Postoperative follow-up visit system and method for maxillofacial malformation orthognathic operation

Abstract

The invention relates to the technical field of maxillofacial deformity treatment, and discloses a maxillofacial deformity orthognathic post-operation follow-up visit system and method, the system comprises an image analysis model and a follow-up visit evaluation model, the image analysis model carries out multi-dimensional feature fusion processing on post-operation initial data, and the data comprises three-dimensional bone seam, occlusion function, soft tissue distribution and other feature data; and the follow-up assessment model performs dynamic interference suppression processing on the real-time physiological data stream and then inputs the feature generation layer to simulate a recovery track so as to generate a follow-up assessment instruction. The feature generation layer comprises an image preprocessing module and a dynamic feature generation module, and the dynamic feature generation module is obtained through combined training of multi-type historical postoperative data and real-time physiological parameters and comprises a feature analysis layer, a feature correction layer and an instruction triggering layer. According to the method, the system is applied, and multi-dimensional and intelligent follow-up assessment of postoperative recovery is achieved. According to the method, the accuracy and the real-time performance of follow-up visit are improved, and effective support is provided for postoperative recovery monitoring and intervention.

Description

Technical Field

[0001] The present invention relates to the technical field of maxillofacial deformity treatment, and specifically to a maxillofacial deformity orthognathic surgery follow-up system and method. Background Art

[0002] Orthognathic surgery is an important means for treating maxillofacial deformities. By adjusting the position and shape of the jawbone, it restores facial beauty and occlusal function. However, the postoperative recovery process of patients is complex, involving the dynamic evolution of physiological indicators in multiple dimensions such as suture healing, occlusal function reconstruction, and soft tissue morphological changes. Traditional follow-up methods have significant limitations.

[0003] From the perspective of data collection and analysis, traditional follow-up mainly relies on doctors' clinical examinations and patients' subjective feedback, lacking accurate quantitative analysis of multi-dimensional data such as three-dimensional suture characteristics, occlusal function parameters, and soft tissue distribution. For example, the dynamic changes in bone block displacement parameters and suture healing parameters are difficult to comprehensively capture through traditional palpation or two-dimensional imaging, resulting in the inability to detect suture healing abnormalities or occlusal function recovery deviations in a timely manner. At the same time, the original image data stream often contains a large amount of invalid signals, and traditional methods are difficult to efficiently complete anatomical region segmentation and noise filtering, affecting data accuracy.

[0004] In terms of the real-time and dynamic nature of follow-up evaluation, existing systems are unable to effectively process real-time physiological data streams. During the postoperative recovery process, small fluctuations in physiological parameters may indicate a deviation in the recovery trajectory. However, traditional methods lack a dynamic interference suppression mechanism and are unable to eliminate the interference of external factors on data collection in real time, resulting in a lag in evaluation. In addition, the ability to fuse and analyze multi-type feature data is insufficient, and the dynamic mapping relationship between each feature vector has not been fully modeled, making it difficult to simulate the real recovery trajectory and resulting in poor pertinence and accuracy of follow-up evaluation instructions.

[0005] From the perspective of model training and parameter optimization, traditional follow-up systems lack a joint training mechanism based on historical postoperative data and real-time physiological parameters. The absence of a dynamic feature generation module makes it impossible to perform personalized feature analysis and correction according to individual differences. For example, the suture healing speed and soft tissue repair ability of different patients vary, and traditional methods are difficult to generate appropriate deviation compensation data. At the same time, the instruction triggering process lacks the ability to optimize dynamic weights and identify abnormal states in real time, unable to minimize the evaluation error through iterative optimization and difficult to accurately locate abnormal healing sites.

[0006] At the clinical application level, the traditional follow-up mode makes it impossible for doctors to comprehensively grasp the subtle changes in the postoperative recovery of patients, and may miss the best intervention opportunity. For example, if the local deviation during the bone suture healing process is not detected in time, it may lead to the recurrence of maxillofacial deformity or the aggravation of occlusal dysfunction; if the abnormal soft tissue distribution is not adjusted in time, it may affect the recovery effect of facial aesthetics. In addition, the subjectivity and low efficiency of manual evaluation make it difficult to meet the follow-up needs of a large number of postoperative patients, and there is an urgent need for an intelligent and precise follow-up system to improve the quality of medical services. Summary of the Invention

[0007] The purpose of the present invention is to provide a follow-up system and method for orthognathic surgery of maxillofacial deformities to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A follow-up system for orthognathic surgery of maxillofacial deformities, the system includes: An image analysis model for performing multi-dimensional feature fusion processing on the initial postoperative data. The initial postoperative data includes a first spatial vector corresponding to three-dimensional bone suture feature data, a second correlation matrix corresponding to occlusal function feature data, and a third mapping sequence corresponding to soft tissue distribution feature data. The three-dimensional bone suture feature data includes a first distribution label generated by bone block displacement parameters and a second quantization label generated by bone suture healing parameters. The multi-dimensional feature fusion processing includes performing spatial coordinate normalization on the first spatial vector, performing time series alignment on the second correlation matrix, performing anatomical region annotation on the third mapping sequence, inputting the processed three types of data into a feature fusion network, and extracting cross-modal fusion features through layer-by-layer feature crossing to form a multi-dimensional fusion feature set; A follow-up evaluation model for performing dynamic interference suppression processing on the real-time physiological data stream and inputting it into a feature generation layer for recovery trajectory simulation, and generating a follow-up evaluation instruction according to the output result of the feature generation layer. The dynamic interference suppression processing includes using an adaptive filtering algorithm to suppress high-frequency noise and removing abnormal fluctuation signals through joint time-domain and frequency-domain analysis. The recovery trajectory simulation includes matching the processed real-time physiological data stream with the corresponding stage data of multiple types of historical postoperative data, extracting the feature change trend at each time node, and generating a postoperative recovery prediction trajectory through time series prediction; The feature generation layer includes an image preprocessing module and a dynamic feature generation module. Among them, the image preprocessing module is used for anatomical interval segmentation and invalid signal filtering of the original image data stream, and the dynamic feature generation module is obtained through joint training based on multiple types of historical postoperative data and real-time physiological parameters. The dynamic feature generation module is used to construct a neural network architecture including a feature analysis layer, a feature correction layer, and an instruction trigger layer.

[0009] Preferably, the feature analysis layer is used to perform anatomical feature association processing on different feature vectors in the original image data stream to generate feature fusion data; the feature correction layer is used to model the dynamic mapping relationship between the feature fusion data corresponding to each feature vector to generate deviation compensation data; the instruction trigger layer is used to perform multi-dimensional feature fusion based on the deviation compensation data and the feature fusion data to generate a follow-up evaluation instruction.

[0010] Preferably, the modeling of the dynamic mapping relationship between the feature fusion data corresponding to each feature vector to generate deviation compensation data includes: Using a suture segmentation algorithm to identify the key healing nodes in the feature fusion data, and determining the correction compensation mapping corresponding to each feature vector based on the type of physiological parameters corresponding to each key healing node; Calculating the offset coefficient between the healing nodes with the same parameters in the correction compensation mappings corresponding to any two feature vectors, and generating deviation compensation data between the any two feature vectors based on the offset coefficient.

[0011] Preferably, the calculating of the offset coefficient between the healing nodes with the same parameters in the correction compensation mappings corresponding to any two feature vectors includes: When the number of healing nodes in the correction compensation mappings corresponding to the any two feature vectors is inconsistent, performing virtual healing point interpolation based on the physiological parameters corresponding to the terminal healing nodes in the one with fewer healing nodes, and calculating the offset coefficient between the healing nodes with the same parameters based on the interpolated data.

