CN121215303A - Multi-mode data fusion-based knee joint protection and knee treatment personalized planning system - Google Patents
Multi-mode data fusion-based knee joint protection and knee treatment personalized planning systemInfo
- Publication number
- CN121215303A CN121215303A CN202511783427.8A CN202511783427A CN121215303A CN 121215303 A CN121215303 A CN 121215303A CN 202511783427 A CN202511783427 A CN 202511783427A CN 121215303 A CN121215303 A CN 121215303A
- Authority
- CN
- China
- Prior art keywords
- treatment
- data
- path
- events
- event
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
本发明涉及医疗技术领域，公开了基于多模态数据融合的膝关节保膝治疗个性化规划系统。所述系统的数据获取与预处理模块，实时获取患者膝关节多模态数据流，生成标准化数据包；治疗事件识别模块，从标准化数据包中自动识别并验证关键治疗事件，生成治疗事件序列；历史治疗路径查询模块，查询分布式数据库，提取相似患者的治疗模式；图匹配与路径生成模块，计算当前治疗事件序列与历史治疗模式的相似度，当超过阈值时生成融合治疗路径并存储至区块链网络；异常监测与分析模块，实时监测融合治疗路径中的异常事件点，触发路径分解机制并进行多维度效能分析；个性化方案生成模块，动态调整治疗参数，生成个性化治疗规划方案。
This invention relates to the field of medical technology and discloses a personalized knee joint preservation treatment planning system based on multimodal data fusion. The system includes a data acquisition and preprocessing module that acquires multimodal data streams of the patient's knee joint in real time and generates standardized data packets; a treatment event identification module that automatically identifies and verifies key treatment events from the standardized data packets and generates treatment event sequences; a historical treatment path query module that queries a distributed database to extract treatment patterns from similar patients; a graph matching and path generation module that calculates the similarity between the current treatment event sequence and historical treatment patterns, and generates a fused treatment path and stores it in a blockchain network when the similarity exceeds a threshold; an anomaly monitoring and analysis module that monitors abnormal event points in the fused treatment path in real time, triggers a path decomposition mechanism, and performs multi-dimensional performance analysis; and a personalized plan generation module that dynamically adjusts treatment parameters to generate personalized treatment plans.
Description
Technical Field
The invention relates to the technical field of medical treatment, in particular to a knee joint and knee protection treatment personalized planning system based on multi-mode data fusion.
Background
Current knee protection and treatment planning mainly relies on doctor experience to judge and standardize treatment schemes. The prior art has many limitations on the utilization of patient data to single-modality image data, and fails to achieve effective integration of multi-source medical data. The treatment event identification process depends on manual labeling, and has low automation degree and strong subjectivity. Historical case data is underutilized and the experience of successful treatment of similar patients is not systematically mined and applied. The treatment path generation method is solidified and cannot be adjusted individually according to individual differences. The abnormality monitoring mechanism is lagged, complications are found by regular reexamination, and real-time early warning capability is lacked. The scheme adjustment process is empirical, and the parameter modification lacks a quantization basis. The conventional method needs to solve key technical problems such as multi-mode data fusion, intelligent event identification, historical experience utilization, personalized scheme generation and the like.
Conventional knee joint therapy planning systems suffer from significant shortcomings in terms of data drive and personalization. The multi-mode data acquisition standard is not uniform, and the problem of space-time inconsistency exists between the imaging data and the clinical examination data. The data preprocessing method is simple, and noise filtering and characteristic alignment effects of data from different sources are limited. The definition of the treatment event is fuzzy, and the automatic identification accuracy of the key treatment nodes is not high. The historical case inquiry efficiency is low, and the matching precision of similar patients is insufficient. The graph matching algorithm is complex in calculation, and the comparison of large-scale case data is too long. The similarity threshold is empirically set, and can not adapt to the characteristic difference of different disease types. The block chain storage performance bottleneck is obvious, and the data writing and verifying efficiency needs to be improved. The abnormal detection sensitivity is fixed, and the risk balance of false alarm and false omission is difficult. The path decomposition granularity is coarse, and the evaluation segment division lacks clinical basis. The efficacy analysis is single in dimension and the therapeutic effect cannot be fully assessed. The parameter adjustment strategy is linearized, and the multi-objective optimization problem cannot be handled.
Disclosure of Invention
The invention aims to provide a knee joint and knee protection treatment personalized planning system based on multi-mode data fusion, so as to solve the problems in the background technology.
To achieve the above object, the present invention provides a knee joint and knee protection therapy personalized planning system based on multi-modal data fusion, the system comprising:
the data acquisition and preprocessing module is used for acquiring the knee joint multi-mode data stream of the patient in real time, and carrying out data fusion preprocessing on the multi-mode data stream to generate a standardized data packet;
the treatment event identification module is used for automatically identifying key treatment events from the standardized data packet, verifying the effectiveness of the treatment events and generating a treatment event sequence;
The historical treatment path inquiring module is used for inquiring the historical treatment paths in the distributed database according to the biological characteristic identifiers of the patients and extracting treatment modes of similar patients;
The map matching and path generating module is used for calculating the similarity between the current treatment event sequence and the historical treatment mode by using a map matching algorithm, generating a fusion treatment path when the similarity exceeds a threshold value, and storing the fusion treatment path into a blockchain network;
the abnormality monitoring and analyzing module is used for monitoring abnormal event points in the fusion treatment path in real time, triggering a path decomposition mechanism when abnormality is detected, dividing the fusion treatment path into a plurality of evaluation segments, and carrying out multidimensional efficiency analysis on each evaluation segment;
and the personalized scheme generating module is used for dynamically adjusting the treatment parameters based on the analysis result to generate a personalized treatment planning scheme.
