Collaborative management system of personalized medical instrument 3d printing design cloud platform
By using a multimodal data acquisition and collaborative adaptation evaluation module, the cross-modal data fusion and role collaboration issues of the personalized medical device 3D printing design cloud platform were solved, achieving high-precision collaborative management and dynamic adaptation, and improving collaborative efficiency and data privacy protection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI NINTH PEOPLES HOSPITAL SHANGHAI JIAO TONG UNIV SCHOOL OF MEDICINE
- Filing Date
- 2025-11-28
- Publication Date
- 2026-06-23
AI Technical Summary
Current cloud platforms for 3D printing design of personalized medical devices suffer from problems such as fragmented cross-modal data fusion, insufficient accuracy and reliability of virtual-real mapping, poor adaptability of role collaboration, and insufficient dynamic scene response capabilities.
By employing a multimodal twin data acquisition module, a role collaboration adaptation evaluation module, and a closed-loop feedback index construction module, features are filtered through mutual information entropy, and cross-modal mapping accuracy and consistency indices are constructed to achieve deep data association and dynamic adaptation, thereby optimizing collaborative management.
It improved the correlation of data across different stages, dynamically adjusted role permissions, enhanced collaboration efficiency and data privacy protection capabilities, optimized the postoperative feedback guidance rate, and achieved high-precision collaborative management.
Smart Images
Figure CN121583489B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical device 3D printing technology, specifically relating to a collaborative management system for a cloud platform for personalized medical device 3D printing design. Background Technology
[0002] With the upgrading of personalized medical needs and the maturity of 3D printing technology, the market demand for personalized medical devices continues to grow. These products require close collaboration across all stages, including clinical practice, design, production, and operation. Cloud platforms, with their advantages of cross-regional data sharing and multi-role collaborative scheduling, have become the core carrier for 3D printing design of personalized medical devices. However, current collaborative management technologies for cloud platforms in 3D printing design of personalized medical devices still face several key bottlenecks, making it difficult to meet the collaborative requirements of high precision, high efficiency, and high reliability. Specific limitations are as follows:
[0003] Multimodal data is often stored in the form of isolated modules. For example, clinical imaging data lacks a deep correlation with design geometric parameters, and sensor data at the production end does not form a mapping and verification with finished product testing results, resulting in cross-stage data being unable to effectively support personalized design and production decisions.
[0004] The system lacks the ability to accurately identify the core needs of the roles and collaborative scenarios. For example, when designers frequently access clinical biomechanical requirements documents, the system cannot automatically perceive the intention to adapt to clinical biomechanical performance. In the face of scenarios where emergency implant design needs to be completed within 48 hours, the system cannot prioritize the allocation of data interaction resources between the clinical and design ends, resulting in a significant reduction in collaborative efficiency.
[0005] Therefore, there is an urgent need for a collaborative management system for a cloud platform for the design of personalized medical device 3D printing to solve the above problems. Summary of the Invention
[0006] The purpose of this invention is to provide a collaborative management system for a cloud platform for the design of personalized medical devices using 3D printing, in order to solve the technical problems in the prior art such as fragmented cross-modal data fusion, insufficient accuracy and reliability of virtual-real mapping, poor adaptability of role collaboration, and insufficient dynamic scene response capability.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] The collaborative management system of the personalized medical device 3D printing design cloud platform includes:
[0009] The multimodal twin data acquisition module is used to build a twin system that supports deep cross-modal data association, reliable evidence storage and dynamic self-optimization. It clarifies the data acquisition type and flow logic, and uses mutual information entropy to screen preprocessed features and highly correlated features of collaborative quality at each stage, providing a highly correlated feature set for mapping accuracy calculation.
[0010] The role-based collaborative adaptation assessment module calculates cross-modal mapping accuracy based on mutual information entropy, then constructs a dual-perception system of intent and scene. Subsequently, it optimizes data quality through edge federation-enhanced data preprocessing, based on role permissions, scene requirements, and role intent. Figure 3 Using structured data and cross-modal mapping accuracy, a role collaboration and adaptation index is constructed to quantify the degree of matching among the three.
