External counterpulsation evaluation system based on AI assistance of cardiopulmonary metabolism multi-source dynamic information

Abstract

The invention discloses an in-vitro counterpulsation evaluation system based on cardiopulmonary metabolism multi-source dynamic information AI assistance. The in-vitro counterpulsation evaluation system comprises a multi-source data acquisition module, a self-adaptive information synchronization module, an AI assistance module and a counterpulsation evaluation module, the multi-source data acquisition module is used for acquiring a real-time data stream of multi-source detection data; the self-adaptive information synchronization module is used for identifying an optimal alignment point according to multi-source detection data to realize data alignment; the AI auxiliary module is used for performing exception supervision according to the data subjected to data alignment; the counterpulsation evaluation module is used for generating an evaluation report according to the abnormal supervision result; according to the invention, real-time dynamic data of different human body systems can be obtained, physiological information of a patient is comprehensively collected from multiple dimensions, a real-time data stream of multi-source detection data is formed, the limitation of traditional single index evaluation is changed, and a rich data basis is provided for comprehensive and accurate auxiliary evaluation.

Description

Technical Field

[0001] This invention relates to the field of intelligent assisted assessment technology, and more specifically to an AI-assisted external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism. Background Technology

[0002] Currently, with the continuous development of medical technology, external counterpulsation (ECP), as a non-invasive circulatory support therapy, has been widely used in the treatment of cardiovascular diseases due to its significant advantages in improving cardiac blood supply and circulation. Early on, the auxiliary assessment of ECP therapy mainly relied on the physician's clinical experience and simple physiological indicator tests. With technological advancements, assessment systems based on single or a few physiological parameters, such as heart rate, blood pressure, and electrocardiogram monitoring, have gradually developed. While these systems can provide basic data for reference, they have significant limitations in assisting treatment decision-making and optimizing the treatment process.

[0003] In external counterpulsation therapy, accurate auxiliary assessment plays a crucial role in treatment efficacy. However, the human cardiopulmonary metabolic system is interconnected and operates collaboratively; relying on a single, limited indicator is insufficient to comprehensively reflect changes in the patient's physiological state during treatment, and cannot provide effective support for physicians to adjust treatment parameters in real time. For example, when adjusting treatment intensity, a single indicator assessment may fail to balance cardiac load, pulmonary ventilation, and metabolic needs, leading to a disconnect between the treatment plan and the patient's actual needs.

[0004] To improve the effectiveness of auxiliary assessments, some systems have attempted to introduce multi-parameter detection. However, due to the lack of an effective data alignment mechanism, data from different sources differ in collection frequency and start time, making direct fusion analysis prone to errors and severely weakening the accuracy of auxiliary assessments. This interferes with doctors' interpretation of data and assessment of treatment progress, making it difficult to optimize treatment plans based on reliable information.

[0005] Furthermore, existing systems also have significant shortcomings in assisting the monitoring of abnormal states. There are complex synergistic relationships among cardiopulmonary metabolic parameters, and traditional algorithms, limited by computational power and logical architecture, cannot capture abnormal changes between these parameters in real time and with precision. When patients experience abnormal cardiopulmonary metabolic synergy, the system cannot provide timely warnings, making it difficult for doctors to adjust treatment strategies promptly. This affects the dynamic optimization and personalized implementation of external counterpulsation therapy, resulting in suboptimal treatment outcomes.

[0006] Currently, the limitations of auxiliary assessment in external counterpulsation therapy, such as its singularity, inaccuracy, and lag in abnormal monitoring, have become key bottlenecks restricting the improvement of treatment effects and hindering the development of personalized medicine. How to overcome technical barriers and build an efficient and accurate auxiliary assessment system is a core issue that urgently needs to be addressed in the field of external counterpulsation. Summary of the Invention

[0007] In view of this, the present invention provides an external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance. It can acquire real-time dynamic data of different human body systems, comprehensively collect patient physiological information from multiple dimensions, and form a real-time data stream of multi-source detection data. This changes the limitations of traditional single-indicator assessment and provides a rich data foundation for comprehensive and accurate auxiliary assessment.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] An external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance includes: a multi-source data acquisition module, an adaptive information synchronization module, an AI assistance module, and a counterpulsation assessment module;

[0010] The multi-source data acquisition module is used to acquire real-time data streams of multi-source detection data;

[0011] The adaptive information synchronization module is used to identify the optimal alignment point based on multi-source detection data to achieve data alignment.