Preferably, the method for acquiring the knee joint multi-mode data stream of the patient in real time comprises the following steps:
Configuring a medical image equipment interface, an electronic health record system interface and a wearable sensor interface, and receiving knee joint image data, a clinical evaluation report and a real-time physiological signal in parallel to form a knee joint multi-mode data stream;
The method comprises the steps of carrying out time stamp alignment and space registration on knee joint multi-mode data streams, adopting a wavelet transformation algorithm to remove noise interference, using a data standardization protocol to convert heterogeneous data into a unified format, carrying out data normalization, unit unification and coding consistency processing, and finally outputting standardized data packets.
Preferably, the method for automatically identifying critical treatment events from standardized data packets includes:
The method comprises the steps of constructing a mixed model of a convolutional neural network and a long-period memory network, inputting multi-mode data in a standardized data packet, extracting spatial features including joint gap width change and cartilage morphological features from a branch of the convolutional neural network, analyzing time sequence features including pain index fluctuation and activity trend from the branch of the long-period memory network, weighting and fusing the spatial features and the time features through an attention mechanism, and outputting feature vectors;
The method comprises the steps of inputting feature vectors into a full-connection layer for classification, identifying treatment event types including medication events, physical treatment events and operation events, setting event validity check rules including event time continuity and physiological parameter rationality, only reserving the treatment events passing the check, and generating a treatment event sequence by sequencing according to time sequence.
Preferably, the method for querying a historical treatment path in a distributed database according to a patient biometric identification comprises:
Extracting biological characteristic identifiers of patients, including age, sex, body mass index and genetic marks, calculating characteristic similarity between the current patient and historical patients in a distributed database, using Euclidean distance measurement numerical characteristics, using Hamming distance measurement classification characteristics, setting a similarity threshold, screening out a subset of historical patients with similarity higher than the threshold, retrieving all available historical treatment paths from the subset of historical patients, wherein each historical treatment path comprises a complete event sequence and treatment results, grouping the historical treatment paths according to treatment modes by using a clustering algorithm, and extracting representative paths in each group as treatment mode templates.
Preferably, the method for calculating the similarity between the current treatment event sequence and the historical treatment mode by using a graph matching algorithm comprises the following steps:
The method comprises the steps of modeling a current treatment event sequence into a directed graph, enabling nodes to represent treatment events and edges to represent time sequence relations among the events, equally converting a historical treatment mode into a directed graph structure, comparing topological structures of the two directed graphs by adopting a graph isomorphism detection algorithm, calculating node matching degree and edge matching degree, introducing a weight factor, adjusting matching degree calculation based on event types and time sequence intervals, judging that matching is successful if the overall similarity score exceeds a preset threshold, aligning the successfully matched historical treatment mode with the current treatment event sequence, and connecting the same event nodes through virtual edges to generate a fusion treatment path.
Preferably, the method of storing the fused treatment path to the blockchain network includes:
the method comprises the steps of serializing a fusion treatment path into a data block, calculating a hash value of the data block, digitally signing the data block by using an asymmetric encryption algorithm, broadcasting the signed data block to a plurality of nodes of a block chain network, and performing a hash algorithm on the data block;
verifying the validity of the data block through a consensus mechanism, and adding the valid data block into the distributed account book; a unique blockchain identifier is generated for each data block and a timestamp and patient identification are recorded.
Preferably, the method for monitoring abnormal event points in the fusion treatment path in real time comprises the following steps:
defining abnormal event types including treatment interruption events, adverse reaction events and treatment effect deviation events;
Setting a control chart based on statistical process control, monitoring key indexes such as joint function scores and inflammation indexes, triggering abnormality detection when the indexes exceed control limits, evaluating abnormal severity grades by using a rule engine, and triggering a path decomposition mechanism only for high severity grade abnormality.
Preferably, the method for dividing the fusion treatment path into a plurality of evaluation segments comprises:
Identifying abnormal event points in the fusion treatment path as division points, adopting a variable point detection algorithm to verify the significance of the division points, dividing the path into continuous time periods by taking the division points as boundaries, and enabling each time period to be called an evaluation period;
And calculating the time trend slope and the effect change rate of each evaluation segment by using a linear regression model.
Preferably, the method for performing multi-dimensional efficacy analysis on each evaluation segment includes:
defining evaluation dimensions including a time efficiency dimension, a clinical effect dimension and an economic cost dimension, calculating a deviation ratio of actual duration to standard duration of an evaluation segment for the time efficiency dimension, quantifying a function improvement rate at the beginning and the end of the evaluation segment for the clinical effect dimension, accumulating resource consumption of all treatment events in the evaluation segment for the economic cost dimension, and synthesizing three dimension scores through a multi-objective optimization algorithm to generate an evaluation segment efficiency index.
Preferably, the method for dynamically adjusting the treatment parameters based on the analysis result comprises the following steps:
Identifying an evaluation segment lower than expected efficacy according to the efficacy index of the evaluation segment, extracting treatment event parameters including drug dosage, treatment frequency and rehabilitation action of the corresponding evaluation segment, exploring a parameter adjustment space by using a reinforcement learning agent, predicting the adjusted efficacy change by simulating environment, selecting a parameter adjustment scheme capable of improving the efficacy index to generate a new treatment event sequence, replacing the corresponding segment in the fusion treatment path, recalculating the overall path efficacy, iteratively optimizing until the termination condition is met, and outputting a final personalized treatment planning scheme.