[0012] Preferably, the image preprocessing module is specifically used for: Performing equal-gradient partitioning on the first spatial vector, the second correlation matrix, and the third mapping sequence according to a preset anatomical interval to generate standardized suture data, standardized occlusion data, and standardized soft tissue data; Performing real-time matching on the standardized suture data and the standardized occlusion data by using a dynamic parameter alignment method, and performing steady-state optimization on the standardized soft tissue data by using a fixed anatomical window mechanism, and outputting a first calibration vector, a second calibration matrix, and a third calibration sequence; wherein, the first calibration vector includes a calibrated first distribution label and a calibrated second quantization label.

[0013] Preferably, the image preprocessing module is further used for: Calculating the healing offset coefficient between the calibrated first distribution label and the calibrated second quantization label in the historical follow-up period; Predicting the expected distribution value of the calibrated second quantization label in the real-time follow-up stage according to the healing offset coefficient and the parameter value of the calibrated first distribution label in the real-time follow-up stage; Generate target correction compensation data based on the calibrated second quantization label and its expected distribution value, and use the feature vector corresponding to the target correction compensation data as the first calibration vector.

[0014] Preferably, the feature correction layer specifically includes: A feature traceability unit for respectively performing healing feature traceability on each feature vector in the feature fusion data to extract the corresponding healing conduction chain from each feature vector; A parameter superposition unit for performing anatomical parameter superposition on the healing conduction chain extracted from each feature vector and the corresponding feature fusion data to generate deviation compensation data.

[0015] Preferably, the feature correction layer further includes: A noise suppression unit for performing phase shift effect elimination processing on the deviation compensation data.

[0016] Preferably, the instruction trigger layer specifically includes: A multi-dimensional feature collaboration unit including multiple instruction decision nodes, and each instruction decision node is connected to each feature vector in the deviation compensation data and the feature fusion data through configuration parameters; A dynamic weight optimization unit for iteratively optimizing the configuration parameters through a dynamic weight adjustment algorithm to minimize the error between the follow-up evaluation instruction and the actual recovery trajectory; An abnormal state recognition unit for performing abnormal healing localization based on the deviation compensation data and the feature fusion data to generate a follow-up evaluation instruction.

[0017] Preferably, the present invention further includes a method for follow-up after orthognathic surgery for maxillofacial deformities, which is applied to the follow-up system for orthognathic surgery for maxillofacial deformities described in any one of the above, and includes the following steps: Perform multi-dimensional feature fusion processing on the initial postoperative data by using an image analysis model. The initial postoperative data includes a first spatial vector corresponding to three-dimensional bone suture feature data, a second correlation matrix corresponding to occlusal function feature data, and a third mapping sequence corresponding to soft tissue distribution feature data. The three-dimensional bone suture feature data includes a first distribution label generated by bone block displacement parameters and a second quantization label generated by bone suture healing parameters; Perform dynamic interference suppression processing on the real-time physiological data stream through a follow-up evaluation model, and input the processed data stream into the feature generation layer for recovery trajectory simulation; Use the image preprocessing module in the feature generation layer to perform anatomical interval segmentation and invalid signal filtering on the original image data stream; The recovery trajectory simulation is carried out by using a dynamic feature generation module jointly trained with multi - type historical postoperative data and real - time physiological parameters. The dynamic feature generation module includes a feature analysis layer, a feature correction layer, and an instruction trigger layer connected in sequence. According to the output result of the feature generation layer, a follow - up evaluation instruction is generated by the follow - up evaluation model.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: In terms of data processing and feature analysis, the image analysis model can perform multi - dimensional feature fusion processing on the initial postoperative data. By integrating the first spatial vector corresponding to the three - dimensional suture feature data, the second correlation matrix corresponding to the occlusal function feature data, and the third mapping sequence corresponding to the soft tissue distribution feature data, combined with the first distribution label generated by the bone block displacement parameter and the second quantization label generated by the suture healing parameter, a refined model of the key indicators for postoperative recovery is realized. The image pre - processing module performs equal - gradient partitioning of the data through a preset anatomical interval, and combines the dynamic parameter alignment and fixed anatomical window mechanism to generate a standardized and calibrated data stream, effectively improving the consistency and reliability of the data, and providing an accurate data basis for subsequent evaluation.

[0019] The dynamic interference suppression processing mechanism of the follow - up evaluation model can eliminate the noise and external interference in the real - time physiological data stream in real time, ensuring that the data input to the feature generation layer truly reflects the physiological state of the patient. The dynamic feature generation module of the feature generation layer is jointly trained based on multi - type historical postoperative data and real - time physiological parameters, and has strong generalization ability and personalized adaptability. The feature analysis layer generates feature fusion data through anatomical feature association processing, the feature correction layer generates deviation compensation data by modeling the dynamic mapping relationship between feature vectors, and the instruction trigger layer generates a follow - up evaluation instruction through multi - dimensional feature fusion, forming a complete intelligent processing link from data analysis to instruction generation.

[0020] In terms of feature correction and evaluation instruction optimization, the feature correction layer extracts the healing conduction chain through the feature tracing unit, and combines parameter superposition and noise suppression processing to accurately capture the dynamic deviation between each feature vector and generate compensation data, effectively improving the accuracy of the recovery trajectory simulation. The multi - dimensional feature coordination unit of the instruction trigger layer connects each feature vector through configuration parameters, the dynamic weight optimization unit minimizes the evaluation error through iterative optimization, and the abnormal state recognition unit accurately locates the abnormal healing site based on the data, ensuring the scientific nature and pertinence of the follow - up evaluation instruction. For example, when the physiological parameters at the key node of suture healing show abnormal deviation, the system can quickly identify and generate an intervention instruction to avoid further expansion of the recovery deviation.

[0021] In terms of clinical application value, the system realizes the full-cycle and multi-dimensional dynamic monitoring of the postoperative recovery process, providing doctors with real-time and accurate evaluation basis. By predicting in advance the trends of suture healing, the trajectories of occlusal function recovery, and the morphological changes of soft tissues, doctors can timely adjust treatment plans, intervene in abnormal recovery processes, reduce the risk of recurrence of maxillofacial deformities, and improve the recovery effects of occlusal function and facial aesthetics. At the same time, the intelligent data processing and evaluation mechanism significantly improves the follow-up efficiency, reduces the subjectivity and errors of manual evaluation, can be widely applied to the follow-up of postoperative patients, optimize the allocation of medical resources, and improve the overall level of medical services. In addition, the scalability of the system provides a basis for combining more physiological parameters and new imaging technologies in the future, contributing to the digital and intelligent development of the field of maxillofacial deformity treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is the working principle diagram of the orthognathic postoperative follow-up system for maxillofacial deformities according to the present invention; Figure 2 is the schematic diagram of the feature analysis-correction-trigger processing flow in the feature generation layer; Figure 3 is the flowchart of the method for generating deviation compensation data in the feature correction layer; Figure 4 is the flowchart of the data calibration process of the image preprocessing module; Figure 5 is the flowchart of the prediction and compensation of healing parameters based on historical data. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] Please refer to Figures 1 - 5 , an orthognathic postoperative follow-up system and method for maxillofacial deformities according to the present invention realizes the accurate evaluation and dynamic follow-up of the postoperative recovery situation through the collaborative work of the image analysis model and the follow-up evaluation model. The specific implementation steps are as follows: The image analysis model is used to perform multi-dimensional feature fusion processing on the initial postoperative data. The initial postoperative data includes a first spatial vector corresponding to three-dimensional suture feature data, a second correlation matrix corresponding to occlusal function feature data, and a third mapping sequence corresponding to soft tissue distribution feature data. The three-dimensional suture feature data further includes a first distribution label generated by bone block displacement parameters and a second quantization label generated by suture healing parameters. The image analysis model extracts key features reflecting the initial postoperative state through the fusion processing of the above multi-dimensional data.

