Display Color Management Method and System Based on Federated Learning and Digital Twin

Abstract

This application relates to a display color management method and system based on federated learning and digital twins. The method includes: collecting local color performance data from multiple medical display nodes to train a color prediction model and obtain local model update parameters; aggregating the encrypted local model parameters using a federated learning server to generate a global optimization model; constructing and maintaining a digital twin for each medical display node using a digital twin platform, driving the digital twin to predict future color drift trajectories based on the global optimization model and real-time ambient light data, and generating a predictive calibration instruction when the prediction result exceeds a preset tolerance; executing the predictive calibration instruction to calibrate the target display, acquiring key target data and generating a data fingerprint, and uploading the transaction record containing the data fingerprint to the blockchain by calling a blockchain smart contract. This application achieves predictive calibration, ensuring color consistency of medical displays.

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and in particular to a display color management method and system based on federated learning and digital twins. Background Technology

[0002] In medical imaging diagnosis, medical imaging monitors serve as crucial output devices for doctors to interpret and diagnose images, and their display quality directly impacts the accuracy and reliability of diagnostic results. Ensuring highly consistent color and grayscale performance across multiple monitors within and across medical institutions is a core aspect of the medical quality control system; however, current technology faces multiple challenges. First, significant differences exist between individual medical monitors. Even when using devices of the same brand and model, variations in manufacturing tolerances, usage intensity, ambient lighting conditions, and LCD panel aging rates can cause differences in brightness output, color gamut coverage, and gamma response curves over time.

[0003] Existing calibration schemes based on the DICOM GSDF standard can only guarantee the display accuracy of a single device at a specific moment. They cannot solve the problem of color consistency degradation caused by asynchronous aging of multiple devices during long-term operation. This is especially true in scenarios where large medical institutions deploy dozens or more monitors, where maintaining cluster-level color uniformity faces significant technical obstacles. Secondly, traditional calibration mechanisms rely on fixed-cycle maintenance strategies, such as manual calibration every three months. This passive response mode has obvious drawbacks: later in the calibration interval, the monitor may have experienced color drift beyond the clinically acceptable range, creating a dangerous calibration gap; at the same time, fixed cycles cannot adapt to the actual usage load of different devices. Some high-load devices may require more frequent calibration, while low-load devices face over-maintenance, resulting in a waste of time for professional technicians and calibration equipment resources. Furthermore, as independent operating units, each medical monitor's operating data, calibration history, and aging trends form isolated data silos, lacking an effective data aggregation and analysis mechanism. This makes it impossible for hospitals to establish a group aging model for the monitors, making it difficult to achieve predictive maintenance decisions based on device status. Finally, the quality control records generated during the calibration process are mostly stored in local spreadsheets or paper documents, which pose risks such as record tampering, accidental loss, or incomplete information. In scenarios such as medical quality audits, equipment certification, or medical disputes, it is difficult to provide a legally valid and tamper-proof chain of evidence, and it cannot meet the strict requirements of medical device quality management systems such as ISO 13485 for data integrity and traceability. Summary of the Invention

[0004] The purpose of this application is to propose a display color management method and system based on federated learning and digital twins, which can efficiently and reliably manage display colors, achieve predictive calibration, ensure color consistency of medical displays, and guarantee the integrity and traceability of calibration data.

[0005] To address the aforementioned technical problems, embodiments of this application provide a display color management method based on federated learning and digital twins, comprising: Local color performance data is collected from multiple medical display nodes, and the local color prediction model is trained based on the color performance data to obtain local model update parameters. The local model update parameters are encrypted and uploaded to the federated learning server. The federated learning server then aggregates the encrypted local model parameters to generate a global optimization model, which is then distributed to the digital twin platform. The digital twin platform constructs and maintains a digital twin for each medical display node, and based on the global optimization model and real-time ambient light data, drives the digital twin to predict the future color drift trajectory of the corresponding physical display. When the prediction result exceeds the preset tolerance, a predictive calibration instruction is generated. The predictive calibration instruction is executed to calibrate the target display and acquire key target data during calibration. A data fingerprint is generated based on the key target data, and the transaction record containing the data fingerprint is uploaded to the blockchain by calling a blockchain smart contract.

[0006] To address the aforementioned technical problems, embodiments of this application provide a display color management system based on federated learning and digital twins, comprising: Multiple medical display nodes, each with built-in sensors and computing units, are used to collect local color performance data and train a locally deployed color prediction model based on the color performance data to obtain local model update parameters. The federated learning server communicates with each terminal to receive the encrypted local model update parameters, and generates a global optimization model based on the weighted aggregation of the encrypted local model parameters, and distributes the global optimization model to the digital twin platform. The digital twin service platform, connected to the federated learning server and the environment perception module, is used to build and maintain a digital twin for each medical display node, and based on the global optimization model and real-time ambient light data, drive the digital twin to predict the future color drift trajectory of the corresponding physical display. When the prediction result exceeds the preset tolerance, a predictive calibration instruction is generated. The blockchain service module, connected to the digital twin service platform and the blockchain network, is used to execute the predictive calibration instructions to calibrate the target display, acquire key target data during calibration, generate a data fingerprint based on the key target data, and upload transaction records containing the data fingerprint to the blockchain by calling the blockchain smart contract.

[0007] This invention provides a display color management method and system based on federated learning and digital twins. It aggregates local data through federated learning to generate a global optimization model, combines digital twins to predict color drift and generate calibration instructions, and uses blockchain to record calibration data. This enables efficient and reliable management of display color, achieving predictive calibration, ensuring color consistency of medical displays, reducing the risk of misdiagnosis, optimizing maintenance resource allocation, and guaranteeing the integrity and traceability of calibration data. Furthermore, through digital twin-driven predictive calibration, this invention transforms calibration from a passive response to proactive intervention, fundamentally eliminating the calibration window period in traditional fixed-cycle calibration models and enabling on-demand maintenance, significantly reducing operational costs. Finally, by integrating environmental perception and visual adaptation models, this invention achieves a leap from physical signal consistency to consistency in physician visual perception, ensuring the reliability of diagnostic images under different viewing environments. Attached Figure Description

[0008] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the implementation of the display color management method based on federated learning and digital twin provided in this application embodiment; Figure 2 This is a flowchart illustrating the implementation of the first sub-process in the display color management method based on federated learning and digital twins provided in this application embodiment; Figure 3 This is a flowchart illustrating the implementation of the second sub-process in the display color management method based on federated learning and digital twins provided in this application embodiment; Figure 4 This is a flowchart illustrating the implementation of the third sub-process in the display color management method based on federated learning and digital twins provided in this application embodiment; Figure 5 This is a flowchart illustrating the implementation of the fourth sub-process in the display color management method based on federated learning and digital twins provided in this application embodiment; Figure 6This is a flowchart illustrating the implementation of the fifth sub-process in the display color management method based on federated learning and digital twins provided in this application embodiment; Figure 7 This is a flowchart illustrating the implementation of the sixth sub-process in the display color management method based on federated learning and digital twins provided in this application embodiment; Figure 8 This is a schematic diagram of a display color management system based on federated learning and digital twin provided in an embodiment of this application. Detailed Implementation

[0010] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0011] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0012] In the traditional field of medical imaging diagnostics, maintaining color consistency among medical imaging monitors is difficult during long-term operation and when multiple devices work together. Furthermore, existing calibration strategies often employ fixed-cycle models, resulting in passive and inefficient calibration with calibration gaps. Simultaneously, isolated equipment management hinders the full realization of data value, and the reliability and traceability of quality control records are insufficient, making it difficult to meet stringent medical quality management requirements.

[0013] To address this, this application proposes a display color management method based on federated learning and digital twins, comprising: collecting local color performance data from multiple medical display nodes, training a locally deployed color prediction model based on the color performance data to obtain local model update parameters; encrypting the local model update parameters and uploading them to a federated learning server, and aggregating the encrypted local model parameters through the federated learning server to generate a global optimization model, and distributing the global optimization model to a digital twin platform; constructing and maintaining a digital twin for each medical display node through the digital twin platform, and driving the digital twin to predict the future color drift trajectory of the corresponding physical display based on the global optimization model and real-time ambient light data, generating a predictive calibration instruction when the prediction result exceeds a preset tolerance; executing the predictive calibration instruction to calibrate the target display and obtain key target data during calibration, generating a data fingerprint based on the key target data, and uploading the transaction record containing the data fingerprint to the blockchain by calling a blockchain smart contract.

[0014] For ease of understanding, the following explains some key terms in this embodiment: Federated learning: a distributed machine learning paradigm that allows multiple participants to collaboratively train a global model without sharing the original data. Each participant only uploads updated parameters of its local model, which are then aggregated by a central server, thus protecting data privacy and improving the model's generalization ability.