[0011] The closed-loop feedback index construction module is used to combine dynamic interaction features and cross-modal mapping accuracy to construct a cross-modal mapping consistency index. Then, based on the follow-up data from the operation and maintenance end, a postoperative feedback time-series feature vector is generated. Combined with the detection data from the production end and the design optimization parameters, an optimized performance time-series feature vector is generated. The closed-loop feedback iteration index is then used to realize hierarchical decision-making for collaborative management.
[0012] Furthermore, the data collection type and flow logic are clearly defined, and the specific methods are as follows:
[0013] The types of data collected cover patient images and demand text data from the clinical end, geometric features, simulation parameters, and historical process data from the design end, equipment timing parameters and finished product testing data from the production end, follow-up text, rehabilitation timing data, and feedback indicator data from the operation and maintenance end, raw data of role permissions, raw data of scenario requirements, and raw data of intent requirements.
[0014] After deduplication, compliance verification, and standardization, all raw data is input into the feature interaction network to mine cross-stage feature interaction relationships and output structured physical data values. The physical data is synchronously mapped to the virtual model to generate virtual data, which is then preprocessed in the same way to obtain twin data values.
[0015] Furthermore, highly correlated features between preprocessing features and collaborative quality at each stage are selected using mutual information entropy. The specific method is as follows:
[0016] Using formula Let P(x, y) represent the mutual information entropy, where X is the preprocessed feature set, which is obtained by summarizing the local preprocessed features of each edge node, Y is the collaborative quality label, which is generated by manual annotation and automatic classification based on historical collaborative results, P(x, y) is the joint probability of feature x and quality label y appearing at the same time, which is obtained by dividing the frequency of x and y appearing together in historical data by the total amount of data, P(x) is the marginal probability of feature x appearing, which is obtained by dividing the frequency of x appearing by the total amount of data, and P(y) is the marginal probability of quality label y appearing, which is obtained by dividing the frequency of y appearing by the total amount of data.
[0017] Based on industry quality standards for personalized medical devices and verification according to actual needs, a preset feature correlation threshold Ia is established. When the time is right, it is judged as a highly correlated feature.
[0018] Furthermore, the cross-modal mapping accuracy is calculated based on mutual information entropy, specifically using the following method:
[0019] Using formula Indicates the cross-modal mapping accuracy, where This represents the modal attention weight, where i represents the i-th modality. Let be the angle between the twin data vector and the physical data vector in the i-th mode, and n be the number of modes. Indicates the accuracy of cross-modal mapping;
[0020] From the formula express, Describes the feature set of the i-th mode. The modal attention weights represent the sum of mutual information entropies of all modal features and collaborative quality. The sum is 1;
[0021] The cosine similarity between twin data and physical data in the i-th modality is calculated using the vector dot product formula. Received, among which Represents twin eigenvectors, The physical feature vector is a multi-dimensional structured data carrier generated after integrating cross-modal data, with each dimension corresponding to a specific data value.
[0022] Furthermore, a dual perception system of intent and scene is constructed, specifically through the following methods:
[0023] Using the character's real-time operation trajectory and historical collaboration habits as input, the system performs semantic parsing and behavioral association through a behavioral intent model to deduce the character's core needs and ultimately form structured character intent tags.
[0024] Based on the current stage of collaboration, external environmental factors, and data interaction characteristics, collaborative scenarios are classified and defined through a scenario recognition model, forming scenario feature labels.
[0025] Furthermore, based on role permissions, scenario requirements, and role intent... Figure 3 Using structured data and combining cross-modal mapping accuracy, a role collaboration and adaptation index is constructed. The specific method is as follows:
[0026] The original data on role permissions is deduplicated, validated for compliance, and standardized in encoding. The resulting structured data is used to construct the current permission set for each role. The system extracts key information, standardizes and transforms the raw data of the scenario requirements, removes redundancy, and finally generates structured data to construct a set of permissions for the scenario requirements. The process involves mapping the execution behavior of the original intent request data with multimodal data, filtering core requirements, standardizing formats, and finally generating structured data to construct the set of permissions required by the role intent. ;
[0027] Using formula This represents the role-playing compatibility index, where... Indicates the role-playing compatibility index. This indicates that the mapping accuracy affects the amplitude control coefficient. These are the baseline thresholds for mapping accuracy, all of which are constant values, and are set by verifying and setting them based on historical data and actual needs.
