Real-time anesthesia depth monitoring system based on deep learning
By using a deep learning-based real-time anesthesia monitoring system, which utilizes data collection, processing, and cloud-based module training models, the problems of human error and human resource requirements in traditional systems are solved, achieving highly accurate and real-time updated anesthesia depth monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-02-26
- Publication Date
- 2026-05-08
AI Technical Summary
Traditional real-time anesthesia monitoring systems rely on predefined standards, which can easily lead to human error, and updating the model requires a lot of human resources and cannot be done in real time.
The system employs a deep learning-based approach, utilizing a data collection module, a data processing module, a cloud module, and an anesthesia prediction and response module. It trains a model using deep learning algorithms to monitor and update the depth of anesthesia in real time, thereby reducing human error and lowering the demand for human resources.
It improved the accuracy of anesthesia depth monitoring, reduced human error, enabled real-time model updates, and lowered human resource costs.
Smart Images
Figure CN121987151A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of anesthesia depth monitoring technology, specifically a real-time anesthesia depth monitoring system based on deep learning. Background Technology
[0002] The implementation principle of real-time anesthesia monitoring systems is usually based on experience and statistical analysis to assess the patient's anesthesia depth and status, using predefined algorithms or models and making judgments based on characteristics and indicators. However, traditional real-time anesthesia monitoring systems often have some drawbacks.
[0003] On the one hand, traditional real-time anesthesia monitoring systems typically use predefined criteria to determine the patient's depth of anesthesia and status, which relies on the subjective choices of experts and may lead to human error, affecting the accuracy of the monitoring system;
[0004] On the other hand, traditional real-time anesthesia monitoring systems require a lot of human resources to analyze and calculate when updating predefined standards or models, and cannot update the required algorithm models in real time. Summary of the Invention
[0005] The purpose of this invention is to provide a real-time anesthesia depth monitoring system based on deep learning to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time anesthesia depth monitoring system based on deep learning, comprising a data collection module, a data processing module, a cloud module, and an anesthesia prediction and response module, wherein:
[0007] The data collection module is used to collect real-time physiological data and historical physiological data, and send the real-time physiological data and historical physiological data to the data processing module;
[0008] The data processing module extracts features from real-time physiological data and extracts features and labels from historical physiological data, and sends the processed real-time physiological data and historical physiological data to the anesthesia prediction and response module and the cloud module, respectively.
[0009] The cloud module receives historical physiological data sent by the data processing module, and uses deep learning algorithms to train a model based on the historical physiological data, and sends the trained model to the anesthesia prediction and response module.
[0010] The anesthesia prediction and response module predicts the depth of anesthesia based on the model trained by the cloud module and the real-time physiological data sent by the data processing module. The prediction results and real-time physiological data are then sent to the cloud module as historical physiological data for model updates.
[0011] As a further improvement to this technical solution, the data collection module includes a real-time data collection unit and an analog-to-digital conversion unit. The data collection module uses sensors to collect real-time physiological data generated during patient anesthesia and sends the data to the analog-to-digital conversion unit. The analog-to-digital conversion unit is used to convert the received analog signal data into digital signal data and send the data to the data cleaning unit in the data processing module.
[0012] As a further improvement to this technical solution, the data collection module includes a historical data collection unit, which collects historical physiological data using a hospital database and sends the data to the data cleaning unit in the data processing module.
[0013] As a further improvement to this technical solution, the data processing module includes a data cleaning unit, a feature and label processing unit, and a training and prediction classification unit. The data cleaning unit receives real-time physiological data and historical physiological data sent by the analog-to-digital conversion unit and the historical data collection unit, respectively, and processes missing and duplicate values in the data before sending the data to the feature and label processing unit. The feature and label processing unit extracts features from the real-time physiological data and extracts features and labels from the historical physiological data before sending the processed data to the training and prediction classification unit. The training and prediction classification unit distinguishes between real-time physiological data and historical physiological data, and sends real-time physiological data to the prediction data receiving unit in the anesthesia prediction and response module and historical physiological data to the training data receiving unit in the cloud module, depending on whether the label column is empty.