[0015] Digital twin: A virtual model of a physical entity or system that enables state monitoring, behavior simulation, performance prediction, and optimized control of the physical entity through real-time data connection.

[0016] Medical display node: refers to a single medical display and its associated sensors and computing units deployed in a medical institution, capable of independently collecting data, performing local model training, and communicating with federated learning servers.

[0017] Color prediction model: A machine learning model used to predict the future color performance or drift trend of a display based on factors such as the display's historical performance data and usage environment.

[0018] Local model update parameters: These refer to the model weights, biases, or other parameters that can be used to update the global model after training the color prediction model using local color performance data on each medical display node.

[0019] Federated Learning Server: A central server responsible for receiving encrypted local model update parameters from each medical display node and performing aggregation operations to generate a globally optimized model.

[0020] Globally optimized model: A unified color prediction model with stronger generalization ability and accuracy, formed by aggregating the updated parameters of all local models through a federated learning server.

[0021] Digital twin platform: A software system used to build, manage and maintain multiple digital twins, and to provide functions such as data interface, model operation and instruction generation.

[0022] Digital twin: In a digital twin platform, a virtual counterpart is created for each physical medical display node, which can reflect the status of the physical display in real time, receive predictive model-driven input, and simulate its behavior.

[0023] Real-time ambient light data: refers to data such as light intensity and color temperature in the environment used by the display, which are collected in real time by environmental sensing modules or sensors.

[0024] Color drift trajectory: refers to the predicted path or trend of how a display's color performance (such as brightness and chromaticity) changes over time in the future.

[0025] Predictive calibration command: When the color shift predicted by the digital twin exceeds the preset tolerance, the digital twin platform generates and sends the command to the target display to guide it in calibration.

[0026] Key target data: refers to the important data related to the calibration process, results, and environment that needs to be collected and recorded during the display calibration operation.

[0027] Data fingerprint: This can be a hash value, Merkle root, or any other string that can uniquely identify data. A hash value is a fixed-length binary string obtained by calculating data of arbitrary length using a hash algorithm; it has the characteristics of data integrity verification and immutability.

[0028] Blockchain smart contracts: A piece of automatically executable and tamper-proof code deployed on a blockchain network to implement preset business logic and data storage functions.

[0029] This application utilizes federated learning to aggregate multi-node data and generate a global optimization model, effectively improving the accuracy and consistency of color prediction for medical displays. A digital twin, combined with real-time ambient light data, enables predictive management of display color drift, transforming passive calibration into proactive intervention and significantly reducing calibration gaps. Simultaneously, blockchain technology ensures the immutability and traceability of calibration records, providing a reliable basis for medical device quality control and comprehensively improving the display quality and management efficiency of medical image diagnosis.

[0030] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0031] Please see Figure 1 , Figure 1 This paper illustrates a specific implementation of a display color management method based on federated learning and digital twins.

[0032] It should be noted that if substantially the same result is obtained, the method of this invention is not based on... Figure 1 Limited to the order of the processes shown, this method includes the following steps: S1: Collect local color performance data through multiple medical display nodes, and train the locally deployed color prediction model based on the color performance data to obtain local model update parameters.

[0033] Specifically, multiple medical display nodes collect local color performance data, and a locally deployed color prediction model is trained based on this data to obtain local model update parameters. Each medical display node can be equipped with a built-in or external color sensor, such as a photometer or colorimeter, to periodically measure the display's brightness, chromaticity, gamma curve, and other color performance parameters. This measurement data is stored in a local computing unit. The local computing unit can deploy a basic machine learning model, such as a linear regression model or a simple multilayer perceptron model. This model is trained using the locally collected color performance data to learn the patterns of color drift in the display. After training, the model generates a set of update parameters, such as the model's weights and biases, which reflect the specific drift characteristics of the local display.

[0034] S2: The local model update parameters are encrypted and uploaded to the federated learning server. The federated learning server then aggregates the encrypted local model parameters to generate a global optimization model, which is then distributed to the digital twin platform.

[0035] Specifically, before uploading, local model update parameters can be encrypted using standard encryption algorithms, such as symmetric or asymmetric encryption, to protect data privacy. The encrypted parameters are then transmitted to the federated learning server via a secure network channel, such as a TLS / SSL-based communication link. Upon receiving the encrypted parameters from all participating nodes, the federated learning server can aggregate them using a simple weighted averaging strategy. For example, it can perform an equal-weighted average of all encrypted parameters or a weighted average based on the number of parameters uploaded by each node. After aggregation, the server decrypts the aggregation result, generating a globally optimized model. This globally optimized model is then distributed to the digital twin platform as the basis for predictions by all digital twins.

[0036] Please see Figure 2 , Figure 2 A specific implementation of step S2 is shown below: S21: The local model update parameters are encrypted and uploaded to the federated learning server to obtain the encrypted local model update parameters. S22: The federated learning server groups the medical display nodes according to their device models or usage scenarios corresponding to preset modes of the medical display nodes. S23: The dynamic weights of the encrypted local model update parameters within each group are determined. S24: A weighted average is calculated on all the encrypted local model update parameters within the same group based on the dynamic weights to obtain an encrypted aggregation result. S25: The encrypted aggregation result within each group is decrypted, generating a global optimization model corresponding to each group, and the global optimization model is distributed to the digital twin platform.

[0037] Specifically, the local model update parameters are encrypted and uploaded to the federated learning server to obtain the encrypted local model update parameters. This step aims to ensure the security and privacy of data transmission. Local model update parameters typically contain sensitive local training information, and direct transmission may lead to data leakage. Through encryption, even if the data is intercepted during transmission, it is difficult for unauthorized third parties to interpret it, thus protecting the data privacy of the medical display nodes. For example, homomorphic encryption technology can be used, allowing the federated learning server to perform aggregation calculations on the model parameters in encrypted form without decryption; alternatively, a secure multi-party computation protocol can be used, enabling multiple medical display nodes to jointly calculate the aggregation result without revealing their respective local model parameters.

[0038] Subsequently, the federated learning server groups the medical display nodes according to their device models or usage scenarios corresponding to their preset modes. This grouping mechanism is crucial for addressing the heterogeneity of medical display nodes. Different device models may have different hardware characteristics, color gamuts, and aging curves, while different usage scenarios (e.g., diagnostic rooms, operating rooms, and image reading rooms) may have varying requirements for color accuracy and consistency. Grouping allows nodes with similar characteristics to be categorized together, making subsequent model aggregation and optimization more targeted and avoiding the problems of insufficient model generalization or poor adaptability to specific groups caused by a "one-size-fits-all" aggregation approach. For example, the federated learning server can maintain a device model database. Upon receiving local model update parameters, it assigns each medical display node to the corresponding device model group based on the device model information reported by each node. Alternatively, when registering or configuring, medical display nodes can preset their primary usage scenario modes and upload this information along with the local model update parameters. The federated learning server then groups the nodes according to these preset modes.

[0039] Based on this, dynamic weights are determined for the updated parameters of the encrypted local models within each group. This step aims to address the issue of uneven contributions from different medical display nodes to the global model within a group. Even within the same group, different nodes may exhibit differences in factors such as color performance data quality, data volume, and historical performance. Dynamic weights allow for a more reasonable allocation of each node's influence during the aggregation process, ensuring that high-quality, high-contribution nodes have a greater impact on the global model, thereby improving the accuracy and robustness of the aggregated model. For example, weights can be calculated based on the amount of color performance data provided by each node, data freshness, and its historical calibration performance; alternatively, weights can be dynamically adjusted based on the performance of the node's local model on the validation set in the previous federated learning iteration.

[0040] Next, a weighted average is calculated on all the encrypted local model update parameters within the same group based on the dynamic weights to obtain the encrypted aggregation result. This step is the core operation of federated learning aggregation. It uses the dynamic weights determined in the previous step to calculate a weighted average on all the encrypted local model update parameters within the same group. The weighted average calculation can comprehensively consider the contribution of each node to the global model, making the aggregation result more representative and effectively suppressing the impact of abnormal or low-quality data on the overall model. Since the local model update parameters are encrypted, this calculation needs to be performed in ciphertext to maintain data privacy. For example, if homomorphic encryption is used, the federated learning server can directly perform the weighted average calculation on the ciphertext to obtain the encrypted aggregation result; or, if secure multi-party computation is used, each node can jointly calculate the weighted average through the protocol without decrypting its own parameters and send the encrypted aggregation result to the federated learning server.