[0012] The AI-assisted module is used to perform anomaly monitoring based on the data aligned with the data.

[0013] The counter-pulse evaluation module is used to generate an evaluation report based on the abnormal monitoring results.

[0014] Preferably, the multi-source data acquisition module connects to multiple data acquisition terminals to acquire real-time dynamic data of different human body systems.

[0015] Preferably, the data acquisition terminal includes a cardiopulmonary system data acquisition device, a metabolic system data acquisition device, and a nervous system data acquisition device.

[0016] Preferably, the adaptive information synchronization module includes a preprocessing submodule, a feature extraction submodule, and an adaptive alignment submodule;

[0017] The preprocessing submodule is used to resample the multi-source detection data using a specific sampling frequency and perform quality assessment to obtain a time-base unified signal matrix and a signal quality assessment matrix.

[0018] The feature extraction submodule is used to extract features from the time-base unified signal matrix to obtain a feature set, and to perform noise reduction based on the signal quality assessment matrix;

[0019] The adaptive alignment submodule is used to establish a global alignment benchmark based on the noise-reduced feature set, and to align various types of data based on the global alignment benchmark.

[0020] Preferably, the adaptive alignment submodule includes a clustering unit, a local alignment unit, and a global alignment unit;

[0021] The clustering unit is used to extract the features of each data point and perform fusion clustering.

[0022] The local alignment unit is used to traverse the clustering categories and perform intra-class time registration by filtering high-quality time periods in the quality assessment matrix through a threshold, so as to obtain the local benchmark for each category.

[0023] The global alignment unit is used to perform correlation analysis on the local benchmarks of each category, generate the global alignment benchmark, and perform alignment.

[0024] Preferably, the AI-assisted module includes an anomaly monitoring submodule and a tiered alarm submodule;

[0025] The anomaly monitoring submodule is used to analyze synergistic anomalies in cardiopulmonary metabolic parameters through multimodal data fusion.

[0026] The hierarchical alarm submodule is used to generate alarms based on different anomaly types.

[0027] Preferably, the anomaly monitoring submodule includes a multi-signal coupling modeling unit and a hierarchical anomaly detection unit;

[0028] The multi-signal coupling modeling unit constructs a dynamic coupling matrix based on multi-source detection data and extracts singular values ​​as the main coupling feature vector;

[0029] The hierarchical anomaly detection unit is used to perform modal stability detection, energy distribution anomaly detection, and spatiotemporal mismatch detection based on the main coupling vector.

[0030] Preferably, the counterpulsation assessment module includes a data processing unit, a report generation unit, and a storage unit;

[0031] The data processing unit is used to perform statistical analysis on the anomaly monitoring results and extract key evaluation indicators;

[0032] The report generation unit automatically generates an external counterpulsation assessment report based on key evaluation indicators and a preset template; the storage unit is used to classify and store the assessment reports for easy retrieval and access later.

[0033] Preferably, the counterpulsation assessment module further includes a visualization unit, which synchronously displays waveforms and numerical change lists of key assessment indicators based on a time axis, distinguishes between normal and abnormal data ranges through color markings, and supports user-defined filtering of different detection time periods and different types of data for comparative analysis, intuitively presenting the external counterpulsation treatment effect and changes in the patient's cardiopulmonary metabolic status.

[0034] Preferably, the counterpulsation assessment module has a built-in assessment parameter database, which pre-stores external counterpulsation assessment standard parameters corresponding to different ages, genders and common underlying diseases;

[0035] The counterpulsation assessment module can match and analyze the patient's actual test data, basic information, and current abnormal monitoring results against the assessment parameter database, and automatically generate adjustment suggestions for external counterpulsation treatment for different patients, including adjustment plans for treatment intensity and treatment duration, as well as subsequent rehabilitation precautions.

