completed the closed loop. The perforator projection set and the measured perforator path are aligned in time and space. The projection correction values ​​and update markers form a structured record that can be directly referenced in subsequent steps, running through the layered approach and instrument safety tunnel, as well as the subsequent processes of microsurgical anastomosis visual-tactile closed loop and virtual perfusion. This completes the construction and solidification of the technical processing link from s210 to s230.

[0076] Step S300 includes at least steps S310-S330:

[0077] S310. Based on the projection correction, the target anchor points of the layered approach and the instrument safety tunnel are generated to obtain the layered path constraint set.

[0078] After the preceding surface projection, ultrasound registration, and correction calculations are completed, the system first retrieves the perforator projection set and projection correction values ​​from the data archive as input for this step. It then verifies the consistency of these inputs with the deformation field, alignment quality index, and projection stability parameters generated by the preceding individual deformable registration and steady-state projection initialization, ensuring that subsequent anchoring and constraint construction remain consistent with the global reference. Specifically, the system marks candidate access routes on the surface mesh based on the surface trajectory of the perforator projection set. These candidate access routes spatially unfold along the skin layer, superficial fascia layer, and intermuscular channel. The system updates the trajectory of the candidate access routes at inflection points and bifurcations using the projection correction values, aligning the candidate access routes with the previously measured perforator paths. Furthermore, the system selects a start point, an end point, and a transition point as target anchor points within the candidate approach zone. The start point corresponds to the skin incision location, the end point points to the superficial fascia window adjacent to the perforation point, and the transition point is located near the geometric midline of the subfascial channel. The spatial positions of these target anchor points are mapped point-by-point from the deformation field to the current body surface coordinates, and time-stamped using the projection update interval of the projection stability parameter, ensuring that the anchor point set can be stably read during field of view switching and slight changes in body position. The system generates a layered channel description around the target anchor points, where the skin layer channel uses curved surface fitting entries to define the incision line direction, the superficial fascia layer channel uses strip-shaped buffer entries to define the retraction and hemostasis operation range, and the intermuscular channel uses tubular constraint entries to define the separation direction and allowed turning amplitude. Understandably, to ensure the operability of subsequent device navigation, the system establishes transition zones at adjacent levels for the aforementioned layered channels. These transition zones record level transition points, transition radii, and allowable attitude change ranges. For areas with vascular crossings, the system generates prohibited crossing entries and deceleration entries around the target anchor point for direct reference in subsequent risk assessment and dynamic radius calculation. Finally, the system structurally encapsulates the target anchor point, layered channels, transition zones, and prohibited crossing and deceleration entries into a layered path constraint set. This set is written to a shared buffer using surface coordinate indices, level indices, and time stamps, serving as input for the next step of risk assessment. Simultaneously, the system retains records of calls to the perforator projection set and projection correction values ​​in this step, enabling the calculation of the subsequent dynamic safety radius curve to trace the anchor point source and trajectory update batch.

[0079] S320. Perform risk assessment of layered access routes and equipment safety tunnels from the layered path constraint set to obtain dynamic safety radius curves and hazard zone coloring maps.

[0080] After obtaining the hierarchical path constraint set, the system conducts a risk assessment process on a hierarchical basis, calculating spatial proximity, constraint satisfaction, and transition continuity for skin layer channels, superficial fascia layer channels, and intermuscular channels, respectively. Specifically, the system first reads the target anchor points and hierarchical channel entries within the hierarchical path constraint set, establishes channel envelopes on the body surface mesh and anatomical reference model, and determines the initial danger zone based on prohibited crossing entries. Subsequently, the system combines the quality entries related to the current region in the preceding alignment quality index to gate the credibility of the envelope boundary, and the boundary points retained after gating serve as the benchmark point set for safety assessment. Further, the system statistically analyzes the distance and attitude change trends between the benchmark point set and adjacent critical structures within each level, and uses the hierarchical transition points and allowable attitude change ranges recorded in the transition zone to calculate attitude tolerance entries within and between layers. When a tolerance entry undergoes abrupt changes in time, the system performs a translation review of the time window based on the drift compensation amount of the projection stability parameters. Entries that still do not meet the continuity requirements after review are marked as warning entries. After completing the item statistics, the system jointly generates a dynamic safety radius curve by combining the reference point sets, tolerance items, and warning items at each level. The dynamic safety radius curve uses the target anchor point as an index and the time stamp as the order, providing the safety radius value for each assessment location and the radius contraction or relaxation rules during interlayer transitions. Simultaneously, the system generates a hazard zone color map within the overlay view of the body surface mesh and the 3D anatomical reference model, centered on prohibited crossing items and warning items. The hazard zone color map uses consistent coloring rules across different levels and adds contraction prompts consistent with the dynamic safety radius curve in the interlayer transition areas. Understandably, to ensure consistency in subsequent instrument navigation calls, the system generates a query index when outputting the dynamic safety radius curve. The query index binds the target anchor point, assessment location, time stamp, and corresponding radius values; when outputting the hazard zone color map, it aligns it with the hierarchical index of the layered path constraint set for layer-by-layer overlay within the view. Finally, the system writes the dynamic safety radius curve and the hazard zone color map into the shared buffer, and associates the version information of the hierarchical path constraint set in the record, so that the instrument navigation in the next step can perceive the correspondence between the constraint set and the safety radius when reading.

[0081] S330. Perform layered approach and instrument safety tunnel navigation on the dynamic safety radius curve, and generate instrument and target distance sequence and boundary crossing event annotation.