[0041] Finally, the encrypted aggregation results within each group are decrypted to generate a global optimization model corresponding to each group, and this global optimization model is then distributed to the digital twin platform. This step is the final step in the aggregation process. After obtaining the encrypted aggregation results, they need to be decrypted to recover the usable model parameters. Since each group has undergone customized aggregation, a global optimization model corresponding to that group is generated. These models are then distributed to the digital twin platform to drive the digital twins within the corresponding groups to perform color drift prediction, thereby achieving targeted and high-precision color management. For example, the decryption key is typically held by the federated learning server or a trusted third party and used to decrypt the aggregation results; after decryption, the federated learning server distributes the global optimization models for these specific groups to the digital twin platform through a secure communication channel. The digital twin platform then associates the model with the corresponding medical display node digital twin based on the group to which it belongs.

[0042] This application introduces a refined grouping mechanism and dynamic weight adjustment strategy in the federated learning aggregation process. Specifically, data transmission security is ensured by encrypting and uploading local model update parameters. The federated learning server groups medical display nodes based on device models or preset usage scenarios, effectively solving the heterogeneity problem and making the aggregation process more targeted, avoiding insufficient model generalization ability that may be caused by uniform aggregation. On this basis, by determining the dynamic weights of the encrypted local model update parameters within each group, the contribution of each node to the global model can be more reasonably evaluated, allowing high-quality, high-contribution nodes to play a greater role in the aggregation, thereby improving the accuracy and robustness of the aggregated model. Finally, a weighted average calculation based on dynamic weights is performed and decrypted to generate a customized global optimization model for each group. After these models are distributed to the digital twin platform, they can more accurately drive the digital twins within the corresponding groups to predict the color drift trajectory of the physical displays. Compared to aggregation methods that do not group or use fixed weights, the solution proposed in this application significantly improves the adaptability and accuracy of the color prediction model, thereby providing medical displays with more refined and accurate color management capabilities and effectively ensuring the accuracy and reliability of medical image diagnosis.

[0043] Please see Figure 3 , Figure 3 A specific implementation of step S23 is shown below: S231: Calculate the data quality score for each of the encrypted local model parameters, wherein the data quality score is the geometric mean of the calibration accuracy value and the sensor reliability value. S232: Calculate a comprehensive weight value using a predefined weighting function based on the data quality score, data sample size, color drift rate of the medical display node, and cumulative power-on time, wherein encrypted local model parameters with data quality scores higher than a first threshold are assigned a superlinear weight reward. S233: Obtain the historical contribution score and model update stability score of the medical display node, and fine-tune the comprehensive weight value based on the historical contribution score and the model update stability score to obtain the dynamic parameters.

[0044] Specifically, when calculating the data quality score for each encrypted local model parameter, the data quality score is the geometric mean of the calibration accuracy value and the sensor reliability value. This step aims to comprehensively evaluate the reliability and accuracy of the updated local model parameters. The calibration accuracy value can be quantified based on the deviation between the local model's predicted results and the actual calibration results (e.g., calculated using the Delta E value or the CIE L*a*b*color difference formula), or determined by assessing whether the calibrated display meets the preset DICOM GSDF standard conformance level. The sensor reliability value can be obtained by evaluating the sensor's own calibration cycle, self-test results, historical failure rate, or comparison results with a standard reference sensor. Using the geometric mean effectively balances the two key indicators of calibration accuracy and sensor reliability, avoiding the excessive influence of extreme values ​​of a single indicator on the overall score, thereby ensuring the comprehensiveness and reliability of the data quality assessment.

[0045] Based on this, a comprehensive weight value is calculated using a predefined weighting function according to the data quality score, data sample size, color drift rate of the medical display node, and cumulative power-on time. Superlinear weight rewards are then assigned to encrypted local model parameters whose data quality scores exceed a first threshold. This step aims to assign an initial weight to the local model update parameters based on multi-dimensional information and effectively incentivize high-quality data. The predefined weighting function can be a linear weighting function, such as summing the data quality score, data sample size, color drift rate, and cumulative power-on time after multiplying them by preset coefficients; or it can be a nonlinear function, such as a sigmoid function or an exponential function, to better reflect the nonlinear relationship between the contributions of each factor to the weight. The superlinear weight reward mechanism, for example, when the data quality score exceeds the first threshold, can use an exponential or polynomial function to increase the weight, making the weight of high-quality data grow faster than linearly, thus giving high-quality data a greater influence during the aggregation process. Among them, the color shift rate can be calculated using historical calibration data or digital twin prediction data, such as the amount of brightness or chromaticity change per unit time; the cumulative power-on time can be obtained through the monitor's built-in timer or system records.

[0046] Further, the historical contribution score and model update stability score of the medical display node are obtained, and the comprehensive weight value is fine-tuned based on the historical contribution score and the model update stability score to obtain the dynamic parameters. This step aims to further optimize the weight allocation, fully considering the long-term performance of the node and the reliability of model updates. The historical contribution score can be calculated based on the number of times the node has participated in federated learning in the past, the magnitude of the improvement of the global model performance by its submitted model updates, or the frequency of its data being adopted. The model update stability score can be determined by evaluating the magnitude of change of the model update parameters submitted by the node between consecutive training rounds, or the performance fluctuation of its model on the local validation set. For example, if a node's model update parameters frequently fluctuate significantly in a short period of time, its stability score is low. The fine-tuning method can use a multiplicative factor or an additive factor to introduce the historical contribution score and the model update stability score into the calculation of the comprehensive weight value, thereby obtaining the final dynamic weight parameters.

[0047] This application addresses the issue of inaccurate weight allocation during federated learning aggregation by introducing a multi-dimensional evaluation and dynamic adjustment mechanism, significantly improving the accuracy of the global optimization model and the reliability of color drift prediction. Specifically, the calculation of the data quality score integrates calibration accuracy and sensor reliability values, ensuring a comprehensive and reliable quality assessment of local model update parameters, laying a solid foundation for subsequent weight allocation. Building upon this, by combining data sample size, color drift rate of medical display nodes, and cumulative power-on time—information on device operating status—a predefined weighting function is used to calculate a comprehensive weight value. High-quality data is assigned a superlinear weight reward, enabling the weight allocation to dynamically adapt to the actual conditions and data contributions of different displays, effectively incentivizing high-quality data participation and reducing aggregation bias. Furthermore, by acquiring historical contribution scores and model update stability scores of medical display nodes and fine-tuning the comprehensive weight value based on these scores, the final dynamic weight parameters not only consider current data quality and device status but also incorporate long-term node performance and model update reliability, enhancing the stability and long-term reliability of the weights. Overall, this refined dynamic weight determination mechanism ensures the fairness and effectiveness of the federated learning aggregation process, enabling the generated global optimization model to more accurately capture the color drift patterns of displays. This provides a more reliable predictive basis for the digital twin platform, ultimately improving the overall accuracy and efficiency of color management for medical displays.

[0048] It should be noted that by assigning superlinear weights to high-quality data, the aggregation mechanism of this invention can effectively incentivize nodes with high-quality data, prevent low-quality data from having an excessively negative impact on the global model, and thus converge to a better global model more quickly.

[0049] S3: Construct and maintain a digital twin for each medical display node through the digital twin platform, and drive the digital twin to predict the future color drift trajectory of the corresponding physical display based on the global optimization model and real-time ambient light data. When the prediction result exceeds the preset tolerance, generate a predictive calibration instruction.

[0050] Specifically, a digital twin is constructed and maintained for each of the aforementioned medical display nodes through the digital twin platform. Based on the global optimization model and real-time ambient light data, the digital twin predicts the future color drift trajectory of the corresponding physical display. When the prediction result exceeds a preset tolerance, a predictive calibration command is generated. The digital twin platform creates a virtual digital twin for each physical medical display node. This digital twin stores the current state information of the corresponding physical display, such as current brightness and chromaticity, and continuously receives real-time data updates from the physical display. The digital twin uses the global optimization model issued to the digital twin platform, combined with real-time ambient light data (such as ambient illuminance and color temperature) obtained from the environmental perception module, to simulate and predict the color drift trend of the physical display over a future period. The prediction result is represented by a series of future brightness and chromaticity values, which together constitute the color drift trajectory. When any predicted value in the predicted color drift trajectory, such as brightness or chromaticity, exceeds a preset tolerance range, the digital twin platform triggers the generation of a predictive calibration command. This instruction can include basic information such as the type of calibration and recommended calibration parameters.

[0051] Please see Figure 4 , Figure 4 A specific implementation of step S3 is shown below: S31: Construct and maintain a digital twin for each medical display node using the digital twin platform. S32: Construct a time-series input vector including the current brightness, chromaticity, panel temperature, cumulative power-on time, and real-time ambient light data of the digital twin. S33: Based on the time-series input vector, output predicted brightness and chromaticity values ​​for a future preset time period using the global optimization model to generate the color drift trajectory and obtain the prediction result. S34: When the prediction result exceeds the preset tolerance, generate the predictive calibration instruction.