[0036] As can be seen from the above technical solutions, compared with the prior art, this invention discloses an AI-assisted external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism. It can acquire real-time dynamic data from different human systems, comprehensively collect patient physiological information from multiple dimensions, and form a real-time data stream of multi-source detection data. This overcomes the limitations of traditional single-indicator assessment and provides a rich data foundation for comprehensive and accurate auxiliary assessment. This invention employs a dynamic adaptive synchronization mechanism to achieve millisecond-level precise alignment of multi-source heterogeneous data and can adjust strategies in real time according to data characteristics. This invention constructs a multimodal data deep coupling analysis model, combined with an AI-based hierarchical early warning mechanism, to achieve early and accurate identification of cardiopulmonary metabolic abnormalities. This invention constructs a dynamic assessment model based on individual patient characteristics to achieve real-time optimization and accurate recommendation of treatment plans. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of an external counterpulsation assessment system based on AI-assisted multi-source dynamic information of cardiopulmonary metabolism, provided by the present invention.

[0039] Figure 2 This is a schematic diagram of the adaptive information synchronization module in an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of the AI-assisted module in an embodiment of the present invention. Detailed Implementation

[0041] 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.

[0042] like Figure 1 This invention discloses an external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance, comprising: a multi-source data acquisition module, an adaptive information synchronization module, an AI assistance module, and a counterpulsation assessment module;

[0043] The multi-source data acquisition module is used to acquire real-time data streams of multi-source detection data; the adaptive information synchronization module is used to identify the best alignment point based on the multi-source detection data to achieve data alignment; the AI-assisted module is used to perform anomaly supervision based on the aligned data; and the counter-pulse evaluation module is used to generate an evaluation report based on the anomaly supervision results.

[0044] In one embodiment, the multi-source data acquisition module connects to multiple data acquisition terminals and acquires real-time dynamic data of different human body systems through each data acquisition terminal.

[0045] The data acquisition terminals include cardiopulmonary data acquisition devices, metabolic system data acquisition devices, and nervous system data acquisition devices.

[0046] Cardiopulmonary data acquisition devices, such as high-precision pulse sensors, ECG electrodes, and respiratory rate sensors, can collect real-time cardiopulmonary function data such as heart rate, heart rhythm, ECG waveform, respiratory rate, and tidal volume. Taking ECG acquisition as an example, the electrodes can accurately sense the weak current generated by the heart's electrical activity, convert it into an electrical signal, and transmit it to the acquisition module, providing raw data for assessing the heart's electrophysiological state.

[0047] Metabolic system data acquisition equipment uses biosensors and metabolic analysis instruments to monitor key indicators in the human metabolic process, such as blood glucose concentration, blood oxygen saturation, exhaled carbon dioxide, and lactate concentration. These data reflect the body's material conversion and energy supply during energy metabolism, and are crucial for assessing the impact of external counterpulsation on the metabolic system.

[0048] Nervous system acquisition equipment uses EEG sensors, EMG sensors, and other technologies to collect data such as brain electrical signals, muscle electrical signals, and nerve conduction velocity, thereby reflecting the functional state of the nervous system and its response to external counterpulsation.

[0049] like Figure 2In one embodiment, the adaptive information synchronization module includes a preprocessing submodule, a feature extraction submodule, and an adaptive alignment submodule;

[0050] The preprocessing submodule is used to resample and perform quality assessment on multi-source detection data using a specific sampling frequency pair to obtain a time-base unified signal matrix and a signal quality assessment matrix.

[0051] The feature extraction submodule is used to extract features from the time-base unified signal matrix to obtain a feature set, and to perform noise reduction based on the signal quality assessment matrix;

[0052] The adaptive alignment submodule is used to establish a global alignment benchmark based on the denoised feature set, and to align various types of data based on the global alignment benchmark.

[0053] In this implementation, the preprocessing submodule performs data preprocessing through multi-rate resampling technology and adopts a synchronization algorithm based on interpolation compensation to unify the time base of the raw signals from multiple sources (ECG, respiration, blood samples, and electromyography).

[0054] For example, when the preset reference frequency is 1000Hz, cubic spline interpolation can be performed on a 30Hz low-frequency breathing waveform signal, and high-frequency signals (such as 2000Hz EEG) can be downsampled to finally generate a time-based unified signal matrix with the dimension of the number of signal channels × the number of unified sampling points.

[0055] Furthermore, the preprocessing process can also assess signal quality, resulting in a signal quality assessment matrix. Specifically, it calculates quality indicators such as signal-to-noise ratio, baseline drift amplitude, and peak-to-peak spacing variance for each signal in real time; it performs dynamic scoring based on a sliding window, and combines weighting coefficients to achieve a comprehensive assessment of multiple quality indicators, outputting a [0-1] interval scoring matrix with the same dimension as the signal matrix.