[0014] As a further improvement to this technical solution, the cloud module includes a training data receiving unit, a model updating unit, and a parameter updating unit. The training data receiving unit receives historical physiological data sent by the training and prediction classification unit, performs data standardization processing, and sends the processed data to the model updating unit. The model updating unit trains the model using a deep learning algorithm based on the historical physiological data sent by the training data receiving unit, and sends the trained model and parameters to the parameter updating unit. The parameter updating unit sends the trained model and parameters to the prediction unit in the anesthesia prediction and response module.
[0015] As a further improvement to this technical solution, the anesthesia prediction and response module includes a prediction data receiving unit, a prediction unit, and a response unit. The prediction data receiving unit receives real-time physiological data sent by the training and prediction classification unit, performs data standardization processing on the data, and sends the processed data to the prediction unit. The prediction unit predicts the depth of anesthesia based on the model sent by the parameter update unit using the real-time physiological data sent by the prediction data receiving unit, sends the prediction result and real-time physiological data to the training data receiving unit, and sends the prediction result to the response unit. The response unit visualizes the prediction result sent by the prediction unit.
[0016] As a further improvement to this technical solution, the training data receiving unit receives the prediction results and real-time physiological data sent by the prediction unit, and uses the prediction results and real-time physiological data as historical physiological data, and sends the data to the model update unit for model update.
[0017] As a further improvement to this technical solution, the training data receiving unit converts the labeled data in the historical physiological data into numerical data and converts the label column values into corresponding integer sizes.
[0018] As a further improvement to this technical solution, the model update unit utilizes a deep learning algorithm to train the model based on historical physiological data, specifically including:
[0019] Forward propagation: The neural network passes input data from the input layer to the output layer;
[0020] Loss calculation: After forward propagation, the prediction results obtained by the neural network are compared with the corresponding label values to calculate the value of a loss function;
[0021] Backpropagation: The error is propagated back into the network using the loss function. The contribution of each parameter to the loss is calculated. The gradient is calculated from the output layer to the input layer using the chain rule. The value of each parameter is updated according to the direction of the gradient. Finally, the parameters in the neural network are updated according to the calculated gradient information.
[0022] Compared with the prior art, the beneficial effects of the present invention are:
[0023] 1. This deep learning-based real-time anesthesia depth monitoring system first collects historical physiological data from a hospital database, then uses a deep learning algorithm to train a model based on the historical physiological data, and then uses sensors to collect real-time physiological data from patients. Based on the trained model, the system predicts the anesthesia depth from the real-time physiological data, thereby reducing the probability of errors caused by manually defined rules. Furthermore, the deep learning algorithm model trained using historical physiological data from the hospital database can increase the accuracy of the detection system.
[0024] 2. This deep learning-based real-time anesthesia depth monitoring system uses a trained model to predict anesthesia from real-time physiological data. After the prediction results and real-time physiological data are sent to the cloud module, the cloud module uses the received prediction results and real-time physiological data as historical physiological data. When the amount of data required for training is met, the model is updated. This real-time update of the algorithm model avoids wasting a lot of human resources and reduces the cost of manually maintaining the rules. Attached Figure Description
[0025] Figure 1 This is a schematic diagram of the overall modules of the present invention;
[0026] Figure 2 This is a schematic diagram of the data collection module unit of the present invention;
[0027] Figure 3 This is a schematic diagram of the data processing module unit of the present invention;
[0028] Figure 4 This is a schematic diagram of the cloud module unit of the present invention;
[0029] Figure 5 This is a schematic diagram of the anesthesia prediction and response module unit of the present invention;
[0030] In the diagram: 100, Data Collection Module; 101, Real-time Data Collection Unit; 102, Analog-to-Digital Conversion Unit; 103, Historical Data Collection Unit; 200, Data Processing Module; 201, Data Cleaning Unit; 202, Feature and Label Processing Unit; 203, Training and Prediction Classification Unit; 300, Cloud Module; 301, Training Data Receiving Unit; 302, Model Update Unit; 303, Parameter Update Unit; 400, Anesthesia Prediction and Response Module; 401, Prediction Data Receiving Unit; 402, Prediction Unit; 403, Response Unit. Detailed Implementation
[0031] 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.