[0082] After generating the dynamic safety radius curve and the hazard zone coloring map, the system enters the instrument navigation process. This step uses the dynamic safety radius curve as the main input and synchronously reads the hierarchical path constraint set and the hazard zone coloring map from the shared buffer as parallel constraints. Combined with the instrument pose information and field of view status information collected on the training site, frame-by-frame navigation and recording are performed. Specifically, the system first locates the current target anchor point and its level based on the hierarchical path constraint set, and retrieves the safety radius value and time stamp matching the target anchor point according to the query index. The safety radius value is then bound to the position of the instrument tip in the instrument pose information to form a safety radius entry for the navigation frame. At the same time, the system retrieves the hazard block corresponding to the current field of view in the hazard zone coloring map and uses it as a prohibited entry for the navigation frame. Subsequently, the system calculates the distance from the instrument tip to the target structure within the navigation frame and defines the following direction using the tangential direction of the curve between the target anchor point and the layered channel. Distance recording points are formed on both the tangential and normal components, continuously recording the instrument-target distance sequence. When a distance recording point does not satisfy the inclusion relationship with a safety radius entry at the same time marker, or overlaps with a prohibited entry, the system generates a boundary violation event annotation at that time marker, recording the corresponding layer, the corresponding target anchor point, the violated radius entry or prohibited entry, and the instrument's attitude information at that time. Further, during layer switching, the system reads the transition zone information from the layered path constraint set, gates the allowable range of instrument attitude changes, and updates the current safety radius entry using the corresponding radius contraction or relaxation rules in the dynamic safety radius curve, thus ensuring continuous recording of the instrument-target distance sequence during layer transitions. When the instrument experiences multiple boundary violation event annotations within a short period, the system merges adjacent annotations into an event cluster and binds the event cluster to the corresponding channel segment of the layered path constraint set, providing retrieval clues for the subsequent definition and review of micromanipulation sections. During navigation, the system writes the instrument-target distance sequence using both fixed time steps and field-of-view refresh steps. After each write, it synchronizes the reference status of the current dynamic safety radius curve and hazard zone coloring map with a version number, ensuring that subsequent reading steps can restore the constraint environment at the time. Finally, at the end of this step, the system outputs the instrument-target distance sequence and boundary violation event annotations, and archives them together with the version information of the hierarchical path constraint set, dynamic safety radius curve, and hazard zone coloring map. This serves as direct input for subsequent microscopic manipulation records and morphological evaluations of microscopic anastomosis-tactile closure and virtual perfusion, and also as a reference for complication event injection and strategy tree evaluation to construct event triggering time limits.

[0083] Step S400 includes at least steps S410-S430:

[0084] S410. Based on the instrument-target distance sequence, perform microscopic anastomosis visual-touch closed loop and virtual perfusion microscopic manipulation records to obtain needle distance and needle width deviation.

[0085] After the layered approach and instrument safety tunnel are completed, the system synchronously reads the instrument-target distance sequence and dynamic safety radius curve from the shared buffer according to time stamps and hierarchical indexes. These two are established as the same reference relationship as the input for this step. Simultaneously, the system reads the layered path constraint set and the danger zone color map as parallel constraints to ensure that the spatial definition of the microscopic working area is consistent with the preceding navigation state. Specifically, during the microscopic field of view initialization phase, the system acquires the microscope object distance, magnification, and field of view attitude, fixes the microscopic working area in body surface coordinates, and binds the microscopic field of view to the instrument tip trajectory in the instrument-target distance sequence frame by frame, ensuring that the subsequently acquired entry and exit points fall within the replayable field of view coordinates. To form a visual-tactile closed loop, the system deploys a tactile module on the microscopic operating table to collect the contact response during needle puncture, traction, and knotting. The resistance change output by the tactile module is time-stamped and synchronized with the microscopic field of view frame sequence; both together constitute the basic stream of the microscopic manipulation record. Subsequently, the system automatically identifies keyframes of needle-tissue interaction within the microscopic field-of-view frame sequence, extracts the entry and exit points of each needle insertion, and forms a needle point set. A suture segment set is generated between these needle point sets. The geometric description of the suture segment set is paired with the distance value from the instrument tip to the target structure at the corresponding moment in the instrument-target distance sequence to determine whether the needle insertion is within the safe radius specified by the dynamic safe radius curve. For needle insertions outside the safe radius, the system marks the time position and hierarchical information of the needle insertion in the micromanipulation record, and registers the corresponding keyframe and tactile resistance peak for subsequent morphological evaluation and rhythm analysis. The system performs geometric quantization on the needle point set, calculates the distance between adjacent needle points as the needle distance, and uses the lateral offset of the needle point relative to the suture segment set as the needle amplitude. The needle distance and needle amplitude form a needle distance-spindle amplitude sequence under the same reference and are checked for consistency with the aforementioned parallel constraints. After verification, the system aligns the needle spacing and width sequence with changes in tactile resistance on a continuous needle-by-needle basis, writing this information sequentially into the micromanipulation record. Finally, the needle spacing and width deviation is generated from the micromanipulation record. This deviation is represented by entries indicating the difference between the needle spacing and width for each needle-by-needle and the standard solution, and is bound to a timestamp, hierarchical index, needle sequence number, and corresponding instrument tip posture. After completing the above processing, the system archives the micromanipulation record and the needle spacing and width deviation as the output of this step. The needle spacing and width deviation will directly serve as the input for the next step, while the micromanipulation record serves as the parallel basis for morphological evaluation in the next step and rhythmic analysis at the end of this step. Simultaneously, the reference status of the instrument-target distance sequence and the dynamic safety radius curve is written into the record header to maintain traceability with previous steps and to reserve reference clues for timeline alignment with subsequent complication event injection and strategy tree evaluation.

[0086] S420. Based on the needle spacing and needle width deviation, perform morphological evaluation of microscopic anastomosis, visual and tactile closed loop and virtual perfusion to obtain patency score and leakage risk prediction.

[0087] After obtaining the needle spacing and needle width deviations, the system initiates the morphological evaluation process using the micromanipulation record as parallel input. Specifically, the system first reads the set of needle points and suture segments from the micromanipulation record, reconstructs the entry and exit points of each stitch into a local anastomosis geometric description, and generates anastomosis contour segments stitch by stitch using the needle spacing and needle width sequence as an index. Each contour segment is spliced ​​and sutured according to time markers to form an anastomosis geometric continuum. The system extracts morphological items such as inner diameter changes, wall adhesion, and knot distribution around the anastomosis geometric continuum, and binds each morphological item to the needle spacing and needle width deviation to present a one-to-one correspondence between "geometry and deviation" in subsequent evaluations. To perform virtual perfusion, the system sets proximal and distal perfusion inlets and outlets on the anastomosis geometric continuum. Using changes in tactile resistance recorded in micromanipulation as a reference for wall compliance, the system divides the anastomosis geometric continuum into several sub-segments. Within each sub-segment, the system constructs connectivity entries and pathway hindrance entries based on changes in inner diameter and knot distribution. The former describes the connectivity between segments, while the latter describes possible local leakage paths. Subsequently, the system performs perfusion calculations segment by segment in chronological order: when a connectivity entry satisfies continuity, the system records the perfusion passage status of that segment; when a pathway hindrance entry is dominant, the system records the perfusion restricted status of that segment. For segments with leakage paths, the system generates leakage candidate regions on the surface of the anastomosis geometric continuum and writes the time location and corresponding needle count of these candidate regions into an intermediate table. After aggregating the passage and restricted states of each segment on the time axis, the system forms a set of patency entries. The patency score is obtained by combining the stability of this set on the time axis with its spatial coverage. Each leakage candidate area is cross-checked spatially with the suture segment set and temporally with the peak tactile resistance. The candidate areas retained after cross-checking are used as leakage risk entries. The system classifies leakage risk entries according to severity and aggregates them to obtain leakage risk predictions. To ensure consistency with previous constraints, the system compares the dynamic safety radius curve with the safety radius entries throughout the morphological assessment: for stitches near the radius boundary, the system marks their corresponding morphological entries as boundary entries for use in the next step of rhythm analysis and subsequent complication event injection and strategy tree evaluation. Finally, the system outputs a patency score and leakage risk prediction as the results of this step, and archives them together with the micromanipulation record, needle distance and width deviation, and boundary entries. The patency score and leakage risk prediction will serve as direct inputs for the next step, and will also be the basis for event setting for complication event injection and strategy tree evaluation, maintaining consistency with the entire process.