[0056] The feature extraction module can achieve multimodal feature fusion extraction and adaptive denoising filtering in combination with the quality assessment matrix.

[0057] For example, windowing is used to extract the mean, variance, zero-crossing rate, and Hjorth parameter of each channel signal to obtain time-domain features; continuous wavelet transform (CWT) is used to extract the energy distribution features of a specific frequency band to obtain frequency-domain features; chaotic features such as sample entropy and Lyapunov exponent are calculated to obtain nonlinear features; these features are fused to obtain a high-dimensional feature set. During this process, the system calculates quality indicators such as signal-to-noise ratio, baseline drift amplitude, and peak-to-peak spacing variance for each signal in real time. A 0-1 interval scoring matrix with the same dimension as the signal matrix is ​​generated through dynamic evaluation using a sliding window and combined with weighting coefficients (e.g., signal-to-noise ratio accounts for 60%, baseline drift accounts for 30%). Low-quality signal segments (e.g., scores <0.5) will be highlighted in subsequent processing.

[0058] Meanwhile, a signal quality assessment matrix is ​​used for adaptive noise reduction: features corresponding to low-quality signal segments are weighted by entropy (e.g., by suppressing noise interference through an exponential decay function), and manifold learning techniques (such as t-SNE or UMAP) are used to perform nonlinear dimensionality reduction on the high-dimensional feature set, preserving the topological relationships of cross-modal features. During this process, the random forest algorithm evaluates feature importance, removes redundant features with correlation coefficients higher than 0.8, and finally obtains an optimized low-dimensional feature set.

[0059] Finally, based on the denoised feature set, the adaptive alignment submodule constructs a global alignment benchmark.

[0060] To further implement the above technical solution, the adaptive alignment submodule includes clustering units, local alignment units, and global alignment units.

[0061] The clustering unit is used to extract the features of each data point and perform fusion clustering; the local alignment unit is used to traverse the clustering categories and filter the high-quality time periods in the quality assessment matrix through a threshold to perform intra-cluster time registration and obtain the local benchmarks for each category; the global alignment unit is used to perform correlation analysis on the local benchmarks of each category, generate a global alignment benchmark, and perform alignment.

[0062] Specifically, the feature fusion clustering unit performs cross-signal joint analysis on the denoised multimodal feature set (including time-domain mean / variance, frequency band energy distribution, sample entropy, etc.). Feature similarity is calculated by defining a hybrid distance metric (e.g., combining dynamic time warping distance, cosine similarity, and Hellinger distance), and a signal quality assessment matrix is ​​introduced as a weight constraint—a distance penalty coefficient (e.g., amplifying the original distance by 1.5 times) is applied to signal segments with a quality score below 0.6. Based on the OPTICS density clustering algorithm, high-density regions in the feature space are automatically divided into K feature clusters (e.g., ECG R-peak features, respiratory wave rising edge features, and electromyography burst features are classified separately). During clustering, the time distribution range of signal segments within each cluster is statistically analyzed using a sliding window, outputting the original signal time interval set {T1, T2, ..., T...} corresponding to each cluster. K}

[0063] Next, the local alignment unit performs intra-class temporal registration for each feature cluster. For the i-th cluster, from the corresponding time interval T... i High-confidence periods with a quality score higher than 0.8 were selected. During this period, the Iterative Dynamic Time Warping (DTW) algorithm is used to calculate the average alignment path of multiple signal segments: First, a reference signal segment S is randomly selected. ref Other signal segments S are transmitted via DTW. jAligning with it, we obtain the time mapping function φ. j (t)(represents S) j The t-th sampling point corresponds to S ref φ j (t) point) and calculate the average waveform:

[0064]

[0065] Where k represents the number of iterations, and N is the number of signal segments in the current cluster. The reference template is iterated and updated repeatedly until convergence, generating the local reference template B for the cluster. i (t). The convergence condition can be set as follows: To improve robustness, outliers with cross-correlation values ​​below 0.7 with the baseline template are removed, and cubic spline interpolation is used to compensate for missing intervals. For high-frequency signal clusters (such as electromyographic bursts), a phase-locked loop (PLL) technique is additionally introduced to calculate the instantaneous phase difference using wavelet phase difference.

[0066] Instantaneous phase difference:

[0067] φ(t)=arg(CWT(S ref (t))·CWT(S target (t)) * )

[0068] Where arg represents the argument, CWT represents the continuous wavelet transform, and S target Indicates the reference signal segment S ref The relative target signal segment.

[0069] To achieve sub-sampling precision phase fine-tuning, local alignment errors are controlled within ±3ms, with the adjustment amount being:

[0070] Δt(t)=Δφ(t) / (2πf c )

[0071] Where Δt(t) is the time adjustment amount calculated based on the instantaneous phase deviation, used to achieve sub-sampling precision phase fine-tuning, f c The wavelet center frequency is used to convert phase deviation into a time quantity when calculating the time adjustment.

[0072] Finally, the global alignment unit performs cross-modal correlation analysis on multiple local benchmarks. A framework is constructed based on key events of each local benchmark (such as the ECG R-peak time t). ECG respiratory peak time t Resp Let be an event sequence graph of nodes, with edge weight ω. mn The event interval Δt between mode m and mode n is represented. mn =t m -tn The statistical significance of ω is determined by calculating the p-value using a t-test; if p < 0.05, then ω mn =1-p, otherwise ω mn =0.) When a conflicting path is detected (e.g., the time difference between the ECG R peak and the pulse oximetry peak exceeds the physiologically reasonable range), the path with the most supporting nodes is preferentially retained, and the global time offset is optimized based on maximum likelihood estimation. Finally, the objective function is solved using dynamic programming:

[0073]

[0074] Where, σ i The confidence level of the i-th local criterion is calculated as follows: σ i =1-exp(-μQ i );Q i Δt represents the average quality score within the cluster, μ is the attenuation coefficient, and λ is the smoothing constraint coefficient used to avoid abrupt shifts in offset between adjacent time periods. i This represents the time offset of the i-th local reference relative to the global reference.

[0075] For example, in a motion monitoring scenario, the module first clusters the high-frequency burst characteristics of the electromyography (EMG) signal, the R-peak characteristics of the electrocardiogram (ECG), and the motion impact characteristics of the accelerometer into three independent clusters. Within each EMG cluster, high-confidence burst segments with a quality score greater than 0.85 are selected. A local benchmark template is generated through iterative DTW (Dynamic Time-to-Wide Wave) to reduce its time jitter from the original ±25ms to ±4ms. In the global alignment stage, an event chain constraint is established based on the physiological conduction sequence of "ECG R-peak → EMG burst → motion impact," constraining the transmission time interval of relevant times. The absolute time offset of the EMG benchmark is then deduced by combining the accelerometer signal, ultimately achieving a multi-signal global alignment error of <10ms. This method, through a hierarchical, progressive clustering-local-global strategy, significantly improves alignment robustness in scenarios with dynamically changing signal quality, avoiding cascading errors caused by the failure of a single signal in traditional methods.

[0076] like Figure 3 To further implement the above technical solution, the AI-assisted module includes an anomaly monitoring submodule and a tiered alarm submodule. The anomaly monitoring submodule is used to analyze the synergistic abnormalities of cardiopulmonary metabolic parameters through multimodal data fusion; the tiered alarm submodule is used to issue alarms based on different anomaly types.

[0077] Furthermore, the anomaly supervision submodule includes a multi-signal coupling modeling unit and a hierarchical anomaly detection unit; the multi-signal coupling modeling unit constructs a dynamic coupling matrix based on multi-source detection data and extracts singular values ​​as the main coupling feature vector; the hierarchical anomaly detection unit is used to perform modal stability detection, energy distribution anomaly detection, and spatiotemporal mismatch detection based on the main coupling vector.

[0078] Specifically, for multi-signal coupling modeling, the multi-source detection data is standardized before constructing the multimodal coupling matrix. Suppose the multi-source detection data contains n signals of different types, such as ECG signals, respiratory signals, metabolic indicator signals, etc. For the i-th signal sequence, the Z-score standardization method is used to convert each signal sequence into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0079]

[0080] Where, μ i and σ i These are the mean and variance of the signal sequence, respectively.