[0032] Please see Figure 1-5 The present invention provides a technical solution: a real-time anesthesia depth monitoring system based on deep learning, including a data collection module 100, a data processing module 200, a cloud module 300, and an anesthesia prediction and response module 400.
[0033] To collect the data required for analysis during patient anesthesia, the real-time data collection unit 101 in the data collection module 100 uses sensors to acquire the patient's real-time physiological data. These sensors include an electrocardiogram (ECG) sensor (for measuring ECG signals), an electroencephalogram (EEG) sensor (for measuring EEG signals), an electromyogram (EMG) sensor (for measuring EMG signals), and a respiration sensor (for measuring respiration signals). The sensors communicate with the system wirelessly, using the HTTP protocol for data transmission. The sensors then send data to the analog-to-digital conversion unit 102 via the HTTP protocol.
[0034] The analog-to-digital conversion unit 102 uses an analog-to-digital converter to convert the received analog signal into a digital signal for application to the model, and sends the converted data to the data cleaning unit 201 in the data processing module 200.
[0035] To ensure data quality and usability, the data cleaning unit 201 processes missing and duplicate values in the dataset sent by the analog-to-digital conversion unit 102, specifically including:
[0036] Missing values: The data cleaning unit 201 handles missing values by filling in the missing values, calculating the average value of the feature column where the missing value is located, and filling the missing value into the missing value.
[0037] Duplicate values: Directly delete rows containing duplicate data to ensure the uniqueness and independence of each data sample.
[0038] To enable the model to make predictions based on data features, the data cleaning unit 201 sends the cleaned data to the feature and label processing unit 202. The feature and label processing unit 202 extracts data features based on the received signal data. For example, if the signal data includes electrocardiogram (ECG), electroencephalogram (EEG), electromyogram (EMG), and respiratory signals, the extracted features would be as follows:
[0039] Heart rate: Calculated by analyzing the position of the R peaks in the signal, the time interval between two adjacent R peaks (i.e., the RR interval), and then the number of R peaks per minute is calculated and divided by the total measurement time.
[0040] RR interval: The time interval between two adjacent R peaks is calculated based on the position of the R peaks in the signal;
[0041] Peak frequency of EEG signal: Perform Fourier transform on the EEG signal to obtain the energy distribution of the signal at different frequencies, calculate the power spectral density of the signal at each frequency, and find the strongest frequency peak by analyzing the spectrum;
[0042] Muscle activity amplitude: The average value obtained by squaring the amplitude of each sampling point of the electromyographic signal and taking the square root;
[0043] Muscle contraction time: calculated by the rise and fall times of the electromyographic signal;
[0044] Respiratory rate: calculated based on the time interval between peak respiratory signals, converted to a frequency per minute;
[0045] Respiratory amplitude: For each respiratory cycle, based on the rise and fall of the signal, find the maximum value of the inspiratory phase and the minimum value of the expiratory phase, and calculate the difference between them.
[0046] The feature and label processing unit 202 sends the extracted data features to the training and prediction classification unit 203, and the training and prediction classification unit 203 sends the patient's real-time physiological data to the prediction data receiving unit 401 in the anesthesia prediction and response module 400 to predict the depth of anesthesia.
[0047] Before making a prediction, an algorithm model needs to be trained. The algorithm model needs to analyze historical physiological data. Therefore, the historical data collection unit 103 in the data collection module 100 collects historical physiological data using the hospital database and sends the historical physiological data to the data cleaning unit 201 for cleaning. Then, the cleaned data is sent to the feature and label processing unit 202 for processing.
[0048] The feature and label processing unit 202 performs feature extraction and label extraction operations on historical physiological data. Feature extraction is the same as for real-time physiological data, while label extraction extracts the anesthesia depth from the data as the label column, where the anesthesia depth is categorized as awake, mild anesthesia, moderate anesthesia, and deep anesthesia. The feature and label processing unit 202 then sends the processed historical physiological data to the training and prediction classification unit 203.