[0088] S430. Perform microscopic matching visual-touch closed-loop and virtual perfusion rhythm analysis on patency score and leakage risk prediction to generate microscopic manipulation rhythm index.

[0089] After obtaining the patency score and leakage risk prediction, the system calls upon the micromanipulation record as parallel input to initiate the rhythm analysis process. Specifically, the system uses the time markers in the micromanipulation record as the main axis, mapping the needle point set, suture segment set, and tactile resistance changes to the same time axis, and forming operation segments according to the number of needles. Each operation segment includes five sub-steps: needle holding, puncture, traction, knotting, and loosening. The system extracts the interval, duration, and peak response time between sub-steps to form rhythm entries. Subsequently, the system aligns the rhythm entries with the patency score on the same time axis: when the patency score is in a low value range, the system retrieves the rhythm entries of the corresponding segment and marks them as rhythm abnormality candidates; simultaneously, the system aligns the rhythm entries with the leakage risk prediction: when the leakage risk prediction shows the existence of leakage candidate areas, the system retrieves the rhythm entries of the corresponding needle count and marks them as leakage-associated candidates. For the two types of candidate entries mentioned above, the system further compares the boundary entries in the dynamic safety radius curve with the boundary entries in the micromanipulation record. If the candidate entry and the boundary entry overlap or are adjacent in time, the system classifies them as boundary-related rhythm entries and writes the corresponding target anchor point and hierarchical index into the structured description of the rhythm entry to maintain consistency with the hierarchical path constraint set. The system performs time aggregation and needle aggregation on all rhythm entries, calculates the rhythm stability across needles and the smoothness of segment switching, and marks repetitive rhythm anomalies occurring in a short period of time as event clusters. The event clusters are bound to the keyframe positions in the micromanipulation record for direct location in subsequent review and strategy setting. Ultimately, the system jointly expresses rhythmic entries, event cluster tags, and boundary-related rhythmic entries as micromanipulation rhythm indicators, and writes them into the archive using time stamps and needle sequence numbers as indexes. The micromanipulation rhythm indicators, along with patency scores and leakage risk predictions, are written into a shared buffer for complication event injection and strategy tree evaluation to set time limits and prioritize events in event setting and intervention records, and also for intelligent coaches and adaptive curriculum engines to retrieve and reference in competency profile construction and recipe generation.

[0090] Step 500 includes at least steps S510-S530:

[0091] S510. Based on the patency score and leakage risk prediction, set the events for complication event injection and strategy tree evaluation to obtain a list of complication events and trigger time limits;

[0092] After the microscopic anastomosis and virtual perfusion are completed, the system automatically archives the patency score and leakage risk prediction as input for this step according to the time stamp and needle sequence number, and calls the microscopic manipulation record and microscopic manipulation rhythm index in parallel to ensure that the event settings are consistent with the geometric shape, perfusion state and manipulation rhythm.

[0093] The patency score specifically includes the aggregated results of perfusion passage and restricted states generated segment by segment based on morphological items such as the change in the internal diameter of the anastomotic geometric continuity, wall fit, and suture knot distribution. The quantitative value is calculated by comprehensively considering the stability on the time axis and the spatial coverage. The leakage risk prediction specifically includes candidate leakage areas identified through virtual perfusion calculations and their severity classifications. This prediction result is verified by cross-checking the spatial location of the suture segment set with the temporal location of the peak tactile resistance, and by comparing the boundary with the safety radius item of the dynamic safety radius curve. Both types of data are accompanied by time stamps, hierarchical indexes, and corresponding needle number sequences, collectively forming the core criteria for triggering complication events.

[0094] Specifically, the system first delineates candidate event regions on the geometric continuum of the anastomosis based on the leakage risk entries in the leakage risk prediction and their corresponding time positions. Then, combining this with low-value segments of the patency score, it identifies the types of complications that are preferentially triggered within the candidate regions. These complications include restricted backflow, worsening leakage, vasospasm, and unstable perfusion. The system binds these complication types with the candidate regions, time positions, and needle sequence numbers to form an event draft. To ensure that event triggering aligns with the trainee's operational rhythm, the system further compares the rhythm entries in the micromanipulation rhythm index with the time positions in the event draft: when a rhythm entry shows a segmental switch or a short pause, the system marks a rhythm-sensitive marker on the event draft for rhythm alignment during the injection phase. Subsequently, the system uses the preceding dynamic safety radius curve and danger zone coloring map to add out-of-bounds gating entries to the complication types in the event draft that may cause out-of-bounds events, automatically coordinating the event triggering with changes in the safety radius. After completing the above annotations, the system refines each event draft into complication event entries. Each complication event entry includes at least the event type, site of action, applicable level, triggering condition, duration window, and termination condition. It also generates trigger time limits for the timing requirements in the triggering conditions. For events requiring alignment with micromanipulation rhythms, the system records the alignment relationship with the corresponding rhythm entry within the trigger time limit. The system versionizes all complication event entries, forming a complication event list. Source references are added to the list header, including the time range for patency scores and leakage risk prediction, the referenced version of the micromanipulation rhythm indicators, and the referenced version of the dynamic safety radius curve, ensuring the traceability of subsequent intervention records and performance analysis. Finally, the system outputs the complication event list and trigger time limits as the results of this step and writes both to a shared buffer as input for the next step.