Compared with the prior art, the invention has the beneficial effects that:
And (3) performing similarity calculation on the current treatment event sequence and the historical treatment mode by using a graph matching algorithm, generating a fusion treatment path when the similarity exceeds a threshold value, and storing the fusion treatment path into a blockchain network. The graph matching algorithm adopts graph isomorphism or subgraph matching technology to quantify the similarity degree of the current treatment event sequence and the historical treatment mode on the topological structure and node attribute. Similarity calculations consider the temporal order, duration, and intensity of treatment events, etc. characteristics. The threshold value setting is determined through statistical learning or clinical verification, so that the reliability of the matching result is ensured. The treatment path is fused to integrate the current patient characteristics with the historical success experience to form a personalized treatment framework. The path generation process adopts a multi-objective optimization algorithm to balance the factors such as curative effect, safety, cost and the like. The blockchain network stores the treatment path by using a distributed ledger technique, and ensures the non-tamper property and traceability of the data. The intelligent contract automatically performs path verification and authority management, and improves data security. The storage process adopts an encryption algorithm to protect privacy of patients, and the hash chain structure ensures data integrity.
And monitoring abnormal event points in the fusion treatment path in real time, triggering a path decomposition mechanism when abnormality is detected, dividing the fusion treatment path into a plurality of evaluation segments, and carrying out multidimensional efficiency analysis on each evaluation segment. The abnormal event points are identified through a rule engine or a machine learning model, and comprise abnormal conditions such as unexpected curative effect, occurrence of complications and the like. The path decomposition mechanism divides the path into logically complete evaluation segments according to the treatment phase and key nodes. The evaluation section divides the evaluation period considering clinical operation habit and treatment effect. The multidimensional performance analysis includes clinical symptom improvement, functional recovery, imaging changes, and patient satisfaction. The analysis party adopts statistical analysis, machine learning or deep learning algorithm to quantify the treatment effect of each evaluation segment. The abnormal detection sensitivity is dynamically adjusted according to the treatment stage, the detection frequency is increased in the acute stage, and the standard is properly relaxed in the recovery stage. And (5) visually displaying the evaluation result to assist doctors in understanding the treatment effect. And dynamically adjusting the treatment parameters based on the analysis result to generate a personalized treatment planning scheme. The parameter adjustment comprises quantifiable indexes such as medicine dosage, rehabilitation training intensity, physical therapy frequency and the like. The adjustment algorithm adopts reinforcement learning or optimization algorithm, and automatically updates the treatment scheme according to the efficiency analysis result. The personalized regimen is generated to take into account individual differences in patient physiological characteristics, lifestyle habits, and treatment preferences, among others. Scheme verification simulates the treatment effect through a digital twin technology, and scheme feasibility and risk are prejudged. The scheme updating mechanism continuously optimizes according to the treatment effect feedback to form a closed-loop optimization system. Through the synergistic effect of graph matching path generation, blockchain evidence storage, anomaly monitoring and parameter dynamic adjustment, accurate planning and personalized implementation of knee joint protection and treatment are realized. Historical experience matching provides a reliable basis, the blockchain storage certificate ensures that the scheme is reliable, abnormal monitoring is realized, timely intervention is realized, and the treatment effect is optimized through parameter adjustment. The integration method significantly improves the standardization and effectiveness of knee protection treatment of the knee joint.
Drawings
FIG. 1 is a schematic diagram of the system for personalized planning of knee joint and knee protection based on multi-modal data fusion;
FIG. 2 is a flow chart of multi-modal data stream acquisition and normalization;
FIG. 3 is a flow chart of historical treatment path query and treatment pattern extraction;
FIG. 4 is a block chain storage process diagram of the FIG. match and path generation module;
FIG. 5 is a schematic diagram of anomaly monitoring and path decomposition.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, the invention provides a knee joint knee-protecting treatment personalized planning system based on multi-mode data fusion, which comprises a data acquisition and preprocessing module for acquiring multi-mode data streams of a knee joint from medical image equipment, an electronic health record system and a wearable sensor in real time, and performing time stamp alignment, spatial registration, noise removal and standardization processing on the data to generate a unified standardized data packet. The treatment event identification module receives the standardized data packet immediately, automatically identifies key drug treatment, physical treatment or operation events by utilizing a convolutional neural network and long-term memory network hybrid model, verifies the effectiveness of the key drug treatment, physical treatment or operation events, and generates a treatment event sequence according to time sequence. The historical treatment path inquiring module inquires the historical treatment paths of similar patients in the distributed database according to the age, sex, body mass index, genetic markers and other biological characteristic identifiers of the current patients, and extracts a representative treatment mode template by using a clustering algorithm. The graph matching and path generating module respectively models the current treatment event sequence and the historical treatment mode into directed graphs, calculates the similarity of the topological structure of the current treatment event sequence and the historical treatment mode through a graph matching algorithm, and when the similarity exceeds a preset threshold value, performs alignment fusion on the successfully matched paths to generate a fusion treatment path and stores the path into a blockchain network to ensure that data cannot be tampered. The abnormality monitoring and analyzing module continuously monitors the executed fusion treatment path, when abnormal events such as treatment interruption, adverse reaction or curative effect deviation are detected, a path decomposition mechanism is triggered, the path is divided into a plurality of evaluation segments at abnormal points, and the efficiency analysis of three dimensions of time efficiency, clinical effect and economic cost is carried out on each evaluation segment. The personalized scheme generating module dynamically adjusts treatment parameters, such as medicine dosage or treatment frequency, by using reinforcement learning technology according to the efficiency analysis results of each evaluation section, and generates and outputs a final personalized treatment planning scheme, thereby completing the closed loop from data to personalized decision.