[0049] The training and prediction classification unit 203 distinguishes between historical physiological data and real-time physiological data based on whether the label column is empty, and sends the historical physiological data and real-time physiological data to the prediction data receiving unit 401 in the anesthesia prediction and response module 400 and the training data receiving unit 301 in the cloud module 300, respectively.
[0050] To reduce the impact of outliers and noise on the model, improve its robustness to outliers, and minimize interference with model training and prediction, thereby enhancing model performance and stability, the training data receiving unit 301 standardizes historical physiological data. This is achieved by calculating the mean and standard deviation of the data's feature columns, subtracting the mean from the original values, and then dividing by the standard deviation, transforming the data into a distribution with a mean of 0 and a standard deviation of 1. This feature weighting maintains balance. The training data receiving unit 301 also converts the data's label column values into corresponding integers, where the integer size represents the number of anesthesia depth types. For example, 0 represents awakeness; 1 represents mild anesthesia; 2 represents moderate anesthesia; and 3 represents deep anesthesia.
[0051] The training data receiving unit 301 sends the processed data to the model update unit 302 for model training. The training process is as follows:
[0052] First, the data is divided into training set and validation set. The training set is sent to the model. The model consists of an input layer, several hidden layers and an output layer. The input layer takes the features as input, connects them to the hidden layers, and finally connects them to the output layer.
[0053] Forward propagation: The neural network passes input data from the input layer to the output layer, where the output is calculated by the activation function and the weights between the layers;
[0054] Loss calculation: After forward propagation, the prediction results obtained by the neural network are compared with the corresponding label values, and a loss function (also known as error) is calculated. The loss function measures the difference between the prediction results and the actual labels.
[0055] Backpropagation: The error is propagated back into the network using the loss function. The contribution of each parameter to the loss is calculated. The gradient is calculated backward from the output layer to the input layer using the chain rule. The value of each parameter is updated according to the direction of the gradient to minimize the loss function. Finally, the parameters in the neural network are updated according to the calculated gradient information.
[0056] The model update unit 302 sends the updated parameters to the parameter update unit 303. If the model is being trained for the first time, the parameter update unit 303 sends the complete model to the prediction unit 402 in the anesthesia prediction and response module 400; otherwise, it only sends the updated parameters.
[0057] The prediction data receiving unit 401 in the anesthesia prediction and response module 400 first standardizes the real-time physiological data and sends the processed data to the prediction unit 402 for prediction. The prediction unit 402 predicts the depth of anesthesia based on the model sent by the model update unit 302, and sends the prediction results and real-time physiological data to the training data receiving unit 301 in the cloud module 300 to increase the training dataset. Finally, the prediction results are sent to the response unit 403.
[0058] The response unit 403 converts the predicted results into the corresponding anesthesia depth type and outputs the anesthesia depth type visually, allowing doctors to understand the patient's anesthesia status in real time.
[0059] The training data receiving unit 301 receives the prediction results and real-time physiological data sent by the prediction unit 402, and uses the prediction results and real-time physiological data as model update data (historical physiological data). When the received dataset is greater than the initial training dataset, it means that the latest dataset has reached the required number of training sets, and then the training set is sent to the model update unit 302 for model update.
[0060] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A real-time anesthesia depth monitoring system based on deep learning, characterized in that: It includes a data collection module (100), a data processing module (200), a cloud module (300), and an anesthesia prediction and response module (400), wherein: The data collection module (100) is used to collect real-time physiological data and historical physiological data, and send the real-time physiological data and historical physiological data to the data processing module (200). The data processing module (200) extracts features from real-time physiological data and extracts features and labels from historical physiological data, and sends the processed real-time physiological data and historical physiological data to the anesthesia prediction and response module (400) and the cloud module (300), respectively. The cloud module (300) receives historical physiological data sent by the data processing module (200), and uses deep learning algorithms to train a model based on the historical physiological data, and sends the trained model to the anesthesia prediction and response module (400). The anesthesia prediction and response module (400) predicts the depth of anesthesia based on the model trained by the cloud module (300) and the real-time physiological data sent by the data processing module (200). The prediction results and real-time physiological data are then sent to the cloud module (300) as historical physiological data for model updates.