[0095] S520. Intervention records for complication event injection and strategy tree evaluation from the complication event list are used to obtain treatment path, time and effect data;

[0096] After obtaining the list of complication events and trigger time limits, the system enters the event injection and recording process. The list of complication events is loaded into the mixed reality scene according to time stamps and applicable levels. Simultaneously, the system reads the micromanipulation records and the instrument-target distance sequence as a monitoring stream to continuously record the trainee's actions after an event is triggered. Specifically, the system cyclically monitors the trigger time limit and current time location during scene operation. When the trigger conditions are met, scene changes matching the event type are injected into the corresponding action site: for restricted reflux, the system reduces distal perfusion performance in the virtual perfusion field of view and overlays a slowly expanding prompt at the edge of the danger zone color map; for aggravated leakage, the system generates a visual surge of leakage candidate areas on the surface of the anastomotic geometric continuum and gradually expands the visible range; for vasospasm, the system generates a rapidly contracting channel performance at the proximal end of the perforator channel and briefly changes color; for unstable perfusion, the system alternately changes the pathway state at the proximal and distal ends and introduces a short flash. All these changes are aligned with the rhythm-sensitive marker in the trigger time limit, ensuring that the injection time is connected to the trainee's segment switching or pause. Upon event injection, the system immediately activates the strategy tree evaluation interface, presenting feasible treatment branches in a hierarchical manner, including suture loosening, additional suturing, irrigation and dilation, repositioning, diversion and drainage, and pathway replacement. The system captures the treatment branch selected by the trainee through voice and gesture input, and records the selection time, duration, and changes in the posture of adjacent instruments at the branch node. During treatment execution, the system simultaneously collects changes in patency scores within a short time window, increases or decreases in the range of leak candidate areas, changes in boundary crossing markers near the safe radius boundary of the instrument-target distance sequence, and changes in the rhythm entries of micromanipulation rhythm indicators within the time period. These collected results are written into the intervention record, corresponding one-to-one with the treatment branch nodes. To form a coherent treatment path, the system connects the branch nodes chronologically within the entire duration window of each event, generating a treatment path, and marking the start node, critical node, and termination node on the path. When an event is resolved prematurely or fails to resolve beyond the trigger time limit, the system records an early termination marker or an overdue marker, respectively. Upon event closure, the system summarizes the event's time and impact data. Time entries include the trigger time, initial selection time, critical node interval, and total treatment duration. Impact entries include the segmental change in patency score, the final extent of the leak candidate area, the addition or removal of boundary violation event markers, and the restoration of rhythm entries. Finally, the system writes the treatment path, time, and impact data as output to a shared buffer, and adds a list of complication events and the version number of the trigger time limit to the record header for alignment and reference in subsequent performance analysis steps.

[0097] S530: Perform performance analysis on the treatment path, time and effect data by injecting complication events and evaluating the strategy tree, and generate a treatment strategy performance matrix and a review timeline;

[0098] After obtaining the treatment path, time, and effect data, the system initiates a performance analysis process and concurrently reads the complication event list, trigger time limits, micromanipulation records, and micromanipulation rhythm indicators to form a unified reference across events, treatment branches, and time periods. Specifically, the system first aligns treatment paths for similar events using event type as the primary index, mapping the start, key, and end nodes of each path to a unified node sequence. On this unified node sequence, the system aggregates time and effect entries node by node to obtain a set of node-level performance entries, and gating the set with the temporal constraints in the trigger time limits, marking nodes that meet or do not meet the time limits. Subsequently, based on the set of node-level performance entries, the system constructs performance entry combinations according to event type and treatment stage, binding the time and effect entries in each combination accordingly to form a strategy-level performance description. For treatment paths with early termination or overdue markers, the system classifies them as special combinations and writes the differences between the trigger and release conditions in the performance description. To create a searchable result structure, the system organizes the aforementioned strategy-level performance descriptions into a performance matrix based on event type (rows) and treatment steps (columns). Each cell in the matrix stores the number of nodes, time entries, and effect entries for the corresponding combination, and establishes a cross-index with the referenced micromanipulation rhythm indicators to directly locate specific segments during debriefing. Simultaneously, the system uses time stamps as the main axis, arranging the trigger time, treatment branch switching time, key node time, and event closure time of each event in sequence. It also overlays changes in patency scores, leakage candidate area ranges, and boundary crossing event labels at the corresponding times to generate a debriefing timeline. For continuous treatment segments across events, the system marks them with segment tags on the debriefing timeline and binds these segment tags to event identifiers in the complication event list to support time-series playback across events. Before outputting, the system adds source references to the treatment strategy performance matrix and debriefing timeline, including version numbers for the treatment path, time, and effect data; version numbers for the complication event list and trigger time limits; and version numbers for the micromanipulation records and micromanipulation rhythm indicators. This ensures that the intelligent coach and adaptive curriculum engine can directly access these references during competency profile construction and recipe generation. Finally, the system outputs the treatment strategy performance matrix and debriefing timeline as the results of this step and archives them in a shared buffer for retrieval in the next process.

[0099] Step S600 includes at least steps S610-S630:

[0100] S610: Obtain the performance matrix of the handling strategy and the review timeline, construct the capability profile of the intelligent coach and adaptive curriculum engine, and obtain the capability radar chart.

[0101] After the complication event injection and strategy tree evaluation are completed, the system synchronously reads the treatment strategy performance matrix and the debriefing timeline from the shared buffer according to the event type index and time stamp index as input for this step. It then aligns these with the previously archived hierarchical path constraint set and target anchor point index to ensure consistent association between behavioral performance and timeline position, spatial hierarchy, and channel segment. Specifically, the system first extracts time entries and effect entries under a unified node sequence based on the event type, treatment stage, and node-level performance entries in the treatment strategy performance matrix. It then establishes time-series segments based on the trigger time, branch switching time, and event closing time of the debriefing timeline, mapping each treatment path to a set of continuous segments. Within each segment set, the system sequentially superimposes node-level performance entries into segment-level performance entries and establishes bidirectional references in the debriefing timeline based on the segment start and end times, thus forming a one-to-one correspondence index between the matrix and the timeline. Subsequently, based on the indicator domain for capability profile construction as described in the claims, the system maps segment-level performance items to capability element items. These capability element items include at least instrument path control, hierarchical channel compliance, boundary violation avoidance, accuracy of complication management, timeliness of complication management, and consistency of strategy switching. To maintain consistency with preceding navigation and alignment assessments, the system adopts a standardized domain for the value domain of each capability element item, consistent with the dynamic safety radius curve and accessibility score, resulting in a capability element sequence. Further, the system performs smoothing and alignment along the segment boundaries on the timeline of the capability element sequence. During smoothing, key nodes in the review timeline are used as stopping points to avoid incorrectly incorporating non-operational intervals across nodes into capability estimation. For handling paths with early termination or overdue markers, the system sets segment occlusion markers in the capability element sequence, ensuring that subsequent profile summarization only includes effective operator behavior segments. After alignment, the system aggregates the capability element sequences into a capability profile set, using event type as the hierarchical key and handling stage as the aggregation key. The set retains the time range and spatial hierarchy index of the original segments. Based on this, the system projects the capability profile set onto a multi-dimensional coordinate domain according to capability profile construction rules, generating a capability radar chart. The radar chart records the summary value, dispersion, and confidence level of each capability element item, and includes references to the handling strategy performance matrix and the review timeline in its graphical structure to ensure that subsequent searches can trace back to specific segments and nodes. The system then outputs the capability radar chart as the result of this step. This radar chart will serve as input for the next step, with version information for the handling strategy performance matrix and the review timeline written into the record header to ensure consistency in the basis for subsequent recipe generation.