[0025] In this embodiment, the multi-dimensional feature fusion processing of the image analysis model can be achieved in the following manner: In the initial postoperative data acquisition stage, the three-dimensional suture feature data is obtained through cone beam CT (CBCT) scanning. The original data containing bone block displacement parameters (such as bone block translation amount, rotation angle) and suture healing parameters (such as suture density, callus coverage rate) is obtained through point cloud extraction and converted into a first spatial vector for storage; the occlusal function feature data is collected by a dynamic occlusal recorder, and parameters such as the distribution of occlusal contact points and occlusal force values at different time points after the operation of the patient are recorded and converted into a second correlation matrix arranged in a time series; the soft tissue distribution feature data is obtained through a three-dimensional surface scanner, and parameters such as the thickness and volume distribution of facial soft tissues are recorded and converted into a third mapping sequence containing anatomical position information.

[0026] For the spatial coordinate normalization processing of the first spatial vector, first, a standard coordinate system is established with the anterior nasal spine (ANS) and the sella turcica point (S) of the patient as the reference points for the original three-dimensional coordinates of the CBCT scan. The original coordinate values of each bone block displacement parameter are converted into standardized coordinate values relative to this reference coordinate system to eliminate spatial deviations caused by different scanning devices or scanning angles.

[0027] For the time series alignment processing of the second correlation matrix, with the surgical date as the time origin, the original timestamps of each time point in the occlusal function feature data are uniformly converted into the number of days after the operation (such as 7 days, 14 days, 30 days after the operation, etc.). The occlusal data at missing time points is supplemented by the linear interpolation method to ensure that the time intervals of each time series in the second correlation matrix are consistent.

[0028] For the anatomical region annotation processing of the third mapping sequence, a soft tissue anatomical atlas template containing key anatomical landmark points such as the orbit, nasolabial fold, and mandibular margin is established in advance. The soft tissue distribution data obtained by three-dimensional surface scanning is registered with this template, and the anatomical names corresponding to each soft tissue region (such as "buccal soft tissue", "mental soft tissue") are automatically annotated to form a third mapping sequence with anatomical labels.

[0029] After the above preprocessing, the normalized first spatial vector, the aligned second correlation matrix, and the labeled third mapping sequence are input into the feature fusion network. This network consists of three layers of feature cross structures: the first layer splices and fuses the local features of each modality data (such as the local displacement features of sutures, the local contact features of occlusion, and the local thickness features of soft tissues); the second layer performs weighted fusion on the key features of each modality (such as the key regions of suture healing, the abnormal points of occlusion function, and the sunken regions of soft tissues) through an attention mechanism; the third layer performs global context correlation on the fusion features output by the previous two layers, and finally forms a multi-dimensional fusion feature set containing spatial position, temporal variation, and anatomical localization information.

[0030] The follow-up evaluation model is used to perform dynamic interference suppression processing on the real-time physiological data stream, removing noise or irrelevant interference during the data acquisition process to ensure the accuracy of the input data. The processed data stream is input into the feature generation layer for recovery trajectory simulation. The implementation process of the follow-up evaluation model is as follows: The real-time physiological data stream is collected in real time through wearable physiological monitoring devices (such as oral pressure sensors, facial motion sensors), including parameters such as bite force, mandibular movement trajectory, and soft tissue deformation. During the dynamic interference suppression processing, first, the high-frequency noise is suppressed through an adaptive filtering algorithm: the filtering parameters are dynamically adjusted according to the noise level of the real-time data stream (such as when a high-frequency fluctuation higher than 20% of the baseline value is detected in the bite force data, the cut-off frequency of the low-pass filter is increased); then, a joint time-frequency domain analysis is performed to extract the time-domain statistical features (such as mean, variance) and frequency-domain power spectrum features (such as main frequency components) of the data stream. Signals that simultaneously meet the conditions that the time-domain variance exceeds 3 times the historical mean and there are non-physiological frequency components in the frequency domain (such as abnormal jitter > 10Hz) are determined as abnormal fluctuations and removed.

[0031] In the recovery trajectory simulation stage, the processed real-time physiological data stream is matched with multiple types of historical postoperative data: The historical data is classified and stored according to the surgical type (such as bimaxillary surgery, unimaxillary surgery), patient age (such as 18 - 30 years old, 31 - 50 years old), and postoperative time stage (such as early healing period, bone remodeling period); during the matching, first, the target historical data set is determined according to the surgical type and age of the current patient, and then the corresponding stage of historical physiological data is extracted according to the postoperative time stage (such as if it is 2 weeks after surgery currently, the time period of 2 weeks ± 3 days after surgery in the historical data is matched). The change trend of features at each time node is extracted through the sliding window method (such as the increasing amplitude of bite force per week, the expanding amplitude of mandibular movement range per week), and the change trend of the current real-time data is compared with the historical trend. If the difference exceeds the preset threshold (such as the increasing amplitude of bite force is lower than 50% of the historical mean), the postoperative recovery prediction trajectory is generated through the time series prediction module, and this trajectory contains the expected change curves of bite force, mandibular movement range, and soft tissue morphology within the next 4 weeks.

[0032] Finally, based on the fused features output by the feature generation layer and the recovery prediction trajectory, the follow-up evaluation model generates specific follow-up evaluation instructions. For example, when the displacement parameters in the key area of suture healing are detected to be abnormal and the increasing trend of bite force is lower than the historical average, an instruction of "It is recommended to increase imaging reexamination 4 weeks after surgery" is generated; when the deformation trend of the soft tissue depression area is consistent with the historical normal recovery trajectory, an instruction of "The current recovery situation is good, and the existing follow-up plan is maintained" is generated.

[0033] The feature generation layer includes an image preprocessing module and a dynamic feature generation module: The image preprocessing module is responsible for segmenting the anatomical intervals and filtering out invalid signals from the original image data stream to standardize the data format and exclude interference information; the dynamic feature generation module is obtained through joint training based on multiple types of historical postoperative data and real-time physiological parameters. The neural network architecture of the dynamic feature generation module is implemented in the following way: A deep learning network including a feature parsing layer, a feature correction layer, and an instruction trigger layer is constructed using a three-level cascaded structure, and a collaborative optimization architecture is formed between each level through a feature transfer and parameter sharing mechanism.

[0034] The feature parsing layer adopts a convolutional neural network structure with a multi-branch attention mechanism to process the preprocessed first calibration vector, second calibration matrix, and third calibration sequence in parallel. Among them, a spatial attention module is designed for the first calibration vector, and channel weights are generated through 1×1 convolution and global pooling operations to strengthen the spatial attention to suture features; a temporal attention module is designed for the second calibration matrix, and a gated recurrent unit is used to capture the temporal variation features of the occlusal function; a hybrid attention module is designed for the third calibration sequence, and the spatial and temporal attention mechanisms are fused to process the soft tissue distribution features. The outputs of each branch are spliced through tensors to form multi-modal feature fusion data.

[0035] The feature correction layer adopts an encoder-decoder architecture with residual connections, including a feature traceability unit and a parameter superposition unit. The feature traceability unit is implemented by a bidirectional long short-term memory network, which performs temporal backtracking on the feature fusion data to extract the feature representation of the healing conduction chain at each time step; the parameter superposition unit performs an element-wise addition operation on the healing conduction chain feature and the original feature fusion data to generate deviation compensation data containing temporal context information. During the parameter superposition process, a learnable weight coefficient is introduced to dynamically adjust the healing conduction chain feature, and the weight coefficient is trained through an adaptive moment estimation optimization algorithm.

[0036] The instruction trigger layer adopts a graph neural network structure, including a multi-dimensional feature collaboration unit, a dynamic weight optimization unit, and an abnormal state recognition unit. The multi-dimensional feature collaboration unit maps the deviation compensation data and feature fusion data into node features, and learns the structural relationship between feature vectors through a graph convolutional network; the dynamic weight optimization unit is based on a reinforcement learning framework, and dynamically adjusts the edge weights in the graph neural network through a policy gradient algorithm to minimize the error between the follow-up evaluation instruction and the actual recovery trajectory; the abnormal state recognition unit uses an isolation forest algorithm to detect outliers in the output features of the graph neural network, and triggers corresponding follow-up evaluation instructions when abnormal healing features are detected.

