Blockchain and multi-modal perception based clinical trial data quality control traceability method

Abstract

This invention provides a clinical trial data quality control and traceability method based on blockchain and multimodal perception, belonging to the field of clinical trial data management technology. The method includes collecting trial environment parameters and data, constructing a dynamic threshold curve to assess the environment status, and generating anomaly identifiers; constructing fingerprint feature vectors using a Bloom filter, implementing hierarchical access control based on threshold signature key pairs, generating a proof sequence through zero-knowledge proofs, and finally constructing a directed acyclic graph concatenated proof sequence to achieve data traceability. This invention can improve the credibility and traceability efficiency of clinical trial data.

Description

Technical Field

[0001] This invention relates to clinical trial data management technology, and more particularly to a clinical trial data quality control and traceability method based on blockchain and multimodal perception. Background Technology

[0002] During clinical trials, environmental fluctuations, operational biases, and data integrity issues often affect the reliability and traceability of trial data. Existing quality control methods mostly rely on manual sampling and static threshold judgments, which are difficult to perceive changes in the trial environment in real time, lack specificity and systematicity, and easily lead to the failure to detect and record key abnormal data in a timely manner.

[0003] On the other hand, as clinical data becomes more multimodal and distributed, traditional data management methods struggle to support cross-node data integrity verification and secure access. Data traceability typically relies on centralized systems, which suffer from issues such as information silos, data tampering risks, and difficulty in flexibly controlling access permissions, thus hindering the reliable sharing and auditing efficiency of clinical data.

[0004] While blockchain technology boasts advantages such as immutability and verifiability, it still lacks efficient data identification and granular access control mechanisms when facing large-scale heterogeneous data and dynamic access management needs, limiting its application in the field of medical data quality control and traceability. Therefore, there is an urgent need for a clinical trial data quality control and traceability method that integrates multimodal environment awareness, distributed data identification, and verifiable access mechanisms. Summary of the Invention

[0005] This invention provides a clinical trial data quality control and traceability method based on blockchain and multimodal perception, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for quality control and traceability of clinical trial data based on blockchain and multimodal perception, comprising:

[0007] Collect environmental parameters and test data from clinical trials, calculate the fluctuation period of environmental parameters, construct a parameter feature matrix and perform singular value decomposition to obtain a dynamic threshold curve, compare the real-time values ​​of environmental parameters with the dynamic threshold curve to generate an environmental status score, generate anomaly identification information containing timestamps based on the environmental status score, and extract corresponding data from the test data based on the anomaly identification information to generate an anomaly monitoring dataset.

[0008] The test data and anomaly monitoring dataset are input into the distributed nodes. A fingerprint feature vector is constructed using a Bloom filter. A threshold signature key pair is generated based on the fingerprint feature vector to construct a hierarchical access control policy. The hierarchical access control policy is written into a smart contract to generate a data access token. The data access token and the fingerprint feature vector are combined to form a data asset identification package and input into the blockchain network. A proof sequence is generated through a zero-knowledge proof protocol.

[0009] A directed acyclic graph structure is constructed based on the proof sequence. The proof sequence is concatenated between nodes using hash pointers and access path information is written in. The traceability request is received, the data asset identification package is obtained based on the access path information, and the traceability result is output after verifying the proof sequence.

[0010] In one alternative embodiment,

[0011] The fluctuation period of environmental parameters is calculated, a parameter feature matrix is ​​constructed, and singular value decomposition is performed to obtain a dynamic threshold curve. The real-time values ​​of environmental parameters are compared with the dynamic threshold curve to generate an environmental status score, including:

[0012] A multimodal sensor array was used to collect environmental parameters for the clinical trial, including temperature, humidity and light parameters. Wavelet denoising was then applied to the environmental parameters to obtain a filtered sequence.

[0013] The autocorrelation function is calculated on the filtered sequence to obtain the correlation coefficient sequence. The local maxima of the correlation coefficient sequence are extracted to obtain the peak position sequence. The peak position sequence is clustered to obtain the peak cluster centers. The fluctuation period is determined based on the time interval of the peak cluster centers.

[0014] The filtered sequence is segmented according to the fluctuation period, the rate of change of each segment is calculated to obtain the rate of change constraint, and the segmented sequence is aligned using a dynamic time warping algorithm based on the rate of change constraint to obtain an aligned sequence. The aligned sequences are then superimposed to form a parameter feature matrix.

[0015] The parameter feature matrix is ​​decomposed into a set of feature vectors and a sequence of singular values. The principal feature vector is determined based on the information entropy curve of the singular value sequence. The standard fluctuation feature matrix is ​​reconstructed based on the principal feature vector. The probability density distribution of the standard fluctuation feature matrix is ​​calculated using the kernel density estimation method. An asymmetric dynamic threshold curve is constructed based on the skewness of the probability density distribution.

[0016] The deviation sequence and rate of change sequence of real-time environmental parameters and dynamic threshold curves are calculated. Based on the temporal correlation of the deviation sequence and rate of change sequence, dynamic weights are determined and weighted to obtain the environmental status score.

[0017] In one alternative embodiment,

[0018] A dynamic time warping algorithm based on rate of change constraints is used to align the segmented sequences to obtain aligned sequences. These aligned sequences are then superimposed to form a parametric feature matrix, which includes:

[0019] The time difference is calculated on the input segmented sequence to obtain the rate of change sequence;

[0020] A fixed-length sliding window is constructed. Extreme points and inflection points are extracted within the sliding window to obtain a set of change points. The time interval, amplitude difference, and curvature of adjacent change points in the set of change points are calculated to obtain a feature vector. Density clustering is performed on the feature vector to obtain change pattern labels.

[0021] The transition frequency matrix is ​​obtained by calculating the transition frequency between adjacent change mode labels. The state transition probability matrix is ​​obtained by normalizing the transition frequency matrix. The change mode combination is determined based on the state transition probability matrix.

[0022] The Euclidean distance between the sequences to be aligned is calculated to obtain the basic cost matrix. The change pattern label and change pattern combination at each position in the basic cost matrix are matched. The alignment cost matrix is ​​obtained by weighting the basic cost matrix according to the matching results.

[0023] The optimal alignment path is calculated based on the alignment cost matrix using a dynamic programming algorithm. The segmented sequences are then time-aligned according to the optimal alignment path to obtain an alignment sequence. The alignment sequences are then superimposed in the time dimension to obtain a parameter feature matrix, where each row of the parameter feature matrix represents the environmental parameter change process of a complete cycle.

[0024] In one alternative embodiment,

[0025] The experimental data and anomaly monitoring dataset are input into distributed nodes. A Bloom filter is used to construct fingerprint feature vectors. Based on the fingerprint feature vectors, threshold signature key pairs are generated to construct a hierarchical access control strategy, including:

[0026] A distributed storage network based on blockchain is constructed to obtain multiple distributed nodes. The distributed nodes are divided into master nodes and slave nodes. A consensus network is built among the master nodes, and slave nodes are assigned to master nodes to form node groups.

[0027] The experimental data and anomaly monitoring dataset are input into the node group. The mutual information between data items is calculated to obtain the association strength matrix. A data dependency graph is constructed based on the association strength matrix. The association subgraphs in the data dependency graph are extracted according to the connectivity between data items to obtain the data type set. The propagation path of the data type set is calculated to obtain the access dependency sequence.

[0028] The access dependency sequence is input into a Bloom filter array, and a feature vector is generated using a group of independent hash functions. The feature vectors are then combined to obtain a fingerprint feature vector.

[0029] The feature polynomial is calculated based on the fingerprint feature vector and mapped to the key generator polynomial. The master private key is obtained based on the coefficient sequence of the key generator polynomial. The master private key is decomposed by Lagrange interpolation to obtain the sub-private key sequence. The corresponding master public key and sub-public key set are generated to form a threshold signature key pair.

[0030] The set of sub-public keys in the threshold signature key pair is grouped to obtain key groups. The coverage of each key group is calculated to obtain the access domain division. Based on the access domain division, hierarchical access constraints are constructed and combined to obtain a hierarchical access control strategy.

[0031] In one alternative embodiment,

[0032] A hierarchical access control policy is written into a smart contract to generate a data access token. This data access token, along with a fingerprint feature vector, forms a data asset identification package, which is then input into the blockchain network. A proof sequence is generated using a zero-knowledge proof protocol, including:

[0033] Extract key verification conditions, timing constraints, and dependency constraints from the hierarchical access control strategy, construct a verification rule set, and write the verification rule set into the smart contract's state verification function.

[0034] Receive clinical trial data access requests, verify the private key combination, access time and access dependency information provided by the requester through a state verification function, and generate a data access token based on the verification result;

[0035] A hash operation is performed on the data access token to obtain the token hash value. The token hash value and the fingerprint feature vector are XORed to obtain a mixed feature sequence. The root hash value is calculated based on the mixed feature sequence as the data asset identifier value. The data asset identifier value, the data access token and the fingerprint feature vector are combined to form a data asset identifier package. The data asset identifier package is written into the blockchain network and a blockchain storage address is generated.

[0036] A token verification module is constructed based on the verification rule set, and a hash verification module is constructed based on the data asset identifier value. The token verification module and the hash verification module are combined to form a verification circuit. The data asset identifier package and the original clinical trial data are extracted based on the blockchain storage address as inputs to the verification circuit, and a proof sequence is generated through a zero-knowledge proof protocol.