[0102] S620: Generate recipes for intelligent coaching and adaptive curriculum engines from the capability radar chart to obtain individual training recipes and retraining parameters.

[0103] After obtaining the capability radar chart, the system initiates the recipe generation process. Using the capability element entries in the radar chart as the primary index, and combining the reference identifiers of the aforementioned event types and handling stages, the system locates the segment and node range where the weak elements are located. Specifically, the system first analyzes the summary value and dispersion of each capability element entry on the capability radar chart, identifies the set of capability element entries that need priority training, and reads the segment time range and spatial hierarchy index associated with this set. The system then compares the version numbers of the previously archived hierarchical path constraint set, dynamic safety radius curve, and through-projection set to verify the availability of these basic references within the corresponding segments of the weak elements. After verification, the reference identifier is written into the combination configuration of this step. Once the combination configuration is established, the system generates individual training recipes driven by the capability element entries. Each individual training recipe includes the level structure, scenario parameters, and evaluation item bindings. The system employs a scene slicing and reorganization mechanism based on the segments and nodes corresponding to weak elements. When the capability element is instrument path control, the system selects segments of the same level and similar channel curvature from the hierarchical path constraint set to generate continuous sub-levels for instrument navigation and micro-operations. When the capability element is the timeliness of complication treatment, the system selects trigger-release node sequences of similar events from the treatment strategy performance matrix, replays and injects complication events of adjacent intensity levels, forming time-limited treatment sub-levels. When the capability element is the coherence of strategy switching, the system cross-connects multiple treatment branches at nodes to form multi-branch switching sub-levels. Scenario parameters are set according to the value range and dispersion of the capability element: for elements with high dispersion, the system enables rhythmic perturbation and slight changes in the field of view within the sub-level; for elements with low aggregate values, the system reduces the intensity of complications and extends the allowed treatment window to form a progressive training ladder. The evaluation item binding consists of three parts: first, capability element items are retrieved from the capability radar chart as level achievement items; second, key nodes are mapped from the review timeline as process items; and third, strategy-level performance descriptions are extracted from the handling strategy performance matrix as scenario items. These three elements form a complete evaluation structure within the level template. Simultaneously with the generation of individual training recipes, the system derives operational parameters related to the base reference to form retraining parameters. These retraining parameters are used for cross-process write-back and influence the initialization and threshold settings of the next training iteration. The retraining parameters include at least the projection update interval, the sampling density of the deformation field initialization point set, the alignment quality index threshold, the target range of the projection stability parameters, the confidence threshold of the perforator projection set, the initial radius and contraction rule of the dynamic safety radius curve, the recording step size of the instrument-target distance sequence, and the tactile resistance sampling step size of the micromanipulation record. After generation, the above retraining parameters and individual training recipes are uniformly written into the output structure. The output structure is referenced in the head illumination capability radar chart and the version mapping relationship with the hierarchical path constraint set, the dynamic safety radius curve and the perforator projection set is also registered to ensure accurate matching during write-back processing.Finally, the system outputs the individual training formula and retraining parameters as the results of this step, which can be used in the next step.

[0104] S630: Perform intelligent coaching and adaptive curriculum engine write-back processing on the retraining parameters to generate data configuration;

[0105] After obtaining the retraining parameters, the system initiates the write-back process, using the joint output of the retraining parameters and the individual training recipe as input for this step. Based on the version mapping relationship of the base reference, the parameters are categorized into three types of data configurations: projection configuration, registration configuration, and acquisition configuration. Specifically, the system first parses the projection configuration, mapping the projection update interval, the target interval of the projection stability parameters, and the confidence threshold of the perforator projection set in the retraining parameters to the initialization key values ​​of the surface projection module. A lookup table is then established between these parameters and the target interval of the projection stability parameters, allowing the update interval and threshold range to be read before the surface projection generation and correction calculation begin after the write-back. Subsequently, the system analyzes the registration configuration, writing the sampling density of the deformation field initialization point set and the alignment quality index threshold from the retraining parameters into the initialization node and quality gating node of the individual deformable registration. In the write-back entries, a regional density table is established according to the body surface grid region, enabling the next round of deformation field calculation to use differentiated densities by region, and to stop early or continue iteration at the alignment evaluation node based on the threshold. Simultaneously, the target interval of the projection stability parameter is incorporated into the stability calculation process of steady-state projection initialization to set the allowable range for stability convergence. Further, the system analyzes the acquisition configuration, writing the initial radius and contraction rule of the dynamic safety radius curve into the generation nodes of the layered approach and instrument safety tunnel, so that the initial safety radius can be calculated after the target anchor point is generated, and the contraction rule can be invoked during risk assessment. The recording step size of the instrument-target distance sequence and the tactile resistance sampling step size of the micromanipulation record are written into the data acquisition node, ensuring that the temporal resolution of subsequent navigation and microscopic recording remains consistent with the individual training recipe. To ensure the atomicity and traceability of the write-back, the system calculates a reference snapshot for each configuration entry after generating the data configuration. This snapshot records the source capability element entry, the corresponding level, and the time range of the segment. It is then bound to the version numbers of the aforementioned hierarchical path constraint set, dynamic safety radius curve, and cross-branch projection set, forming a rollback-compatible and comparable data configuration version. During the write-back, the system sequentially pauses the entry queues for individual deformable registration and steady-state projection initialization, writing the projection configuration, registration configuration, and acquisition configuration to the corresponding nodes in dependency order. After writing is complete, the queues are restored, and the time stamp of this write-back and the target module's acceptance confirmation are recorded in the log. At this point, the system outputs the data configuration and completes the write-back to individual deformable registration and steady-state projection initialization. The data configuration, as the result of this step, is simultaneously archived in a shared buffer for direct reading during subsequent training rounds at startup. It also serves as the historical write-back basis for the intelligent coach and adaptive curriculum engine in version selection for subsequent capability profile construction.