Referring to fig. 2, the data acquisition and preprocessing module is responsible for interfacing with various data source interfaces, the medical imaging device interface is responsible for receiving knee joint image data generated by a magnetic resonance imaging or computed tomography device, the electronic health record system interface is responsible for extracting clinical evaluation reports and medical history records of a patient from a hospital information system, and the wearable sensor interface is responsible for acquiring real-time physiological signals transmitted by a sensor attached to a knee joint part of the patient, including joint movement angles, surface electromyographic signals and local temperature data. In particular implementations, the three interfaces receive data in parallel, forming a continuous, multi-source, multi-modal data stream of the knee joint. The preprocessing of the knee joint multi-modal data stream comprises several key steps, the time stamp alignment operation ensures that all data points have a uniform time reference, and the spatial registration operation mainly aims at medical image data and calibrates images acquired by different time points or different devices to the same coordinate system. A wavelet transform algorithm is applied to the data stream to remove noise interference, which can effectively distinguish signal features from random noise. The data normalization protocol is started later, the data normalization processing is covered in the protocol content to scale the numerical value to a specific interval, the unit unified processing ensures that all physical quantities adopt international standard units, the coding consistency processing solves the problem of data coding difference among different systems, and finally a standardized data packet with unified format and reliable quality is output.
The core of the treatment event identification module is a convolutional neural network and long-term memory network mixed model, and a standardized data packet is used as the input of the mixed model. In a specific implementation, the convolutional neural network branches specifically process data having a spatial structure, such as extracting the change features of the joint gap width and the morphological features of the cartilage surface from knee joint image data. The long-term and short-term memory network branches then process the time series data and analyze the fluctuation mode of pain indexes obtained from the wearable sensor and the clinical report and the change trend of the joint activity. The features extracted by the two neural network branches are sent to an attention mechanism layer, and the attention mechanism layer dynamically calculates the weight coefficients of the spatial features and the temporal features, and performs weighted fusion to generate a comprehensive feature vector. It will be appreciated that the feature vectors are then input to a fully connected layer for classification, with the output nodes of the fully connected layer corresponding to different treatment event types, including medication events, physical treatment events, and surgical events. The set event validity checking rules start to work, the rule base comprises checking the event time continuity, preventing the occurrence of events with time logic contradiction, judging the rationality of the physiological parameters related to the events, and filtering out the data which obviously do not accord with medical common sense. Only the treatment events passing all the verification rules can be reserved, the effective events are ordered according to the sequence of the occurrence time, and finally a standard treatment event sequence is generated. In some embodiments, the architecture of the convolutional neural network may employ a combination of multi-layer convolutional layers and pooled layers, and the long-term memory network may include multiple layers of hidden units to capture longer-term time dependencies. A specific implementation of the attention mechanism layer may be a structure based on scaled dot product attention. The classifier of the fully connected layer may use a Softmax activation function to output the probability of each event type. The event validity check rules may be implemented with a set of predefined logical condition statements.
In the specific implementation of the data acquisition and preprocessing module, the medical imaging device interface needs to support digital imaging and communication standard protocols for medical science to ensure compatibility with mainstream imaging devices. The electronic health record system interface generally needs to follow a health-level seventh-layer protocol or a rapid medical interoperability resource standard to realize interconnection and interworking between systems. The wearable sensor interface may use bluetooth low energy technology or zigbee protocol for wireless data transmission. The timestamp alignment operation requires a high precision time synchronization mechanism, for which network time protocols are often used. The spatial registration operation may involve rigid body transformation or affine transformation algorithms to achieve accurate alignment between images. The specific choice of wavelet transform algorithm depends on the noise characteristics, and discrete wavelet transform is commonly used for digital signal processing. Normalization processing in the data normalization protocol may employ a min-max scaling method or a normalization method. The unit unification process needs to build a unit conversion mapping table. The code consistency process may require mapping of local codes to standard medical term systems, such as international disease classification codes or systematic clinical medical terms. It will be appreciated that these technical details together ensure the quality and efficiency of the generation of standardized data packets.
During operation of the treatment event recognition module, the input data of the convolutional neural network branches need to be preprocessed, and the medical image data may be resampled to a uniform resolution and converted to a tensor format. The size, step length and filling mode of the convolution kernel in the convolution neural network branch need to be designed according to the feature scale so as to optimize the feature extraction effect. The length of the input sequence of the long-short-term memory network branch needs to be determined, and the performance of the model can be influenced by the overlong or the overlong sequence. The weight calculation process of the attention mechanism layer is learnable and is continuously optimized in the model training process. The number of nodes of the fully connected layer needs to match the number of treatment event types. The formulation of event validity check rules requires deep participation of domain experts to ensure clinical rationality of the rules. In some embodiments, training of the entire hybrid model requires a large amount of labeling data, i.e., multi-modal data samples containing accurate therapy event signatures. The training process typically employs a back-propagation algorithm and a gradient descent optimizer. The selection of the loss function may be a cross entropy loss function. To avoid the overfitting phenomenon, regularization techniques such as dropping or weight decay may be employed. After model training is completed, an evaluation needs to be performed on a separate test data set to confirm the identification accuracy and generalization capability. Alternatively, the model may be incrementally learned periodically with new clinical data to preserve the advances in performance.
Example 2 referring to fig. 3, after the historical treatment path query module is initiated, a set of biometric identifiers including specific values, such as 58 years old, male sex, body mass index 26.5, and genetic markers such as the rs143383 locus genotype of the GDF5 gene, are first extracted from the current patient record. In a specific implementation, the module calculates the feature similarity between the current patient and each historical patient in the distributed database by using the biological feature identifiers, measures the numerical features such as age and body mass index by using Euclidean distance, calculates the square sum of the difference values of the corresponding feature values, and measures the classified features such as gender and genetic markers by using Hamming distance, wherein the calculation mode is the inconsistent feature category number. The system sets a similarity threshold, e.g., 0.85, and screens out all historic patients with similarity to the current patient characteristic calculation result higher than 0.85, so as to form a similar patient subset. All available historic treatment paths are retrieved from the subset of similar patients, each historic treatment path containing the complete sequence of treatment events and the final treatment outcome record. The module groups the retrieved plurality of historical treatment paths using a clustering algorithm, which may be used for this purpose, the algorithm divides the paths into different clusters based on similarity of treatment event sequences, and selects a central path or the most representative path from each cluster as a treatment pattern template.