[0037] The entire neural network architecture adopts an end-to-end training method, and optimizes the parameters by minimizing the following objective function: using the known recovery trajectory in the historical postoperative data as a supervision signal, constructing a composite loss function including feature consistency loss, temporal continuity loss, and anomaly detection loss, and using the stochastic gradient descent algorithm to iteratively update the network parameters. During the training process, batch normalization technology is used to accelerate convergence, and dropout is used to prevent overfitting, and finally a dynamic feature generation module that can accurately predict the recovery trajectory after orthognathic surgery for maxillofacial deformities is formed.

[0038] Embodiment 1: This embodiment relates to the specific working mechanism of the feature analysis layer, and its core function is to perform anatomical feature association processing on different feature vectors in the original image data stream to generate feature fusion data. The specific implementation method is as follows: The original image data stream comes from the multi-modal examination data of patients after orthognathic surgery for maxillofacial deformities, including three-dimensional suture feature data, occlusal function feature data, and soft tissue distribution feature data, which are presented in the form of a first spatial vector, a second correlation matrix, and a third mapping sequence respectively. These data have different physical meanings and data structures: the first spatial vector represents the spatial distribution and dynamic changes of the three-dimensional suture, including the first distribution label generated by the bone block displacement parameter (such as the discretized label of the displacement direction and displacement distance) and the second quantization label generated by the suture healing parameter (such as the continuous quantization values of the healing density and healing speed); the second correlation matrix reflects the mechanical conduction relationship of the occlusal function, and records the force conduction intensity and cooperation between different tooth positions and occlusal contact points; the third mapping sequence describes the spatial distribution pattern and the change trajectory over time of soft tissues (such as skin, muscle, mucosa).

[0039] The processing flow of the feature analysis layer begins with the anatomical relevance analysis of multi-source feature vectors. First, for the three-dimensional suture feature data and occlusal function feature data, the system establishes a spatial mapping relationship between bone and occlusal function based on maxillofacial anatomical principles. For example, the forward displacement of the maxillary bone block may cause the forward movement of the anterior tooth occlusal contact point. The feature analysis layer establishes the spatial position correlation between the two by identifying the displacement direction of the maxillary bone block (a key parameter in the first spatial vector) and the position change of the anterior tooth occlusal contact point (the corresponding parameter in the second correlation matrix). Specifically, the system pre-designs an anatomically recognizable mapping table, divides the maxilla into multiple sub-regions (such as the anterior tooth region bone segment, molar region bone segment), and each sub-region corresponds to a specific occlusal contact area (such as the anterior tooth occlusal surface, posterior tooth occlusal surface). Through the coordinate mapping algorithm, the bone block displacement parameters and the occlusal contact point parameters are registered in three-dimensional space to form a linkage feature group of bone-occlusal function.

[0040] For the three-dimensional suture feature data and soft tissue distribution feature data, the feature analysis layer analyzes the interaction between the suture healing area and the surrounding soft tissues based on the anatomical hierarchical relationship. For example, during the mandibular suture healing process, the change in the tension of the surrounding masseter muscle may affect the stability of the bone block. The feature analysis layer establishes an association model between the soft tissue mechanical state and the suture healing process by extracting the healing density parameter of the mandibular suture (the second quantization label) and the parameter of the change in the masseter muscle thickness (the soft tissue thickness parameter in the third mapping sequence). In the specific operation, the system divides the masseter muscle area around the mandibular suture into multiple analysis units, and each unit corresponds to a specific suture sub-region. Through time series analysis, the temporal correlation between the change in the masseter muscle thickness and the increase in the suture healing density is identified to generate a collaborative feature group of soft tissue-bone healing.

[0041] After completing the anatomical feature association analysis of different feature vectors, the feature analysis layer enters the stage of generating feature fusion data. For the linkage feature group of bone-occlusal function, the system uses the tensor fusion algorithm to perform dimensional expansion and matrix multiplication operations on the bone block displacement direction and displacement distance parameters in the first spatial vector and the occlusal contact point position and bite force conduction intensity parameters in the second correlation matrix to generate a three-dimensional bone-occlusal function fusion tensor. Each element of this tensor represents the joint eigenvalue of the displacement of a specific bone segment and the force on the corresponding occlusal contact point. For example, the element value can reflect the increase in the bite force at the anterior tooth occlusal contact point when the anterior tooth region bone segment of the maxilla moves forward by 0.5 mm.

[0042] For the soft tissue - bone healing synergy feature group, the system uses a sequence alignment algorithm to align the soft tissue thickness change sequence in the third mapped sequence with the bone suture healing density change sequence in the first spatial vector along the time axis. The dynamic time warping (DTW) algorithm is used to eliminate the temporal deviation caused by the difference in data acquisition time, and then a two - dimensional soft tissue - bone healing fusion matrix is generated. The row dimension of the matrix represents time nodes, and the column dimension represents anatomical position nodes. Each matrix element records the combined eigenvalue of the soft tissue thickness and bone suture healing density at the corresponding time point and anatomical position. For example, the element value can reflect the product of the change rate of the masseter muscle thickness and the growth rate of the bone suture healing density in the mandibular angle bone suture area on the 7th day after surgery.

[0043] In addition, the feature analysis layer also needs to process the correlation between the occlusal function feature data and the soft tissue distribution feature data. For example, abnormal occlusal function may cause uneven stress on the buccal soft tissue, resulting in changes in soft tissue texture. The feature analysis layer extracts the occlusal contact point distribution uniformity parameter in the second correlation matrix and the buccal soft tissue texture feature parameter (such as the gray - level co - occurrence matrix eigenvalue) in the third mapped sequence to establish the occlusal - soft tissue mechanical conduction feature group. The system first divides the buccal soft tissue into grids, and each grid unit corresponds to a specific dental occlusal area. Through the finite element analysis method, the stress distribution of the soft tissue under different occlusal contact modes is simulated, and then the occlusal contact point distribution uniformity parameter is mapped to the stress parameter of the soft tissue grid unit to generate an occlusal - soft tissue stress fusion vector. Each component of this vector represents the combined eigenvalue of the stress state of the corresponding grid unit and the occlusal uniformity.

[0044] During the feature fusion process, the system needs to normalize the feature parameters of different dimensions to eliminate the influence of dimensional differences on the fusion result. For the displacement distance parameter (unit: mm) in the first spatial vector and the occlusal force parameter (unit: N) in the second correlation matrix, the Z - score standardization method is used to convert each parameter into standard normal distribution data with a mean of 0 and a standard deviation of 1. For the soft tissue thickness parameter (unit: mm) and the bone suture healing density parameter (dimensionless) in the third mapped sequence, the min - max standardization method is used to scale the parameter range to the interval [0, 1]. The normalized data is combined into a unified feature vector space through the feature concatenation method, providing standardized input data for subsequent feature correction and instruction generation.

[0045] The processing of the feature analysis layer is completely based on anatomical principles and data-driven correlation analysis, without relying on any empirical assumptions or preset thresholds. Through the anatomical feature correlation and fusion of multi-dimensional feature vectors, the system can extract complex correlation features reflecting the recovery after orthognathic surgery for maxillofacial deformities from the original data. These features not only contain independent information of a single modality, but also cover the synergistic information between multi-modalities, providing a more comprehensive feature expression for the dynamic interference suppression processing and recovery trajectory simulation of the follow-up evaluation model. For example, the three-dimensional fusion features of bone-bite-soft tissue can comprehensively reflect the overall impact of bone structure changes on function and appearance, enabling the system to comprehensively evaluate the postoperative recovery from three levels: anatomical structure, functional state, and tissue morphology, avoiding the limitations of single-dimensional evaluation.