[0106] Example 2: Figure 2 A structural block diagram of a mixed reality-based anterolateral thigh free flap surgical training system according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0107] The registration initialization module 01 is used to acquire preoperative images and surface meshes, perform zonal labeling and hierarchical division of the skin layer, superficial fascia layer, and intermuscular channels, and establish local correspondence constraints and deformable mappings between the control area and the image subdomain to generate a deformation field. Specifically, it receives preoperative images and surface meshes from the training subject as input. The preoperative images include layer-by-layer slice data and surface reference markers, and the surface mesh is composed of a continuous curved surface mesh generated by a surface scanning device. The preoperative images undergo import, format unification, and marker removal processing to establish the correspondence between image coordinates and surface coordinates. Zonal labeling and hierarchical division are performed on the preoperative images, structurally marking the skin layer, superficial fascia layer, and intermuscular channels within the image domain. Noise point removal and hole repair are performed on the surface mesh to maintain continuity at points of abrupt curvature changes. Several traceable marker points are set on the surface, and corresponding anatomical reference points are simultaneously marked in the preoperative images to form an initial correspondence set. By employing a layer-by-layer dense sampling method, the distribution density of the initial correspondence set is improved, extending the correspondence information to the entire anterolateral thigh region. A data pipeline is constructed: control regions are defined on the body surface grid, and image subdomains are established in the preoperative images; local correspondence constraints are established between each pair of control regions and image subdomains, forming a continuous deformable mapping. A deformation field is output, describing the point-by-point mapping relationship from the preoperative image space to the body surface grid space, and is passed to the error estimation module as a data source.

[0108] Error estimation module 02, connected to the registration initialization module, is used to compare the mapped position set with the reference set within the sampling band, generating fitting error, interlayer error, strain error, and fluctuation entries, and outputting an alignment quality index. Specifically, it receives the deformation field from the registration initialization module as input. Representative sampling bands are selected on the body surface grid according to the distribution of control areas, covering the raised areas of the skin layer, the flat areas of the superficial fascia, and the adjacent intermuscular communication areas. The deformation field is applied point-by-point to the structured markers of the preoperative image to generate a mapped position set, which is then compared with the corresponding reference set on the body surface grid. During the comparison, errors corresponding to different layers are classified and summarized to form fitting error entries, interlayer error entries, strain error entries, and fluctuation entries. For areas prone to drift, a short-time window is established in temporal order, and the fluctuation degree of the mapped position set within the short-time window is aggregated to obtain stability-related fluctuation entries. The alignment quality index, containing the entry set, is output and passed to the stability calculation module as input.

[0109] The stability calculation module 03, connected to the error estimation module, is used to group control areas based on fitting error and inter-layer error, and establish a buffer window, drift compensation amount, and projection update interval within each group to generate projection stability parameters. Specifically, it receives the alignment quality index from the error estimation module as input. Based on the fitting error and inter-layer error entries in the alignment quality index, local control areas are grouped, with control areas having similar fitting degrees and inter-layer continuity grouped together. A time-series buffer window is established within each group; within the buffer window, the entry values ​​corresponding to the alignment quality index are read sequentially and correlated with the mapping results of the deformation field at the same time position to form a time-space joint reference sequence. For fluctuation entries in the reference sequence, the matching drift compensation amount is calculated based on the control area and layer from which they originate, and a projection update interval coordinated with the fluctuation degree is set within the group. For control areas with boundary transitions, a buffer sideband is established, and a buffer window scale adapted to the grouping is configured within the buffer sideband. The projection stabilization parameters are generated, which include a threshold set, drift compensation amount, projection update interval and buffer window, and are passed to the body surface projection module as gating input.

[0110] The surface projection module 04, connected to the stability calculation module, is used to generate a perforator skeleton set under threshold gating and perform deformation field mapping. It employs drift compensation and projection update intervals for temporal gating and freeze control, outputting the perforator projection set and the confidence map of the exit point. Specifically, it receives projection stability parameters from the stability calculation module as input, and simultaneously imports preoperative images and a surface mesh. The anterolateral thigh region of the preoperative image is analyzed hierarchically, sequentially completing the structured annotation of the skin layer, superficial fascia layer, and intermuscular channels. Under the constraints of the structured annotation, continuous bright and dark channels with tubular morphology are traced for connectivity, forming a perforator candidate set. Gating is performed based on the threshold gating in the projection stability parameters, retaining entries that satisfy connectivity and course consistency. The retained entries are then skeletonized to generate the perforator skeleton set. The deformation field is invoked to map the point-by-point positions of the perforator skeleton assembly in the preoperative image space, and the mapping results are then mapped to the corresponding positions on the body surface grid. Temporal gating and freeze control are applied to the mapping process using drift compensation and projection update intervals in the projection stability parameters. The mapped trajectory is then smoothed in its neighborhood and broken points are sutured to form the perforator projection set. Simultaneously, considering the continuity of the perforator skeleton in the preoperative image, the stability during the body surface mapping process, and the matching degree with the alignment quality index, perforation point position entries are generated according to the body surface coordinates, and each entry is assigned a confidence scale to form a perforation point confidence map. The perforator projection set and the perforation point confidence map are output and transmitted to the ultrasound exploration module as input.

[0111] The ultrasound exploration module 05, connected to the body surface projection module, is used to delineate priority exploration sectors based on the penetration point confidence map, perform frame-sector binding, constraint matching, and travel consistency checks, and output the measured penetrating path and hit and duration indicators. Specifically, it receives the penetrating projection set and penetration point confidence map from the body surface projection module as input, and simultaneously accesses the ultrasound image stream and probe pose information from the training site. The time sequence of the ultrasound image stream and the time stamp of the penetrating projection set are synchronized using the projection update interval in the projection stability parameters. The probe pose and the body surface grid are surface-constrained to ensure that the ultrasound sector and the body surface projection are consistent. Based on the penetration point confidence map, several priority exploration sectors are delineated on the body surface grid, and a matching sector-shaped scanning path is configured for each priority exploration sector. During the scanning process, ultrasound image frames and probe poses are bound to specific sectors within the same time frame. Using the surface trajectory of the corresponding sector in the perforator projection set as a priori, the tubular acoustic echoes within the image frames are bound and matched frame-by-frame with the surface trajectory to obtain candidate echo sequences. Connectivity analysis and course consistency checks are performed on the candidate echo sequences, eliminating segments that deviate from the surface trajectory beyond a threshold. Adjacent segments that meet consistency are spliced ​​temporally and spatially to form a continuous echo segment set covering the depth range of the sector. This continuous echo segment set is projected onto a surface reference to generate the measured perforator path. The duration from entering the sector to the first identification of the measured perforator path is recorded, and the number of intersections between the measured perforator path and the corresponding surface trajectory is counted to form hit and duration indices. The measured perforator path and hit and duration indices are output and transmitted to the correction calculation module as input.