[0052] Specifically, the digital twin platform is a software system or service architecture used to create, manage, and operate digital twins. Its function is to provide a virtual space for real-time mapping and simulation of the state, behavior, and performance data of the medical display nodes in the physical world. This platform can be deployed as a cloud-based service, utilizing the powerful computing and storage capabilities of the cloud to process massive amounts of data and support complex model calculations; alternatively, it can adopt a hybrid model combining edge computing and cloud computing, performing partial data preprocessing and real-time response on edge devices close to the medical display nodes to reduce latency and improve efficiency. The medical display node refers to a physical display device deployed in medical institutions for medical image diagnosis, with built-in sensors and computing units capable of collecting local color performance data. In the digital twin system, each medical display node corresponds to a digital twin, which is the data source and command execution object of the digital twin, serving as a bridge between the physical and digital worlds. The digital twin is a virtual mapping of the medical display node in the digital twin platform, and it maintains synchronization with the corresponding physical display through real-time data streaming. The digital twin described herein not only includes the static attributes of the display, but more importantly, dynamically reflects its operating status and environmental factors. The digital twin can be a complex data model that represents the various parameters and historical behavior of the physical display through structured data storage and association; or it can be a simulation model containing simulation logic and prediction algorithms, capable of simulating the physical display's response under different conditions and its future drift trends.

[0053] Based on this, a time-series input vector is constructed, including the current brightness, chromaticity, panel temperature, cumulative power-on time, and real-time ambient light data of the digital twin. This time-series input vector is an ordered set of data used to drive the global optimization model for prediction. It contains key state parameters and environmental data of the digital twin at different points in time. Its function is to provide the prediction model with comprehensive and dynamic contextual information to capture the inherent patterns and external influences of display color drift. This vector can be composed of data collected in real time by sensors, such as obtaining the current brightness and chromaticity through a built-in light sensor, obtaining the panel temperature through a temperature sensor, recording the cumulative power-on time through system logs, and obtaining the real-time ambient light data through an external ambient light sensor; or it can be combined with historical calibration data, usage records, etc., to form richer time-series features. The current brightness refers to the actual luminous intensity of the medical display node at a certain moment, and the chromaticity refers to the color coordinates it displays. These parameters are core indicators for evaluating the color performance of the display and are also the main manifestations of color drift. These measurements can be taken in real time using the monitor's built-in brightness and color sensors, or periodically through precise measurements using an external colorimeter, and then uploaded to the digital twin platform for updates. The panel temperature refers to the real-time operating temperature of the medical display node's display panel. The temperature of the display panel directly affects its luminous efficiency and color performance, and is one of the important physical factors causing color drift. This data is typically monitored and collected in real time using a temperature sensor integrated within the monitor. The cumulative power-on time refers to the total operating time of the medical display node since its commissioning. The degree of monitor aging is closely related to the cumulative power-on time and is a key indicator for predicting its future performance degradation and color drift trends. This data is typically recorded and maintained by the monitor's firmware or operating system. The real-time ambient light data refers to the current light intensity and color temperature of the environment in which the medical display node is located. Changes in ambient light affect the human eye's perception of the display's colors and also have a certain impact on the display's own color performance. This data can be collected in real time using an ambient light sensor deployed near the monitor, or obtained through the ambient light sensor built into the monitor.

[0054] Furthermore, the global optimization model outputs predicted brightness and chromaticity values ​​for a future preset time period based on the time-series input vector to generate the color drift trajectory, thus obtaining the prediction result. The global optimization model is a unified model obtained by aggregating model parameters trained locally on multiple medical display nodes using federated learning technology. Its function is to leverage collective intelligence to learn the general patterns and individual differences in display color drift, thereby providing more accurate and robust prediction capabilities. This model can be a deep learning model based on recurrent neural networks (RNNs) or long short-term memory networks (LSTMs), which excels at processing time-series data and capturing long-term dependencies; it can also be a model based on the Transformer architecture, which better models the complex relationships between different input features through a self-attention mechanism. The future preset time period refers to the time span for the global optimization model to predict color drift. Its function is to provide sufficient foresight for predictive calibration so that calibration can be scheduled before actual drift occurs. This time period can be flexibly set according to actual needs and the aging speed of the display; for example, it can be preset to a week, a month, or longer to adapt to different management strategies and calibration cycles. The predicted luminance and chromaticity values ​​are estimates of the future color performance of the medical display node by the global optimization model based on the time-series input vector. These values ​​directly reflect the expected color state of the display at a future point in time. They are specific components of the color drift trajectory, and these predicted values ​​can quantify the future color change trend of the display. The color drift trajectory is a curve or trend graph of the predicted luminance and chromaticity values ​​over the preset future time period. Its function is to intuitively show the direction and magnitude of the evolution of the display's color performance over time. This trajectory can be presented as a sequence of points on a two-dimensional chromaticity graph, or as a curve graph of luminance and chromaticity changes over time, facilitating analysis and judgment by users or the system. The prediction result refers to the color drift trajectory output by the global optimization model, or key information extracted from the trajectory, such as the maximum deviation value of luminance or chromaticity within the preset future time period. Its function is to provide a quantifiable indicator for comparison with the preset tolerance, thereby determining whether calibration needs to be triggered.

[0055] Finally, when the predicted result exceeds the preset tolerance, the predictive calibration instruction is generated. The preset tolerance refers to the maximum allowable deviation range of the color performance of the medical display node. Its function is to define the acceptable standard for the display's color performance; when the predicted result exceeds this range, the display is considered to require calibration. This tolerance can be determined based on industry standards (such as DICOM GSDF), clinical needs, or user-defined settings. For example, it can be set to a luminance deviation not exceeding a certain percentage, or a chromaticity deviation within a certain ΔE value. The predictive calibration instruction is an operation instruction automatically generated by the digital twin platform and sent to the target display when the predicted result exceeds the preset tolerance. Its function is to achieve proactive, on-demand display calibration, avoiding the diagnostic risks and resource waste caused by passive calibration. This instruction can include information such as the calibration trigger time, target calibration parameters, and calibration method, and can be further encapsulated into an executable automated script or notification information.

[0056] This application achieves precise mapping of the physical display in the digital space by constructing and maintaining a digital twin for each medical display node, laying the foundation for subsequent status monitoring and prediction. By constructing a time-series input vector containing the current brightness, chromaticity, panel temperature, cumulative power-on time, and real-time ambient light data, this application can comprehensively and dynamically capture internal and external factors affecting the display's color performance, solving the problems of isolated data and incomplete information in traditional methods, and providing rich and accurate contextual information for the prediction model. Based on this, the global optimization model outputs the predicted brightness and chromaticity values ​​for the future preset time period based on the time-series input vector, thereby generating the color drift trajectory, making the prediction of the display's future color performance more accurate and reliable. When the prediction result exceeds the preset tolerance, the system can promptly generate the predictive calibration command, thereby transforming the traditional passive, periodic calibration into active, on-demand calibration, effectively avoiding the diagnostic risks caused by the "calibration gap," improving calibration efficiency and timeliness, ensuring that the medical display maintains optimal color performance throughout its entire life cycle, and significantly improving the accuracy and reliability of medical image diagnosis.

[0057] Please see Figure 5 , Figure 5 A specific implementation of step S34 is shown below: S341: When the predicted result exceeds the preset tolerance, based on the ambient illuminance and ambient color temperature in the real-time ambient light data, the visual adaptation model is invoked to calculate the contrast sensitivity function adjustment and color temperature compensation. S342: The contrast sensitivity function adjustment and the color temperature compensation are fused with the target physical parameters based on the display standard to generate target calibration parameters. S343: The target calibration parameters, along with the optimal calibration time window determined based on the predicted drift exceedance time point and the preset work schedule, are encapsulated into the predictive calibration instruction.

[0058] Specifically, when the prediction result exceeds the preset tolerance, the system invokes the visual adaptation model based on the ambient illuminance and ambient color temperature from the real-time ambient light data to calculate the contrast sensitivity function adjustment and color temperature compensation. Here, the ambient illuminance in the real-time ambient light data refers to the brightness level of the environment in which the display is located, and the ambient color temperature refers to the color characteristics of ambient light, usually measured in Kelvin (K). This data can be acquired in real-time by ambient light sensors integrated inside or outside the medical display node, for example, through photodiode arrays or spectrometers. The visual adaptation model aims to simulate the visual perception characteristics of the human eye under different ambient light conditions. For example, it can employ visual adaptation models based on CIE (International Commission on Illumination) recommendations, such as CIECAM02 or CAM16, or a machine learning-based visual adaptation model trained on a large amount of human visual experiment data to predict changes in the human eye's perception of display color and brightness under specific ambient light conditions. The contrast sensitivity function adjustment amount represents the amount of adjustment required for the display contrast to match the target contrast in order to make the contrast perceived by the human eye consistent with the current ambient light conditions; the color temperature compensation amount represents the amount of compensation required for the display color temperature to counteract the influence of ambient light color temperature on the display's white point perception.