[0037] In one alternative embodiment,

[0038] A hash operation is performed on the data access token to obtain the token hash value. The token hash value is then XORed with the fingerprint feature vector to obtain a hybrid feature sequence. The root hash value, calculated based on the hybrid feature sequence, serves as the data asset identifier.

[0039] The access permission scope and validity period in the data access token are standardized, and a secure hash algorithm is used to perform hash calculation on the standardized data access token. In the hash calculation process, timestamp information is introduced as a random factor to generate a token hash value.

[0040] The integrity of the fingerprint feature vector is detected, and the missing dimensions are supplemented. The supplemented feature vector is then normalized to obtain a standardized fingerprint feature vector. The standardized fingerprint feature vector is then aligned and padded with zeros according to the length of the token hash value to obtain an aligned feature vector.

[0041] Construct a sliding window, segment the token hash value according to the sliding window size, and perform a bitwise XOR operation between the segmented token hash value and the aligned feature vector to obtain a mixed feature sequence;

[0042] A random perturbation factor sequence is generated. The random perturbation factor sequence is mixed with the mixed feature sequence to obtain a perturbation sequence. The perturbation sequence is rearranged according to a preset rearrangement rule to obtain an enhanced mixed feature sequence. The enhanced mixed feature sequence is divided into data blocks of fixed size. A multi-layer hash calculation structure is constructed. Iterative hash calculation is performed on adjacent data blocks to obtain the root hash value as the data asset identifier value.

[0043] In one alternative embodiment,

[0044] A directed acyclic graph structure is constructed based on the proof sequence. The proof sequence is concatenated between nodes using hash pointers, and access path information is written in. A tracing request is received, and the data asset identification package is obtained based on the access path information. After verifying the proof sequence, the tracing result is output, including:

[0045] The timestamp, operation type, and data identifier are extracted from the proof sequence and standardized to obtain an attribute feature vector. Node identifiers are generated based on the attribute feature vectors, and nodes with the node identifiers are used as the basic units of the directed acyclic graph.

[0046] Based on the temporal relationship of the proof sequence, nodes are arranged in layers. Horizontal hash pointers are established between nodes in the same layer, and vertical hash pointers are established between nodes in adjacent layers. The nodes are connected into a directed acyclic graph structure through the horizontal and vertical hash pointers.

[0047] For each proof sequence, an access path identifier is generated, written to the corresponding node, and added to the horizontal hash pointer and the vertical hash pointer to form a proof sequence containing access path information.

[0048] Upon receiving a tracing request, the target node is located in the directed acyclic graph structure based on the access path information in the request. The corresponding data asset identification package is obtained, and the proof sequence is traversed along the horizontal and vertical hash pointers to verify the integrity and validity of the proof sequence. The verified proof sequence is then output as the tracing result.

[0049] A second aspect of the present invention provides an electronic device, comprising:

[0050] processor;

[0051] Memory used to store processor-executable instructions;

[0052] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0053] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0054] In this embodiment, a clinical trial data quality control and traceability method based on blockchain and multimodal perception was adopted to achieve dynamic monitoring and anomaly identification of environmental parameters, effectively improving the quality management level and reliability of clinical trial data and reducing the risk of environmental factors interfering with trial results. A fingerprint feature vector was constructed using distributed storage and a Bloom filter, and a hierarchical access control strategy was established in conjunction with threshold signature technology, enhancing the security and privacy protection capabilities of clinical trial data, effectively preventing unauthorized access or tampering, and improving the system's efficiency in processing large-scale clinical data. Full-process traceability of clinical trial data was achieved based on zero-knowledge proofs and a directed acyclic graph structure. A complete traceability chain was formed by concatenating proof sequences using hash pointers, ensuring the authenticity and integrity of data traceability, improving the auditability and regulatory compliance of clinical trial data, and providing reliable data support for drug development and clinical decision-making. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the clinical trial data quality control and traceability method based on blockchain and multimodal perception, as described in an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram illustrating the alignment of segmented temperature parameter sequences based on the dynamic time warping algorithm in an embodiment of the present invention.

[0057] Figure 3 This is a flowchart illustrating the data asset tracing process based on a directed acyclic graph, as described in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0060] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention, as shown below. Figure 1 As shown, the clinical trial data quality control and traceability method based on blockchain and multimodal perception includes:

[0061] Collect environmental parameters and test data from clinical trials, calculate the fluctuation period of environmental parameters, construct a parameter feature matrix and perform singular value decomposition to obtain a dynamic threshold curve, compare the real-time values ​​of environmental parameters with the dynamic threshold curve to generate an environmental status score, generate anomaly identification information containing timestamps based on the environmental status score, and extract corresponding data from the test data based on the anomaly identification information to generate an anomaly monitoring dataset.

[0062] The test data and anomaly monitoring dataset are input into the distributed nodes. A fingerprint feature vector is constructed using a Bloom filter. A threshold signature key pair is generated based on the fingerprint feature vector to construct a hierarchical access control policy. The hierarchical access control policy is written into a smart contract to generate a data access token. The data access token and the fingerprint feature vector are combined to form a data asset identification package and input into the blockchain network. A proof sequence is generated through a zero-knowledge proof protocol.

[0063] A directed acyclic graph structure is constructed based on the proof sequence. The proof sequence is concatenated between nodes using hash pointers and access path information is written in. The traceability request is received, the data asset identification package is obtained based on the access path information, and the traceability result is output after verifying the proof sequence.

[0064] In one alternative embodiment,

[0065] The fluctuation period of environmental parameters is calculated, a parameter feature matrix is ​​constructed, and singular value decomposition is performed to obtain a dynamic threshold curve. The real-time values ​​of environmental parameters are compared with the dynamic threshold curve to generate an environmental status score, including:

[0066] A multimodal sensor array was used to collect environmental parameters for the clinical trial, including temperature, humidity and light parameters. Wavelet denoising was then applied to the environmental parameters to obtain a filtered sequence.

[0067] The autocorrelation function is calculated on the filtered sequence to obtain the correlation coefficient sequence. The local maxima of the correlation coefficient sequence are extracted to obtain the peak position sequence. The peak position sequence is clustered to obtain the peak cluster centers. The fluctuation period is determined based on the time interval of the peak cluster centers.

[0068] The filtered sequence is segmented according to the fluctuation period, the rate of change of each segment is calculated to obtain the rate of change constraint, and the segmented sequence is aligned using a dynamic time warping algorithm based on the rate of change constraint to obtain an aligned sequence. The aligned sequences are then superimposed to form a parameter feature matrix.

[0069] The parameter feature matrix is ​​decomposed into a set of feature vectors and a sequence of singular values. The principal feature vector is determined based on the information entropy curve of the singular value sequence. The standard fluctuation feature matrix is ​​reconstructed based on the principal feature vector. The probability density distribution of the standard fluctuation feature matrix is ​​calculated using the kernel density estimation method. An asymmetric dynamic threshold curve is constructed based on the skewness of the probability density distribution.

[0070] The deviation sequence and rate of change sequence of real-time environmental parameters and dynamic threshold curves are calculated. Based on the temporal correlation of the deviation sequence and rate of change sequence, dynamic weights are determined and weighted to obtain the environmental status score.

[0071] In this implementation, a multimodal sensor array is used to collect environmental parameters for the clinical trial, specifically including temperature, humidity, and illumination parameters. The temperature sensor is a resistance temperature detector (RTD) sensor with an accuracy of ±0.1℃ and a sampling frequency of once every 60 seconds; the humidity sensor is a capacitive humidity sensor with an accuracy of ±2%RH and a sampling frequency of once every 60 seconds; the illumination sensor is a photoresistor sensor with a range of 0-10000 lux and a sampling frequency of once every 60 seconds. The collected environmental parameters are then subjected to wavelet denoising to obtain a filtered sequence. Wavelet denoising uses wavelet basis functions to perform multi-scale decomposition of the original signal, and soft thresholding is applied to high-frequency coefficients. The threshold value is adaptively adjusted according to the signal-to-noise ratio. The signal is then reconstructed to obtain the denoised filtered sequence. For example, the original temperature data collected in the drug storage area during a clinical trial was {21.3, 21.5, 21.4, 22.0, 22.5, 22.4, 22.3, 21.9, 21.5, 21.3}℃. After wavelet denoising, the filtered sequence was obtained as {21.35, 21.45, 21.6, 22.1, 22.4, 22.35, 22.15, 21.85, 21.45, 21.35}℃.

[0072] The correlation coefficient sequence is obtained by calculating the autocorrelation function of the filtered sequence. The autocorrelation function is calculated by taking the dot product of the filtered sequence and its time-shifted sequence. The correlation coefficient sequence reflects the similarity of the signals under different time delays. Local maxima are extracted from the correlation coefficient sequence to obtain the peak position sequence. A local maximum is a point where all adjacent points are smaller than the value of the peak. The peak position sequence is then clustered to obtain the peak cluster centers. The density peak clustering algorithm is used to determine the cluster centers by calculating the local density and distance of the peak positions. The fluctuation period is determined based on the time interval between the peak cluster centers. For example, the correlation coefficient sequence of the filtered temperature parameter shows a local maximum at a time delay of {0, 360, 720, 1080, 1440} minutes, with corresponding correlation coefficient values ​​of {1.0, 0.82, 0.75, 0.79, 0.85}. Density peak clustering finds the cluster center at 1440 minutes, therefore the fluctuation period of the temperature parameter is determined to be 1440 minutes, or 24 hours.