[0046] Embodiment 2: This embodiment details the process of the feature correction layer modeling the dynamic mapping relationship and generating deviation compensation data. The specific steps are as follows: First, a suture segmentation algorithm is used to identify the key healing nodes in the feature fusion data. The suture segmentation algorithm automatically identifies the key position points during the suture healing process, such as the bone fragment ends and the healing frontiers, based on the spatial distribution and gray value differences of the three-dimensional suture feature data. The algorithm first preprocesses the three-dimensional suture feature data, including denoising, smoothing, and normalization operations, to improve the clarity and distinguishability of the features. Then, using threshold segmentation and region growing techniques, the suture region is separated from the surrounding tissues, and a binary mask of the suture is generated. Next, through morphological operations and contour extraction, the suture boundary is further refined to determine the precise position and morphology of the suture.

[0047] After identifying the suture region, the algorithm further analyzes the internal structure and features of the suture to determine the key healing nodes. These nodes usually correspond to specific anatomical positions in the suture, such as the junctions of bone fragments, stress concentration areas, or the passage sites of blood vessels and nerve bundles. The algorithm identifies these key nodes by calculating the curvature, gray gradient, and texture features within the suture region, combined with preset anatomical knowledge and rules. Each key healing node corresponds to a specific type of physiological parameter, such as displacement speed, healing density, etc. According to these parameter types, a corresponding correction compensation mapping is determined for each feature vector, that is, the correspondence between the parameters in the feature vector and the key healing nodes is established.

[0048] Calculate the offset coefficient between the healing nodes with the same parameters in the correction compensation mappings corresponding to any two eigenvectors. When the number of healing nodes in the correction compensation mappings of two eigenvectors is inconsistent, virtual healing point interpolation needs to be performed based on the physiological parameters corresponding to the terminal healing nodes in the one with fewer healing nodes. For example, if the correction compensation mapping of eigenvector A contains 5 healing nodes and the mapping of eigenvector B contains 3 healing nodes, then based on the parameters of the terminal node of eigenvector B, virtual parameter values are generated at the positions of the missing nodes through an interpolation algorithm to align the number and positions of the nodes of both.

[0049] After interpolation, for the healing nodes with the same parameter (such as displacement velocity), calculate the percentage difference or spatial distance difference of their parameter values as the offset coefficient. The offset coefficient reflects the synergy or difference between different eigenvectors in the same parameter dimension. Based on all the offset coefficients, deviation compensation data between any two eigenvectors is generated, which is used to adjust the parameter weights or mapping relationships in the feature fusion process to compensate for the evaluation deviation caused by the differences in eigenvectors.

[0050] In the process of generating the deviation compensation data, the feature correction layer also considers the factor of the time dimension. Since the recovery after orthognathic surgery for maxillofacial deformities is a dynamic process, there may be temporal correlations between the eigenvectors at different time points. To capture this temporal correlation, the feature correction layer uses time series analysis methods to model and analyze the eigenvectors at different time points. Specifically, the algorithm first represents the eigenvector at each time point as a point in a high-dimensional space, and then constructs a time series model of the eigenvectors by calculating the distances and similarities between these points.

[0051] Based on this time series model, the algorithm can predict the eigenvector values at future time points and compare them with the actual observed values to detect whether there are abnormal changes. If abnormal changes are found, the algorithm will automatically adjust the deviation compensation data to adapt to this change. In addition, time series analysis can also help identify the lag effect and causal relationship between eigenvectors, further improving the accuracy and effectiveness of deviation compensation.

[0052] In addition to considering the factor of the time dimension, the feature correction layer also considers the factor of the spatial dimension. Due to the complexity of the maxillofacial structure, there may be spatial correlations between the eigenvectors in different regions. To capture this spatial correlation, the feature correction layer uses spatial statistical analysis methods to model and analyze the eigenvectors in different regions. Specifically, the algorithm first divides the maxillofacial region into multiple sub-regions, and then constructs a spatial association model of the eigenvectors by calculating the spatial distances and similarities between these sub-regions.

[0053] Based on this spatial association model, the algorithm can identify regions with similar feature patterns and integrate and analyze the feature vectors of these regions. In addition, spatial statistical analysis can also help detect the spatial heterogeneity and spatial autocorrelation between feature vectors, further improving the accuracy and effectiveness of bias compensation.

[0054] After generating the bias compensation data, the feature correction layer also verifies and optimizes it. Specifically, the algorithm first applies the bias compensation data to the feature fusion process, and then verifies the effectiveness of the bias compensation data by comparing the fusion result with the evaluation result of clinical experts. If a large difference is found between the fusion result and the evaluation result of clinical experts, the algorithm will automatically adjust the parameters of the bias compensation data to improve the accuracy and reliability of the fusion result.

[0055] In addition, the feature correction layer also adopts cross-validation and model selection techniques to compare and evaluate different bias compensation methods and parameters to select the optimal bias compensation scheme. In this way, the feature correction layer can continuously optimize the generation process of the bias compensation data, improve its accuracy and effectiveness, and thus provide a more reliable basis for the generation of subsequent follow-up evaluation instructions.

[0056] Example 3: This example elaborates on the specific processing flow of the image preprocessing module. Through operations such as anatomical interval division, data standardization, dynamic matching, and steady-state optimization, this module realizes the standardized processing and noise filtering of the original image data stream, providing standardized input data for subsequent analysis. The specific implementation is as follows: The input data of the image preprocessing module are the first spatial vector, the second correlation matrix, and the third mapping sequence in the initial postoperative data, corresponding to three-dimensional suture feature data, occlusal function feature data, and soft tissue distribution feature data respectively. The processing process first performs spatial division on the multi-source data based on the preset anatomical intervals. The preset anatomical intervals are constructed according to maxillofacial anatomical standards. For example, the jawbone structure is divided into sub-regions such as the body of the maxilla, the alveolar process of the maxilla, the ascending ramus of the mandible, and the body of the mandible; the occlusal function regions are divided into the anterior tooth occlusion area, the posterior tooth occlusion area, the left occlusal contact point group, the right occlusal contact point group, etc.; the soft tissue distribution regions are divided into the soft tissue of the cheek, the soft tissue of the lip, the soft tissue around the temporomandibular joint, etc. Each anatomical interval corresponds to a specific physiological parameter monitoring range, such as the bone block displacement parameter of the body of the maxilla, the bite force conduction parameter of the anterior tooth occlusion area, the thickness change parameter of the soft tissue of the cheek, etc.

[0057] Based on the spatial division, the module performs equal-gradient division on each feature data. For the first spatial vector corresponding to the three-dimensional suture feature data, the system divides the sutures within each anatomical interval into several gradient layers according to the bone block displacement direction and displacement distance. For example, taking every 0.2 mm of displacement distance as a gradient, standardized suture data is generated; for the second correlation matrix corresponding to the occlusal function feature data, gradient division is performed according to the magnitude of the occlusal force conduction intensity (such as intervals of 0 - 50 N, 50 - 100 N, etc.) to generate standardized occlusal data; for the third mapping sequence corresponding to the soft tissue distribution feature data, gradient stratification is carried out based on the variation range of parameters such as soft tissue thickness and blood flow volume to generate standardized soft tissue data. The purpose of equal-gradient division is to convert continuous physiological parameters into discrete feature levels, facilitating subsequent data matching and analysis.

[0058] After completing the spatial division and gradient stratification, the module performs dynamic parameter alignment processing on the standardized suture data and occlusal data. The dynamic parameter alignment is based on the synchronous calibration logic of the time series. By establishing a two-dimensional mapping table of "timestamp - anatomical position", the suture displacement event is temporally correlated with the change of occlusal function parameters. For example, when the three-dimensional reconstruction image shows that the maxillary bone block has a 0.3 mm forward displacement on the 3rd day after surgery, the system automatically retrieves the pressure change value of the anterior tooth occlusal contact point in the synchronous occlusal function monitoring data at the same period, and fills the missing values in the data acquisition interval through the interpolation algorithm to ensure the one-to-one correspondence between the two on the time axis. This process uses the dynamic time warping (DTW) algorithm to eliminate the temporal deviation caused by the sampling frequency difference of different monitoring devices, so that the dynamic changes of suture displacement and occlusal function form a comparable synchronous sequence.