The map matching and path generation module receives the sequence of treatment events generated from the treatment event recognition module and the treatment pattern template provided by the historical treatment path query module. In particular implementations, the module models the current sequence of treatment events as a directed graph structure, the nodes of the directed graph representing specific treatment events, and the edges of the directed graph representing timing relationships between events. The historical treatment pattern template is also converted into a structurally similar directed graph. The module compares the topological structure similarity between two directed graphs using a graph isomorphism detection algorithm that calculates the node matching degree and the edge matching degree. A weight factor is introduced in the similarity calculation, the weight factor being adjusted based on the importance of the event type and the difference in actual timing interval between events. If the calculated overall similarity score exceeds a preset threshold (e.g., 0.9), then a successful match is determined. After the matching is successful, the module executes the alignment operation, the directed graph of the current treatment event sequence is overlapped with the directed graph of the history treatment mode successfully matched, and the nodes representing the same treatment event in the two graphs are connected by adding the virtual edges, so that a new fusion treatment path is generated through fusion. It will be appreciated that this fused treatment path incorporates both the events that have occurred for the current patient and the subsequent treatment steps that were validated in the similar cases of history.
In some embodiments, the comprehensive calculation of feature similarity may use a weighted formula to integrate different distance metrics. An exemplary integrated similarity calculation formula is as follows:
Wherein the symbols are Representing a composite similarity score between the current patient and a historic patient. Sign symbolEuclidean distance representing the normalized numerical feature. Sign symbolRepresenting the hamming distance of the classification feature. Sign symbolRepresenting the weight coefficient given to the euclidean distance term. Sign symbolRepresenting weight coefficients given to hamming distance terms, and the weight coefficients satisfyIs a relationship of (3). It will be appreciated that by this formula, the distance measures of different dimensions and properties can be unified to a similarity score between 0 and 1 for comparison with the threshold. In the specific implementation of the clustering algorithm, the initial clustering center selection of the K-means clustering algorithm can adopt a K-means++ method to avoid sinking into a local optimal solution. The distance metric may use a dynamic time warping distance or an edit distance to better measure similarity between treatment event sequences. In graph matching, node matching calculation may need to take into account similarity of event parameters, such as proximity of drug dose or treatment intensity. Edge matching computation may involve comparison after discretization of the time interval. The addition of virtual edges requires following the principles of temporal logical consistency, ensuring that the fused paths are feasible on a timeline. Optionally, for a plurality of history treatment modes successfully matched, weights can be given according to the quality of the original treatment results, and weighted fusion can be performed, so that a better fusion treatment path can be generated.
Example 3. The map matching and path generation module after successful generation of a fused treatment path, for example a path comprising a sequence of events of "low frequency electrotherapy → joint loosening → oral celecoxib → ultrasound guided injection", initiates the process of storing it in the blockchain network. In implementations, the module first serializes this fused treatment path into a structured data block containing information such as a list of treatment events, a time stamp, a patient anonymity identifier, etc. The hash value of the data block is then calculated, which may employ the SHA-256 algorithm to generate a fixed length unique digital fingerprint, such as a 64-bit hexadecimal string. Next, the data block is digitally signed using an asymmetric encryption algorithm, and the operator encrypts the hash value of the data block using its private key to generate a digital signature. The signed data block, i.e., containing the original data block and the attached digital signature, is broadcast to a plurality of participating nodes in the blockchain network, which may be servers of a hospital, research institution, or regulatory agency.
After receiving the broadcasted data block, the nodes in the blockchain network start a consensus mechanism to verify the validity of the data block. In particular implementations, the verification process includes decrypting the digital signature using the public key of the data uploader to obtain a hash value H1, and locally recomputing the hash value H2 of the received original data block, comparing H1 to H2 to verify that the data was not tampered with during transmission and indeed originated from the purported sender. In some embodiments, the consensus mechanism may employ a practical Bayesian fault tolerance algorithm requiring more than two-thirds of nodes to agree on a data format specification, a valid signature, and a reasonable business logic to determine that a data block is valid. Once validated, this valid data block is added to a new block and linked to the blockchain's distributed ledger, forming a non-tamperable record. Each successfully stored data block generates a unique blockchain identifier, typically consisting of a blockhash value, and records a timestamp accurate to the millisecond level and a corresponding patient anonymity identifier.
The abnormality monitoring and analysis module is responsible for continuous monitoring of the fusion treatment path being clinically performed. The module continues to receive a flow of treatment performance data from the pre-treatment line, which may include a patient's daily pain self-score, joint activity measurements, medication intake records, and physical therapist's assessment notes. The module clearly defines the type of abnormal event to be monitored, the treatment interruption event refers to the cancellation of the planned treatment session or the failure of the patient to visit, the adverse reaction event refers to the drug allergy or operation related discomfort occurring in the treatment process, and the treatment deviation event refers to the actual improvement degree of the key clinical index which is obviously lower than the expected target of the fusion treatment path. To objectively identify these anomalies, the module sets a control chart based on statistical process control, such as a mean-range control chart, to monitor changes in knee function scores. The upper and lower control limits of the control map are typically set based on historical data or clinical criteria, e.g., the upper control limit is set to the expected improvement trend line plus three times the standard deviation, and the lower control limit is set to the expected improvement trend line minus three times the standard deviation.