[0073] The filtered sequence is segmented according to the fluctuation period. For the temperature parameter, each segment is 1440 minutes long. The rate of change of each segment is calculated to obtain the rate of change constraint. The rate of change is calculated by dividing the difference between adjacent time points by the time interval. The rate of change constraint is used to limit the range of time elastic deformation during sequence alignment. A dynamic time warping algorithm based on the rate of change constraint is used to align the segmented sequences. This algorithm constructs a cumulative distance matrix and searches for the optimal path under the rate of change constraint to achieve sequence alignment, resulting in aligned sequences. The aligned sequences are then superimposed to form a parameter feature matrix. For example, a seven-day temperature filtered sequence from a clinical trial is divided into seven segments, each with 1440 data points. After alignment using the dynamic time warping algorithm, a standardized aligned sequence of 1440 time points is obtained. The seven aligned sequences are superimposed to form a 7×1440 parameter feature matrix.

[0074] Singular value decomposition (SVD) is performed on the parameter eigenma matrix to obtain eigenvector sets and a sequence of singular values. SVD decomposes the matrix into the product of a left singular vector matrix, a diagonal singular value matrix, and a right singular vector matrix. An information entropy curve is calculated based on the singular value sequence. Information entropy represents the amount of information contained in each singular value, obtained by normalizing the singular values ​​and calculating their negative logarithms. The principal eigenvectors are determined based on the information entropy curve. The eigenvector corresponding to the cumulative information entropy exceeding a preset threshold (e.g., 95%) is selected as the principal eigenvector. A standard fluctuation feature matrix is ​​reconstructed based on the principal eigenvectors, reflecting the typical fluctuation patterns of environmental parameters. For example, the singular value sequence of the temperature parameter feature matrix is ​​{45.2, 12.5, 5.8, 2.1, 1.3, 0.9, 0.4}, and the calculated cumulative information entropy curve is {66.5%, 84.9%, 93.4%, 96.5%, 98.4%, 99.7%, 100%}. The first four eigenvectors are selected as the principal eigenvectors for reconstruction to obtain the standard fluctuation feature matrix.

[0075] The probability density distribution of the standard fluctuation characteristic matrix is ​​calculated using the kernel density estimation method. Kernel density estimation is achieved by weighted averaging of data points using a Gaussian kernel function, and the kernel bandwidth is adaptively determined based on the data distribution. An asymmetric dynamic threshold curve is constructed based on the skewness of the probability density distribution. Skewness reflects the asymmetry of the distribution; positive skewness indicates a right-side tail, and negative skewness indicates a left-side tail. The skewness value affects the degree of asymmetry between the upper and lower limits of the threshold curve. For example, the probability density distribution of the temperature parameter's standard fluctuation characteristic matrix has a skewness of 0.65, exhibiting a right-skewed distribution. The upper limit of the constructed asymmetric dynamic threshold curve is set to the mean plus 2.3 times the standard deviation, and the lower limit is set to the mean minus 1.7 times the standard deviation. During daytime working hours (8:00-18:00), the threshold bandwidth narrows by 20%, and during nighttime (22:00-6:00), the threshold bandwidth expands by 15%, forming a threshold curve that dynamically changes over time.

[0076] The deviation sequence between real-time environmental parameter values ​​and the dynamic threshold curve is calculated. The deviation sequence is the difference sequence between the real-time value and the center line of the threshold curve. Simultaneously, the rate of change sequence is calculated, which is the time derivative sequence of the real-time value. Dynamic weights are determined based on the temporal correlation between the deviation and rate of change sequences. The temporal correlation is obtained through autocorrelation and cross-correlation analysis within a sliding window, and the dynamic weights are adaptively adjusted according to the strength of the correlation. The deviation and rate of change sequences are weighted according to the dynamic weights to obtain an environmental state score, ranging from 0 to 100. A higher score indicates a more satisfactory environmental state. For example, if the real-time temperature is 23.5℃, the center value of the dynamic threshold curve is 22.0℃, the upper limit is 24.0℃, and the lower limit is 20.0℃, the calculated normalized deviation value is 0.75, the rate of change is 0.2℃ / minute, and the dynamic weights are 0.65 and 0.35, respectively. The calculated environmental state score is 82.5, indicating that the current temperature state is good but with slight fluctuations.

[0077] In this embodiment, the accuracy and stability of environmental parameter acquisition are improved through a multimodal sensor array and wavelet denoising technology; the periodic fluctuation characteristics of environmental parameters are accurately identified through autocorrelation function calculation and peak clustering analysis; precise alignment of environmental data from different periods is achieved based on a dynamic time warping algorithm, effectively eliminating the influence of time scale distortion; singular value decomposition and principal eigenvector reconstruction extract the essential features of environmental parameters, reducing the impact of random interference; the asymmetric dynamic threshold curve is adaptively adjusted according to the skewness characteristics of the probability density distribution, improving the sensitivity and specificity of anomaly detection; and the dynamic weight allocation mechanism based on temporal correlation makes the environmental state score more objective and comprehensive, enabling timely detection of potential risks. The overall solution significantly improves the accuracy and reliability of clinical trial environmental data quality control, providing a solid guarantee for the traceability of clinical trial data quality.

[0078] Figure 2 This is a schematic diagram illustrating the alignment of segmented temperature parameter sequences based on the dynamic time warping algorithm in an embodiment of the present invention, as shown below. Figure 2As shown in the figure, the horizontal axis represents the 24-hour time period, and the vertical axis represents the temperature value (°C). The orange dashed line represents three segments of the original temperature sequence. It can be seen that although the overall trend is similar, there are phase differences and local fluctuations at specific time points. This technical solution uses a dynamic time warping algorithm with a rate of change constraint (blue solid line) to successfully align the original sequence, forming a smooth standard curve. The aligned standard curve clearly shows the daily temperature variation pattern: it remains around 21.5°C in the early morning, begins to rise in the morning, reaches a peak of 24.2°C at 12 noon, begins to decrease in the afternoon, and gradually falls back to 21.5°C in the evening. The purple dashed line represents the rate of change constraint condition applied by the algorithm (upper limit +0.5°C / hour, lower limit -0.5°C / hour), ensuring the physical rationality of temperature changes during the alignment process. The figure also marks three key temperature points: 23.5°C at 9:40 AM, 24.2°C (daily high) at 12:20 PM, and 22.6°C at 4:40 PM. This technical solution, by introducing a dynamic time warping algorithm with a rate of change constraint, not only ensures high precision in sequence alignment but also ensures that the alignment result conforms to physical laws, providing a reliable foundation for constructing the parameter feature matrix.

[0079] In one alternative embodiment,

[0080] A dynamic time warping algorithm based on rate of change constraints is used to align the segmented sequences to obtain aligned sequences. These aligned sequences are then superimposed to form a parametric feature matrix, which includes:

[0081] The time difference is calculated on the input segmented sequence to obtain the rate of change sequence;

[0082] A fixed-length sliding window is constructed. Extreme points and inflection points are extracted within the sliding window to obtain a set of change points. The time interval, amplitude difference, and curvature of adjacent change points in the set of change points are calculated to obtain a feature vector. Density clustering is performed on the feature vector to obtain change pattern labels.

[0083] The transition frequency matrix is ​​obtained by calculating the transition frequency between adjacent change mode labels. The state transition probability matrix is ​​obtained by normalizing the transition frequency matrix. The change mode combination is determined based on the state transition probability matrix.

[0084] The Euclidean distance between the sequences to be aligned is calculated to obtain the basic cost matrix. The change pattern label and change pattern combination at each position in the basic cost matrix are matched. The alignment cost matrix is ​​obtained by weighting the basic cost matrix according to the matching results.

[0085] The optimal alignment path is calculated based on the alignment cost matrix using a dynamic programming algorithm. The segmented sequences are then time-aligned according to the optimal alignment path to obtain an alignment sequence. The alignment sequences are then superimposed in the time dimension to obtain a parameter feature matrix, where each row of the parameter feature matrix represents the environmental parameter change process of a complete cycle.

[0086] For example, the time difference of the input segmented sequence is first calculated to obtain the rate of change sequence. In a specific implementation, this is calculated by dividing the numerical difference between adjacent time points by the time interval. The rate of change sequence reflects the rate of change of environmental parameters at each time point, providing a basis for subsequent identification of key change points. For example, the temperature parameter in a clinical drug storage environment has sampling values ​​of {21.3, 21.5, 21.8, 22.1, 22.4, 22.6, 22.7, 22.5, 22.2, 21.9}℃ over 10 consecutive minutes. The difference sequence between adjacent time points is calculated as {0.2, 0.3, 0.3, 0.3, 0.2, 0.1, -0.2, -0.3, -0.3}℃. With a sampling interval of 1 minute, the rate of change sequence is obtained as {0.2, 0.3, 0.3, 0.3, 0.2, 0.1, -0.2, -0.3, -0.3}℃ / minute.

