Remote medical online real-time biomedical examination system
By adopting a cloud-edge layered architecture in the telemedicine system, and utilizing edge computing nodes for real-time event detection and cloud-based long-term trend analysis, the system addresses the latency and insufficient diagnostic capabilities of existing systems in time-sensitive clinical scenarios. This achieves a balance between immediate response and long-term prediction, enhancing the system's real-time diagnostic capabilities and data privacy protection.
Patent Information
- Application Number
- CN202511062531.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-31
- Publication Date
- 2025-11-25
AI Technical Summary
Existing remote patient monitoring systems suffer from significant delays and limited real-time diagnostic capabilities in time-sensitive clinical scenarios, failing to meet immediate emergency needs and effectively perform long-term trend analysis.
It adopts a cloud-edge layered architecture, deploys lightweight machine learning models on edge computing nodes for real-time event detection, and trains complex models on remote cloud servers for long-term trend analysis. It combines knowledge distillation and federated learning to optimize the model, thereby achieving data diversion and privacy protection.
It achieves a balance between immediate response and long-term analysis, reduces network latency, improves the ability to detect critical events in real time, and protects personal privacy while improving the accuracy of disease progression prediction.
Smart Images
Figure CN121011372A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of remote health monitoring technology, specifically to a remote medical online real-time biomedical testing system. Background Technology
[0002] The development of existing remote patient monitoring (RPM) systems began with simple telephone follow-ups and gradually evolved into more sophisticated telemetry technologies, such as Holter monitors and cyclic event recorders, which store data for later offline analysis. While these foundational systems were significant at the time, they lacked real-time intervention capabilities.
[0003] With the rise of the internet and cloud computing technologies, centralized RPM systems have been developed. In this architecture, data from sensors worn by patients is directly transmitted to remote cloud servers for storage and processing. This model provides scalability and centralized data access, enabling clinicians to view patient data at any time. The core of such systems lies in wirelessly transmitting physiological signals such as electrocardiograms (ECGs) to remote servers or smartphones.
[0004] While cloud-based RPM systems play a significant role currently, they still have some limitations in time-sensitive clinical scenarios, including:
[0005] Significant latency: In a pure cloud architecture, data is transmitted over wide area networks (such as cellular networks and the internet), resulting in substantial latency. In critical application scenarios requiring immediate intervention, such as detecting life-threatening arrhythmias, this latency is unacceptable.
[0006] Limited real-time diagnostic capabilities: Due to latency, alerts are often retrospective rather than predictive or immediate. Existing systems may send low-resolution or downsampled data to the cloud or rely on patient-triggered event tagging, both of which can miss transient, asymptomatic, but crucial events.
[0007] Current solutions generally treat all data analysis as a single task and execute it in the cloud. However, clinical needs are actually twofold: on the one hand, there is the immediate and critical need for sub-second detection of acute events; on the other hand, there is the need to predict disease progression by analyzing long-term trends. Pure cloud systems cannot meet the first need. Although edge computing has been proposed to address latency issues, simply migrating cloud logic to the edge cannot meet the second need, as the latter requires massive data aggregation and complex modeling. Summary of the Invention
[0008] To overcome the limitations of the prior art, the present invention provides a remote medical online real-time biomedical testing system and a corresponding method.
[0009] A remote medical online real-time biomedical testing system includes a data collection module, a preprocessing module, an edge processing module, an analysis module, and a visualization module;
[0010] The data collection module includes one or more biomedical sensors for collecting patients' physiological data in real time.
[0011] The preprocessing module preprocesses the collected physiological data to support subsequent analysis.
[0012] The edge processing module includes an edge computing node that is communicatively connected to the biomedical sensor. The edge computing node stores a first machine learning model, which is configured to perform real-time event classification on physiological data from the biomedical sensor.
[0013] The analysis module includes a remote cloud server that is communicatively connected to the edge computing node. The cloud server is configured to train a second machine learning model based on data selectively received from the edge computing node, and to generate an updated version of the first machine learning model based on the second machine learning model for deployment to the edge computing node.
[0014] The visualization module includes a web interface for displaying the analysis results and long-term trend information of the analysis module to medical professionals.
[0015] Preferably, the first machine learning model is a quantized convolutional neural network optimized for efficient execution on resource-constrained edge computing nodes.
[0016] Preferably, the second machine learning model is a hybrid CNN-Long Short-Term Memory network model, which is trained to model long-term time dependencies for disease progression or risk prediction.