[0059] Subsequently, the contrast sensitivity function adjustment and the color temperature compensation are fused with the target physical parameters based on the display standard to generate target calibration parameters. The display standard can be the DICOM GSDF standard, which (Digital Imaging and Communications in Medicine Grayscale Standard Display Function) is a widely adopted grayscale display standard in the field of medical imaging. It defines the grayscale response curves that medical displays should present under different brightness inputs to ensure the consistency of image diagnosis. Target physical parameters based on the DICOM GSDF standard typically include target brightness, target chromaticity (white point chromaticity coordinates), and target gamma curve. The fusion process aims to combine the visual adaptive adjustments caused by ambient light with the physical display characteristics required by the DICOM GSDF standard to generate calibration parameters that both meet medical standards and adapt to the current environment. For example, fusion can be performed using a weighted average method, assigning different weights to the adjustment amount according to the degree of influence of ambient light; or through an iterative optimization algorithm to minimize the impact of ambient light on visual perception while meeting the DICOM GSDF standard. The generated target calibration parameters can be the display's lookup table (LUT) data, gamma curve adjustment parameters, white point color coordinate adjustment values, or brightness gain / offset, etc. These parameters will be directly used for the display's color calibration.

[0060] Based on this, the target calibration parameters and the optimal calibration time window determined according to the predicted drift excess time point and the preset work schedule are encapsulated into the predictive calibration instruction. The predicted drift excess time point refers to the future time point at which the display color drift will exceed the preset tolerance, based on the prediction results of the digital twin. The preset work schedule refers to the regular work schedule, rest time, maintenance window, etc., set by the medical institution for the display or related departments. For example, it can be stored as time period records in a database or calendar events. The determination of the optimal calibration time window aims to select a time period with the least impact on medical work for calibration. For example, based on the predicted drift excess time point and combined with the preset work schedule, time periods with idle equipment, low usage, or non-peak diagnostic periods can be intelligently identified as the optimal calibration window. The predictive calibration instruction is a structured data packet that contains the target calibration parameters for calibrating the display and the recommended optimal time window for performing the calibration operation. For example, it can include information such as calibration type, target brightness, target chromaticity, gamma curve parameters, calibration start time, and calibration end time.

[0061] This application effectively addresses the issues of inaccurate calibration parameters and unreasonable calibration scheduling. By introducing real-time ambient light data and a visual adaptation model, the generation of calibration parameters fully considers the visual perception characteristics of the human eye in different environments. This ensures that the calibrated display provides more accurate color and grayscale performance in actual use, thus improving the reliability of diagnostic images. Simultaneously, by combining predicted drift deviation times with preset work schedules, the optimal calibration time window is intelligently determined. This allows calibration operations to be performed within a timeframe that minimizes or eliminates disruption to medical workflows, avoiding equipment downtime caused by calibration during peak diagnostic periods. This significantly improves equipment utilization efficiency and the overall intelligence level of operation and maintenance. This predictive and adaptive calibration command generation mechanism further enhances the accuracy and practicality of the medical display color management system.

[0062] S4: Execute the predictive calibration instruction to calibrate the target display, acquire the target key data during calibration, generate a data fingerprint based on the target key data, and upload the transaction record containing the data fingerprint to the blockchain by calling the blockchain smart contract.

[0063] Specifically, the system executes the aforementioned predictive calibration instructions to calibrate the target display, acquires key target data during calibration, generates a data fingerprint from this key data, and uploads the transaction record containing the data fingerprint to the blockchain via a smart contract. Upon receiving the predictive calibration instructions, the target display can automatically or with manual confirmation initiate the calibration process. The calibration process adjusts the display's internal parameters to restore its color performance to a standard state. During or after calibration, the system acquires a series of key target data related to the calibration, such as brightness measurements before and after calibration, and calibration time. This collected key data can be directly input into a preset hash algorithm, such as SHA-256, to calculate a unique data fingerprint. Subsequently, the system submits the transaction information containing this data fingerprint to the blockchain network by calling the smart contract interface in the blockchain service module. The blockchain network verifies and records the transaction, thus achieving tamper-proof evidence storage of the calibration record.

[0064] Please see Figure 6 , Figure 6 A specific implementation of step S4 is shown below: S41: Execute the predictive calibration command to calibrate the target display and acquire the target key data during calibration, wherein the target key data includes calibration trigger type, luminance and chromaticity measurements before calibration, luminance and chromaticity measurements after calibration, ambient light intensity and color temperature during calibration, digital identity of the operator performing the calibration, and digital twin prediction data packet triggering this calibration. S42: Standardize the target key data to obtain standardized target key data. S43: Combine the standardized target key data into a string of a preset format and calculate the data fingerprint using a preset hash algorithm. S44: Upload the transaction record containing the data fingerprint to the blockchain by calling the blockchain smart contract.

[0065] Specifically, in the process of executing the predictive calibration instructions to calibrate the target display and acquiring the target key data during calibration, the core is to ensure the actual execution of the calibration process and to simultaneously and comprehensively collect all key information related to this calibration. Calibration can be performed through an automated calibration system that automatically adjusts the display's color parameters based on the predictive calibration instructions, for example, through closed-loop control using a built-in color sensor and calibration software. Alternatively, it can be performed manually in conjunction with a data acquisition module, whereby professionals manually calibrate the display according to instructions, while simultaneously recording relevant data through connected measurement equipment and software. The target key data is a comprehensive record of the calibration process, including the calibration trigger type (e.g., whether triggered by predictive drift deviation or periodic maintenance), luminance and chromaticity measurements before and after calibration (to quantify the calibration effect), ambient illuminance and color temperature during calibration (reflecting environmental conditions during calibration), the digital identity of the operator performing the calibration (ensuring operator traceability), and the digital twin predictive data package that triggered the calibration (providing the background and basis for the calibration). The comprehensiveness of this data is the foundation for subsequent traceability and auditing.

[0066] The target key data is standardized to eliminate data heterogeneity and ensure consistency in format, units, and structure across all calibration data. This can be achieved in several ways, such as defining a unified data model (e.g., JSON or XML format) and mapping all collected data to that model; or by performing unit conversion and dimension unification on data collected from different sensors or devices, for example, unifying brightness values ​​to cd / m². 2 The chromaticity values ​​are standardized to CIE Lab or CIE xyY coordinates. Standardization effectively avoids subsequent processing errors or inefficiencies caused by inconsistent data formats, laying the foundation for data integration and analysis.

[0067] Combining the standardized target key data into a string with a preset format, and then calculating the data fingerprint using a preset hash algorithm, is a crucial step in ensuring data integrity and preventing tampering. Specifically, a string concatenation rule can be predefined; for example, all standardized key data fields can be concatenated in a specific order and with separators (such as commas and semicolons) to form a unique string. Subsequently, a cryptographic hash algorithm, such as SHA-256 or SHA-3, is selected to calculate the string, generating a fixed-length data fingerprint. This data fingerprint is characterized by unidirectionality (the original data cannot be deduced from the data fingerprint) and collision resistance (the probability of different input data generating the same data fingerprint is extremely low). Any minor modification to the original key data will cause a significant change in the calculated data fingerprint, thus easily detecting whether the data has been tampered with.

[0068] By invoking the blockchain smart contract, transaction records containing the data fingerprint are uploaded to the blockchain, leveraging the decentralized, immutable, and traceable characteristics of blockchain technology. In practice, a dedicated smart contract for evidence storage can be pre-deployed on the blockchain network. After the data fingerprint is generated, the system constructs a transaction, using the data fingerprint as part of the transaction payload, and invokes the evidence storage function of the smart contract through the blockchain network interface. This transaction is verified by the blockchain network's consensus mechanism and packaged into a new block, thus permanently recorded on the blockchain. Once uploaded to the chain, the data fingerprint and its associated transaction records cannot be tampered with or deleted, providing a highly reliable traceability credential for record calibration.