[0087] A fixed-length sliding window is constructed, with the window length set to 1 / 12 of the parameter period, approximately 120 minutes in the temperature parameter. Extreme points and inflection points are extracted within the sliding window to obtain a set of change points. Extreme points are those with the largest or smallest values ​​within a local range, identified by comparing the positive or negative rate of change of adjacent points. Inflection points are those where the direction of change of the rate of change reverses, determined by calculating the direction of change using the second-order difference. The time interval, amplitude difference, and curvature of adjacent change points in the set of change points are calculated to obtain an eigenvector. The time interval refers to the distance between adjacent change points on the time axis; the amplitude difference is the absolute value of the numerical difference between adjacent change points; and the curvature is approximated by the angle formed by three adjacent change points, reflecting the degree of curvature of the change trend. For example, within a sliding window of a temperature sequence, the set of change points identified at time positions {15, 42, 78, 95} minutes corresponds to temperature values ​​{21.5, 23.2, 23.0, 21.8}℃. The calculated feature vector includes time intervals {27, 36, 17} minutes, amplitude differences {1.7, 0.2, 1.2}℃, and curvatures {0.13, 0.09, 0.11}.

[0088] Density clustering was performed on the feature vectors to obtain change pattern labels. The density clustering used the DBSCAN algorithm, which determines the cluster centers by setting the neighborhood radius and the minimum number of samples, and maps the feature vectors to the change pattern space for cluster analysis. The change pattern labels represent different types of environmental parameter change patterns, such as "rapid rise," "slow fall," and "steady fluctuation." In the cluster analysis of temperature parameters, a neighborhood radius of 0.5 and a minimum number of samples of 3 were set, resulting in the following change pattern labels: Label 1 represents "rapid rise," characterized by short time intervals, large amplitude differences, and large curvature; Label 2 represents "slow rise," characterized by long time intervals, medium amplitude differences, and small curvature; Label 3 represents "steady fluctuation," characterized by medium time intervals, small amplitude differences, and small curvature; Label 4 represents "rapid fall," characterized by short time intervals, large amplitude differences, and large curvature; and Label 5 represents "slow fall," characterized by long time intervals, medium amplitude differences, and small curvature.

[0089] The transition frequency matrix is ​​obtained by calculating the transition frequency between adjacent change mode labels. A 5×5 transition frequency matrix is ​​formed by counting the number of transitions between adjacent change mode labels in all segmented sequences. The transition frequency matrix is ​​then normalized to obtain the state transition probability matrix. Normalization is achieved by dividing each element of a row by the sum of the elements in that row. The resulting state transition probability matrix reflects the transition probabilities between different change modes. Change mode combinations are determined based on the state transition probability matrix, and mode pairs with transition probabilities higher than a threshold of 0.15 are selected as effective change mode combinations. For example, in the state transition probability matrix of an environmental temperature parameter in a clinical trial, the transition probability from "rapid increase" (label 1) to "stable fluctuation" (label 3) is 0.62, and the transition probability from "stable fluctuation" (label 3) to "rapid decrease" (label 4) is 0.58. These two transitions are identified as high-probability change mode combinations, denoted as combination A and combination B, respectively.

[0090] The basic cost matrix is ​​obtained by calculating the Euclidean distance between the sequences to be aligned. The Euclidean distance is calculated as the square root of the sum of the squares of the differences between the corresponding points of the two sequences. The change pattern label at each position in the basic cost matrix is ​​matched with the change pattern combination. The matching method is to determine whether the change pattern label sequences at the current position and its preceding and following positions are consistent with the determined change pattern combination. The basic cost matrix is ​​weighted according to the matching results. The cost of a successfully matched position decreases, and the decrease is proportional to the transition probability of the combination; the cost of a failed match increases, and the increase is inversely proportional to the transition probability of the combination. For example, for two temperature sequences to be aligned, the basic cost value at position (i, j) is 0.8. If the change pattern label sequences at this position and its preceding and following positions match combination A, and the transition probability of combination A is 0.62, then the adjusted cost value is 0.8 × (1 - 0.62) = 0.304; if they do not match, then the adjusted cost value is 0.8 × (1 + 0.62) = 1.296.

[0091] A dynamic programming algorithm is used to calculate the path with the minimum cumulative cost based on the alignment cost matrix, which is then taken as the optimal alignment path. Starting from the top left corner of the matrix, the dynamic programming algorithm selects a direction of movement (right, down, or down-right) at each step, accumulating the minimum cost until the bottom right corner is reached, and then backtracking to obtain the optimal alignment path. The segmented sequences are then time-aligned according to the optimal alignment path to obtain the aligned sequence. The alignment process inserts null values ​​by horizontal or vertical movement along the path, while diagonal movements maintain the original values, achieving time scaling transformation of the sequence. For example, for two temperature segmented sequences with lengths of 145 and 152 respectively, the optimal alignment path obtained by the dynamic programming algorithm passes through the points {(1,1), (2,2), (3,4), (4,5), ..., (145, 152)}. Based on this path, the two sequences are aligned to the standard length of 150, resulting in the time-aligned sequence.

[0092] The aligned sequences are superimposed over time to obtain the parameter feature matrix, where each row represents the environmental parameter change process of a complete cycle. For continuous 7-day temperature parameter monitoring, a standard length of 1440 aligned sequences is generated each day (corresponding to 24 hours, one data point per minute). The 7 aligned sequences are superimposed to form a 7×1440 parameter feature matrix. This matrix captures the complete fluctuation characteristics and periodic change patterns of the temperature parameter within a week, laying the foundation for subsequent singular value decomposition and dynamic threshold curve construction.

[0093] In this embodiment, a dynamic time warping algorithm constrained by the rate of change achieves precise alignment of segmented sequences of environmental parameters, effectively solving the problem of inconsistent periodic fluctuations in clinical trial environmental parameters. The use of a sliding window-based extreme point and inflection point identification method accurately captures key features of parameter changes. Density clustering-based change pattern labeling improves the ability to identify similar change patterns. The construction of the state transition probability matrix reveals the inherent laws governing environmental parameter changes. A cost-weighted mechanism for change pattern combination matching enhances the accuracy of aligning key change points. A dynamic programming algorithm ensures the acquisition of the globally optimal alignment path. The overall scheme significantly improves the accuracy and robustness of environmental parameter sequence alignment, providing a reliable foundation for quality control and anomaly detection of clinical trial environmental data.

[0094] In one alternative embodiment,

[0095] The experimental data and anomaly monitoring dataset are input into distributed nodes. A Bloom filter is used to construct fingerprint feature vectors. Based on the fingerprint feature vectors, threshold signature key pairs are generated to construct a hierarchical access control strategy, including:

[0096] A distributed storage network based on blockchain is constructed to obtain multiple distributed nodes. The distributed nodes are divided into master nodes and slave nodes. A consensus network is built among the master nodes, and slave nodes are assigned to master nodes to form node groups.

[0097] The experimental data and anomaly monitoring dataset are input into the node group. The mutual information between data items is calculated to obtain the association strength matrix. A data dependency graph is constructed based on the association strength matrix. The association subgraphs in the data dependency graph are extracted according to the connectivity between data items to obtain the data type set. The propagation path of the data type set is calculated to obtain the access dependency sequence.

[0098] The access dependency sequence is input into a Bloom filter array, and a feature vector is generated using a group of independent hash functions. The feature vectors are then combined to obtain a fingerprint feature vector.

[0099] The feature polynomial is calculated based on the fingerprint feature vector and mapped to the key generator polynomial. The master private key is obtained based on the coefficient sequence of the key generator polynomial. The master private key is decomposed by Lagrange interpolation to obtain the sub-private key sequence. The corresponding master public key and sub-public key set are generated to form a threshold signature key pair.

[0100] The set of sub-public keys in the threshold signature key pair is grouped to obtain key groups. The coverage of each key group is calculated to obtain the access domain division. Based on the access domain division, hierarchical access constraints are constructed and combined to obtain a hierarchical access control strategy.

[0101] For example, a blockchain-based distributed storage network is first constructed to obtain multiple distributed nodes. The distributed storage network adopts a consortium blockchain architecture, with a total of N nodes, including M master nodes and NM slave nodes. The distributed nodes are divided into master nodes and slave nodes. Master nodes have the authority for data verification, consensus maintenance, and block generation, while slave nodes have the functions of data storage, querying, and forwarding. A consensus network is built among the master nodes, employing an improved practical Byzantine fault-tolerant algorithm, with a maximum number of fault-tolerant nodes of (M-1) / 3. Slave nodes are assigned to master nodes to form node groups. Each master node manages no more than (NM) / M slave nodes, and a master-slave replication mechanism is used within the node group to maintain data consistency. For example, in a clinical trial data quality control system, 20 distributed nodes are set up, including 7 master nodes and 13 slave nodes. The maximum fault tolerance for a master node is 2, and each master node manages an average of 1-2 slave nodes.