[0112] The correction calculation module 06, connected to the ultrasonic exploration module, is used to perform point-to-point comparison between the measured penetrating path and the penetrating projection set, using priority exploration sectors as units. After window smoothing, it generates an offset entry set and outputs the projection correction amount. Specifically, it receives the measured penetrating path and hit / duration indices from the ultrasonic exploration module as input, using the penetrating projection set as a reference. On the body surface mesh, using priority exploration sectors as units, the measured penetrating path and the penetrating projection set are paired with the same name: if multiple measured segments exist within the same sector, the segment with the shorter duration and higher number of intersections in the hit / duration indices is determined as the main segment, and the other segments are registered as candidate segments. Within each sector, the main segment and the corresponding body surface trajectory are compared point-to-point, and displacement entries are recorded according to the tangential and normal directions. The displacement entries are then window-smoothed based on the drift compensation amount in the projection stability parameters to obtain a constrained offset entry set. For candidate segments, only the portion with high consistency with the main segment's trajectory is retained after window smoothing and incorporated into the offset entry set. After the offset entry set is determined, based on the spatial coverage of the entries within the sector, they are converted into adjustment entries for the surface mapping layer and fine-tuning entries for the shallow structure. These two types of entries together form a candidate set for projection correction quantities. Quality entries related to the current sector are read from the alignment quality index, and each quality entry is matched one-to-one with the candidate set. Adjustment entries that conflict with quality entries are deleted, and the application order of the remaining entries is rearranged so that the adjustments first apply to the surface mapping layer and then to the shallow structure. Using the projection update interval in the projection stability parameters as the cycle, the rearranged adjustment entries are applied to the perforation projection set in batches, completing the batch update of the surface trajectory and the position of the exit point, generating projection correction quantities. The projection correction quantities are output and passed to the path constraint module as input.

[0113] The path constraint module 07, connected to the correction calculation module, is used to select target anchor points within the candidate entry zone and generate surface fitting entries, strip buffer entries, and tubular constraint entries, outputting a hierarchical path constraint set. Specifically, it receives the projection correction amount from the correction calculation module as input and performs consistency verification with the deformation field, alignment quality index, and projection stability parameters. Candidate entry zones are marked on the body surface mesh according to the surface trajectory of the perforator projection set. The candidate entry zones are spatially expanded sequentially along the skin layer, superficial fascia layer, and intermuscular channel. The inflection points and bifurcation points of the candidate entry zones are updated using the projection correction amount to align the candidate entry zones with the measured perforator paths. Within the candidate entry zone, a start point, an end point, and a transition point are selected as target anchor points. The start point corresponds to the skin incision location, the end point points to the superficial fascia window adjacent to the perforator exit point, and the transition point is located near the geometric midline of the subfascial channel. The spatial position of the target anchor point is mapped point-by-point from the deformation field to the current body surface coordinates and time-marked using the projection update interval of the projection stability parameters. A layered channel description is generated around the target anchor point. The skin layer channel uses curved surface fitting entries to define the cutting line direction; the superficial fascia layer channel uses strip-shaped buffer entries to define the retraction and hemostasis operation range; and the intermuscular channel uses tubular constraint entries to define the separation direction and allowed turning amplitude. Transition zones are established at adjacent layers, recording the layer transition points, transition radii, and allowed posture changes. For areas with vascular crossings, prohibition and deceleration entries are generated around the target anchor point. The output layered path constraint set, including the target anchor point, layered channels, transition zones, and prohibition and deceleration entries, is passed to the safe navigation module as input.

[0114] The safety navigation module 08, connected to the path constraint module, is used to calculate the dynamic safety radius curve and danger zone coloring map based on the hierarchical path constraint set, and bind safety radius entries and prohibited entry entries to perform device navigation, outputting the device-target distance sequence and boundary crossing event annotations; specifically, it receives the hierarchical path constraint set from the path constraint module as input. The risk assessment process is carried out hierarchically, calculating spatial proximity, constraint satisfaction, and transition continuity relationships for skin layer channels, superficial fascia layer channels, and intermuscular channels respectively. Target anchor points and hierarchical channel entries within the hierarchical path constraint set are read, and channel envelopes are established on the body surface mesh and anatomical reference model. Initial danger zones are determined based on prohibited crossing entries. The credibility of the envelope boundary is gating by combining quality entries related to the current region in the alignment quality index. The boundary points retained after gating are used as the benchmark point set for safety assessment. Within each level, the distance and attitude change trends between the reference point set and adjacent critical structures are statistically analyzed. Using the level transition points and allowable attitude change ranges recorded in the transition zone, attitude tolerance entries within and between levels are calculated. When a tolerance entry experiences abrupt changes in time, the time window is shifted and verified based on the drift compensation amount of the projection stability parameters. Entries that still do not meet the continuity requirements after verification are marked as warning entries. The reference point set, tolerance entries, and warning entries for each level are combined to generate a dynamic safety radius curve. The dynamic safety radius curve uses the target anchor point as an index and the time marker as the order, providing the safety radius value for each evaluation location and the radius contraction or relaxation rules during inter-level transitions. Simultaneously, using prohibited crossing entries and warning entries as the core, a danger zone color map is generated within the overlay view of the body surface mesh and the 3D anatomical reference model. The danger zone color map uses consistent coloring rules across different levels, and a contraction indicator consistent with the dynamic safety radius curve is added to the inter-level transition area. The dynamic safety radius curve and danger zone color map are output and used for instrument navigation, generating instrument-target distance sequences and boundary crossing event annotations.