[0059] For the standardized soft tissue data, the module adopts a fixed anatomical window mechanism for steady-state optimization. The fixed anatomical window takes the surgical incision as the center, and delimits a cubic analysis area with a side length of 5 mm in the surrounding soft tissue area (the window size can be adjusted according to the anatomical structure), and continuously monitors the soft tissue parameters within the window (such as the gray value in the ultrasound image and the blood flow volume parameter of the laser Doppler flowmeter). The steady-state optimization is achieved through a low-pass filtering algorithm, setting the cut-off frequency to 0.1 Hz to filter out high-frequency noise and retain the low-frequency signal reflecting the long-term healing trend of the soft tissue. For example, in the early postoperative period, the soft tissue may show a short-term surge in blood flow due to the inflammatory reaction. After filtering, the baseline change trend of the blood flow can be extracted to avoid interfering with the judgment of the suture healing state. The processed soft tissue data generates a third calibration sequence, which includes the parameter mean, variance, and trend eigenvalue of each anatomical window.

[0060] In addition, the image preprocessing module also needs to perform cross-parameter correlation analysis on the calibrated feature data. Taking the first distribution label (bone block displacement label) and the second quantitative label (bone suture healing density) in the first calibration vector as an example, the system calculates the healing offset coefficient of the two during the historical follow-up period. The healing offset coefficient is defined as the Pearson correlation coefficient between the bone block displacement distance and the bone suture healing density in the same anatomical interval, reflecting the synergy between bone structure changes and the healing process. When the correlation coefficient of a certain area is lower than the preset threshold (such as 0.5), it indicates that there may be delayed healing or abnormal displacement, which needs to be monitored in the real-time follow-up stage.

[0061] Based on the historical offset coefficient and real-time displacement parameters, the module predicts the expected distribution value of suture healing density. The specific method is: use the linear regression model to fit the functional relationship between displacement distance and healing density in historical data (such as healing density = α × displacement distance + β), and substitute the real-time displacement parameters into the model to calculate the expected value. If the difference between the real-time monitored healing density value and the expected value exceeds 15%, the target correction compensation data is generated to adjust the weight parameter of the characteristic vector of the region. For example, when a mandibular segment is displaced by 0.8mm, the model predicts that the healing density should be 0.6g / cm³. If the measured value is 0.4g / cm³, a compensation factor is generated to perform weighted correction on the characteristic vector of the region to highlight the abnormal signal of the healing density parameter.

[0062] In the data output link, the module converts the processed standardized data into a unified data format: the first calibration vector integrates the spatial position, displacement gradient and healing density characteristics of the suture, and is expressed in the form of a three-dimensional coordinate vector; the second calibration matrix records the spatial distribution and force conduction synergy of the occlusal contact points, and uses an N×N matrix (N is the number of occlusal contact points) to store the correlation strength between each point; the third calibration sequence arranges the steady-state eigenvalues of the soft tissue parameters in chronological order to form a one-dimensional time series array. The three types of data are associated through timestamp indexes to ensure the spatiotemporal consistency of multimodal data.

[0063] During the entire processing process, the module strictly follows the principle of anatomical zoning to ensure the accuracy of spatial positioning; through gradient division and parameter alignment, the dimensional unification and time-series synchronization of multi-source data are achieved; with the help of fixed window filtering and cross-parameter prediction, the stability and clinical interpretability of the data are enhanced. For example, when analyzing patients with zygomatic complex fractures after surgery, the module can dynamically match the suture displacement data of the zygomatic alveolar ridge area with the force conduction data of the occlusal contact point of the ipsilateral posterior teeth, and at the same time evaluate the effect of local blood supply on healing through the blood flow parameters of the cheek soft tissue window. The final output includes multidimensional calibration data of spatial position, functional status, and tissue physiology, providing high-quality input for the subsequent feature generation layer.

[0064] Embodiment 4: This embodiment focuses on the feature tracing and parameter superposition process of the feature correction layer. The feature correction layer includes a feature tracing unit, a parameter superposition unit, and a noise suppression unit. Through in-depth analysis and dynamic correction of the feature fusion data, deviation compensation data reflecting the dynamic mapping relationship of the healing process is generated. The specific implementation is as follows: The core function of the feature tracing unit is to perform healing feature tracing on each feature vector in the feature fusion data to extract the corresponding healing conduction chain. The feature fusion data contains multi-dimensional information from three-dimensional bone sutures, occlusal function, and soft tissue distribution. For example, a bone-occlusal function fusion tensor formed by fusing the first spatial vector and the second correlation matrix, a soft tissue-bone healing fusion matrix formed by fusing the third mapping sequence and the first spatial vector, etc. For each feature vector (such as the bone block displacement feature vector in the fusion tensor, the soft tissue thickness feature vector in the fusion matrix), the feature tracing unit identifies the key feature chain directly related to the healing process through time series analysis and spatial path tracking.

[0065] Taking the three-dimensional bone suture feature vector as an example, the feature tracing unit first performs time series expansion on the bone block displacement parameters in the first spatial vector, and extracts the displacement trajectories of each anatomical sub-region (such as the bone segment in the anterior maxillary tooth area) at different follow-up time points (such as a displacement of 0.2 mm on the first day after surgery and a cumulative displacement of 0.8 mm on the seventh day). Combining the bone suture healing parameters (the second quantization label), the system constructs a "displacement-healing" correlation map, tracks the healing density change nodes on the displacement path (such as the healing density at point A in the middle of the displacement path reaches 0.4 g / cm³ on the third day after surgery, and the healing density at point B reaches 0.7 g / cm³ on the tenth day after surgery), and forms a healing conduction chain from the initial position of the bone block to the current position. This conduction chain contains a series of ordered (displacement position, healing density, time) triples.

[0066] For the occlusal function feature vector (such as the bite force transmission vector in the second correlation matrix), the feature tracing unit analyzes the transfer trajectory of the occlusal contact points over time. For example, in the early stage after orthognathic surgery, due to the unstable displacement of the bone blocks, the occlusal contact points may be concentrated in the molar area; as the bone sutures heal, the contact points gradually shift to the anterior tooth area. The system generates a healing conduction chain of the occlusal function by identifying the spatial migration path of the occlusal contact points (such as from the first mandibular molar to the mandibular central incisor) and the bite force values at the corresponding time points (such as the bite force in the molar area drops from 80 N to 50 N, and the bite force in the anterior tooth area rises from 20 N to 60 N). This conduction chain reflects the recovery process of the occlusal function from the compensatory state to the normal state.

[0067] The tracing of the healing characteristics of the soft tissue distribution feature vector is based on the spatio-temporal changes of the third mapping sequence. For example, the soft tissue thickness parameter around the surgical incision thickens due to edema in the early postoperative period and then gradually subsides. The feature tracing unit extracts the soft tissue thickness change curve within 5 mm around the incision (such as the thickness increasing by 1.5 mm on the 1st postoperative day and returning to the preoperative level on the 14th day), and combines it with the blood flow parameter in this area (such as the blood flow peak value being 2.3 times the baseline value on the 3rd postoperative day and dropping to 1.2 times the baseline value on the 7th day) to construct a soft tissue healing conduction chain, which records the sequential characteristics of edema subsidence and blood circulation reconstruction.

[0068] The function of the parameter superposition unit is to perform anatomical parameter superposition on the healing conduction chain extracted from each feature vector and the corresponding feature fusion data to generate deviation compensation data. Anatomical parameter superposition is not simply data splicing, but based on the relevance of anatomical structures, cross-integrating the dynamic features in the conduction chain and the static features in the fusion data. For example, in the bone-occlusal function fusion tensor, the bone block displacement node (such as the anterior tooth segment displacement of 0.6 mm) in the healing conduction chain is superposed with the corresponding occlusal contact point parameter (the occlusal force in the anterior tooth area is 55 N) in the fusion tensor to generate a new composite parameter - the "displacement-occlusal force synergy value", which is equal to the product of the displacement distance and the occlusal force (0.6 mm × 55 N = 33 N mm), reflecting the load state of the occlusal function at this displacement.