[0081] Then, an n×n dynamic coupling matrix C(t) is constructed, where the matrix elements C ij (t) represents the coupling strength between the i-th signal and the j-th signal at time t. The coupling strength is calculated using a cross-correlation function.

[0082]

[0083] in, and The signal sequences x' are respectively i (t) and x' j (t) is the mean value of the signal within the time window [t-τ, t+τ], where τ is a set time delay parameter used to capture the time dependency between signals.

[0084] Finally, singular value decomposition is performed on the dynamic coupling matrix:

[0085] C(t)=U∑V T

[0086] Here, U and V are orthogonal matrices, and ∑ is a diagonal matrix, with the elements on the diagonal being the singular values. The first m largest singular values ​​are selected to form the principal coupling eigenvector.

[0087] For anomaly detection analysis, including modal stability detection, energy distribution anomaly detection, and spatiotemporal mismatch detection.

[0088] Modal stability detection includes calculating the rate of change of the coupled eigenvectors over time. For each singular value, the rate of change at its adjacent time points is calculated.

[0089]

[0090] Set a threshold ∈ 1, when r iWhen (t)>∈1, the i-th mode is considered unstable. By statistically analyzing the number and distribution of unstable modes, the overall modal stability is determined to be abnormal. If the number of unstable modes exceeds a set threshold, a modal stability anomaly alarm is issued.

[0091] Energy distribution anomaly detection includes: calculating the energy proportion of each singular value in the main coupling eigenvector.

[0092]

[0093] Define a normal energy distribution range, for example, by obtaining the average μ of the energy proportion of each singular value through historical normal data statistics. ei and standard deviation σ ei When the energy percentage exceeds [μ ei -kσ ei μ ei +kσ ei When a singular value is detected, the energy distribution corresponding to that singular value is considered abnormal. By analyzing the singular values ​​representing the proportion of abnormal energy and their changing trends, it is determined whether the overall energy distribution is abnormal. If the total proportion of abnormal energy exceeds a set threshold, an energy distribution anomaly alarm is issued.

[0094] Spatiotemporal mismatch detection includes: constructing a spatiotemporal correlation matrix, where each element represents the spatial correlation between the i-th and j-th signals at time t, which can be obtained through the spatial location and time delay information of the signals.

[0095] S ij (t) = spatial_correlation(x) i (t), x j (-Δt ij ))

[0096] Where spatial_correlation represents the spatial correlation function, Δt ij The time delay between the i-th signal and the j-th signal can be determined by methods such as cross-correlation analysis.

[0097] Finally, the matrix norm is used to measure the degree of difference between it and the spatiotemporal correlation matrix under normal conditions:

[0098]

[0099] A threshold ∈2 is set, and when d(t)>∈2, it is considered that there is a spatiotemporal mismatch anomaly.

[0100] In this embodiment, the hierarchical alarm submodule is used to drive the generation of corresponding alarms based on the anomaly type determined by the anomaly supervision submodule.

[0101] To further implement the above technical solution, the counter-pulse assessment module includes a data processing unit, a report generation unit, and a storage unit.

[0102] The data processing unit is used to perform statistical analysis on the abnormal monitoring results and extract key evaluation indicators; the report generation unit automatically generates an external counterpulsation evaluation report based on the key evaluation indicators and a preset template; the storage unit is used to classify and store the evaluation reports for easy retrieval and access later.

[0103] Specifically, a sliding time window is used for anomaly frequency calculation, duration statistics, and signal strength analysis. Each time the time window slides across, a corresponding SQL statement is executed to count the occurrences of each anomaly type within that window. For anomaly duration calculation, the timestamp of each anomaly record is continuously tracked during the window's sliding process, and the anomaly is determined by calculating the difference between adjacent records.

[0104] Next, key indicators are extracted based on time series data. This embodiment uses the Exponentially Weighted Moving Average (EWMA) algorithm.

[0105] After calculating the statistical indicators, the EWMA algorithm is applied to each indicator, assigning different weights based on time sequence to give more weight to recent data as the primary indicator. For example, for the anomaly frequency indicator f... t Its EWMA value is:

[0106] EWMA t =α×f t +(1-α)×EWMA t-1

[0107] Where α is the weighting coefficient.

[0108] Then, the principal component analysis (PCA) algorithm is run periodically (e.g., every minute) to reduce the dimensionality of the updated indicators and re-extract the key evaluation indicators to adapt to the dynamic changes in the data.