[0017] Preferably, the preprocessing module preprocesses the collected physiological data, including: applying a digital bandpass filter to eliminate noise and using a heartbeat detection algorithm to segment the data stream into discrete windows.
[0018] Preferably, the step of generating an updated version of the first machine learning model based on the second machine learning model specifically employs a knowledge distillation method, wherein the second machine learning model serves as the teacher model and the updated first machine learning model serves as the student model.
[0019] Preferably, the training of the second machine learning model based on data selectively received from the edge computing node specifically employs a federated learning method, wherein model updates, rather than the original patient data, are transmitted to the cloud server for aggregation.
[0020] Preferably, the physiological data is an electrocardiogram signal, the first machine learning model is configured for real-time arrhythmia classification, and the second machine learning model is configured for predicting the risk of long-term adverse cardiac events.
[0021] Preferably, the edge computing node is configured to generate an alarm message when a critical event is detected, and to display the alarm message to medical professionals through a visualization module.
[0022] A remote medical online real-time biomedical testing method includes the following steps:
[0023] At edge computing nodes, physiological data streams are acquired from biomedical sensors;
[0024] Using a first machine learning model deployed on the edge computing node, the data stream is processed in real time to generate a first inference result for immediate event detection;
[0025] Based on the first inference result, an instant alert is generated on the edge computing node;
[0026] Based on the first reasoning result, data related to the physiological data stream is selectively transmitted to a remote cloud server;
[0027] On the cloud server, a second machine learning model is trained using transmission data aggregated from one or more edge nodes for long-term trend analysis;
[0028] An updated version of the first machine learning model is generated based on the second machine learning model, and the updated first machine learning model is deployed back to the edge computing node.
[0029] Preferably, the selective data transmission specifically involves determining the data to be transmitted based on the confidence score of the first inference result, and transmitting only data segments related to low-confidence inference results or predefined anomaly patterns.
[0030] Compared with the prior art, the advantages of this invention are:
[0031] This solution employs a cloud-edge layered architecture, which simultaneously optimizes both immediate response and long-term analysis needs. At the edge, by deploying a lightweight first machine learning model, the system can perform real-time, localized analysis of critical events such as arrhythmias, buying time for clinical intervention. In the cloud, leveraging powerful computing capabilities and aggregated data, a more complex second machine learning model is trained to identify long-term, subtle patterns indicating disease progression, thereby transforming the system from a passive alerting tool into a proactive health management platform with both immediate response and forward-looking predictive capabilities.
[0032] Furthermore, this solution offloads data. Edge nodes only upload data when necessary (e.g., when anomalies are detected or model confidence is low), and most of the raw data is discarded after local processing, greatly reducing reliance on network bandwidth. Because only feature vectors or a small segment of relevant data are transmitted, rather than the complete raw data stream, personal privacy is protected to a great extent. Attached Figure Description
[0033] Figure 1 This is a system architecture diagram of a remote medical online real-time biomedical testing system proposed in this invention. Detailed Implementation
[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0035] Example 1:
[0036] Reference Figure 1 This invention provides a remote medical online real-time biomedical testing system. The system includes a data collection module, a preprocessing module, an edge processing module, an analysis module, and a visualization module.
[0037] The data collection module includes one or more medical-grade sensors for acquiring physiological data. These sensors may include, but are not limited to, multi-lead electrocardiogram (ECG) sensors, photoplethysmography (PPG) sensors for measuring blood oxygen saturation (SpO2) and heart rate variability (HRV), body temperature sensors, accelerometers for fall detection, and blood pressure monitors. These sensors can be integrated into a single wearable device, such as a patch, smartwatch, or chest strap, or can communicate with edge nodes as standalone devices. The sensors are configured for continuous or high-frequency data acquisition (e.g., ECG sampling frequency of 100-500 Hz) to capture high-fidelity signals.
[0038] The preprocessing module preprocesses the collected physiological data to support subsequent analysis.
[0039] The edge processing module includes an edge computing node that is communicatively connected to the biomedical sensor. The edge computing node stores a first machine learning model, which is configured to perform real-time event classification on physiological data from the biomedical sensor.
[0040] The analysis module includes a remote cloud server that is communicatively connected to the edge computing node. The cloud server is configured to train a second machine learning model based on data selectively received from the edge computing node, and to generate an updated version of the first machine learning model based on the second machine learning model for deployment to the edge computing node.
[0041] The visualization module includes a web interface for displaying the analysis results and long-term trend information of the analysis module to medical professionals.
[0042] Example 2:
[0043] This embodiment provides a remote medical online real-time biomedical testing method, specifically including:
[0044] Real-time edge processing:
[0045] Signal acquisition: Edge nodes continuously receive raw physiological data streams from sensor modules.
[0046] Signal preprocessing: Real-time processing of the raw signal to remove noise and extract relevant components.
[0047] For ECG signals, this includes:
[0048] Noise cancellation: Apply digital filters, such as bandpass filters (e.g., 5-15Hz), to eliminate power line interference, baseline drift, and muscle artifacts.
[0049] Heartbeat segmentation: Using algorithms such as Hamiltonian mean algorithm to identify individual heartbeats (e.g. QRS complexes), continuous signals are segmented into analyzable windows.
[0050] Edge machine learning inference: Input the preprocessed data segment into the first machine learning model.
[0051] Model Type: The first machine learning model is a lightweight, computationally efficient model. A preferred embodiment is a lightweight convolutional neural network (CNN), such as the "LiteNet" architecture or a model based on the inverted residual structure of MobileNetV2. These models excel at pattern recognition of signal data.
[0052] Quantization: In order to run on resource-constrained edge nodes, the first machine learning model is quantized after training (e.g., converting 32-bit floating-point weights to 8-bit integers) using tools such as TensorFlow Lite, which greatly reduces model size and computational cost while maintaining high accuracy.
[0053] Inference task: The first machine learning model performs real-time classification tasks, such as identifying arrhythmias (e.g., normal sinus rhythm, atrial fibrillation, premature ventricular contractions) from ECG segments.
[0054] Intelligent Alerts and Data Offloading: Based on the inference results of the first machine learning model:
[0055] If a high-confidence critical event is detected (e.g., the probability of detecting atrial fibrillation > 95%), the edge node immediately generates a local alarm (such as an audible alarm or haptic feedback) and transmits a high-priority alarm data packet to the cloud computing platform.
[0056] If no critical event is detected, or the model output confidence is low, the edge node enters the selective data transmission step.
[0057] Selective data transmission: To save bandwidth and protect privacy, edge nodes do not stream all raw data, but selectively transmit it.
[0058] Feature vectors extracted during the preprocessing stage.
[0059] The original data segment corresponding to low-confidence predictions or newly observed, unclassifiable patterns.
[0060] Periodic, downsampled summary data used for baseline tracking.
[0061] It should be noted that this invention employs two different and complementary model architectures: a lightweight model at the edge and a complex hybrid model in the cloud. The lightweight model at the edge provides real-time classification, while the complex hybrid model in the cloud provides long-term prognosis and disease progression modeling. Data transmitted from edge nodes is used by the cloud for long-term prognosis and disease progression modeling.
[0062] Cloud-based model training:
[0063] The cloud computing platform will aggregate data selectively transmitted from thousands of patients into a large, diverse dataset.
[0064] This dataset is used to train a second machine learning model. A preferred embodiment of this second machine learning model is a hybrid CNN-LSTM model or a deep RNN with LSTM / GRU units. CNN layers can extract spatial features from signal fragments, while LSTM layers can model temporal dependencies between events over long periods (days or weeks), which is crucial for modeling disease progression.
[0065] The second machine learning model is trained to perform tasks that edge nodes cannot accomplish, such as predicting the risk of readmission for heart failure at 90 days or modeling the progression of Alzheimer's disease based on biomarker data.
[0066] Dynamic model synchronization and optimization:
[0067] The first machine learning model is improved by leveraging the knowledge of a more powerful second machine learning model. This can be achieved through various techniques; in this embodiment, a knowledge distillation method is used, where the second machine learning model acts as a "teacher" model to train a new, more accurate "student" first machine learning model.
[0068] The newly trained first machine learning model is quantized, compressed, and validated before being securely pushed to the edge node cluster via OTA updates. This ensures that the entire system becomes more accurate as more data is collected.
[0069] Example 3:
[0070] This embodiment provides a data privacy-preserving approach where a second machine learning model is trained using federated learning principles. In this mode, the global second machine learning model is sent to edge nodes instead of sending data to the cloud. Each edge node trains its model locally using its local data. Only the resulting model weight updates (gradients), not the original data, are sent back to the cloud. The cloud server aggregates these updates to create an improved global model, which is then redistributed. This approach enables collaborative model training without the need for centralized storage of sensitive patient data, thus addressing user privacy concerns.
[0071] Example 4:
[0072] This embodiment illustrates the specific application of this solution in the field of remote cardiac monitoring.
[0073] This system was installed on a patient discharged after a cardiac event. Edge nodes continuously analyze their ECG signals.
[0074] Instant Alert: A first-level machine learning model (lightweight CNN) on the edge node detects an asymptomatic atrial fibrillation episode. Within milliseconds, the module triggers a local alert and sends a notification to the cloud. The on-call clinician receives the notification through the visualization module's user interface, sees the event classification and associated ECG waveform, and takes intervention measures (e.g., calling the patient to adjust medication), thus acting before more serious complications such as stroke occur.
[0075] Long-term prognosis: Over several weeks, a second machine learning model (CNN-LSTM) in the cloud analyzed the frequency and duration of arrhythmias, trends in heart rate variability, and other subtle features extracted from selectively uploaded data. The second machine learning model identified a deterioration pattern consistent with an increased risk of readmission for heart failure. The system flagged the patient as requiring follow-up. Based on the analysis, clinicians adjusted the patient's long-term care plan, potentially avoiding a costly and dangerous hospital readmission.
[0076] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0077] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A remote medical online real-time biomedical testing system, characterized in that, It includes a data collection module, a preprocessing module, an edge processing module, an analysis module, and a visualization module; The data collection module includes one or more biomedical sensors for collecting patients' physiological data in real time. The preprocessing module preprocesses the collected physiological data to support subsequent analysis. The edge processing module includes an edge computing node that is communicatively connected to the biomedical sensor. The edge computing node stores a first machine learning model, which is configured to perform real-time event classification on physiological data from the biomedical sensor. The analysis module includes a remote cloud server that is communicatively connected to the edge computing node. The cloud server is configured to train a second machine learning model based on data selectively received from the edge computing node, and to generate an updated version of the first machine learning model based on the second machine learning model for deployment to the edge computing node. The visualization module includes a web interface for displaying the analysis results and long-term trend information of the analysis module to medical professionals.
2. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The first machine learning model is a quantized convolutional neural network optimized for efficient execution on resource-constrained edge computing nodes.
3. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The second machine learning model is a hybrid CNN-Long Short-Term Memory network model, which is trained to model long-term time dependencies for disease progression or risk prediction.
4. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The preprocessing module preprocesses the collected physiological data, including applying a digital bandpass filter to eliminate noise and using a heartbeat detection algorithm to segment the data stream into discrete windows.
5. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The process of generating an updated version of the first machine learning model based on the second machine learning model specifically employs a knowledge distillation method, wherein the second machine learning model serves as the teacher model and the updated first machine learning model serves as the student model.
6. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The second machine learning model is trained based on data selectively received from the edge computing node, specifically using a federated learning method, where model updates, rather than the original patient data, are transmitted to the cloud server for aggregation.
7. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The physiological data is an electrocardiogram signal. The first machine learning model is configured for real-time arrhythmia classification, and the second machine learning model is configured for predicting the risk of long-term adverse cardiac events.
8. The remote medical online real-time biomedical testing system according to claim 1, characterized in that, The edge computing node is configured to generate an alarm message when a critical event is detected, and to display the alarm message to medical professionals through a visualization module.
9. A remote medical online real-time biomedical testing method, characterized in that, Includes the following steps: At edge computing nodes, physiological data streams are acquired from biomedical sensors; Using a first machine learning model deployed on the edge computing node, the data stream is processed in real time to generate a first inference result for immediate event detection; Based on the first inference result, an instant alert is generated on the edge computing node; Based on the first reasoning result, data related to the physiological data stream is selectively transmitted to a remote cloud server; On the cloud server, a second machine learning model is trained using transmission data aggregated from one or more edge nodes for long-term trend analysis; An updated version of the first machine learning model is generated based on the second machine learning model, and the updated first machine learning model is deployed back to the edge computing node.
10. A remote medical online real-time biomedical testing method according to claim 2, characterized in that, The selective data transmission specifically involves determining the data to be transmitted based on the confidence score of the first inference result, and transmitting only data segments related to low-confidence inference results or predefined anomaly patterns.