[0028] Furthermore, by combining dynamic interaction features with cross-modal mapping accuracy, a cross-modal mapping consistency index is constructed. The specific method is as follows:
[0029] The specific formula for calculating the cross-modal mapping consistency index is as follows: ,in This represents the twin data value of the j-th dimension in the dynamic interaction features. This represents the physical data value of the j-th dimension in the dynamic interaction features, where j is the dimension index within a certain modality, and J is the number of dimensions within that modality. Indicates mapping deviation, This represents the corrected weighting coefficient, with a value range of [0, 1]. The specific value is determined by verification based on actual needs.
[0030] Furthermore, a postoperative feedback time-series feature vector is generated based on follow-up data from the operations and maintenance side. This is combined with production-side detection data and design optimization parameters to generate an optimized performance time-series feature vector. A closed-loop feedback iteration index is then comprehensively constructed. The specific method is as follows:
[0031] Based on follow-up text data and rehabilitation time series data collected from the operation and maintenance terminal, key feedback information is extracted through semantic parsing, rehabilitation indicators are sorted by time dimension to form a time series sequence, and after standardization and feature extraction processing, a structured postoperative feedback time series feature vector is obtained.
[0032] Based on finished product testing data and equipment runtime sequence data from the production end, and combined with simulation parameters optimized by design, the performance monitoring data of the product from production to operation and maintenance stages are integrated along the time axis. After screening and normalization, a time sequence feature vector reflecting the optimized performance of the product is formed.
[0033] The cross-modal mapping consistency index and the role collaboration adaptation index are combined and expressed using the formula. Closed-loop feedback iteration exponent, where This represents the postoperative feedback time-series feature vector. The time-series feature vector represents the performance characteristics, and DTW represents the dynamic time warping distance. This represents the historical maximum value obtained from the statistical calculation of DTW distances for all similar time-series data pairs. and To maintain a fixed weight, extract data from past collaborative case studies on the platform. , And the corresponding final synergistic effect indicators, were calculated through multiple regression analysis. , The variance contribution ratios of the two synergistic effect indicators are normalized so that their sum is 1, resulting in the following: and The true value, This represents the minimum value (in this embodiment, we take...). ).
[0034] Furthermore, hierarchical decision-making for collaborative management is achieved through a closed-loop feedback iterative index. The specific method is as follows:
[0035] Based on historical data, statistics Based on the distribution pattern and the correlation of failure rates, thresholds I1 and I2 are determined, where I1 is... The minimum value of the first a1% interval in the historical distribution interval, I2 is The maximum value in the last a2% interval of the historical distribution interval, where a1 and a2 are constant values, and the specific values are set based on actual needs.
[0036] when At the same time, maintain the current collaboration strategy;
[0037] when At this time, non-critical tasks are paused at the edge, and the cloud is combined with... Optimize the logic linking feedback rules with the preprocessed data from step one;
[0038] when When necessary, an emergency response is initiated to identify the core issues of cross-modal mapping deviation and role adaptation conflict, and collaboration is restored after rectification meets the standards.
[0039] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0040] 1. This invention supports cross-modal data linkage such as clinical images, design geometric data, production sensor parameters, and operation and maintenance follow-up text through a fusion engine, which effectively improves the correlation of cross-linked data. Based on multimodal features, a two-way mapping between twin data and physical data is constructed. The cross-modal mapping accuracy is calculated by weighted cross-modal cosine similarity. Combined with dimension-by-dimensional deviation correction, the matching accuracy deviation between twin data and physical data is effectively reduced.
[0041] 2. This invention analyzes the core needs of roles through a behavioral intent model and defines collaborative scenarios through a scene recognition model. It combines a role collaboration adaptation index to quantify the matching degree of permissions, scenarios, and intents, thereby achieving dynamic adjustment of permissions and avoiding collaboration interruptions caused by insufficient permissions. Based on scene feature tags, it prioritizes the scheduling of core resources, thereby achieving dynamic adaptation between roles and scenarios and improving collaboration efficiency and data privacy protection capabilities.