[0102] The experimental data and anomaly monitoring dataset are input into the node group. The experimental data includes basic subject information, medication records, and test results; the anomaly monitoring dataset includes records of abnormal environmental parameters, abnormal operation procedures, and abnormal data input labels. The mutual information between data items is calculated to obtain the association strength matrix. Mutual information is calculated by statistically analyzing the difference between the joint distribution and marginal distribution of two data items, used to quantify the correlation strength between data items. For example, the mutual information value between medication records and blood pressure test results is 0.75, between medication records and body temperature is 0.62, and between blood pressure and body temperature is 0.43. A data dependency graph is constructed based on the association strength matrix, establishing connections between data item pairs with mutual information values ​​greater than a threshold of 0.4, forming an undirected weighted graph structure. Based on the connectivity between data items, the association subgraphs in the data dependency graph are extracted to obtain the data type set. Connectivity is calculated using the degree centrality of nodes, selecting the top 30% of nodes by degree centrality as core nodes. Community structures are extracted around these core nodes to obtain the association subgraph. The data type set includes five main types: patient basic information, treatment records, test results, environmental monitoring, and operation records. The propagation path of a set of data types is calculated to obtain an access dependency sequence. The propagation path uses a shortest path algorithm to calculate the reachability and dependencies between each data type, forming an access dependency sequence represented by a directed acyclic graph. For example, the sequence "Patient basic information → treatment record → test result" forms an access dependency sequence, indicating that access to the test result requires prior access to the treatment record and patient basic information.

[0103] The access dependency sequence is input into a Bloom filter array, which consists of K independent Bloom filters, each with a bit array length of L bits, and an error rate controlled within 0.001. Feature vectors are generated using independent hash function groups. Each Bloom filter uses H independent hash functions to map each element in the access dependency sequence to H positions in the bit array, with these positions set to 1. The feature vectors are then combined to obtain a fingerprint feature vector by sequentially concatenating the bit arrays of the K Bloom filters to form a binary fingerprint feature vector of length K×L. For example, for an access dependency sequence of a patient's treatment data, using 5 Bloom filters, each 256 bits long, and each filter using 3 hash functions, the generated fingerprint feature vector is 1280 bits long, containing approximately 15% bits of 1 value.

[0104] The feature polynomial is calculated based on the fingerprint feature vector. The fingerprint feature vector is segmented, and the number of bits in each segment is converted into integers as the coefficients of the polynomial, constructing a (t-1)th order feature polynomial, where t is the security threshold. The feature polynomial is mapped to the key generator polynomial. The mapping process is implemented using a secure hash algorithm and an elliptic curve cryptography algorithm to ensure that the polynomial coefficients meet cryptographic security requirements. The master private key is obtained from the coefficient sequence of the key generator polynomial. The master private key is generated by combining the coefficient sequence and performing modular arithmetic. The master private key is decomposed by Lagrange interpolation to obtain the sub-private key sequence. Lagrange interpolation ensures that any t sub-private keys can reconstruct the master private key, while any t-1 sub-private keys cannot obtain any information about the master private key. The corresponding master public key and sub-public key sets are generated to form a threshold signature key pair. The public key is derived from the private key through elliptic curve multiplication. For example, for clinical trial data with a security threshold of t=3, 5 sub-private keys are generated through Lagrange interpolation to form a (3,5) threshold signature scheme, requiring at least 3 authorized parties to sign in order to access the complete data.

[0105] The set of sub-public keys in the threshold signature key pair is grouped to obtain key groups. The grouping principle is based on the data type set and access control requirements, grouping keys corresponding to highly related data types into the same group. The coverage of each key group is calculated to obtain the access domain division. The coverage is determined by analyzing the set of data resources that each key group can unlock. Based on the access domain division, hierarchical access constraints are constructed, forming a hierarchical structure of access domains according to their inclusion relationship. The upper-level access domain has data access rights to the lower-level access domain. The combination results in a hierarchical access control strategy, which includes role definitions, key allocation rules, and access restrictions. For example, in clinical trial data access control, there are four access levels: administrator, researcher, reviewer, and operator. The administrator level has all five sub-private keys and can access all data; the researcher level has three sub-private keys and can access patient information, treatment records, and test results; the reviewer level has two sub-private keys and can only access environmental monitoring and operation record data; the operator level has one sub-private key and can only view its own related operation records. This hierarchical access control strategy effectively ensures the security and access compliance of clinical trial data.

[0106] Figure 3 This is a flowchart illustrating the data asset tracing process based on a directed acyclic graph, as described in an embodiment of the present invention. Figure 3 As shown in the diagram, this system's core process is divided into two main stages: data preparation and traceability verification. The preparation stage involves building nodes (extracting features to generate identifiers), constructing a directed acyclic graph (with a time-series hierarchical arrangement), and generating access path identifiers, forming a DAG storage structure. The traceability stage receives a request, locates the target node, obtains the asset identifier package, and uses a Depth-First Search (DFS) algorithm to traverse the proof sequence for integrity and validity verification. Finally, it outputs the traceability result, which includes the operation history and proof of origin.

[0107] This technical solution effectively addresses the problems of low data security, coarse access control, and weak traceability in traditional clinical trial data management by constructing a blockchain-based distributed storage network and a hierarchical access control strategy. Existing technologies typically employ centralized database storage and fixed-role access control mechanisms, which struggle to cope with complex and ever-changing clinical trial scenarios and fail to meet the refined access needs of different stakeholders. To address these issues, this application innovatively introduces mutual information computation and data dependency graph analysis to accurately capture the inherent relationships between data items; it uses a Bloom filter array to construct fingerprint feature vectors, significantly improving the efficiency and accuracy of data feature extraction; it generates threshold signature key pairs based on feature vectors, realizing a flexible and controllable distributed authorization mechanism; and it constructs refined hierarchical access constraints through key grouping and access domain division. Compared to existing technologies, this solution's improvement focuses on achieving refined control and secure traceability of data access. By introducing cryptographic technology and the immutability of blockchain, it ensures the security, integrity, and traceability of clinical trial data throughout its entire lifecycle, while supporting differentiated access control for different roles, thus enhancing data security while maintaining system availability and flexibility.

[0108] In one alternative embodiment,

[0109] A hierarchical access control policy is written into a smart contract to generate a data access token. This data access token, along with a fingerprint feature vector, forms a data asset identification package, which is then input into the blockchain network. A proof sequence is generated using a zero-knowledge proof protocol, including:

[0110] Extract key verification conditions, timing constraints, and dependency constraints from the hierarchical access control strategy, construct a verification rule set, and write the verification rule set into the smart contract's state verification function.

[0111] Receive clinical trial data access requests, verify the private key combination, access time and access dependency information provided by the requester through a state verification function, and generate a data access token based on the verification result;

[0112] A hash operation is performed on the data access token to obtain the token hash value. The token hash value and the fingerprint feature vector are XORed to obtain a mixed feature sequence. The root hash value is calculated based on the mixed feature sequence as the data asset identifier value. The data asset identifier value, the data access token and the fingerprint feature vector are combined to form a data asset identifier package. The data asset identifier package is written into the blockchain network and a blockchain storage address is generated.

[0113] A token verification module is constructed based on the verification rule set, and a hash verification module is constructed based on the data asset identifier value. The token verification module and the hash verification module are combined to form a verification circuit. The data asset identifier package and the original clinical trial data are extracted based on the blockchain storage address as inputs to the verification circuit, and a proof sequence is generated through a zero-knowledge proof protocol.

[0114] This implementation extracts key verification conditions, temporal constraints, and dependency constraints from a hierarchical access control strategy to construct a set of verification rules. Key verification conditions specify the number and combination of sub-private keys required to access different types of data, implemented using a threshold signature mechanism. For example, accessing basic patient information requires at least three valid sub-private keys. Temporal constraints define the effective time window and access frequency limits for data access. For instance, clinical researchers can only access test data between 8:00 AM and 6:00 PM on weekdays, and the number of accesses per hour cannot exceed 10. Dependency constraints define the preconditions and associated authorization requirements for data access. For example, access to adverse reaction records requires prior informed consent from the patient. The set of verification rules is written into a smart contract to generate a state verification function. The smart contract adopts an upgradeable design pattern, supporting dynamic updates of verification rules. The state verification function implements complex rule verification through conditional judgments and logical combinations. For example, access rules for adverse drug reaction data can be defined as: verifying that the number of sub-private keys is ≥3 and includes the administrator's private key; verifying that the access time is within the allowed period; and verifying that access to basic patient information and medication records has been obtained.

[0115] The system receives clinical trial data access requests, which include the requester's identity, target data type, access time, and sub-private key combination. A state verification function verifies the private key combination, access time, and access dependency information provided by the requester. The verification process consists of three steps: First, verifying the validity of the private key by using an elliptic curve digital signature algorithm to verify whether the private key matches the corresponding public key; second, verifying time sequence compliance by checking whether the current access time is within the allowed time period and does not exceed the access frequency limit; and third, verifying dependency integrity by checking whether prior data access permissions have been authorized. Based on the verification results, a data access token is generated. The data access token contains information such as the requester's identity, authorized data scope, validity period, access restrictions, and digital signature. For example, a clinical researcher requests access to a specific patient's blood pressure monitoring data, providing three valid sub-private keys. After successful verification, the generated data access token includes: the researcher ID is "Researcher_12345", authorizing access to "BloodPressure_Data" of "Patient_67890", valid from 8:00 to 18:00 on the same day, with access restricted to "read-only and not exportable", and is accompanied by the digital signature of the master node.