[0069] For the soft tissue-bone healing fusion matrix, the parameter superposition unit superposes the blood flow peak time point (the 3rd postoperative day) in the soft tissue healing conduction chain with the healing density parameter (0.3 g / cm³) in the corresponding bone suture area in the fusion matrix to generate a "blood circulation-healing correlation value", which reflects the degree of influence of the blood circulation state on the healing speed after normalization. In addition, for the occlusal-soft tissue stress fusion vector, the system superposes the contact point transfer node (such as the contact point transferring from the premolar area to the molar area on the 7th postoperative day) in the occlusal conduction chain with the soft tissue stress parameter (the soft tissue stress in the molar area increases by 15 kPa) to generate an "occlusal stress transfer amount" for evaluating the mechanical influence of occlusal function changes on soft tissues.

[0070] The noise suppression unit performs phase shift effect elimination processing on the superimposed deviation compensation data. The phase shift effect stems from time delays or spatial registration errors during multi-modal data acquisition. For example, there is a 5-second time difference between CT image acquisition and bite force sensor data acquisition, resulting in a timing misalignment between the sutural displacement parameters and the bite data. The noise suppression unit uses the cross-correlation algorithm to calculate the time delay amount of different modal data, and performs time translation on the lagged data through linear interpolation to achieve phase alignment. For spatial registration errors (such as coordinate deviations between the three-dimensional suture model and soft tissue ultrasound images), the system uses the Iterative Closest Point (ICP) algorithm for spatial transformation to unify the coordinate systems of different modalities into the craniofacial anatomical reference coordinate system and eliminate spatial phase deviations.

[0071] Throughout the processing flow of the feature correction layer, the system follows the logical framework of "anatomical structure constraint - time series association - parameter dynamic superposition". For example, when analyzing a patient after sagittal split ramus osteotomy, the feature tracing unit first tracks the posterior movement trajectory of the mandibular ramus segment and the change in the suture healing density in the corresponding area to generate a suture healing conduction chain; at the same time, it analyzes the thickness change of the soft tissue around the ipsilateral temporomandibular joint and the tension change of the masseter muscle to generate a soft tissue conduction chain. The parameter superposition unit superimposes the posterior movement amount of the bone segment (1.2 mm) and the masseter muscle tension value (45 N) to form a deviation compensation parameter reflecting the skeletal-muscular mechanical balance; the noise suppression unit eliminates the phase error caused by the difference in device sampling frequencies by calibrating the timestamps of CT and electromyogram signals to ensure the spatio-temporal consistency of the compensation parameters.

[0072] The output data of the feature correction layer - the deviation compensation data - is stored in the form of a multi-dimensional tensor, and each tensor element corresponds to the feature correction value at a specific anatomical location and a specific time point. For example, the three-dimensional tensor element (x, y, z, t) represents the deviation compensation value at the anatomical coordinates (x, y, z) at time t, which is generated through conduction chain superposition and noise elimination and is used to adjust the weight coefficients of each parameter in the feature fusion data. For example, when the suture healing density in a certain area is lower than the expected value, the deviation compensation data will increase the weight of this parameter in the feature fusion, enabling the system to prioritize attention to this abnormal area in subsequent evaluations.

[0073] Example 5: This example describes the specific composition and working principle of the instruction trigger layer. The instruction trigger layer includes a multi-dimensional feature collaboration unit, a dynamic weight optimization unit, and an abnormal state recognition unit, and realizes the generation process from feature data to follow-up evaluation instructions through the collaboration of multiple units. The specific implementation is as follows: The multi-dimensional feature collaboration unit consists of multiple instruction decision nodes. Each node corresponds to a clinical evaluation logic and is connected to each feature vector in the configuration parameters, deviation compensation data, and feature fusion data. For example, a certain node associates the healing density of the three-dimensional sutures (from the first calibration vector) with the uniformity of the bite force conduction of the occlusal function (from the second calibration matrix). Its configuration parameters are the preset density threshold (such as 0.6 g / cm³) and the uniformity threshold (such as 85%). When the suture healing density in a certain anatomical area is lower than the threshold and the bite force uniformity is lower than the threshold, the node triggers an evaluation instruction of "delayed healing and abnormal function". The evaluation logic of the node is constructed based on clinical guidelines, such as the correlation criteria for skeletal healing and functional recovery in the "Orthognathic Surgery Postoperative Follow-up Standards".

[0074] The dynamic weight optimization unit iteratively optimizes the configuration parameters through a dynamic weight adjustment algorithm to minimize the error between the follow-up evaluation instruction and the actual recovery trajectory. This algorithm uses the gradient descent method, and the objective function is defined as:

[0075] where is the error value, is the number of samples, is the actual recovery status vector annotated by clinical experts, is the evaluation instruction vector generated by the system. The algorithm updates the configuration parameters (such as thresholds, weight coefficients) of each instruction decision node through backpropagation to make the objective function gradually decrease. For example, for the "suture healing density - bite force uniformity" node, the algorithm may adjust the density threshold from 0.6 g / cm³ to 0.55 g / cm³ to adapt to the healing characteristics of a specific patient group.

[0076] The abnormal state recognition unit locates abnormal healing based on the deviation compensation data and feature fusion data. The system first establishes a feature template for the normal healing mode, which is generated by analyzing the mean and standard deviation of a large amount of historical postoperative data. For the real-time input feature data, the unit calculates the Mahalanobis distance between the feature vector of each anatomical sub-region and the template:

[0077] where is the real-time feature vector, is the template mean vector, is the template covariance matrix. When the Mahalanobis distance of a certain area exceeds the preset critical value (such as ), it is determined that there is abnormal healing in this area, and it is accurately located to the specific anatomical position (such as the suture area of the anterior wall of the maxilla, the group of occlusal contact points of the right posterior teeth), and a follow-up evaluation instruction containing the abnormal type (such as delayed healing, stress concentration) and location information is generated.

[0078] In the instruction generation process, the node outputs of the multi-dimensional feature collaboration unit are combined through logic gates (such as "AND", "OR") to form a preliminary evaluation conclusion. For example, the node of "suture healing density lower than the threshold" and the node of "soft tissue edema lasting more than 2 weeks" are logically combined through "AND" to generate an instruction of "possible infection or blood circulation disorder". The dynamic weight optimization unit adjusts the trigger conditions of the logic gates according to historical evaluation data, for example, increasing the trigger probability of the "OR" logic from 40% to 60% to adapt to the high-risk patient group.

[0079] The positioning accuracy of the abnormal state recognition unit depends on the spatial resolution of the feature data. For example, in three-dimensional suture data, the system divides the jaw bone into voxel units of 0.5mm×0.5mm×0.5mm, and each unit corresponds to a feature vector. By calculating the Mahalanobis distance for each voxel, millimeter-level positioning of the abnormal area is achieved. For occlusal function data, each occlusal contact point is used as the minimum analysis unit, and the risk area of occlusal trauma is identified through the abnormal pattern of force conduction at the contact point (such as a 50% sudden increase in unilateral biting force).

[0080] The output instructions of the instruction trigger layer adopt a structured format, including the following fields: anatomical location (such as "posterior edge of the ascending ramus of the mandible"), abnormal type (such as "delayed suture healing"), recommended measure (such as "increase the frequency of CT reexamination"), and timestamp (such as "14th day after surgery"). For example, when the system detects that the suture healing density at the posterior edge of the ascending ramus of the mandible is continuously lower than 80% of the template mean for two weeks, and the blood flow of the masseter muscle in the corresponding area is 1.5 times higher than the mean, the instruction is generated: "Anatomical location: Posterior edge of the ascending ramus of the mandible; Abnormal type: Delayed suture healing with soft tissue inflammation; Recommended measure: Perform local ultrasound examination and adjust antibiotic medication; Timestamp: May 22, 2025, 14:30".