[0069] This application effectively addresses the issues of insufficient credibility and lack of traceability in calibration records. By comprehensively acquiring key target data during calibration, it ensures that complete information about the calibration process is recorded. Standardizing this data unifies the data format, improving the efficiency and accuracy of data processing. Furthermore, generating unique data fingerprints from the standardized data provides a robust anti-tampering mechanism for calibration records; any unauthorized modifications will be immediately detected. Finally, transaction records containing data fingerprints are uploaded to the blockchain, leveraging its decentralized and immutable characteristics to provide permanent, transparent, and verifiable traceability credentials for medical monitor calibration records. This not only significantly enhances the compliance of the medical device quality management system and meets stringent audit requirements but also provides irrefutable evidence in the event of medical disputes, thereby significantly enhancing the reliability and security of medical imaging diagnosis.

[0070] Please see Figure 7 , Figure 7 A specific implementation of step S44 is shown below: S441: Construct a transaction message for invoking the notarized smart contract. The payload of the transaction message includes at least: the data fingerprint, a timestamp certificate generated by a trusted timestamp service, and a unique identifier for the target display. S442: Verify and sort the transaction message using a consensus algorithm, and package it into a new block, generating a transaction hash and block number. S443: Return and store the transaction hash and block number as traceability credentials.

[0071] Specifically, when constructing a transaction that invokes a smart contract for evidence storage, this step aims to encapsulate the calibration data to be uploaded to the blockchain into a transaction format that conforms to the requirements of the blockchain network. Specifically, this can be achieved by using blockchain client software deployed on the medical display node or its associated edge computing device to invoke the interface of a smart contract pre-deployed on the blockchain and populate the transaction data according to the parameter structure defined in the contract. Alternatively, a separate blockchain gateway service can be used. This service receives data from a digital twin service platform or other management system, and then constructs and submits the transaction information to the blockchain network, thereby achieving decoupling from the underlying blockchain.

[0072] The payload of the transaction information is the core content of the blockchain transaction, used to carry the key data of the calibration record. The data fingerprint is a fixed-length digest obtained by hashing the target key data at the time of calibration. This data fingerprint has one-wayity, collision resistance, and avalanche effect, ensuring the integrity and immutability of the calibration data. For example, mainstream cryptographic hash algorithms such as SHA-256 (a 256-bit secure hash algorithm) or Keccak-256 can be used to calculate the standardized target key data string to generate a unique hash fingerprint. The timestamp certificate generated by the trusted timestamp service is a time proof provided by an independent, authoritative third-party timestamp service agency or decentralized timestamp protocol (such as OpenTimestamps). This certificate typically contains time information, the timestamped data fingerprint, and the service provider's digital signature, used to prove that the calibration data existed at a specific point in time and has not been tampered with, thereby enhancing the legal validity and credibility of the record. The unique identifier of the target display is a code used to uniquely identify a specific medical display throughout the management system. This identifier can be the display's factory serial number, a unique asset number assigned by the asset management system, the device's MAC address, or a combination identifier that combines the device model, production batch, and internal management code, ensuring that calibration records can be accurately traced back to the specific physical display.

[0073] The transaction information is verified and sorted using a consensus algorithm, and then packaged into a new block, generating a transaction hash and block number. This step is the core mechanism for ensuring transaction validity and data consistency in a blockchain network. A consensus algorithm is a set of rules by which nodes in a blockchain network reach a consensus on the validity of transactions and the generation of blocks. For example, in a public blockchain, a Proof-of-Work (PoW) mechanism can be used, where miners compete for the right to record transactions by solving computational problems, and verified transactions are packaged into new blocks. In consortium or private blockchains, consensus algorithms such as Practical Byzantine Fault Tolerance (PBFT) or RAFT can be used, where transactions are verified and blocks are generated through voting and confirmation by multiple nodes, achieving higher transaction throughput and determinism. Transactions verified and sorted by the consensus algorithm are collected by miners or validators and encapsulated into a new data structure, i.e., a new block. Each block contains the data fingerprint of the previous block, forming an immutable chain structure. During the block generation process, the system generates a unique transaction hash for each transaction and assigns an incrementing block number to each newly generated block as its unique identifier on the blockchain.

[0074] The transaction hash and block number are returned and stored as traceability credentials. This step aims to provide clear and tamper-proof on-chain credentials for subsequent queries, audits, and tracing. Once a transaction is successfully recorded on the blockchain, the blockchain service module returns the transaction hash and the block number of the block to the application or service that initiated the transaction (e.g., a digital twin service platform). These credentials can then be stored in a local database, a distributed file system, or the data model of the digital twin as metadata for the calibration record. Using these credentials, users or auditors can query the corresponding transaction details and block information on a blockchain explorer at any time, thereby verifying the authenticity, completeness, and on-chain time of the calibration record, greatly improving traceability and data credibility.

[0075] This application effectively addresses the issues of insufficient credibility, auditing capabilities, and compliance in calibration records. Specifically, when constructing transaction information, it explicitly requires the payload to include at least a data fingerprint, a timestamp certificate generated by a trusted timestamp service, and a unique identifier for the target display. The data fingerprint ensures the integrity and immutability of critical calibration data; any alteration to the original data will result in a mismatch in the data fingerprint. The trusted timestamp certificate provides authoritative third-party time verification, effectively preventing time reversal or tampering and enhancing the legal validity of the record. The unique identifier for the target display accurately associates the calibration record with a specific physical display, providing a clear basis for subsequent equipment management and traceability. Based on this, a consensus algorithm verifies and sorts the transaction information, packaging it into a new block to generate a transaction hash and block number. This process leverages the decentralized and immutable characteristics of blockchain to ensure the validity, order, and finality of transactions, preventing invalid or malicious records from being uploaded to the chain, and providing a distributed, highly reliable storage foundation for calibration records. Finally, the generated transaction hash and block number are returned and stored as traceability credentials, enabling users or auditors to quickly and accurately query the corresponding calibration records on the blockchain, greatly improving the record's searchability and auditing efficiency. In summary, this application, through a refined blockchain on-chain process—from transaction information construction and on-chain verification to credential storage—comprehensively enhances the completeness, temporal accuracy, device correlation, transaction reliability, and traceability of medical monitor calibration records. This significantly improves the credibility, auditing capabilities, and compliance of calibration records, providing solid data support for quality control in medical imaging diagnosis.

[0076] Please refer to Figure 8 As a response to the above Figure 1 The implementation of the method shown in this application provides an embodiment of a display color management system based on federated learning and digital twins.

[0077] like Figure 8 As shown, the display color management system based on federated learning and digital twins in this embodiment includes: multiple medical display nodes, a federated learning server, a digital twin service platform, and a blockchain service module, wherein: Multiple medical display nodes, each with built-in sensors and computing units, are used to collect local color performance data and train a locally deployed color prediction model based on the color performance data to obtain local model update parameters. The federated learning server communicates with each terminal to receive the encrypted local model update parameters, and generates a global optimization model based on the weighted aggregation of the encrypted local model parameters, and distributes the global optimization model to the digital twin platform. The digital twin service platform, connected to the federated learning server and the environment perception module, is used to build and maintain a digital twin for each medical display node, and based on the global optimization model and real-time ambient light data, drive the digital twin to predict the future color drift trajectory of the corresponding physical display. When the prediction result exceeds the preset tolerance, a predictive calibration instruction is generated. The blockchain service module, connected to the digital twin service platform and the blockchain network, is used to execute the predictive calibration instructions to calibrate the target display, acquire key target data during calibration, generate a data fingerprint based on the key target data, and upload transaction records containing the data fingerprint to the blockchain by calling the blockchain smart contract.

[0078] Specifically, multiple medical display nodes serve as the system's front-end execution units, each with built-in sensors and computing units. These built-in sensors, such as high-precision colorimeters or spectroradiometers, periodically or on-demand collect color performance data from the local display, including key parameters such as brightness, chromaticity, and gamma curves. The built-in computing units then use this collected data to train the deployed color prediction model locally, thereby generating local model update parameters. This distributed data acquisition and local training mechanism effectively avoids the privacy leakage risks associated with centralized uploading of raw sensitive data, while also reducing the computational burden on the central server. For example, the computing units can run lightweight neural network models or support vector machine models, iteratively updating model weights based on local data. Furthermore, external color calibration devices can be connected to the local computing terminal via standard interfaces, with the external devices handling data acquisition and the local computing terminal responsible for model training and parameter generation.

[0079] The federated learning server, acting as the core coordinator of the system, communicates with each medical display node. Its primary function is to receive encrypted local model update parameters from each node and perform weighted aggregation based on these encrypted parameters to generate a globally optimized model. To ensure the security of data transmission and aggregation, the federated learning server can employ techniques such as homomorphic encryption or secure multi-party computation to process the received parameters. Aggregation algorithms can include federated averaging (FedAvg) or weighted averaging to synthesize the learning outcomes of each node. For example, the federated learning server can be deployed in the cloud or a private data center, receiving and processing parameters through a secure aggregation protocol. Alternatively, the federated learning server can be a distributed cluster, handling parameters uploaded by a large number of nodes through load balancing and fault tolerance mechanisms, and utilizing GPUs to accelerate aggregation computation, ensuring efficient model updates. The generated globally optimized model is then distributed to the digital twin service platform to drive the predictive capabilities of the digital twin.

[0080] The digital twin service platform serves as the system's intelligent decision-making center, connecting to the federated learning server and environmental awareness module. This platform constructs and maintains a digital twin—a virtual copy of the physical display—for each medical monitor node. The digital twin receives real-time operational data from the physical display (such as current brightness, chromaticity, panel temperature, and cumulative power-on time) and real-time ambient light data (such as ambient illuminance and ambient color temperature) from the environmental awareness module. Combined with a global optimization model issued by the federated learning server, the digital twin predicts the future color drift trajectory of the corresponding physical display. When the prediction exceeds a preset tolerance, the platform generates predictive calibration instructions. For example, the digital twin service platform can be built on a microservice architecture, with each digital twin acting as an independent service instance, receiving data in real-time and using simulation models to predict color drift. Furthermore, the platform can integrate advanced analysis modules, such as hybrid prediction algorithms based on physical models and data-driven models, combining historical data and real-time ambient light data to predict long-term trends in key parameters such as brightness, chromaticity, and gamma curves of the display.

[0081] The blockchain service module serves as the system's trusted evidence storage guarantee, connecting to the digital twin service platform and the blockchain network. This module executes predictive calibration instructions generated by the digital twin service platform to calibrate the target display. Upon calibration, the module acquires key target data, such as calibration trigger type, brightness and chromaticity measurements before and after calibration, ambient light data, and the operator's digital identity. Subsequently, the module generates a data fingerprint from this key target data and uploads the transaction record containing the data fingerprint to the blockchain by calling a blockchain smart contract. For example, the blockchain service module can be a standalone application programming interface (API) service, responsible for interacting with the digital twin service platform, receiving calibration instructions and key data, and calling pre-deployed smart contracts (such as evidence storage contracts) to upload the data fingerprint and related metadata of the calibration record to the blockchain. Alternatively, the module can be integrated within the digital twin service platform, interacting directly with the blockchain network through an SDK or client to construct, sign, broadcast, and query transactions, ensuring the transparency and immutability of the calibration data.

[0082] This application constructs a complete, closed-loop display color management system. Multiple medical display nodes achieve distributed data acquisition and local model training, effectively solving the problems of isolated device management and data privacy. The federated learning server, through a secure aggregation mechanism, gathers the learning experience of each node while protecting data privacy, generating a more generalizable global optimization model and improving its accuracy and robustness. The digital twin service platform utilizes the global optimization model and real-time environmental data to achieve predictive management of display color drift, transforming passive calibration into proactive intervention, significantly reducing the "calibration gap," and improving calibration efficiency and display consistency. The blockchain service module provides tamper-proof, traceable, and authoritative evidence for calibration records, meeting medical quality audit and compliance requirements and greatly enhancing the credibility of quality control records. Overall, through the collaborative work of its modules, this system effectively solves the problems in traditional medical display color management, such as the difficulty in achieving and maintaining color consistency between displays, passive and inefficient calibration strategies, isolated device management and insufficient exploitation of data value, and insufficient credibility and traceability of quality control records, providing more reliable and efficient display assurance for medical image diagnosis.

[0083] Furthermore, the digital twin service platform is configured as follows: The local model update parameters are encrypted and uploaded to the federated learning server to obtain the encrypted local model update parameters. The federated learning server groups the medical display nodes according to their device models or according to the usage scenarios corresponding to the preset modes of the medical display nodes. The dynamic weights of each encrypted local model update parameter within a group are determined. Based on the dynamic weights, a weighted average is calculated on all the encrypted local model update parameters within the same group to obtain an encrypted aggregation result. The encrypted aggregation result within each group is decrypted to generate a global optimization model corresponding to each group, and the global optimization model is distributed to the digital twin platform.

[0084] The digital twin service platform is configured as follows: A digital twin is constructed and maintained for each medical display node; a time-series input vector is constructed including the current brightness, chromaticity, panel temperature, cumulative power-on time, and real-time ambient light data of the digital twin; the global optimization model outputs predicted brightness and chromaticity values ​​for a future preset time period based on the time-series input vector to generate the color drift trajectory and obtain the prediction result; when the prediction result exceeds the preset tolerance, the predictive calibration instruction is generated.

[0085] Furthermore, the blockchain service module is as follows: The predictive calibration command is executed to calibrate the target display, and the target key data during calibration is acquired. The target key data includes the calibration trigger type, pre-calibration luminance and chromaticity measurements, post-calibration luminance and chromaticity measurements, ambient light intensity and color temperature during calibration, the digital identity of the operator performing the calibration, and the digital twin predictive data packet that triggered the calibration. The target key data is then standardized to obtain standardized target key data. The standardized target key data is combined into a string with a preset format, and the string is calculated using a preset hash algorithm to generate the data fingerprint; the transaction record containing the data fingerprint is uploaded to the blockchain by calling the blockchain smart contract.

[0086] The following example provides a more detailed explanation of the application process: In a large medical institution, multiple medical display nodes are deployed, distributed across different diagnostic rooms and reading rooms. To ensure that these displays maintain highly consistent color performance over long-term use, and to address issues such as the difficulty in achieving and maintaining color consistency between displays, passive and inefficient calibration strategies, isolated equipment management, and insufficient traceability of quality control records in traditional calibration methods, this method was applied to the institution's display color management.

[0087] First, sensors built into each medical display node continuously collect local color performance data, such as brightness, chromaticity, and gamma curves. This data is used to train a color prediction model on the local computing unit to learn the display's own color shift patterns. After training, each node generates a set of local model update parameters. To protect data privacy and security, these local model update parameters are encrypted before being uploaded.

[0088] The encrypted local model update parameters are uploaded to the federated learning server. Upon receiving these parameters, the federated learning server groups the medical display nodes according to their device model (e.g., grouping all devices with the model name "Diagnostic Display A" into one group) or the usage scenario corresponding to a preset mode (e.g., grouping all devices used for "Radiology Diagnosis" into another group). Within each group, the federated learning server calculates the dynamic weights of each encrypted local model update parameter. These dynamic weights are the result of considering multiple factors: First, a data quality score is calculated for each parameter, which is the geometric mean of the calibration accuracy value and the sensor reliability value; second, a comprehensive weight value is calculated using a predefined weighting function based on the data quality score, data sample size, color drift rate of the medical display node, and cumulative uptime. Encrypted local model parameters with data quality scores higher than a first threshold are given a superlinear weight bonus; finally, the historical contribution score and model update stability score of the medical display node are obtained, and the comprehensive weight value is fine-tuned based on these scores to obtain the final dynamic weights. Based on these dynamic weights, the federated learning server performs a weighted average of all encrypted local model update parameters within the same group to obtain an encrypted aggregation result. Subsequently, the encrypted aggregation result within each group is decrypted to generate a global optimization model for each group. These global optimization models are then distributed to the digital twin platform.

[0089] The digital twin platform builds and maintains a digital twin for each medical display node. The platform constructs a time-series input vector containing the digital twin's current brightness, chromaticity, panel temperature, cumulative power-on time, and real-time ambient light data (provided by the environment sensing module). Based on this time-series input vector, the global optimization model outputs predicted brightness and chromaticity values ​​for a preset future time period (e.g., the next week or month), thereby generating the future color drift trajectory of the physical display. In this way, the system can proactively predict the display's color drift trend, rather than passively waiting for problems to occur. When the prediction results exceed a preset tolerance, the system generates a predictive calibration command. For example, if it is predicted that the brightness of a display will exceed the tolerance range specified by the DICOM GSDF standard within the next three days, a calibration command will be generated immediately.

[0090] When the predicted result exceeds the preset tolerance, the system uses the ambient illuminance and ambient color temperature from real-time ambient light data to call the visual adaptation model and calculate the contrast sensitivity function adjustment and color temperature compensation. These adjustments are then fused with the target physical parameters based on display standards to generate target calibration parameters. Simultaneously, the system determines the optimal calibration time window based on the predicted drift exceedance time point and the preset work schedule (e.g., the monitor has a free period on Wednesday afternoons). Finally, the target calibration parameters and the optimal calibration time window are encapsulated into predictive calibration instructions. This predictive calibration strategy is significantly superior to traditional fixed-period calibration, avoiding the risk of "calibration gaps" and enabling precise on-demand maintenance, thus improving resource utilization efficiency.

[0091] The system executes predictive calibration instructions to calibrate the target display. Upon calibration, the system acquires key target data, including the calibration trigger type (e.g., "predictive drift error"), luminance and chromaticity measurements before and after calibration, ambient light and color temperature during calibration, the digital identity of the operator performing the calibration, and the digital twin prediction data packet that triggered the calibration. This key target data is standardized, combined into a pre-formatted string, and a data fingerprint is generated using a pre-defined hash algorithm. Subsequently, the system constructs a transaction invoking a notarized smart contract. The payload of this transaction includes at least the generated data fingerprint, a timestamp certificate generated by a trusted timestamp service, and a unique identifier for the target display. The transaction is verified and sorted using a consensus algorithm and packaged into a new block, generating a transaction hash and block number. These transaction hashes and block numbers are returned and stored as traceability credentials. Through blockchain technology, calibration records achieve immutability and traceability, solving the problems of easy tampering and loss of traditional paper or spreadsheet records, and meeting the requirements of medical quality auditing and compliance.

[0092] Obviously, the embodiments described above are merely some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the scope of this application. This application can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of protection of this application.

Claims

1. A display color management method based on federated learning and digital twins, characterized in that, include: Local color performance data is collected from multiple medical display nodes, and the local color prediction model is trained based on the color performance data to obtain local model update parameters. The local model update parameters are encrypted and uploaded to the federated learning server. The federated learning server then aggregates the encrypted local model parameters to generate a global optimization model, which is then distributed to the digital twin platform. The digital twin platform constructs and maintains a digital twin for each medical display node, and based on the global optimization model and real-time ambient light data, drives the digital twin to predict the future color drift trajectory of the corresponding physical display. When the prediction result exceeds the preset tolerance, a predictive calibration instruction is generated. The predictive calibration instruction is executed to calibrate the target display and acquire key target data during calibration. A data fingerprint is generated based on the key target data, and the transaction record containing the data fingerprint is uploaded to the blockchain by calling a blockchain smart contract.

2. The display color management method based on federated learning and digital twins according to claim 1, characterized in that, The process of encrypting the local model update parameters and uploading them to the federated learning server, then aggregating them based on the encrypted local model parameters through the federated learning server to generate a global optimization model, and finally distributing the global optimization model to the digital twin platform includes: The local model update parameters are encrypted and uploaded to the federated learning server to obtain the encrypted local model update parameters. The medical display nodes are grouped by the federated learning server according to their device models or by the usage scenarios corresponding to the preset modes of the medical display nodes. Determine the dynamic weights of the encrypted local model update parameters within each group; Based on the dynamic weights, a weighted average calculation is performed on all the encrypted local model update parameters within the same group to obtain the encrypted aggregation result. The encrypted aggregation results within each group are decrypted to generate a global optimization model for each group, and the global optimization model is then sent to the digital twin platform.

3. The display color management method based on federated learning and digital twins according to claim 2, characterized in that, The determination of the dynamic weights of the encrypted local model update parameters within each group includes: Calculate the data quality score for each of the encrypted local model parameters, wherein the data quality score is the geometric mean of the calibration accuracy value and the sensor reliability value; Based on the data quality score, data sample size, color drift rate of the medical display node, and cumulative power-on time, a comprehensive weight value is calculated using a predefined weighting function. Among these, encrypted local model parameters with data quality scores higher than a first threshold are given a superlinear weight reward. The historical contribution score and model update stability score of the medical display node are obtained, and the comprehensive weight value is fine-tuned based on the historical contribution score and the model update stability score to obtain the dynamic parameters.

4. The display color management method based on federated learning and digital twins according to claim 1, characterized in that, The process involves constructing and maintaining a digital twin for each medical display node through the digital twin platform. Based on the global optimization model and real-time ambient light data, the digital twin is driven to predict the future color drift trajectory of the corresponding physical display. When the prediction result exceeds a preset tolerance, a predictive calibration instruction is generated, including: A digital twin is constructed and maintained for each of the medical display nodes through the digital twin platform; Construct a time-series input vector including the current brightness, chromaticity, panel temperature, cumulative power-on time of the digital twin, and the real-time ambient light data; The global optimization model outputs predicted brightness and chromaticity values ​​for a future preset time period based on the time series input vector to generate the color drift trajectory and obtain the prediction result. When the prediction result exceeds the preset tolerance, the predictive calibration instruction is generated.

5. The display color management method based on federated learning and digital twins according to claim 4, characterized in that, When the prediction result exceeds the preset tolerance, the predictive calibration instruction is generated, including: When the prediction result exceeds the preset tolerance, the visual adaptation model is invoked based on the ambient illuminance and ambient color temperature in the real-time ambient light data to calculate the contrast sensitivity function adjustment amount and color temperature compensation amount. The contrast sensitivity function adjustment amount and the color temperature compensation amount are fused with the target physical parameters based on the display standard to generate target calibration parameters; The target calibration parameters, along with the optimal calibration time window determined based on the predicted drift deviation time point and the preset work schedule, are encapsulated into the predictive calibration instruction.

6. The display color management method based on federated learning and digital twins according to any one of claims 1 to 5, characterized in that, The process of executing the predictive calibration instruction to calibrate the target display, acquiring key target data during calibration, generating a data fingerprint based on the key target data, and uploading transaction records containing the data fingerprint to the blockchain by calling a blockchain smart contract includes: The predictive calibration command is executed to calibrate the target display and to acquire the target key data during calibration. The target key data includes calibration trigger type, luminance and chromaticity measurements before calibration, luminance and chromaticity measurements after calibration, ambient light intensity and color temperature during calibration, digital identity of the operator performing the calibration, and digital twin predictive data package that triggered this calibration. The target key data is standardized to obtain standardized target key data; The standardized target key data is combined into a string in a preset format, and the string is calculated using a preset hash algorithm to generate the data fingerprint; The transaction record containing the data fingerprint is uploaded to the blockchain by invoking the blockchain smart contract.

7. The display color management method based on federated learning and digital twins according to claim 6, characterized in that, The step of uploading transaction records containing the data fingerprint to the blockchain by calling the blockchain smart contract includes: Construct a transaction information for calling a notarized smart contract, wherein the payload of the transaction information includes at least: the data fingerprint, a timestamp certificate generated by a trusted timestamp service, and a unique identifier for the target display; The transaction information is verified and sorted using a consensus algorithm, and then packaged into a new block to generate a transaction hash and block number. The transaction hash and the block number are returned and stored as traceability credentials.

8. A display color management system based on federated learning and digital twins, used in the display color management method based on federated learning and digital twins as described in any one of claims 1 to 7, characterized in that, include: Multiple medical display nodes, each with built-in sensors and computing units, are used to collect local color performance data and train a locally deployed color prediction model based on the color performance data to obtain local model update parameters. The federated learning server communicates with each terminal to receive the encrypted local model update parameters, and generates a global optimization model based on the weighted aggregation of the encrypted local model parameters, and distributes the global optimization model to the digital twin platform. The digital twin service platform, connected to the federated learning server and the environment perception module, is used to build and maintain a digital twin for each medical display node, and based on the global optimization model and real-time ambient light data, drive the digital twin to predict the future color drift trajectory of the corresponding physical display. When the prediction result exceeds the preset tolerance, a predictive calibration instruction is generated. The blockchain service module, connected to the digital twin service platform and the blockchain network, is used to execute the predictive calibration instructions to calibrate the target display, acquire key target data during calibration, generate a data fingerprint based on the key target data, and upload transaction records containing the data fingerprint to the blockchain by calling the blockchain smart contract.

9. The display color management system based on federated learning and digital twins according to claim 8, characterized in that, The digital twin service platform is configured as follows: The local model update parameters are encrypted and uploaded to the federated learning server to obtain the encrypted local model update parameters. The medical display nodes are grouped by the federated learning server according to their device models or by the usage scenarios corresponding to the preset modes of the medical display nodes. Determine the dynamic weights of the encrypted local model update parameters within each group; Based on the dynamic weights, a weighted average calculation is performed on all the encrypted local model update parameters within the same group to obtain the encrypted aggregation result. The encrypted aggregation results within each group are decrypted to generate a global optimization model for each group, and the global optimization model is then sent to the digital twin platform.

10. The display color management system based on federated learning and digital twins according to claim 8, characterized in that, The digital twin service platform is configured as follows: A digital twin is constructed and maintained for each of the medical display nodes; Construct a time-series input vector including the current brightness, chromaticity, panel temperature, cumulative power-on time of the digital twin, and the real-time ambient light data; The global optimization model outputs predicted brightness and chromaticity values ​​for a future preset time period based on the time series input vector to generate the color drift trajectory and obtain the prediction result. When the prediction result exceeds the preset tolerance, the predictive calibration instruction is generated.

Images