[0109] For report generation, Natural Language Generation (NLG) technology is used to generate textual descriptions based on key assessment indicator data, following pre-defined language templates and logical rules. For example, when heart rate indicators are abnormal, a description is automatically generated: "The patient's current heart rate is XX beats / minute, higher than the normal range (60-100 beats / minute), which may indicate excessive cardiac load." For chart generation, a data visualization library (such as D3.js) is used. Appropriate chart types (line charts, bar charts, etc.) are selected based on data characteristics, and visual charts are generated through data binding and rendering technologies and embedded in the report. Regarding report review and export, a web-based report review system is built, using WebSocket technology for real-time communication. After reviewers modify the report content, other relevant personnel can immediately see the changes. The PDF.js library is used to convert the report content to PDF format, supporting multiple export formats (such as PDF and Word), and file download functionality is implemented through file stream operations.

[0110] For data storage, a distributed file system is used to store evaluation report files. A storage cluster including management nodes and storage nodes is constructed, and report files are uploaded and downloaded through a specific interface. A relational database is used to store patient information and basic report information. Different data tables are created, and patients and reports are mapped through association fields. The report data table is partitioned according to time to speed up data retrieval.

[0111] The decision tree algorithm is used to classify the report data, using patient age, gender, underlying diseases, and treatment stage as classification criteria. A classification model is trained to predict the category of new reports and store them in the corresponding storage area. In the database, indexes are created for frequently used query fields, and in the file storage system, category tags are added to each report file for easy and quick data retrieval.

[0112] Develop a daily incremental backup and weekly full backup schedule. Daily incremental backups only copy changed data blocks to the backup storage device, while weekly full backups package and compress the entire storage content and store it off-site. During data recovery, locate the point in time to be restored based on the backup records. For database data, restore the full backup file first, then apply the incremental backup data; for files in the file storage system, directly copy the files from the backup device to their original location at the corresponding point in time.

[0113] Furthermore, it also includes visualization units and a built-in evaluation parameter database.

[0114] The visualization unit synchronously displays waveforms and numerical change lists of key assessment indicators based on a time axis. It distinguishes between normal and abnormal data ranges through color markings and supports user-defined filtering of different detection time periods and different types of data for comparative analysis, intuitively presenting the effects of external counterpulsation therapy and changes in the patient's cardiopulmonary metabolic status.

[0115] The database pre-stores external counterpulsation assessment standard parameters corresponding to different ages, genders, and common underlying diseases. The counterpulsation assessment module can match and analyze the patient's actual test data, basic information, and current abnormal monitoring results against the assessment parameter database, and automatically generate external counterpulsation treatment adjustment suggestions for different patients, including adjustment plans for treatment intensity and treatment duration, as well as subsequent rehabilitation precautions.

[0116] Specifically, for visualization, when displaying waveforms, the time-series data is smoothed, and the waveforms are drawn using plotting techniques, updating the data every second to ensure real-time changes. When displaying lists of numerical changes, virtual scrolling technology is used, showing only data rows within the visible screen area to improve display efficiency when dealing with large amounts of data. Furthermore, the filtering function records filter conditions on the front end. When a user selects a filter, a request is sent to the back end, which retrieves data based on the conditions and returns the matching data to the front end for re-display. The comparative analysis function calculates the similarity between different time periods or different types of data, displaying the comparison results using a heatmap, with color intensity indicating the magnitude of data differences.

[0117] For the built-in evaluation parameter database, a parameter management system is first developed using a microservice architecture, including service modules for parameter storage, querying, and updating. These service modules collaborate to implement parameter storage, querying, and updating operations, manage parameter versions, and record information on each parameter change.

[0118] Then, the K-nearest neighbor algorithm is used for parameter matching. The actual test data of patients are grouped into data sets, and the five most similar standard parameter data are searched in the database. By calculating the distance between data sets, similar data are sorted, and different weights are assigned according to the reliability of the data to calculate the matching reference value, which is then compared with the actual values ​​of patients.

[0119] Finally, a suggestion generation system is built using a rule engine. Treatment adjustment rules are written in a specific format, and the patient's actual test data and parameter matching analysis results are input into the rule engine. Through the matching and reasoning of the rule engine, personalized treatment adjustment suggestions are generated, including specific adjustment values ​​for treatment intensity and duration, as well as rehabilitation precautions.

[0120] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0121] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-assisted external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism, characterized in that, include: Multi-source data acquisition module, adaptive information synchronization module, AI-assisted module, and counterpulsation evaluation module; The multi-source data acquisition module is used to acquire real-time data streams of multi-source detection data; The adaptive information synchronization module is used to identify the optimal alignment point based on multi-source detection data to achieve data alignment; the AI-assisted module is used to perform anomaly monitoring based on the aligned data. The counter-pulse evaluation module is used to generate an evaluation report based on the abnormal monitoring results.

2. The external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 1, characterized in that, The multi-source data acquisition module connects to multiple data acquisition terminals to obtain real-time dynamic data from different human body systems.

3. The extracorporeal counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 2, characterized in that, The data acquisition terminal includes a cardiopulmonary system data acquisition device, a metabolic system data acquisition device, and a nervous system data acquisition device.

4. The external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 1, characterized in that, The adaptive information synchronization module includes a preprocessing submodule, a feature extraction submodule, and an adaptive alignment submodule; The preprocessing submodule is used to resample the multi-source detection data using a specific sampling frequency and perform quality assessment to obtain a time-base unified signal matrix and a signal quality assessment matrix. The feature extraction submodule is used to extract features from the time-base unified signal matrix to obtain a feature set, and to perform noise reduction based on the signal quality assessment matrix; The adaptive alignment submodule is used to establish a global alignment benchmark based on the noise-reduced feature set, and to align various types of data based on the global alignment benchmark.

5. The external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 4, characterized in that, The adaptive alignment submodule includes clustering units, local alignment units, and global alignment units; The clustering unit is used to extract features from each data point and perform fusion clustering. The local alignment unit is used to traverse the clustering categories and perform intra-class time registration by filtering high-quality time periods in the quality assessment matrix through a threshold, so as to obtain the local benchmark for each category. The global alignment unit is used to perform correlation analysis on the local benchmarks of each category, generate the global alignment benchmark, and perform alignment.

6. The external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 1, characterized in that, The AI-assisted module includes an anomaly monitoring submodule and a graded alarm submodule; the anomaly monitoring submodule is used to analyze the synergistic abnormalities of cardiopulmonary metabolic parameters through multimodal data fusion; the graded alarm submodule is used to issue alarms based on different anomaly types.

7. The external counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 6, characterized in that, The anomaly supervision submodule includes a multi-signal coupling modeling unit and a hierarchical anomaly detection unit; The multi-signal coupling modeling unit constructs a dynamic coupling matrix based on multi-source detection data and extracts singular values ​​as the main coupling feature vector; The hierarchical anomaly detection unit is used to perform modal stability detection, energy distribution anomaly detection, and spatiotemporal mismatch detection based on the main coupling vector.

8. The extracorporeal counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 7, characterized in that, The counter-pulse assessment module includes a data processing unit, a report generation unit, and a storage unit; The data processing unit is used to perform statistical analysis on the anomaly monitoring results and extract key evaluation indicators; The report generation unit automatically generates an external counterpulsation assessment report based on key evaluation indicators and a preset template; the storage unit is used to classify and store the assessment reports.

9. The extracorporeal counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance as described in claim 8, characterized in that, The counterpulsation assessment module also includes a visualization unit, which synchronously displays waveforms and numerical change lists of key assessment indicators based on a time axis. It distinguishes between normal and abnormal data ranges through color markings and supports user-defined filtering of different detection time periods and different types of data for comparative analysis, in order to present the effects of external counterpulsation treatment and changes in the patient's cardiopulmonary metabolic status.

10. The extracorporeal counterpulsation assessment system based on multi-source dynamic information of cardiopulmonary metabolism and AI assistance according to claim 9, characterized in that, The counterpulsation assessment module has a built-in assessment parameter database, which pre-stores external counterpulsation assessment standard parameters corresponding to different ages, genders, and various common underlying diseases. The counterpulsation assessment module can match and analyze the patient's actual test data, basic information, and current abnormal monitoring results against the assessment parameter database, and automatically generate adjustment suggestions for external counterpulsation treatment for different patients, including adjustment plans for treatment intensity and treatment duration, as well as subsequent rehabilitation precautions.

Images