[0081] The processing process of the entire instruction trigger layer integrates data-driven algorithm optimization and rule constraints of clinical knowledge. The dynamic weight optimization unit improves the system's adaptability to individual differences through machine learning, while the multi-dimensional feature collaboration unit and the abnormal state recognition unit ensure that the evaluation logic conforms to clinical diagnosis and treatment norms. For example, when dealing with patients with temporomandibular joint disorders, the system optimizes the weight of the "suture displacement in the joint area - condylar movement trajectory" node, gives priority to the feature data in the joint area, and avoids possible missed diagnosis problems in traditional evaluations.

[0082] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0083] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A follow-up system for orthognathic surgery of maxillofacial deformities, characterized in that, Including: An image analysis model for performing multi-dimensional feature fusion processing on postoperative initial data, where the postoperative initial data includes a first spatial vector corresponding to three-dimensional bone suture feature data, a second correlation matrix corresponding to occlusal function feature data, and a third mapping sequence corresponding to soft tissue distribution feature data. The three-dimensional bone suture feature data includes a first distribution label generated by bone block displacement parameters and a second quantization label generated by bone suture healing parameters. The multi-dimensional feature fusion processing includes performing spatial coordinate normalization processing on the first spatial vector, performing time series alignment processing on the second correlation matrix, performing anatomical region annotation processing on the third mapping sequence, inputting the processed three types of data into a feature fusion network, and extracting cross-modal fusion features through layer-by-layer feature crossing to form a multi-dimensional fusion feature set; A follow-up evaluation model for performing dynamic interference suppression processing on real-time physiological data streams and inputting them into a feature generation layer for recovery trajectory simulation, and generating a follow-up evaluation instruction according to the output result of the feature generation layer. The dynamic interference suppression processing includes using an adaptive filtering algorithm to suppress high-frequency noise and removing abnormal fluctuation signals through joint time-domain and frequency-domain analysis. The recovery trajectory simulation includes matching the processed real-time physiological data streams with corresponding stage data of multiple types of historical postoperative data, extracting the feature change trends at each time node, and generating a postoperative recovery prediction trajectory through time series prediction; The feature generation layer includes an image preprocessing module and a dynamic feature generation module. Among them, the image preprocessing module is used for anatomical interval segmentation and invalid signal filtering of the original image data stream, and the dynamic feature generation module is obtained through joint training based on multiple types of historical postoperative data and real-time physiological parameters. The dynamic feature generation module is used to construct a neural network architecture including a feature parsing layer, a feature correction layer, and an instruction trigger layer.

2. The orthognathic surgery follow-up system for maxillofacial deformities according to claim 1, characterized in that, The feature parsing layer is used for performing anatomical feature association processing on different feature vectors in the original image data stream to generate feature fusion data; The feature correction layer is used for modeling the dynamic mapping relationship between the feature fusion data corresponding to each feature vector to generate deviation compensation data. The instruction trigger layer is used for performing multi-dimensional feature fusion based on the deviation compensation data and the feature fusion data to generate a follow-up evaluation instruction.

3. The orthognathic surgery follow-up system for maxillofacial deformities according to claim 2, wherein The modeling of the dynamic mapping relationship between the feature fusion data corresponding to each feature vector to generate deviation compensation data includes: Using a bone suture segmentation algorithm to identify the healing key nodes in the feature fusion data, and determining the correction compensation mapping corresponding to each feature vector based on the type of physiological parameters corresponding to each healing key node; Calculating the offset coefficient between the healing nodes with the same parameters in the correction compensation mappings corresponding to any two feature vectors, and generating the deviation compensation data between the any two feature vectors based on the offset coefficient.

4. The follow-up system for orthognathic surgery of maxillofacial deformities according to claim 3, wherein The calculating the offset coefficient between the healing nodes with the same parameters in the correction compensation mappings corresponding to any two feature vectors includes: When the number of healing nodes in the correction compensation mappings corresponding to any two eigenvectors is inconsistent, virtual healing point interpolation is performed based on the physiological parameters corresponding to the terminal healing nodes in the one with fewer healing nodes, and the offset coefficient between the healing nodes of the same parameter is calculated based on the interpolated data.

5. The follow-up system for orthognathic surgery of maxillofacial deformities according to claim 1, wherein The image preprocessing module is specifically configured to: Perform equal-gradient partitioning on the first spatial vector, the second correlation matrix, and the third mapping sequence according to a preset anatomical interval to generate standardized suture data, standardized occlusion data, and standardized soft tissue data; Perform real-time matching on the standardized suture data and the standardized occlusion data by using a dynamic parameter alignment method, and perform steady-state optimization on the standardized soft tissue data by using a fixed anatomical window mechanism, and output a first calibration vector, a second calibration matrix, and a third calibration sequence; wherein, the first calibration vector includes a calibrated first distribution label and a calibrated second quantization label.

6. The follow-up system for orthognathic surgery of maxillofacial deformities according to claim 5, wherein The image preprocessing module is further configured to: Calculate the healing offset coefficient between the calibrated first distribution label and the calibrated second quantization label during the historical follow-up period; Predict the expected distribution value of the calibrated second quantization label during the real-time follow-up stage according to the healing offset coefficient and the parameter value of the calibrated first distribution label during the real-time follow-up stage; Generate target correction compensation data based on the calibrated second quantization label and its expected distribution value, and use the eigenvector corresponding to the target correction compensation data as the first calibration vector.

7. The follow-up system for orthognathic surgery of maxillofacial deformities according to claim 2, characterized in that, The feature correction layer specifically includes: A feature traceability unit, configured to perform healing feature traceability on each eigenvector in the feature fusion data respectively, so as to extract the corresponding healing conduction chain from each eigenvector; A parameter superposition unit, configured to perform anatomical parameter superposition on the healing conduction chain extracted from each eigenvector and the corresponding feature fusion data to generate deviation compensation data.

8. The follow-up system for orthognathic surgery of maxillofacial deformities according to claim 7, characterized in that, The feature correction layer further includes: A noise suppression unit, configured to perform phase shift effect elimination processing on the deviation compensation data.

9. The orthognathic surgery follow-up system for maxillofacial deformities according to claim 2, wherein The instruction trigger layer specifically includes: A multi-dimensional feature collaboration unit, including a plurality of instruction decision nodes, and each instruction decision node is connected to each eigenvector in the deviation compensation data and the feature fusion data through configuration parameters; A dynamic weight optimization unit, configured to iteratively optimize the configuration parameters through a dynamic weight adjustment algorithm to minimize the error between the follow-up evaluation instruction and the actual recovery trajectory; An abnormal state recognition unit, configured to perform abnormal healing positioning based on the deviation compensation data and the feature fusion data to generate a follow-up evaluation instruction.

10. A follow-up method for orthognathic surgery of maxillofacial deformities, applied to the follow-up system for orthognathic surgery of maxillofacial deformities described in any one of claims 1 to 9, characterized in that, Including the following steps: Perform multi-dimensional feature fusion processing on the postoperative initial data by using an image analysis model, where the postoperative initial data includes a first spatial vector corresponding to three-dimensional suture feature data, a second correlation matrix corresponding to occlusal function feature data, and a third mapping sequence corresponding to soft tissue distribution feature data, and the three-dimensional suture feature data includes a first distribution label generated by bone block displacement parameters and a second quantization label generated by suture healing parameters; Perform dynamic interference suppression processing on the real-time physiological data stream through the follow-up evaluation model, and input the processed data stream into the feature generation layer for recovery trajectory simulation; Use the image preprocessing module in the feature generation layer to perform anatomical interval segmentation and invalid signal filtering on the original image data stream; Adopt a dynamic feature generation module jointly trained by multi-type historical postoperative data and real-time physiological parameters for recovery trajectory simulation. The dynamic feature generation module includes a feature analysis layer, a feature correction layer, and an instruction trigger layer connected in sequence; According to the output result of the feature generation layer, generate a follow-up evaluation instruction by the follow-up evaluation model.

Images