When the treatment data input in real time indicate that the monitoring index exceeds the control limit preset by the control chart, the system immediately triggers an abnormality detection alarm. It will be appreciated that not all alarms represent a serious problem requiring intervention, and thus the module enables a built-in rules engine to evaluate the anomaly event deeply. The rules engine comprises a predefined rule base containing a series of logic conditions for evaluating the severity of an anomaly. For example, a rule may specify that an abnormal event with a "three consecutive pain score above the upper control limit and accompanied by a red swelling report" is classified as high severity. Whereas "single joint activity measurement is slightly below the lower control limit but no other symptoms" may be judged as low severity level. The module formally triggers the subsequent path resolution mechanism only if the rule engine evaluates the severity level of the anomaly event to "high". The design avoids unnecessary intervention of the system on temporary and slight data fluctuation, and ensures the accuracy of the response of the system. Alternatively, the rule base of the rule engine may be dynamically updated and maintained by clinical professionals based on up-to-date medical guidelines and clinical experience. The output of the anomaly monitoring and analyzing module is high-grade anomaly event points subjected to severity level screening, and the points are used as trigger signals and segmentation basis of path decomposition.
Referring to fig. 4, a secure storage process of knee treatment paths in a blockchain network is illustrated. A plurality of interconnected blocks of data are shown, each block containing an anonymous identifier of the patient, an accurate time stamp, a unique hash value, and specific treatment event information. The blocks are mutually linked through the encryption hash value to form a non-tamperable chained structure. This design ensures the integrity and traceability of the treatment path data, and any modifications to the history will be detected by the system. Arrows in the chart represent the connection relation among blocks, and black marked treatment events show complete treatment sequences from low-frequency electrotherapy to ultrasound-guided injection, so that the practical application of the blockchain technology in medical data safety management is embodied.
Example 4 when the anomaly monitoring and analysis module detects a high severity level of anomaly event point, such as a severe joint swelling adverse reaction recorded on day 45 of the treatment process, the module then triggers the path resolution mechanism. In a specific implementation, the module first identifies the abnormal event point as a potential division point, in order to verify the statistical significance of the division point instead of random fluctuation, the module adopts a variable point detection algorithm for analysis, the Bayesian variable point detection algorithm can be applied to the scene, the algorithm calculates the probability of mean or variance mutation of the treatment key index time sequence before and after the point, and when the probability exceeds a preset confidence level, the point is confirmed to be the significant division point. After validation, the module segments the complete fusion treatment path into successive time segments, e.g., into an "initial day to 44 day" evaluation segment and a "45 day to current" evaluation segment, each time segment being referred to as an evaluation segment, bounded by this salient segmentation point. For each evaluation segment generated by segmentation, the module extracts its key attributes, including the start time point attribute of the evaluation segment, the end time point attribute of the evaluation segment, and the complete sequence of treatment events occurring within the evaluation segment. The module uses a linear regression model to model index changes in each evaluation segment, the linear regression model uses time as independent variable and clinical indexes as dependent variable, the calculated regression coefficient is the time trend slope of the evaluation segment, and the index difference value at the beginning and the end of the evaluation segment is calculated to obtain the effect change rate.
The multi-dimensional efficacy analysis of each evaluation segment is the core step in evaluating therapeutic benefit. The analysis process defines three independent assessment dimensions, the time efficiency dimension concerns the progress of the treatment, the clinical effect dimension concerns the health outcome, and the economic cost dimension concerns the resource consumption. In a specific implementation, for the time efficiency dimension, a deviation ratio is calculated between the actual duration of the evaluation segment and a standard duration expected for the segment of treatment according to clinical pathway criteria, e.g., 15 days for one evaluation segment and 10 days for a standard duration, the deviation ratio is (15-10)/10=0.5. For the clinical effect dimension, the function improvement rate at the beginning and the end of the evaluation segment is quantified and calculated by adopting the following formula:
Wherein the symbols are Representing the rate of functional improvement of the evaluation segment. Sign symbolRepresenting the functional score (e.g., WOMAC score) at the beginning of the evaluation segment. Sign symbolRepresenting the functional score at the end of the evaluation segment. Sign symbolRepresenting the functional scoring target value under ideal conditions. For the economic cost dimension, the resources consumed by all treatment events in the evaluation segment are accumulated, and the costs of medicine, consumables, equipment use and manpower are converted into monetary units according to a standard price list and summed. It can be understood that the measurement units and meanings of the three dimensions are different, and for comprehensive judgment, the module integrates the scores of the three dimensions into a single evaluation segment efficiency index through a multi-objective optimization algorithm, and the algorithm distributes weights for each dimension, performs weighted summation and normalizes. Referring to table 1, the intermediate results of the extracted attributes and calculations in the multidimensional performance analysis are shown.
TABLE 1 evaluation segment multidimensional Performance analysis Table
In some embodiments, the variability point detection algorithm may also use cumulative and control graph algorithms or autoregressive model residual analysis to identify structural variability points in the sequence, in addition to bayesian methods. When the linear regression model calculates the time trend slope, if the data has significant nonlinear characteristics, fitting using piecewise linear regression or nonlinear models can be considered. When the multi-objective optimization algorithm integrates dimension scores, the setting of the weight coefficients can be determined by clinical experts by adopting a analytic hierarchy process or can be dynamically adjusted according to the management objective of a hospital. It will be appreciated that the segment performance index provides a quantified, integrated benefit index for laterally comparing performance of different segments and identifying weaknesses that require optimization. Alternatively, the results of the performance analysis may be visually presented to the physician, for example, using a thermodynamic diagram to show the scores of the different assessment segments in each dimension, to aid in making clinical decisions.