[0116] A hash operation is performed on the data access token to obtain the token hash value. The hash algorithm uses SHA-256, outputting a fixed-length 256-bit hash value. The token hash value and the fingerprint feature vector are XORed to obtain a mixed feature sequence. The XOR operation is performed bit-by-bit to ensure that the original information cannot be directly extracted from the mixed feature sequence. A root hash value is calculated based on the mixed feature sequence as the data asset identifier. The calculation method involves segmenting the mixed feature sequence, constructing a Merkle tree, and extracting the root node hash value. The data asset identifier value, data access token, and fingerprint feature vector are combined to form a data asset identifier package. The data asset identifier package is encapsulated in JSON format and includes metadata fields and data fields. The data asset identifier package is written to the blockchain network, generating a blockchain storage address. The storage process is implemented by calling the blockchain network's transaction interface, and the generated storage address consists of a transaction hash and an index. For example, the data access token hash value of a patient's blood pressure data is "7f83b1657ff1fc53b92dc18148a1d65dfc2d4b1fa3d677284addd200126d9069". After XORing it with a fingerprint feature vector of length 1280 bits, the root hash value "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" is calculated and used as the data asset identifier value. The data asset identifier package is written into the blockchain network and a storage address "0x8b5...a72:3" is generated.

[0117] A token verification module is constructed based on a set of verification rules. This module includes a sub-private key verification circuit, a timestamp verification circuit, and a dependency verification circuit to verify the validity of the data access token. A hash verification module is constructed based on the data asset identifier value. This module includes a hash calculation circuit and a comparison circuit to verify data integrity and consistency. The token verification module and the hash verification module are combined to form a verification circuit, which is represented using arithmetic circuits and supports zero-knowledge proof protocols. The data asset identifier package and original clinical trial data are extracted from the blockchain storage address and used as inputs to the verification circuit. The verification circuit performs the following verifications: verifying the validity of the data access token signature, verifying whether the fingerprint feature vector matches the original data, and verifying whether the data asset identifier value is calculated correctly. A proof sequence is generated using a zero-knowledge proof protocol. The zero-knowledge proof adopts the zk-SNARK protocol, and its implementation includes a setup phase, a proof phase, and a verification phase. In the setup phase, public parameters are generated based on the verification circuit. In the proof phase, the prover generates a proof using private inputs (such as the original data and private key) and public inputs (such as the data asset identifier value). In the verification phase, the verifier verifies the validity of the proof using the public parameters and public inputs without needing to know the content of the private inputs. For example, if a researcher needs to prove that they have permission to access a specific patient's blood pressure data and that the data has not been tampered with, the system generates a proof sequence through a zero-knowledge proof protocol. This proof sequence can be verified by any verifier, but it will not reveal the patient's private data or the researcher's private key information.

[0118] Through the aforementioned technologies, this invention offers significant advantages in clinical trial data access control and traceability verification. Smart contracts automatically execute access control policies, ensuring data access compliance; data asset identification packages bind access permissions to data characteristics, enabling precise traceability; and zero-knowledge proof protocols provide verifiability while protecting privacy, making data access and usage traceable and tamper-proof. This approach is particularly suitable for multi-center clinical trial data sharing and auditing scenarios, ensuring both data security and patient privacy while meeting regulatory compliance requirements.

[0119] In one alternative embodiment,

[0120] A hash operation is performed on the data access token to obtain the token hash value. The token hash value is then XORed with the fingerprint feature vector to obtain a hybrid feature sequence. The root hash value, calculated based on the hybrid feature sequence, serves as the data asset identifier.

[0121] The access permission scope and validity period in the data access token are standardized, and a secure hash algorithm is used to perform hash calculation on the standardized data access token. In the hash calculation process, timestamp information is introduced as a random factor to generate a token hash value.

[0122] The integrity of the fingerprint feature vector is detected, and the missing dimensions are supplemented. The supplemented feature vector is then normalized to obtain a standardized fingerprint feature vector. The standardized fingerprint feature vector is then aligned and padded with zeros according to the length of the token hash value to obtain an aligned feature vector.

[0123] Construct a sliding window, segment the token hash value according to the sliding window size, and perform a bitwise XOR operation between the segmented token hash value and the aligned feature vector to obtain a mixed feature sequence;

[0124] A random perturbation factor sequence is generated. The random perturbation factor sequence is mixed with the mixed feature sequence to obtain a perturbation sequence. The perturbation sequence is rearranged according to a preset rearrangement rule to obtain an enhanced mixed feature sequence. The enhanced mixed feature sequence is divided into data blocks of fixed size. A multi-layer hash calculation structure is constructed. Iterative hash calculation is performed on adjacent data blocks to obtain the root hash value as the data asset identifier value.

[0125] For example, the access permission scope and validity period in the data access token are standardized. The standardization process includes permission scope encoding and time format unification. The permission scope encoding uses a hierarchical permission mapping table, mapping different types of permissions to fixed-length codes, such as "read-only permission" mapped to "R01", "read-write permission" mapped to "RW02", and "administrator permission" mapped to "ADM03". The time format is unified using UTC timestamps, accurate to the second. A secure hash algorithm, SHA-256, is used to hash the standardized data access token, with the standardized token content string as input. Timestamp information is introduced as a random factor during the hash calculation process; specifically, the current system timestamp is concatenated with the token content string before hash calculation, enhancing resistance to replay attacks. The generated token hash value is a 256-bit (32-byte) binary sequence. For example, the standardized token content for a clinical researcher accessing patient blood sample data is "ID: Researcher_456; Access: R01; Data: BloodSample_789; ValidFrom: 1626249600; ValidTo: 1626278400". After incorporating the timestamp "1626250125", the calculated token hash value is "8f4e7f6c5d2a1b3e9f8d7c6b5a4e3d2c1b0a9f8e7d6c5b4a3f2e1d0c9b8a7f6".

[0126] The dimensionality integrity of the fingerprint feature vector is detected by calculating the difference between the vector length and the expected standard length. Missing dimensions are then supplemented using nearest-neighbor interpolation, which estimates the missing dimensions based on the feature values ​​of adjacent dimensions. The supplemented feature vector is then normalized to obtain a standardized fingerprint feature vector. Normalization maps the values ​​of each dimension of the vector to the [0, 1] interval, preserving the feature distribution. The standardized fingerprint feature vector is then aligned and padded with zeros according to the length of the token hash value to obtain an aligned feature vector. The alignment method reduces the dimension by feature dimensionality reduction (e.g., principal component analysis) when the feature vector length is greater than the hash value length; when the feature vector length is less than the hash value length, zero values ​​are added to the end of the vector. For example, if a patient's original fingerprint feature vector has a dimension of 240, and the detection finds that dimensions 53-57 are missing, nearest-neighbor interpolation is used to supplement the missing dimensions to obtain a complete 240-dimensional feature vector. After normalization, it is mapped to the [0, 1] interval, and then alignment is performed to obtain a 256-bit aligned feature vector.

[0127] A sliding window is constructed with a size of 8 bits and a step size of 4 bits. The token hash value is segmented according to the sliding window size, generating 63 overlapping hash segments. A bitwise XOR operation is performed between the segmented token hash value and the aligned feature vector to obtain a mixed feature sequence. The XOR operation is implemented by comparing two binary numbers bit by bit, outputting 0 for the same bit and 1 for different bits. The mixed feature sequence retains information from both the token hash value and the feature vector, but it is difficult to directly deduce the original data. For example, the first 8-bit segment of the token hash value is "10110010", and the corresponding aligned feature vector segment is "01101101". After XORing, it becomes "11011111". This process is repeated for all segments to complete the XOR operation, resulting in a 256-bit mixed feature sequence.

[0128] A random perturbation factor sequence is generated using a cryptographically secure random number generator, with the same length as the mixed feature sequence. The random perturbation factor sequence is then mixed with the mixed feature sequence to obtain a perturbation sequence. The mixing method uses a weighted sum operation, summing corresponding bits of the two sequences with a weighted ratio of 0.7:0.3. A result greater than or equal to 0.5 is set to 1, and a result less than 0.5 is set to 0. The perturbation sequence is then rearranged according to a predefined rearrangement rule to obtain an enhanced mixed feature sequence. The rearrangement rule uses the Fisher-Yates shuffling algorithm, and a fixed seed ensures the repeatability of the rearrangement. The enhanced mixed feature sequence is divided into fixed-size data blocks, each 16 bits in size, resulting in 16 data blocks. A multi-layered hash computation structure is constructed, similar to a Merkle tree but using an improved hash connection method. Iterative hash calculations are performed on adjacent data blocks. In the first round, the hash values ​​of the 16 data blocks are combined pairwise to obtain 8 intermediate hash values. In the second round, the 8 intermediate hash values ​​are combined pairwise to obtain 4 new intermediate hash values. In the third round, the 4 intermediate hash values ​​are combined pairwise to obtain 2 intermediate hash values. In the final round, the 2 intermediate hash values ​​are combined to obtain the root hash value. The root hash value serves as the data asset identifier, with a length of 256 bits, which is represented as 64 characters in hexadecimal. For example, performing multi-level hash calculations on the enhanced mixed feature sequence of a patient's electronic health record, the final root hash value obtained is "a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6e7f8a9b0c1d2e3f4a5b6c7d8e9f0a1b2", which serves as the unique data asset identifier value for this health record.

[0129] The data asset identifiers generated using the above methods possess the following characteristics: uniqueness (different data access tokens or fingerprint feature vectors will generate different identifiers); irreversibility (the original token content or feature vector cannot be deduced from the identifier); tamper-proofness (minor changes in data content will lead to significant changes in the identifier); and verifiability (authorized parties holding the same data and access tokens can independently calculate and verify the correctness of the identifier). These characteristics make data asset identifiers reliable identifiers for clinical trial data in blockchain networks, supporting data access control and integrity verification, and providing a solid foundation for the quality control and traceability of clinical trial data.

[0130] In practical applications, case report data from a multi-center clinical drug trial, after processing, generates data asset identifier values ​​that are recorded in a blockchain network. When it is necessary to verify the authenticity and integrity of the data, the identifier value of the current data can be calculated and compared with the blockchain record to confirm whether the data has been tampered with. Simultaneously, by verifying the validity of data access tokens, it is possible to track who accessed which data at what time, achieving end-to-end data quality control and traceability.

[0131] In one alternative embodiment,

[0132] A directed acyclic graph structure is constructed based on the proof sequence. The proof sequence is concatenated between nodes using hash pointers, and access path information is written in. A tracing request is received, and the data asset identification package is obtained based on the access path information. After verifying the proof sequence, the tracing result is output, including:

[0133] The timestamp, operation type, and data identifier are extracted from the proof sequence and standardized to obtain an attribute feature vector. Node identifiers are generated based on the attribute feature vectors, and nodes with the node identifiers are used as the basic units of the directed acyclic graph.

[0134] Based on the temporal relationship of the proof sequence, nodes are arranged in layers. Horizontal hash pointers are established between nodes in the same layer, and vertical hash pointers are established between nodes in adjacent layers. The nodes are connected into a directed acyclic graph structure through the horizontal and vertical hash pointers.

[0135] For each proof sequence, an access path identifier is generated, written to the corresponding node, and added to the horizontal hash pointer and the vertical hash pointer to form a proof sequence containing access path information.

[0136] Upon receiving a tracing request, the target node is located in the directed acyclic graph structure based on the access path information in the request. The corresponding data asset identification package is obtained, and the proof sequence is traversed along the horizontal and vertical hash pointers to verify the integrity and validity of the proof sequence. The verified proof sequence is then output as the tracing result.

[0137] For example, firstly, the timestamp, operation type, and data identifier are extracted from the proof sequence and standardized to obtain the attribute feature vector. The timestamp uses UTC format, accurate to the millisecond level. The operation type includes data entry, data query, data modification, and data review. The data identifier includes a data type code and a unique identifier. The standardized encoding uses a fixed-length field processing method: the timestamp encoding length is 17 bits, the operation type encoding length is 4 bits, and the data identifier encoding length is 32 bits. The attribute feature vector is composed of these three encoded parts concatenated in sequence, with a total length of 53 bits. Node identifiers are generated based on the attribute feature vectors. The node identifiers are calculated using the SHA-256 hash algorithm on the attribute feature vectors, and the output is a 64-bit hexadecimal string. Nodes with node identifiers are used as the basic units of a directed acyclic graph. Each node contains a node identifier, an attribute feature vector, proof content, and a pointer field. For example, in a clinical trial, the data entry operation for patient blood pressure is timestamped as "20250620153045123", the operation type is "entry", the code is "ENT1", the data identifier is the combination of patient ID and test item "P4567_BP_SYSTOLIC_01", the standardized attribute feature vector is "20250620153045123ENT1P4567_BP_SYSTOLIC_01", and the generated node identifier is "7f83b1657ff1fc53b92dc18148a1d65dfc2d4b1fa3d677284addd200126d9069".

[0138] Based on the temporal relationship of the proof sequences, nodes generated within the same time interval are placed in the same layer, forming a bottom-up hierarchical structure with time arranged in natural ascending order. Horizontal hash pointers are established between nodes in the same layer, recording the node identifiers and relationships of adjacent nodes. Horizontal connections are achieved by combining the current node identifier with the next node identifier to calculate the hash value, which serves as the pointer. Vertical hash pointers are established between nodes in adjacent layers, recording the reference relationships from lower-level nodes to upper-level nodes. This is achieved by combining the current node identifier with the referenced node identifier to calculate the hash value, which serves as the pointer. These horizontal and vertical hash pointers connect the nodes into a directed acyclic graph structure, ensuring both the integrity of data traceability and effective verification of operational history. For example, in a clinical trial tracking system, the first layer, recording patient enrollment information, has 5 proof sequences; the second layer, recording baseline examination data, has 8 proof sequences; and the third layer, recording treatment follow-up data, has 12 proof sequences. Nodes on the same level are connected sequentially via horizontal hash pointers. For example, the first node in the first level is connected to the second node via a horizontal hash pointer, and so on. The patient enrollment information node in the first level is connected to the corresponding baseline examination data node in the second level via a vertical hash pointer, forming a complete traceability link.

[0139] An access path identifier is generated for each proof sequence. This identifier is generated by combining hierarchical information, the node's position within the layer, and the operation type, in the format "L{hierarchy}_N{node position}_O{operation type}". The access path identifier is written to the corresponding node, and an access path field is added to the node's metadata area. The access path identifier is then added to the horizontal and vertical hash pointers by using it as an additional input when calculating the pointer hash value, forming a proof sequence containing access path information. This design makes the access path information part of the hash calculation; any tampering with the path information will result in a change in the hash value, enhancing system security. For example, a patient's blood pressure data is located at the 4th node in the 3rd layer, and its access path identifier is "L3_N4_OENT1". This identifier is written to the node's metadata and added to the horizontal and vertical hash pointers connecting that node.

[0140] Upon receiving a tracing request, the system parses the request content to obtain the description information and access credentials of the target data. Based on the access path information in the request, the system locates the target node in the directed acyclic graph (DAG) structure. The access path information can be a specific access path identifier or data description information. When an access path identifier is provided, the node is located directly based on its level and position; when data description information is provided, the node is located by mapping to the corresponding access path identifier through an index structure. The system obtains the corresponding data asset identifier package, which contains a data access token, a fingerprint feature vector, and a data asset identifier value. The system traverses the proof sequence along the horizontal and vertical hash pointers using a depth-first search algorithm. Starting from the target node, it first traces the data source downwards along the vertical pointers, and then traces the operations at the same level along the horizontal pointers. The system verifies the integrity and validity of the proof sequence. Integrity verification is achieved by checking whether the hash pointer chain is complete and without breaks; validity verification is achieved by verifying the validity of the proof at each node using zero-knowledge proofs. The verified proof sequence is output as the tracing result, which includes the data operation history, data source proof, and integrity proof. For example, when a request is made to trace adverse reaction report data for a specific patient ID "P4567", the access path identifier "L5_N7_OADV2" is provided. The system locates the 7th node at the 5th level, obtains the data asset identifier package of the adverse reaction report, traverses along the hash pointer to obtain the complete proof sequence from patient enrollment, baseline examination, medication records to adverse reaction records, verifies the validity of all proofs, and outputs a complete traceability result containing 22 operation nodes, proving the authenticity of the adverse reaction data and the complete operation history.

[0141] This technical solution effectively addresses the problems of data silos, broken traceability chains, and low verification efficiency in existing clinical trial data traceability systems by constructing a proof sequence traceability structure based on a directed acyclic graph. Existing technologies typically employ linear log recording or simple blockchain storage for data traceability. These methods lack the ability to express complex data relationships and are prone to creating traceability silos in multi-center clinical trial scenarios, making it difficult to construct a complete view of the data operation history. This application, starting from improving the integrity and verification efficiency of data traceability, innovatively introduces an attribute feature vector standardized encoding mechanism to accurately capture the key features of data operations; it adopts a hierarchical directed acyclic graph structure to clearly display the temporal and dependency relationships of data operations; it achieves secure concatenation and integrity assurance of proof sequences through bidirectional connections of horizontal and vertical hash pointers; and it introduces an access path identification mechanism, significantly improving data location and traceability efficiency. Compared to existing technologies, this solution not only enhances the immutability and integrity of traceability data, but also achieves efficient targeted traceability and full-path verification through the combination of hash pointers and access path identifiers. This makes every operation of clinical trial data a traceable and verifiable link in the chain of evidence, providing solid technical support for clinical trial data quality control, while also meeting regulatory compliance requirements and data auditing needs.

[0142] A second aspect of the present invention provides an electronic device, comprising:

[0143] processor;

[0144] Memory used to store processor-executable instructions;

[0145] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0146] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0147] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A clinical trial data quality control and traceability method based on blockchain and multimodal perception, characterized in that, include: Collect environmental parameters and test data from clinical trials, calculate the fluctuation period of environmental parameters, construct a parameter feature matrix and perform singular value decomposition to obtain a dynamic threshold curve, compare the real-time values ​​of environmental parameters with the dynamic threshold curve to generate an environmental status score, generate anomaly identification information containing timestamps based on the environmental status score, and extract corresponding data from the test data based on the anomaly identification information to generate an anomaly monitoring dataset. The test data and anomaly monitoring dataset are input into the distributed nodes. A fingerprint feature vector is constructed using a Bloom filter. A threshold signature key pair is generated based on the fingerprint feature vector to construct a hierarchical access control policy. The hierarchical access control policy is written into a smart contract to generate a data access token. The data access token and the fingerprint feature vector are combined to form a data asset identification package and input into the blockchain network. A proof sequence is generated through a zero-knowledge proof protocol. A directed acyclic graph structure is constructed based on the proof sequence. The proof sequence is concatenated between nodes using hash pointers and access path information is written in. The traceability request is received, the data asset identification package is obtained based on the access path information, and the traceability result is output after verifying the proof sequence. The fluctuation period of environmental parameters is calculated, a parameter feature matrix is ​​constructed, and singular value decomposition is performed to obtain a dynamic threshold curve. The real-time values ​​of environmental parameters are compared with the dynamic threshold curve to generate an environmental status score, including: A multimodal sensor array was used to collect environmental parameters for the clinical trial, including temperature, humidity and light parameters. Wavelet denoising was then applied to the environmental parameters to obtain a filtered sequence. The autocorrelation function is calculated on the filtered sequence to obtain the correlation coefficient sequence. The local maxima of the correlation coefficient sequence are extracted to obtain the peak position sequence. The peak position sequence is clustered to obtain the peak cluster centers. The fluctuation period is determined based on the time interval of the peak cluster centers. The filtered sequence is segmented according to the fluctuation period, the rate of change of each segment is calculated to obtain the rate of change constraint, and the segmented sequence is aligned using a dynamic time warping algorithm based on the rate of change constraint to obtain an aligned sequence. The aligned sequences are then superimposed to form a parameter feature matrix. The parameter feature matrix is ​​decomposed into a set of feature vectors and a sequence of singular values. The principal feature vector is determined based on the information entropy curve of the singular value sequence. The standard fluctuation feature matrix is ​​reconstructed based on the principal feature vector. The probability density distribution of the standard fluctuation feature matrix is ​​calculated using the kernel density estimation method. An asymmetric dynamic threshold curve is constructed based on the skewness of the probability density distribution. The deviation sequence and rate of change sequence of real-time environmental parameters and dynamic threshold curves are calculated. Based on the temporal correlation of the deviation sequence and rate of change sequence, dynamic weights are determined and weighted to obtain the environmental status score.

2. The method according to claim 1, characterized in that, A dynamic time warping algorithm based on rate of change constraints is used to align the segmented sequences to obtain aligned sequences. These aligned sequences are then superimposed to form a parametric feature matrix, which includes: The time difference is calculated on the input segmented sequence to obtain the rate of change sequence; A fixed-length sliding window is constructed. Extreme points and inflection points are extracted within the sliding window to obtain a set of change points. The time interval, amplitude difference, and curvature of adjacent change points in the set of change points are calculated to obtain a feature vector. Density clustering is performed on the feature vector to obtain change pattern labels. The transition frequency matrix is ​​obtained by calculating the transition frequency between adjacent change mode labels. The state transition probability matrix is ​​obtained by normalizing the transition frequency matrix. The change mode combination is determined based on the state transition probability matrix. The Euclidean distance between the sequences to be aligned is calculated to obtain the basic cost matrix. The change pattern label and change pattern combination at each position in the basic cost matrix are matched. The alignment cost matrix is ​​obtained by weighting the basic cost matrix according to the matching results. The optimal alignment path is calculated based on the alignment cost matrix using a dynamic programming algorithm. The segmented sequences are then time-aligned according to the optimal alignment path to obtain an alignment sequence. The alignment sequences are then superimposed in the time dimension to obtain a parameter feature matrix, where each row of the parameter feature matrix represents the environmental parameter change process of a complete cycle.

3. The method according to claim 1, characterized in that, The experimental data and anomaly monitoring dataset are input into distributed nodes. A Bloom filter is used to construct fingerprint feature vectors. Based on the fingerprint feature vectors, threshold signature key pairs are generated to construct a hierarchical access control strategy, including: A distributed storage network based on blockchain is constructed to obtain multiple distributed nodes. The distributed nodes are divided into master nodes and slave nodes. A consensus network is built among the master nodes, and slave nodes are assigned to master nodes to form node groups. The experimental data and anomaly monitoring dataset are input into the node group. The mutual information between data items is calculated to obtain the association strength matrix. A data dependency graph is constructed based on the association strength matrix. The association subgraphs in the data dependency graph are extracted according to the connectivity between data items to obtain the data type set. The propagation path of the data type set is calculated to obtain the access dependency sequence. The access dependency sequence is input into a Bloom filter array, and a feature vector is generated using a group of independent hash functions. The feature vectors are then combined to obtain a fingerprint feature vector. The feature polynomial is calculated based on the fingerprint feature vector and mapped to the key generator polynomial. The master private key is obtained based on the coefficient sequence of the key generator polynomial. The master private key is decomposed by Lagrange interpolation to obtain the sub-private key sequence. The corresponding master public key and sub-public key set are generated to form a threshold signature key pair. The set of sub-public keys in the threshold signature key pair is grouped to obtain key groups. The coverage of each key group is calculated to obtain the access domain division. Based on the access domain division, hierarchical access constraints are constructed and combined to obtain a hierarchical access control strategy.

4. The method according to claim 1, characterized in that, A hierarchical access control policy is written into a smart contract to generate a data access token. This data access token, along with a fingerprint feature vector, forms a data asset identification package, which is then input into the blockchain network. A proof sequence is generated using a zero-knowledge proof protocol, including: Extract key verification conditions, timing constraints, and dependency constraints from the hierarchical access control strategy, construct a verification rule set, and write the verification rule set into the smart contract's state verification function. Receive clinical trial data access requests, verify the private key combination, access time and access dependency information provided by the requester through a state verification function, and generate a data access token based on the verification result; A hash operation is performed on the data access token to obtain the token hash value. The token hash value and the fingerprint feature vector are XORed to obtain a mixed feature sequence. The root hash value is calculated based on the mixed feature sequence as the data asset identifier value. The data asset identifier value, the data access token and the fingerprint feature vector are combined to form a data asset identifier package. The data asset identifier package is written into the blockchain network and a blockchain storage address is generated. A token verification module is constructed based on the verification rule set, and a hash verification module is constructed based on the data asset identifier value. The token verification module and the hash verification module are combined to form a verification circuit. The data asset identifier package and the original clinical trial data are extracted based on the blockchain storage address as inputs to the verification circuit, and a proof sequence is generated through a zero-knowledge proof protocol.

5. The method according to claim 4, characterized in that, A hash operation is performed on the data access token to obtain the token hash value. The token hash value is then XORed with the fingerprint feature vector to obtain a hybrid feature sequence. The root hash value is calculated based on the hybrid feature sequence as the data asset identifier, including: The access permission scope and validity period in the data access token are standardized, and a secure hash algorithm is used to perform hash calculation on the standardized data access token. In the hash calculation process, timestamp information is introduced as a random factor to generate a token hash value. The integrity of the fingerprint feature vector is detected, and the missing dimensions are supplemented. The supplemented feature vector is then normalized to obtain a standardized fingerprint feature vector. The standardized fingerprint feature vector is then aligned and padded with zeros according to the length of the token hash value to obtain an aligned feature vector. Construct a sliding window, segment the token hash value according to the sliding window size, and perform a bitwise XOR operation between the segmented token hash value and the aligned feature vector to obtain a mixed feature sequence; A random perturbation factor sequence is generated. The random perturbation factor sequence is mixed with the mixed feature sequence to obtain a perturbation sequence. The perturbation sequence is rearranged according to a preset rearrangement rule to obtain an enhanced mixed feature sequence. The enhanced mixed feature sequence is divided into data blocks of fixed size. A multi-layer hash calculation structure is constructed. Iterative hash calculation is performed on adjacent data blocks to obtain the root hash value as the data asset identifier value.

6. The method according to claim 1, characterized in that, A directed acyclic graph structure is constructed based on the proof sequence. The proof sequence is concatenated between nodes using hash pointers, and access path information is written in. A tracing request is received, and the data asset identification package is obtained based on the access path information. After verifying the proof sequence, the tracing result is output, including: The timestamp, operation type, and data identifier are extracted from the proof sequence and standardized and encoded to obtain an attribute feature vector. Node identifiers are generated based on the attribute feature vectors, and nodes with the node identifiers are used as the basic units of the directed acyclic graph. Based on the temporal relationship of the proof sequence, nodes are arranged in layers. Horizontal hash pointers are established between nodes in the same layer, and vertical hash pointers are established between nodes in adjacent layers. The nodes are connected into a directed acyclic graph structure through the horizontal and vertical hash pointers. For each proof sequence, an access path identifier is generated, written to the corresponding node, and added to the horizontal hash pointer and the vertical hash pointer to form a proof sequence containing access path information. Upon receiving a tracing request, the target node is located in the directed acyclic graph structure based on the access path information in the request. The corresponding data asset identification package is obtained, and the proof sequence is traversed along the horizontal and vertical hash pointers to verify the integrity and validity of the proof sequence. The verified proof sequence is then output as the tracing result.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images