[0042] 3. This invention generates a postoperative feedback time-series feature vector from follow-up data at the operation and maintenance end, and generates an optimized performance time-series feature vector by combining it with detection data at the production end. The similarity between the two is measured by dynamic time warping distance, and then a closed-loop feedback iteration index is constructed to effectively improve the postoperative feedback's guidance rate for design and production optimization, thereby achieving collaborative quality dynamic assessment and risk warning. Attached Figure Description
[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0044] Figure 1 The collaborative management system module of the personalized medical device 3D printing design cloud platform of the present invention is shown;
[0045] Figure 2 The figure shows the steps of the method for calculating the role collaboration adaptation index of the present invention;
[0046] Figure 3 The diagram illustrates the steps of the collaborative management hierarchical decision-making method of the present invention. Detailed Implementation
[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0048] like Figure 1 , Figure 2 and Figure 3 The collaborative management system of the personalized medical device 3D printing design cloud platform shown includes the following:
[0049] The multimodal twin data acquisition module builds a twin system that supports deep cross-modal data association, reliable evidence storage, and dynamic self-optimization. It clarifies the data acquisition type and flow logic, and uses mutual information entropy to screen preprocessed features and highly correlated features of collaborative quality at each stage, providing a highly correlated feature set for mapping accuracy calculation.
[0050] A digital twin foundation is built, addressing multiple dimensions of data including clinical needs, design models, production equipment, product entities, and postoperative follow-up. Raw data directly collected from physical devices (such as patient images from clinical imaging equipment, process parameters from production sensors, and rehabilitation data from maintenance follow-up records) undergoes deduplication, compliance verification, and standardization before being input into a feature interaction network. This network mines cross-stage feature interaction relationships, ultimately outputting multi-dimensional structured feature values, denoted as physical data values. A virtual model is mapped in real-time based on the physical data, synchronously simulating the state and characteristics of the physical entity (e.g., the design twin model replicates the dimensional parameters of the physically collected data, and the production equipment twin module simulates the process operation status). The generated virtual data, after preprocessing consistent with the physical data, is input into the feature interaction network to extract multi-dimensional feature values, denoted as twin data values. Ensuring a one-to-one correspondence with the physical data dimensions, a multimodal fusion engine and blockchain evidence nodes are provided. The multimodal fusion engine supports cross-modal association of images, text, sensor data, 3D models, and collaborative state-related data. The blockchain evidence nodes record key collaborative data (including core preprocessed data) in real-time and generate hash values to ensure traceability.
[0051] In the clinical setting, patient images and requirement texts are linked to generate a three-dimensional constraint model, and change records are uploaded to the blockchain.
[0052] On the design side, design geometric data and historical process data are integrated to calculate the appropriate attention weight, and the design version hash is uploaded to the blockchain.
[0053] On the production side, link equipment timing data and finished product inspection data to identify dimensions affecting quality, and record process adjustments on the blockchain.
[0054] On the operations and maintenance side, follow-up texts and recovery time series data are mined to extract decay patterns and synchronized to the self-evolution engine and blockchain.
[0055] Core raw data collection involves three types of core raw data: role and permission raw data (platform-preset basic permission list for roles, dynamic adjustment records of collaborative compliance, and temporary permission application records for emergency tasks), scenario requirement raw data (historical permission usage records for the same scenario, current scenario task objectives and constraints), and intent requirement raw data (real-time operation trajectory of roles, historical collaborative habits, and semantic descriptions of behaviors). All raw data is synchronized in real time to the blockchain notarization node and the multimodal fusion engine.
[0056] By using mutual information entropy to filter highly correlated features between preprocessing features (such as clinical imaging features and equipment time-series features) and collaborative quality (collaboration efficiency and product qualification rate) at each stage, a highly correlated feature set is provided for mapping accuracy calculation. The specific formula is shown below:
[0057] ;
[0058] Wherein, X is the preprocessing feature set (including image size and requirement parameters at the clinical end, geometric features and simulation parameters at the design end, process temperature and printing speed at the production end, rehabilitation time series data and feedback indicator data at the operation and maintenance end), which is obtained by summarizing the local preprocessing data of each edge node; Y is the collaborative quality label (based on the historical collaborative process results, including whether the collaborative task was completed on time, whether the product test was qualified, and whether the fault was effectively avoided, etc., the historical data is first labeled with basic labels, such as efficient collaboration, qualified product, and fault collaboration, and then the label pattern is learned through the classification algorithm to realize the automatic label generation of new data, and finally form a standardized collaborative quality label set), which is generated by manual labeling and automatic classification by algorithm based on historical collaborative results; P(x, y) is the joint probability of feature x and quality label y appearing at the same time, which is obtained by dividing the frequency of x and y appearing together in the historical data by the total amount of data; P(x) is the marginal probability of feature x appearing, which is obtained by dividing the frequency of x appearing by the total amount of data; P(y) is the marginal probability of quality label y appearing, which is obtained by dividing the frequency of y appearing by the total amount of data.
[0059] A larger I(X, Y) indicates a higher feature correlation. Based on industry quality standards for personalized medical devices and verification according to actual needs, a preset feature correlation threshold Ia is established. When a feature is identified as highly correlated, the filtered highly correlated features will serve as the core input for cross-modal mapping.
[0060] The role-based collaborative adaptation assessment module calculates cross-modal mapping accuracy based on mutual information entropy, then constructs a dual-perception system of intent and scene. Subsequently, it optimizes data quality through edge federation-enhanced data preprocessing, based on role permissions, scene requirements, and role intent. Figure 3 Using structured data and cross-modal mapping accuracy, a role collaboration and adaptation index is constructed to quantify the degree of matching among the three.
[0061] The modal attention weights are obtained by normalizing the mutual information entropy I(X,Y), and the cross-modal mapping accuracy is calculated using weighted cross-modal cosine similarity. The specific calculation formula is as follows:
[0062] ;
[0063] in This represents the modal attention weight, where i represents the i-th modality. Let be the angle between the twin data vector and the physical data vector in the i-th modality, and n be the number of modalities, which is determined by the actual modal types involved in the fusion (e.g., if there are three modalities: image, text, and sensor, then n=3). This indicates that the cross-modal mapping accuracy ranges from [0, 1], constraining the calculation boundaries of subsequent core indices.
[0064] Furthermore, From the formula express, Describes the feature set of the i-th mode. The sum of mutual information entropy of all modal features and collaborative quality is used to normalize the weights and ensure the modal attention weights. The sum is 1.
[0065] Furthermore, The cosine similarity between twin data and physical data in the i-th modality is calculated using the vector dot product formula. Received, among which Represents twin eigenvectors, The physical feature vector is a multi-dimensional structured data carrier generated after integrating cross-modal data, with each dimension corresponding to a specific data value;
[0066] A dual perception system of intent and scenario is constructed. Based on the real-time operation trajectory of the role (such as designers frequently accessing clinical requirement documents and modifying key parameters of the model) and historical collaboration habits (such as a doctor's preference for specific mechanical performance requirements), semantic parsing and behavioral association are performed through behavioral intent models to deduce the core demands of the role (such as adjusting design parameters to adapt to clinical mechanical requirements and troubleshooting equipment failures to ensure production progress), forming structured role intent tags and completing the construction of role intent. Based on the current collaboration stage (clinical requirement docking, design and development, production and manufacturing, postoperative operation and maintenance), external environmental factors (such as emergency surgery requirements, equipment failure alarms, and routine batch production), and data interaction characteristics (such as high-frequency data transmission and intensive cross-role communication), collaborative scenarios are classified and defined through scenario recognition models (such as emergency adaptation scenarios, fault troubleshooting scenarios, and routine collaboration scenarios), forming scenario feature tags and completing the construction of collaborative scenarios.
[0067] Enhanced data preprocessing is performed through edge federation, and distributed processing is carried out at the edge. Each node performs data noise reduction and spatiotemporal alignment locally, avoiding the centralized transmission of sensitive data.
[0068] Through federated learning and collaborative optimization, preprocessed model parameters are shared to improve data consistency across the entire chain.
[0069] By using scenario-based feature filtering, combined with dual labels of intent and scenario, features highly relevant to the current intent and scenario are prioritized for retention (such as clinical need features and design modification features in emergency adaptation scenarios), thereby improving the relevance of subsequent calculations.
[0070] The original data on role permissions is deduplicated (removing duplicate permission records), validated (filtering valid permissions that conform to the platform's permission rules), and standardized (unified to the platform's standard permission codes). The resulting structured data is used to construct the current permission set for each role. For example, the current permission set for a doctor role is {clinical needs entry permission code, patient image viewing permission code, treatment plan review permission code, postoperative feedback submission permission code}. Key information extraction (extracting core requirements related to permissions and tasks), standardization transformation (converting natural language scene descriptions into recognizable related data by a multimodal fusion engine), and redundancy removal (deleting scene information irrelevant to cross-modal collaboration) are performed on the raw data of the scene requirements. The resulting structured data is used to construct the scene requirement permission set. The process involves mapping the execution behavior of the original data related to intent requirements to multimodal data (binding role operation trajectories to multimodal data such as clinical needs and design models), filtering core requirements (retaining the associated data necessary to realize the role intent), and unifying the format (aligning with other multimodal data formats). The resulting structured data is used to construct the set of permissions required for the role intent. .
[0071] A role collaboration adaptation index is constructed by combining cross-modal mapping accuracy to quantify the degree of matching between role permissions, collaboration scenario requirements, and role core intent. The specific formula is shown below:
[0072] ;
[0073] in, Indicates the role-playing compatibility index. This indicates that the mapping accuracy affects the amplitude control coefficient, and the setting is verified by combining historical data with actual needs. The baseline threshold for mapping accuracy is set based on historical data statistics. At this time, it demonstrates the positive empowerment of cross-modal mapping accuracy for collaborative adaptation. At the same time, the overestimation bias of the adaptation index is corrected.
[0074] The closed-loop feedback index construction module combines dynamic interaction features and cross-modal mapping accuracy to construct a cross-modal mapping consistency index. Then, it generates a postoperative feedback time-series feature vector based on follow-up data from the operation and maintenance end, and generates an optimized performance time-series feature vector by combining production end detection data and design optimization parameters. The closed-loop feedback iteration index is then comprehensively constructed to achieve hierarchical decision-making for collaborative management.
[0075] A cross-modal mapping consistency index is set based on dynamic interaction features and cross-modal mapping accuracy. By considering the feature interaction effects of multi-dimensional links (such as design, production, and feedback), while inheriting the judgment of basic mapping accuracy, key features are selected through comparative learning. This avoids redundant interference caused by cross-modal mapping accuracy covering all modalities. The specific calculation formula of the cross-modal mapping consistency index is as follows:
[0076] ;
[0077] in This represents the twin data value of the j-th dimension in the dynamic interaction features. This represents the physical data value of the j-th dimension in the dynamic interaction feature, where j is the dimension index within the modality (e.g., the j-th dimension in an image modality, such as the size dimension), and J is the number of dimensions within the modality. Indicates mapping deviation, This represents the corrected weighting coefficient, with a value range of [0, 1]. The specific value is determined by verification based on actual needs.
[0078] Based on follow-up text data and rehabilitation time-series data (such as records of rehabilitation indicators at different stages after surgery) collected from the operation and maintenance end, key feedback information is extracted through semantic parsing. The rehabilitation indicators are sorted by time dimension to form a time-series sequence. After standardization and feature extraction, a structured postoperative feedback time-series feature vector is obtained. Based on finished product testing data (such as dimensional accuracy and mechanical strength test results) and equipment operation time-series data from the production end, combined with simulation parameters after design optimization, the performance monitoring data of the product from production to operation and maintenance stages are integrated along the time axis. After screening and normalization, a time-series feature vector reflecting the optimized performance of the product is formed.
[0079] By combining the cross-modal mapping consistency index and the role collaboration adaptation index, a closed-loop feedback iteration index is set, as shown in the following formula:
[0080] ;
[0081] in This represents the postoperative feedback time-series feature vector. Representing the performance time-series feature vector, DTW stands for Dynamic Time Warping Distance, a distance metric used to measure the similarity between two time-series data of different lengths and with misaligned time axes. Ultimately, it calculates the minimum cumulative distance reflecting the differences in core change trends. The smaller the value, the closer the trend of postoperative feedback matches the optimized performance effect. This represents the historical maximum value among all similar time-series data pairs (i.e., the combination of postoperative feedback time-series characteristics and performance time-series characteristics of all products). and For fixed weights, and The sum of the values is 1, extracting data from past collaborative cases on the platform. , And the corresponding final collaborative effect indicators (such as product qualification rate and collaborative cycle compliance rate), were calculated through multiple regression analysis. , The variance contribution ratios of the two synergistic effect indicators are normalized so that their sum is 1, resulting in the following: and The true value, This represents the minimum value (in this embodiment, we take...). ).
[0082] Based on historical data, statistics Based on the distribution pattern and the correlation of failure rates, thresholds I1 and I2 are determined, where I1 is... The minimum value of the first a1% interval in the historical distribution interval, I2 is The maximum value in the last a2% interval of the historical distribution range. a1 and a2 are constant values, and the specific values are set based on actual needs.
[0083] when At the same time, maintain the current collaboration strategy;
[0084] when At this time, non-critical tasks are paused at the edge, and the cloud is combined with... Optimize the logic linking feedback rules with the preprocessed data from step one;
[0085] when At that time, an emergency response is initiated to locate the cross-modal mapping deviation (based on...). ), role compatibility conflict (based on The core issue is to restore collaboration after rectification and compliance.
[0086] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
[0087] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A collaborative management system for a cloud platform for personalized medical device 3D printing design, characterized in that, include: The multimodal twin data acquisition module is used to build a twin system that supports deep cross-modal data association, reliable evidence storage and dynamic self-optimization. It clarifies the data acquisition type and flow logic, and uses mutual information entropy to screen preprocessed features and highly correlated features of collaborative quality at each stage, providing a highly correlated feature set for mapping accuracy calculation. The types of data collected cover patient images and demand text data from the clinical end, geometric features, simulation parameters, and historical process data from the design end, equipment timing parameters and finished product testing data from the production end, follow-up text, rehabilitation timing data, and feedback indicator data from the operation and maintenance end, raw data of role permissions, raw data of scenario requirements, and raw data of intent requirements. After all the original data is deduplicated, compliance-checked, and standardized, it is input into the feature interaction network to mine cross-link feature interaction relationships and output structured physical data values. The physical data is synchronously mapped to the virtual model to generate virtual data, and after the same preprocessing, twin data values are obtained. The role collaboration adaptation assessment module calculates cross-modal mapping accuracy based on mutual information entropy, then constructs a dual perception system of intent and scene, and subsequently optimizes data quality through edge federation enhanced data preprocessing. Based on three types of structured data—role permissions, scene requirements, and role intent—and combined with cross-modal mapping accuracy, it constructs a role collaboration adaptation index to quantify the degree of matching among the three. The closed-loop feedback index construction module is used to combine dynamic interaction features and cross-modal mapping accuracy to construct a cross-modal mapping consistency index. Then, based on the follow-up data from the operation and maintenance end, it generates a postoperative feedback time-series feature vector. Combined with the production end detection data and design optimization parameters, it generates an optimized performance time-series feature vector. The closed-loop feedback iteration index is then used to realize hierarchical decision-making for collaborative management. The specific method for constructing a closed-loop feedback iteration index is as follows: Based on the follow-up text data and rehabilitation time series data collected by the operation and maintenance end, key feedback information is extracted through semantic parsing, rehabilitation indicators are sorted by time dimension to form a time series sequence, and after standardization and feature extraction processing, a structured postoperative feedback time series feature vector is obtained. Based on finished product testing data and equipment runtime sequence data from the production end, and combined with simulation parameters optimized by design, the performance monitoring data of the product from production to operation and maintenance stages are integrated along the time axis. After screening and normalization, a time sequence feature vector reflecting the optimized performance of the product is formed. The cross-modal mapping consistency index and the role collaboration adaptation index are combined and expressed using the formula. Closed-loop feedback iteration exponent, where This represents the postoperative feedback time-series feature vector. The time-series feature vector represents the performance characteristics, and DTW represents the dynamic time warping distance. This represents the historical maximum value obtained from the statistical calculation of DTW distances for all similar time-series data pairs. and To maintain a fixed weight, extract data from past collaborative case studies on the platform. , And the corresponding final synergistic effect indicators, were calculated through multiple regression analysis. , The variance contribution ratios of the two synergistic effect indicators are normalized so that their sum is 1, resulting in the following: and The true value, This represents the minimum value and should be set according to actual needs.
2. The collaborative management system of the personalized medical device 3D printing design cloud platform according to claim 1, characterized in that, The method for selecting highly correlated features between preprocessing features and collaborative quality at each stage using mutual information entropy is as follows: Using formula Let P(x, y) represent the mutual information entropy, where X is the preprocessed feature set, which is obtained by summarizing the local preprocessed features of each edge node, Y is the collaborative quality label, which is generated by manual annotation and automatic classification based on historical collaborative results, P(x, y) is the joint probability of feature x and quality label y appearing at the same time, which is obtained by dividing the frequency of x and y appearing together in historical data by the total amount of data, P(x) is the marginal probability of feature x appearing, which is obtained by dividing the frequency of x appearing by the total amount of data, and P(y) is the marginal probability of quality label y appearing, which is obtained by dividing the frequency of y appearing by the total amount of data. Based on industry quality standards for personalized medical devices and verification according to actual needs, a preset feature correlation threshold Ia is established. When the time is right, it is judged as a highly correlated feature.
3. The collaborative management system of the personalized medical device 3D printing design cloud platform according to claim 1, characterized in that, The method for calculating cross-modal mapping accuracy based on mutual information entropy is as follows: Using formula Indicates the cross-modal mapping accuracy, where This represents the modal attention weight, where i represents the i-th modality. Let be the angle between the twin data vector and the physical data vector in the i-th mode, and n be the number of modes. Indicates the accuracy of cross-modal mapping; From the formula express, Describes the feature set of the i-th mode. The modal attention weights represent the sum of mutual information entropies of all modal features and collaborative quality. The sum is 1; The cosine similarity between twin data and physical data in the i-th modality is calculated using the vector dot product formula. Received, among which Represents twin eigenvectors, The physical feature vector is a multi-dimensional structured data carrier generated after integrating cross-modal data, with each dimension corresponding to a specific data value.
4. The collaborative management system of the personalized medical device 3D printing design cloud platform according to claim 1, characterized in that, The specific method for constructing a dual perception system of intent and scene is as follows: Using the character's real-time operation trajectory and historical collaboration habits as input, the system performs semantic parsing and behavioral association through a behavioral intent model to deduce the character's core needs and ultimately form structured character intent tags. Based on the current stage of collaboration, external environmental factors, and data interaction characteristics, collaborative scenarios are classified and defined through a scenario recognition model, forming scenario feature labels.
5. The collaborative management system of the personalized medical device 3D printing design cloud platform according to claim 3, characterized in that, Based on three types of structured data—role permissions, scenario requirements, and role intent—a role collaboration and adaptation index is constructed by combining cross-modal mapping accuracy. The specific method is as follows: The original data on role permissions is deduplicated, validated for compliance, and standardized in encoding. The resulting structured data is used to construct the current permission set for each role. The system extracts key information, standardizes and transforms the raw data of the scenario requirements, removes redundancy, and finally generates structured data to construct a set of permissions for the scenario requirements. The process involves mapping the execution behavior of the original intent request data with multimodal data, filtering core requirements, standardizing formats, and finally generating structured data to construct the set of permissions required by the role intent. ; Using formula This represents the role-playing compatibility index, where... Indicates the role-playing compatibility index. This indicates that the mapping accuracy affects the amplitude control coefficient. These are the baseline thresholds for mapping accuracy, all of which are constant values, and are set by verifying and setting them based on historical data and actual needs.
6. The collaborative management system of the personalized medical device 3D printing design cloud platform according to claim 3, characterized in that, By combining dynamic interaction features and cross-modal mapping accuracy, a cross-modal mapping consistency index is constructed. The specific method is as follows: The specific formula for calculating the cross-modal mapping consistency index is as follows: ,in This represents the twin data value of the j-th dimension in the dynamic interaction features. This represents the physical data value of the j-th dimension in the dynamic interaction features, where j is the dimension index within the modality, and J is the number of dimensions within the modality. Indicates mapping deviation, This represents the corrected weighting coefficient, with a value range of [0, 1]. The specific value is determined by verification based on actual needs.
7. The collaborative management system of the personalized medical device 3D printing design cloud platform according to claim 1, characterized in that, Hierarchical decision-making for collaborative management is achieved through a closed-loop feedback iterative index. The specific method is as follows: Based on historical data, statistics Based on the distribution pattern and the correlation of failure rates, thresholds I1 and I2 are determined, where I1 is... The minimum value of the first a1% interval in the historical distribution interval, I2 is The maximum value in the last a2% interval of the historical distribution interval, where a1 and a2 are constant values, and the specific values are set based on actual needs. when At the same time, maintain the current collaboration strategy; when At this time, non-critical tasks are paused at the edge, and the cloud is combined with... Optimize the logic linking feedback rules with the preprocessed data from step one; when When necessary, an emergency response is initiated to identify the core issues of cross-modal mapping deviation and role adaptation conflict, and collaboration is restored after rectification meets the standards.