Referring to fig. 5, a real-time monitoring and anomaly analysis system during knee joint preservation therapy is shown. The graph shows the patient's score change versus expected trend of improvement over the course of actual treatment with the treatment time on the horizontal axis and knee function score on the vertical axis. The green area of the graph represents the normal range of score fluctuation, and the red solid line indicates the upper and lower limits of statistical process control. The blue solid line records the actual daily scoring of the patient and the system triggers an anomaly detection alarm when the score exceeds the control limit. The purple vertical lines mark significant segmentation points identified based on a bayesian variability detection algorithm, dividing the entire treatment process into two evaluation segments. The treatment periods of all the evaluation sections are distinguished by the background areas with different colors, and the system provides data support for the dynamic adjustment of the treatment scheme by carrying out multi-dimensional efficiency analysis on each evaluation section, so that the safety and the effectiveness of the treatment process are ensured.
Embodiment 5 the personalized solution generation module receives output from the anomaly monitoring and analysis module, i.e., a multi-dimensional performance analysis report for each of the evaluation segments, the report containing the performance index for each of the evaluation segments and a detailed dimension score. In practice, the module first identifies an evaluation segment with a performance index lower than the expected performance based on the evaluation segment performance index, for example, the qualification threshold for the performance index is set to 0.7, and then all evaluation segments with performance index lower than 0.7 will be marked as low performance evaluation segments requiring optimization. The module then extracts detailed parameters of all treatment events recorded within these low efficacy evaluation segments, including, for medication treatment events, the medication name, specific dose, frequency of administration, and route of administration, for physical treatment events, the parameters of the extraction including treatment type (e.g., ultrasound, mid-frequency electrotherapy), treatment intensity, duration of each treatment, and frequency of weekly execution, for rehabilitation training events, the parameters of action combination, number of repetitions per set, number of daily training sets, etc.
After extracting the treatment event parameters of the low-performance evaluation segment, the module starts a reinforcement learning agent to explore the space for parameter adjustment. In implementations, the reinforcement learning agent takes the current treatment plan state (i.e., the current set of parameters) as a state input, and the possible parameter adjustment actions as an action space. The reinforcement learning agent performs these adjustments in a simulated treatment environment model that is constructed based on historical data and physiological mechanisms to predict potential trends in the patient's clinical indices (e.g., pain scores, joint activity) after the parameter adjustment actions are performed. The goal of reinforcement learning agents is to maximize a bonus function whose design is directly related to the improvement expectations of the performance index of the evaluation segment. An exemplary bonus function is designed as follows:
Wherein the symbols are Representing in stateTake action downwardsThe instant rewards obtained. Sign symbolRepresenting the amount of change in the predicted clinical effect dimension score. Sign symbolRepresenting the amount of variation in the predicted economic cost dimension score (typically an increase in cost is considered negative). Sign symbolRepresenting the amount of change in the predicted time efficiency dimension score (typically time extension is considered negative). Sign symbol,,Is a weight coefficient used to balance the importance of different dimensions in the bonus calculation. The reinforcement learning agent learns a strategy by continually trial and error in the simulation environment that is capable of selecting those parameter tuning actions that are most likely to result in a high prize, i.e., significantly improving the overall performance index.
When the reinforcement learning agent finds one or more parameter adjustment schemes that can improve the predictive performance index, the module selects an optimal scheme from the schemes. The selection criteria may be based on the highest prize value or by a combination of clinical feasibility and safety of the regimen. In particular implementations, the module uses the selected parameter adjustment scheme to generate a new, optimized sequence of treatment events, e.g., "oral celecoxib 200 mg/day," physical therapy 2 times per week is replaced with "oral celecoxib 250 mg/day," physical therapy 3 times per week "in the original low efficacy evaluation segment, thereby forming a new sequence of treatment events. The module then replaces the corresponding low efficacy evaluation segment in the original fusion treatment path with this new treatment event sub-sequence. After the replacement is completed, the module recalculates the overall performance index of the entire adjusted treatment path, which may be obtained by weighted averaging the performance indices of the individual evaluation segments. It will be appreciated that this process is iterative and the module will evaluate the new path again and initiate a new round of parameter tuning optimization if there are still low performance segments or if the overall performance has not reached the termination condition. Iterative optimization continues until the overall path efficacy meets a preset termination condition, at which point the module outputs a final determined version of the personalized treatment planning scheme. In some embodiments, the reinforcement learning agent may be implemented based on a deep Q network or a near-end policy optimization algorithm. The construction of the simulated environment model requires integration of pathophysiological knowledge, pharmacokinetic models and evidence of rehabilitation efficacy of knee osteoarthritis. The design of the parameter adjustment action space requires setting reasonable boundaries, for example, the adjustment of the drug dosage must be within the safety range prescribed by pharmacopoeia, and the adjustment of the treatment frequency needs to take the tolerability of the patient into consideration.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202511783427.8A CN121215303A (en) | 2025-12-01 | 2025-12-01 | Multi-mode data fusion-based knee joint protection and knee treatment personalized planning system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202511783427.8A CN121215303A (en) | 2025-12-01 | 2025-12-01 | Multi-mode data fusion-based knee joint protection and knee treatment personalized planning system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN121215303A true CN121215303A (en) | 2025-12-26 |