Claims

1. A training method for anterolateral thigh free flap surgery based on mixed reality, characterized in that, include: Preoperative images and body surface meshes are acquired, and the skin layer, superficial fascia layer and intermuscular channel are partitioned and hierarchically divided. Local correspondence constraints and deformable mappings are established between the control area and the image subdomain. Then, error estimation and stability are performed, individual deformable registration and steady-state projection initialization are performed, and projection stability parameters are generated. Based on the projection stability parameters, perforator detection and ultrasound fusion projection are performed to generate projection correction values. Based on the projection correction, a layered approach and instrument safety tunnel are constructed, and a distance sequence between the instrument and the target is generated. Based on the distance sequence between the instrument and the target, microscopic anastomosis visual-touch closed loop and virtual perfusion are performed to generate microscopic manipulation rhythm indicators; Based on the micromanipulation rhythm index, complication events are injected and strategy tree evaluation is performed to generate a treatment strategy performance matrix and a review timeline. The system acquires a performance matrix of handling strategies and a review timeline. Based on event type and handling process, it aggregates the sequence of capability elements into a set of capability profile items, performs scene slicing and reorganization to bind level structure, scenario parameters and evaluation items, then generates recipes and writes them back, and uses an intelligent coaching and adaptive curriculum engine to generate data configuration.

2. The method according to claim 1, characterized in that, The process of generating projection stability parameters includes: Preoperative images and body surface meshes were acquired. Several traceable markers were set on the body surface and corresponding anatomical reference points were simultaneously marked in the preoperative images to form an initial correspondence set. Local correspondence constraints were established in order from superficial to deep. The skin layer, superficial fascia layer and intermuscular channels were partitioned and hierarchically divided. Local correspondence constraints and deformable mappings were established between the control area and the image subdomain. Data preparation was carried out for individual deformable registration and steady-state projection initialization to obtain the deformation field.

3. The method according to claim 2, characterized in that, The process of generating projection stability parameters includes: The location set of the representative areas covered by the deformation field in the sampling zone, including the raised area of ​​the skin layer, the flat area of ​​the superficial fascia, and the adjacent intermuscular communication area, is compared with the reference set. The errors corresponding to different levels are classified and summarized, and a short time window in time sequence is established to aggregate the degree of fluctuation. Adhesion error, interlayer error, strain error and fluctuation items are generated. Error estimation of individual deformable registration and steady-state projection initialization is performed to obtain the alignment quality index.

4. The method according to claim 3, characterized in that, The process of generating projection stability parameters includes: The alignment quality index is controlled by grouping the control areas based on the bonding error and interlayer error. A time-series buffer window is established in each group and the entry values ​​are read to form a time-space joint reference sequence. The drift compensation amount is calculated for the fluctuation entries and the projection update interval is set. The buffer window, drift compensation amount and projection update interval are established in the group. The stability of individual deformable registration and steady-state projection initialization is calculated to generate projection stability parameters.

5. The method according to claim 2, characterized in that, Preoperative imaging and surface mesh include: Preoperative imaging specifically includes medical image slice data covering the anterolateral thigh region, as well as coordinate information of reference markers distributed on the body surface; The surface mesh is specifically composed of a high-precision continuous curved surface mesh generated by optical or laser surface scanning equipment, and its vertex data includes spatial coordinates, normal vectors and local curvature attributes.

6. The method according to claim 1, characterized in that, The process of generating data configuration includes: The system acquires a performance matrix of handling strategies and a review timeline. Using event type as the hierarchical key and handling process as the aggregation key, it aggregates the sequence of capability elements into a set of capability profile items and projects them onto a multi-dimensional coordinate domain. Based on event type and handling process, the system aggregates the sequence of capability elements into a set of capability profile items, constructs capability profiles for intelligent coaching and adaptive curriculum engines, and obtains a capability radar chart.

7. The method according to claim 6, characterized in that, The process of generating data configuration includes: Based on the capability radar chart, the scene is sliced ​​and reorganized according to the segments and nodes corresponding to the weak elements. When the capability element item is the instrument path control, the same level and similar channel curvature segments are selected to generate continuous sub-levels. When it is the timeliness of complication treatment, the adjacent intensity level events are replayed and injected to form time-limited sub-levels. The level structure, scenario parameters and evaluation items are bound together to generate recipes for intelligent coaching and adaptive curriculum engines, resulting in individual training recipes and retraining parameters.

8. The method according to claim 7, characterized in that, The process of generating data configuration includes: The retraining parameters are merged according to the projection, registration and acquisition configurations. The projection update interval, the target interval of the projection stability parameters and the confidence threshold of the cross-branch projection set are mapped to the initialization key value of the body surface projection module. The sampling density of the deformation field initialization point set and the alignment quality index threshold are written into the individual deformable registration initialization node and the quality gate node, and written into the regional density table and the convergence allowable range. The data configuration is generated by writing back the intelligent coach and adaptive curriculum engine.

9. The method according to claim 1, characterized in that, The process of generating projection stability parameters also includes: Data preparation, error estimation, and stability determination for individual deformable registration and steady-state projection initialization are performed sequentially.

10. A mixed reality-based surgical training system for anterior lateral thigh free flap, applied to the method described in any one of claims 1-9, characterized in that, include: The registration initialization module is used to acquire preoperative images and body surface mesh, perform partitioning and hierarchical division of the skin layer, superficial fascia layer and intermuscular channel, and establish local correspondence constraints and deformable mapping between the control area and the image subdomain to generate a deformation field. The error estimation module, connected to the registration initialization module, is used to compare the mapped position set with the reference set within the sampling band, generate fitting error, interlayer error, strain error and fluctuation items, and output alignment quality index. The stability calculation module, connected to the error estimation module, is used to group the control area based on the bonding error and inter-layer error, and to establish a buffer window, drift compensation amount and projection update interval within the group to generate projection stability parameters. The surface projection module, connected to the stability calculation module, is used to generate the perforator skeleton set and perform deformation field mapping under threshold total table gating. It uses drift compensation and projection update interval to implement timing gating and freeze control, and outputs the perforator projection set and the confidence map of the perforation point. The ultrasound exploration module, connected to the body surface projection module, is used to delineate priority exploration sectors based on the confidence map of the penetration point, perform frame sector binding, constraint matching and course consistency check, and output the measured penetrating path and hit and duration indicators. The correction calculation module, connected to the ultrasonic exploration module, is used to compare the measured penetrating path with the penetrating projection set in the priority exploration sector as the unit, generate an offset entry set through window smoothing, and output the projection correction amount. The path constraint module, connected to the correction calculation module, is used to select target anchor points within the candidate entry zone and generate surface fitting entries, strip buffer entries, and tubular constraint entries, outputting a hierarchical path constraint set; The safety navigation module, connected to the path constraint module, is used to calculate the dynamic safety radius curve and hazard zone coloring map based on the hierarchical path constraint set, and bind safety radius entries and prohibited entries to perform instrument navigation, outputting the distance sequence between the instrument and the target and boundary crossing event annotations.

Images