2. The real-time anesthesia depth monitoring system based on deep learning according to claim 1, characterized in that: The data collection module (100) includes a real-time data collection unit (101) and an analog-to-digital conversion unit (102). The data collection module (100) uses sensors to collect real-time physiological data generated during patient anesthesia and sends the data to the analog-to-digital conversion unit (102). The analog-to-digital conversion unit (102) is used to convert the received analog signal data into digital signal data and send the data to the data cleaning unit (201) in the data processing module (200).
3. The real-time anesthesia depth monitoring system based on deep learning according to claim 1, characterized in that: The data collection module (100) includes a historical data collection unit (103), which collects historical physiological data using a hospital database and sends the data to the data cleaning unit (201) in the data processing module (200).
4. The real-time anesthesia depth monitoring system based on deep learning according to claim 1, characterized in that: The data processing module (200) includes a data cleaning unit (201), a feature and label processing unit (202), and a training and prediction classification unit (203). The data cleaning unit (201) receives real-time physiological data and historical physiological data sent by the analog-to-digital conversion unit (102) and the historical data collection unit (103), respectively, and processes missing and duplicate values in the data, and sends the data to the feature and label processing unit (202). The feature and label processing unit (202) is used to extract features from real-time physiological data, extract features and labels from historical physiological data, and send the processed data to the training and prediction classification unit (203). The training and prediction classification unit (203) is used to distinguish between real-time physiological data and historical physiological data, and sends real-time physiological data to the prediction data receiving unit (401) in the anesthesia prediction and response module (400) and sends historical physiological data to the training data receiving unit (301) in the cloud module (300) according to whether the label column is empty.
5. The real-time anesthesia depth monitoring system based on deep learning according to claim 1, characterized in that: The cloud module (300) includes a training data receiving unit (301), a model update unit (302), and a parameter update unit (303). The training data receiving unit (301) receives historical physiological data sent by the training and prediction classification unit (203), performs data standardization processing, and sends the processed data to the model update unit (302). The model update unit (302) trains the model using a deep learning algorithm based on the historical physiological data sent by the training data receiving unit (301), and sends the trained model and parameters to the parameter update unit (303). The parameter update unit (303) sends the trained model and parameters to the prediction unit (402) in the anesthesia prediction and response module (400).
6. The real-time anesthesia depth monitoring system based on deep learning according to claim 1, characterized in that: The anesthesia prediction and response module (400) includes a prediction data receiving unit (401), a prediction unit (402), and a response unit (403). The prediction data receiving unit (401) receives real-time physiological data sent by the training and prediction classification unit (203), performs data standardization processing, and sends the processed data to the prediction unit (402). The prediction unit (402) predicts the depth of anesthesia based on the model sent by the parameter update unit (303) and sends the prediction results and real-time physiological data to the training data receiving unit (301) and the prediction results to the response unit (403). The response unit (403) visualizes the prediction results sent by the prediction unit (402).
7. The real-time anesthesia depth monitoring system based on deep learning according to claim 5, characterized in that: The training data receiving unit (301) receives the prediction results and real-time physiological data sent by the prediction unit (402), and uses the prediction results and real-time physiological data as historical physiological data, and sends the data to the model update unit (302) for model update.
8. The real-time anesthesia depth monitoring system based on deep learning according to claim 4, characterized in that: The training data receiving unit (301) converts the label data in the historical physiological data into numerical data and converts the label column values into corresponding integer sizes.
9. The real-time anesthesia depth monitoring system based on deep learning according to claim 5, characterized in that: The model update unit (302) uses a deep learning algorithm to train the model based on historical physiological data, specifically including: Forward propagation: The neural network passes input data from the input layer to the output layer; Loss calculation: After forward propagation, the prediction results obtained by the neural network are compared with the corresponding label values to calculate the value of a loss function; Backpropagation: The error is propagated back into the network using the loss function. The contribution of each parameter to the loss is calculated. The gradient is calculated from the output layer to the input layer using the chain rule. The value of each parameter is updated according to the direction of the gradient. Finally, the parameters in the neural network are updated according to the calculated gradient information.