Family
ID=98121612
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202511783427.8A Pending CN121215303A (en) | 2025-12-01 | 2025-12-01 | Multi-mode data fusion-based knee joint protection and knee treatment personalized planning system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN121215303A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160321427A1 (en) * | 2015-04-28 | 2016-11-03 | Siemens Medical Solutions Usa, Inc. | Patient-Specific Therapy Planning Support Using Patient Matching |
| US20200237452A1 (en) * | 2018-08-13 | 2020-07-30 | Theator inc. | Timeline overlay on surgical video |
| US11256760B1 (en) * | 2018-09-28 | 2022-02-22 | Automation Anywhere, Inc. | Region adjacent subgraph isomorphism for layout clustering in document images |
| CN114944209A (en) * | 2022-03-11 | 2022-08-26 | 青岛百洋智能科技股份有限公司 | An integrated computing method and system for medical similar medical records |
| CN118975778A (en) * | 2024-08-05 | 2024-11-19 | 西安市红会医院（西安市骨科研究所） | An anesthesia depth monitoring system based on EEG signals |
| CN119377894A (en) * | 2024-12-27 | 2025-01-28 | 北京紫云智能科技有限公司 | Multimodal data dynamic fusion and annotation method and system for medical decision-making |
| CN119964786A (en) * | 2025-04-11 | 2025-05-09 | 四川互慧软件有限公司 | Dynamic critical care treatment pathway construction method and system based on historical medical record analysis |
| US20250174337A1 (en) * | 2023-11-27 | 2025-05-29 | Cephalgo Sas | Artificial intelligence-based personalized predictive treatment system |
| CN120151370A (en) * | 2025-03-25 | 2025-06-13 | 江苏省中医药研究院(江苏省中西医结合医院) | Medical monitoring data transmission method, monitoring method and application method |
| CN120356610A (en) * | 2025-02-24 | 2025-07-22 | 景宁畲族自治县人民医院(县域医共体) | Gynecopathy diagnosis and treatment method and system integrating electronic medical records |
-
2025
- 2025-12-01 CN CN202511783427.8A patent/CN121215303A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160321427A1 (en) * | 2015-04-28 | 2016-11-03 | Siemens Medical Solutions Usa, Inc. | Patient-Specific Therapy Planning Support Using Patient Matching |
| US20200237452A1 (en) * | 2018-08-13 | 2020-07-30 | Theator inc. | Timeline overlay on surgical video |
| US11256760B1 (en) * | 2018-09-28 | 2022-02-22 | Automation Anywhere, Inc. | Region adjacent subgraph isomorphism for layout clustering in document images |
| CN114944209A (en) * | 2022-03-11 | 2022-08-26 | 青岛百洋智能科技股份有限公司 | An integrated computing method and system for medical similar medical records |
| US20250174337A1 (en) * | 2023-11-27 | 2025-05-29 | Cephalgo Sas | Artificial intelligence-based personalized predictive treatment system |
| CN118975778A (en) * | 2024-08-05 | 2024-11-19 | 西安市红会医院（西安市骨科研究所） | An anesthesia depth monitoring system based on EEG signals |
| CN119377894A (en) * | 2024-12-27 | 2025-01-28 | 北京紫云智能科技有限公司 | Multimodal data dynamic fusion and annotation method and system for medical decision-making |
| CN120356610A (en) * | 2025-02-24 | 2025-07-22 | 景宁畲族自治县人民医院(县域医共体) | Gynecopathy diagnosis and treatment method and system integrating electronic medical records |
| CN120151370A (en) * | 2025-03-25 | 2025-06-13 | 江苏省中医药研究院(江苏省中西医结合医院) | Medical monitoring data transmission method, monitoring method and application method |
| CN119964786A (en) * | 2025-04-11 | 2025-05-09 | 四川互慧软件有限公司 | Dynamic critical care treatment pathway construction method and system based on historical medical record analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN118609743B (en) | Medical data management method, system and storage medium based on artificial intelligence | |
| Liu et al. | Healthcare fraud detection: A survey and a clustering model incorporating geo-location information | |
| CN119889617B (en) | Multi-level attribution and recommendation method and system for medical management decision | |
| CN119028609B (en) | Otorhinolaryngology patient management follow-up system and method | |
| CN116662375B (en) | HIS-based prescription data verification method and system | |
| CN118016270A (en) | Internet-based cervical cancer perioperative traditional Chinese medicine nursing monitoring method and system | |
| CN105005709A (en) | Process mining method based on single-disease treatment process | |
| CN119479982A (en) | A method and system for intelligent processing of hospital pharmacy dispensing information | |
| CN120072246A (en) | Neurosurgery nursing management system based on cloud computing | |
| CN118841187A (en) | Patient compliance assessment system | |
| CN117854738B (en) | Medical care system based on intelligent terminal | |
| CN118824582A (en) | A method and system for remote graded follow-up and precise management of chronic cardiovascular disease | |
| CN118800482A (en) | An electronic follow-up management system for monitoring the efficacy of topical drugs in dermatology | |
| CN116313144B (en) | A disease risk prediction system based on physical examination data with non-equidistant time intervals | |
| CN117297549A (en) | Anesthesia state monitoring method and system based on neural network | |
| CN119418951B (en) | Adverse drug reaction monitoring method and system based on big data | |
| CN119230134B (en) | Model training method, system and medium for acute respiratory tract infection prediction | |
| CN120299733A (en) | A multi-dimensional medical data analysis and decision-making support system and method | |
| CN121215303A (en) | Multi-mode data fusion-based knee joint protection and knee treatment personalized planning system | |
| CN119650029A (en) | A smart health management system and method based on the Internet of Things | |
| CN120108737B (en) | Health management data processing method and system based on patient dynamic feedback | |
| CN121034515B (en) | A method, device and medium for processing data on hyperuricemia and gout | |
| CN120236782B (en) | A digital tumor prevention and treatment management platform and management method | |
| CN120048500B (en) | Anesthesia recovery state intelligent assessment method and system | |
| CN117524434B (en) | Expert information management optimization method and system based on vein treatment data platform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |