A non-invasive blood pressure prediction method, device, and real-time monitoring system based on convolutional neural networks.
By using a convolutional neural network-based method to preprocess and predict blood pressure using fingertip PPG signals, the accuracy and portability issues of non-invasive real-time blood pressure prediction are solved, achieving efficient and accurate blood pressure prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANHUI UNIV
- Filing Date
- 2022-10-24
- Publication Date
- 2026-04-17
AI Technical Summary
Existing technologies cannot achieve high-precision, portable, non-invasive real-time blood pressure prediction, and the prediction accuracy is low.
A convolutional neural network-based approach is adopted to collect fingertip PPG signals, preprocess them, and then input them into the backbone network and the attention mechanism information distillation network for blood pressure prediction. This includes data segmentation, bandpass filtering, and data screening, using a dual-channel encoder and an improved attention mechanism information distillation module.
It achieves high-precision and rapid non-invasive blood pressure prediction, improving prediction performance and portability.
Smart Images
Figure CN115470716B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of bioinformatics and Internet of Things (IoT) technology, specifically to a non-invasive blood pressure prediction method, device, and real-time monitoring system based on convolutional neural networks. Background Technology
[0002] Cardiovascular disease (CVD) has become the leading cause of death worldwide. In 2012, CVD caused 17.5 million deaths, making it the leading cause of death from non-communicable diseases. Blood pressure (BP), as a crucial parameter reflecting human health, is also a fundamental hemodynamic parameter for the prevention and treatment of cardiovascular diseases. Therefore, regular blood pressure monitoring is essential for preventing hypertension and other cardiovascular diseases. While individuals at risk of hypertension need regular blood pressure measurements, traditional cuff blood pressure monitors require experienced personnel for accurate measurement and are inconvenient to carry. With the rapid development of deep learning, non-invasive blood pressure prediction methods based on convolutional neural networks have attracted widespread attention. These methods primarily use neural networks to directly predict blood pressure waveforms based on PPG signal data, and then calculate systolic and diastolic blood pressure using the systolic and diastolic peaks of the blood pressure waveform.
[0003] PPG (Photoplethysmography) signals are waveform signals obtained through photoplethysmography. The main principle is that a PPG sensor emits green light through a light-emitting diode (LED) that penetrates skin tissue and arteries / veins. A portion of this light is absorbed by the tissue (muscle, skin, etc.), while the remainder is reflected back to the photodetector. Because the blood flow velocity varies in arteries, the absorption and reflection of light will also differ.
[0004] Therefore, due to changes in the absorption of light by the blood flowing in the arteries, the light intensity received by the photodetector also changes accordingly. When we convert the light signal into an electrical signal, we obtain the photoplethysmogram (PPG). A person's blood pressure rises because the heart muscle contracts during systole, pushing blood to the periphery of the body. This propagation of the PPG wave causes blood pressure to rise. Similarly, blood pressure drops when the heart relaxes during diastole to fill with blood. This results in a periodic signal with a clear and defined systolic peak (the first flow out of the heart during systole) and a subtle diastolic peak (blood returning to the heart from the periphery during diastole). Each periodic repetition contains one systolic peak and one diastolic peak. Systolic and diastolic blood pressure can be calculated from the systolic and diastolic peaks.
[0005] Existing technologies cannot achieve real-time prediction and analysis of blood pressure, and their accuracy is low. Therefore, there is an urgent need to propose a non-invasive blood pressure prediction solution with advantages such as automatic prediction, high prediction accuracy, low cost, and portability. Summary of the Invention
[0006] In view of this, the purpose of this invention is to improve the accuracy and speed of blood pressure prediction while enhancing prediction performance, thereby achieving efficient and accurate blood pressure prediction.
[0007] A first aspect of the present invention provides a non-invasive blood pressure prediction method based on a convolutional neural network, the method comprising:
[0008] Collect PPG signals from the user's fingertips;
[0009] The PPG signal is preprocessed to obtain a first processed signal;
[0010] The first processed signal is input into a blood pressure prediction model based on a convolutional neural network to predict and generate a user's blood pressure signal; the blood pressure prediction model includes a backbone network and an information distillation network based on an attention mechanism.
[0011] Further, the preprocessing of the PPG signal to obtain a first processed signal includes:
[0012] The preprocessing includes data segmentation, bandpass filtering, and data filtering operations.
[0013] The data segmentation operation includes segmenting the acquired continuous PPG signal into segments of a predetermined length.
[0014] The bandpass filtering operation includes filtering low-frequency and high-frequency noise from the segmented data using bandpass filtering;
[0015] The data filtering operation includes removing PPG signals that are damaged by noise by calculating the autocorrelation of the data.
[0016] Furthermore, the backbone network includes a dual-channel encoded Unet network; the Unet network employs a dual-channel encoder, that is, encoding is performed again after the first encoder of each layer, so that even shallow layers can have deep semantic information; the dual encoders of each layer are concatenated together using an append operation;
[0017] In this process, the first encoder of each layer receives data from the first encoder of the previous layer, and the second encoder of each layer receives data from the second encoder of the previous layer and the first encoder of the current layer. The two encoders of each layer are then fed into the decoder of the current layer through an add operation.
[0018] Furthermore, the encoder includes an information distillation network based on an attention mechanism;
[0019] The information distillation network based on the attention mechanism includes:
[0020] The input data is sequentially processed through one-dimensional convolution operations with kernel sizes of 3, 5, and 7, and the semantic information at a deeper level is extracted through information distillation.
[0021] The data from the three outputs are concatenated in parallel on the channels through a concatenation operation, so that the output contains semantic information at different scales. At the same time, global average pooling is applied in parallel to the original input to compress each channel. After that, the data passes through two fully connected layers, which triples the original number of channels. Finally, the output is multiplied by the information distillation output to obtain the final result.
[0022] Furthermore, before collecting the PPG signal from the user's fingertip, the method also includes:
[0023] Pre-train a blood pressure prediction model based on a convolutional neural network;
[0024] The pre-trained blood pressure prediction model based on a convolutional neural network includes:
[0025] Step 1: Obtain a dataset consisting of PPG signals and corresponding ABP signals, and perform data sampling and segmentation on the signals in the dataset;
[0026] Step 2: Denoise the PPG and ABP signal data in the dataset;
[0027] Step 3: Filter the denoised PPG and ABP signal data.
[0028] Step 4: Divide the selected PG signal data and ABP signal data into training set, validation set and test set according to a predetermined ratio;
[0029] Step 5: Train the blood pressure prediction model based on the training set data into a convolutional neural network to generate the optimal prediction model.
[0030] Furthermore, in a second aspect, the present invention also proposes a non-invasive blood pressure prediction device based on a convolutional neural network, the device comprising an acquisition module, a preprocessing module, and a blood pressure prediction module; wherein:
[0031] The acquisition module is used to acquire PPG signals from the user's fingertips;
[0032] A preprocessing module is used to perform preprocessing on the PPG signal to obtain a first processed signal;
[0033] The blood pressure prediction module is used to input the first processed signal into a blood pressure prediction model based on a convolutional neural network to predict and generate a user's blood pressure signal; the blood pressure prediction model includes a backbone network and an information distillation network based on an attention mechanism.
[0034] Furthermore, the preprocessing module includes a data segmentation module, a bandpass filtering submodule, and a data filtering submodule; wherein:
[0035] The data segmentation module is used to segment the acquired continuous PPG signal into segments of a predetermined length;
[0036] The bandpass filter submodule is used to filter low-frequency and high-frequency noise in the segmented data by bandpass filtering;
[0037] The data filtering submodule is used to remove PPG signals that are damaged by noise by calculating the autocorrelation of the data.
[0038] Furthermore, in a third aspect, the present invention also proposes a non-invasive real-time blood pressure monitoring system based on a convolutional neural network, wherein the non-invasive blood pressure prediction device based on the convolutional neural network acquires PPG signals in real time and predicts and generates user blood pressure signals in real time; and a platform subsystem.
[0039] After the non-invasive blood pressure prediction device based on convolutional neural network predicts and generates the user's blood pressure signal, it uploads it to the platform subsystem; the platform subsystem includes a database module, a blood pressure analysis module, and a message module;
[0040] The database module is used to store uploaded blood pressure signals based on the MQTT server;
[0041] The blood pressure analysis module is used to calculate and obtain the user's blood pressure analysis information based on the blood pressure signal; the blood pressure analysis information includes diastolic blood pressure, systolic blood pressure, and mean arterial pressure information;
[0042] The message module is used to transmit blood pressure signals to the MQTT server of the database module via the MQTT intermediate message protocol; and in response to user requests, to send the blood pressure signals and blood pressure analysis information to the client.
[0043] In this invention, PPG signals from a user's fingertip are collected; these PPG signals are preprocessed to obtain a first processed signal; and the first processed signal is input into a blood pressure prediction model based on a convolutional neural network to predict and generate the user's blood pressure signal. The blood pressure prediction model includes a backbone network and an information distillation network based on an attention mechanism. By using a dual-channel encoding section, dual-channel feature extraction is employed to improve accuracy. An improved attention mechanism information distillation module is used in the encoding section to obtain feature information at different scales. The dual-channel encoding section uses an addition operation to connect the input to the decoding section, accelerating blood pressure prediction speed while improving the model's predictive performance. Attached Figure Description
[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0045] Figure 1 This is a flowchart of the non-invasive blood pressure prediction method based on convolutional neural networks disclosed in Embodiment 1 of the present invention;
[0046] Figure 2 This is a diagram showing the overall framework of the blood pressure prediction model based on a convolutional neural network disclosed in Embodiment 1 of the present invention.
[0047] Figure 3 This is a structural diagram of the information distillation module based on the improved attention mechanism of the convolutional neural network model disclosed in Embodiment 1 of the present invention;
[0048] Figure 4 The figure shown is a schematic diagram of the overall topology of the non-invasive blood pressure prediction method based on convolutional neural networks provided in Embodiment 1.
[0049] Figure 5 This is the overall framework diagram of the non-invasive real-time blood pressure monitoring system disclosed in Embodiment 3 of the present invention. Detailed Implementation
[0050] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.
[0051] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.
[0052] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0053] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0054] It should be noted that "multiple" as mentioned in this article refers to two or more.
[0055] The implementation details of the technical solutions in the embodiments of this application are described in detail below:
[0056] Example 1
[0057] This invention provides a non-invasive blood pressure prediction method based on convolutional neural networks, such as... Figure 1 As shown, this embodiment provides a non-invasive blood pressure prediction method based on a convolutional neural network, the method comprising:
[0058] S1 collects the PPG signal from the user's fingertip.
[0059] In this embodiment, a PPG sensor is used to detect the user's PPG signal. The user's PPG signal is collected by pressing their fingertip into the PPG sensor. Unlike traditional cuff-type blood pressure monitoring, this method is simpler and more convenient to operate.
[0060] Furthermore, the PPG signals acquired by the PPG sensor can be transmitted to a nearby embedded development board or other modules or devices with data processing capabilities, currently via either Bluetooth or serial port protocols. It is clear that other modules or devices with data processing capabilities include, but are not limited to, computers, terminals, remote servers, or cloud computing.
[0061] S2, perform preprocessing on the PPG signal to obtain the first processed signal.
[0062] Specifically, in this embodiment, preprocessing is performed on the PPG signal to obtain a first processed signal, including: the preprocessing includes data segmentation operation, bandpass filtering operation, and data filtering operation.
[0063] The data segmentation operation includes segmenting the acquired continuous PPG signal into segments of a predetermined length; specifically, the data segmentation can segment the acquired continuous PPG signal into segments of 1024 data points, which facilitates subsequent model prediction.
[0064] Bandpass filtering is used to handle situations where segmented data is contaminated with noise. A bandpass filter of [0.5, 8] Hz is used to filter out both low-frequency and high-frequency noise.
[0065] The data filtering operation includes removing PPG signals damaged by noise by calculating the autocorrelation of the data. Specifically, data filtering removes PPG signals severely damaged by noise by calculating the autocorrelation of the data. A normal PPG signal should have good periodicity. By calculating the autocorrelation of the PPG signal, if its first peak value is greater than 0.7, it is considered to have good periodicity and is not severely contaminated by noise, and is thus retained; otherwise, it is discarded.
[0066] S3, the first processed signal is input into a blood pressure prediction model based on a convolutional neural network to predict and generate a user's blood pressure signal; the blood pressure prediction model includes a backbone network and an information distillation network based on an attention mechanism.
[0067] Specifically, in this embodiment, Figure 2 This is the overall framework of the blood pressure prediction model based on convolutional neural networks provided in Embodiment 1 of the present invention, namely the backbone network of this prediction model. Figure 3 This is the information distillation module based on an improved attention mechanism of a convolutional neural network model provided in Embodiment 1 of the present invention. Figure 3 The information distillation module of the improved attention mechanism is Figure 2 The encoder module in a convolutional neural network (i.e., the downsampling part in the backbone network) aims to increase the receptive field and fuse more information. The experiment uses the Adam optimizer for optimization, with 250 epochs and a batch size of 32. We use MSE as the loss function, where the error term is squared, and larger errors are penalized more severely.
[0068] Furthermore, the backbone network includes a dual-channel encoded Unet network; the Unet network employs a dual-channel encoder, meaning that encoding is performed again after the first encoder in each layer, thus enabling shallow layers to possess deep semantic information; an addition operation is used to concatenate the dual encoders of each layer together (i.e., append). Figure 2 The algorithm employs an Add operation (Add, Addition) to the decoder of a given layer. In this layer, the first encoder receives data from the first encoder of the previous layer, and the second encoder receives data from both the second encoder of the previous layer and the first encoder of the current layer. The data from both encoders in each layer is then passed to the decoder through an Add operation. This accelerates blood pressure prediction while enabling the decoding layer to acquire semantic information at different scales, thus improving the model's predictive performance.
[0069] Furthermore, the encoder includes an information distillation network based on an attention mechanism;
[0070] The information distillation network based on the attention mechanism extracts deeper semantic information by sequentially passing the input data through one-dimensional convolution operations with kernel sizes of 3, 5, and 7. The three output data are then concatenated in parallel across channels using a concatenation operation, ensuring that the output contains semantic information at different scales. Simultaneously, global average pooling (GPA) is applied in parallel to the original input to compress each channel. After passing through two fully connected layers, the number of channels is tripled. Finally, this output is multiplied by the information distillation output to obtain the final result.
[0071] In this embodiment, the information distillation module of the improved attention mechanism not only contains semantic information of different depths, but also enhances task-related channels and suppresses channels that are not related to the channels.
[0072] Furthermore, before acquiring the PPG signal from the user's fingertip, the method further includes: S0, pre-training a blood pressure prediction model based on a convolutional neural network. The pre-trained blood pressure prediction model based on a convolutional neural network includes:
[0073] Step 1: Obtain a dataset consisting of PPG signals and corresponding ABP signals, and perform data sampling and segmentation on the signals in the dataset.
[0074] Specifically, in this embodiment, the ABP signal of the PPG signal in the dataset is sampled and segmented. The sampling rate is set to 125Hz and the segment length is 1024 sampling points. This approach ensures that each PPG and ABP signal segment has an appropriate number of periods, while also facilitating the input to the subsequent model.
[0075] Step 2: Denoise the PPG and ABP signal data in the dataset.
[0076] Specifically, in this embodiment, the PPG and ABP data of the dataset are denoised to avoid noise interference with the prediction results. PPG and ABP signals are composed of low-frequency signals. However, motion artifacts and baseline drift occur due to movement, breathing, or improper wearing of the collection device. The denoising method is bandpass filtering, using a bandpass filter of [0.5, 8] Hz to filter out both low-frequency and high-frequency noise.
[0077] Step 3: Filter the denoised PPG and ABP signal data.
[0078] Specifically, in this embodiment, the denoised ABP and PPG data are filtered by calculating the autocorrelation of the data to remove PPG signals that are severely damaged by noise. A normal PPG signal should have good periodicity. By calculating the autocorrelation of the PPG signal, if its first peak value is greater than 0.7, it is considered to have good periodicity and is not severely contaminated by noise, and is thus retained; otherwise, it is discarded.
[0079] Step 4: Divide the selected PG signal data and ABP signal data into training set, validation set and test set according to a predetermined ratio.
[0080] Specifically, in this embodiment, the selected ABP and PPG data are divided into training, validation, and test sets in a 7:1:2 ratio. The PPG signal serves as the input for training, validation, and testing, while the ABP signal serves as the output. The main function of the model is to fit the ABP signal to the PPG signal; the lengths of the PPG and ABP signals are the same. Finally, diastolic blood pressure (DBP), systolic blood pressure (SBP), and mean arterial pressure (MAP) are calculated based on the output ABP signal. The calculation formula is as follows:
[0081] SBP = max(ABP)
[0082] DBP = min(ABP)
[0083] MAP = mean(ABP)
[0084] Step 5: Train the blood pressure prediction model based on the training set data into a convolutional neural network to generate the optimal prediction model.
[0085] The Adam optimizer was used in the experiment, with 250 epochs and a batch size of 32. MSE was used as the loss function, and the error term was squared, with larger errors receiving a larger penalty.
[0086]
[0087] The optimal prediction model is generated by training the model using the training set data. Specifically, the model is first trained on the training set, then tested on the validation set. If the result is better than previously saved results, the model weights for this epoch are saved. After multiple iterations, the optimal data generated from the training and validation sets is obtained. Finally, the accuracy of the saved model in predicting blood pressure is evaluated using the test set data.
[0088] like Figure 4The diagram shows the overall topology of the non-invasive blood pressure prediction method based on convolutional neural networks provided in this embodiment. The method involves acquiring and preprocessing the raw PPG signal, generating a preprocessed PPG signal, performing autocorrelation filtering, and then feeding it into the model to predict blood pressure signals, thereby calculating diastolic blood pressure (DBP), systolic blood pressure (SBP), and mean arterial pressure (MAP).
[0089] Compared to existing technologies, this method improves accuracy by using a dual-channel encoding section and extracting features through dual channels. The encoding section employs an information distillation module with an improved attention mechanism to obtain feature information at different scales. The dual-channel encoding section uses an addition operation to connect the input to the decoding section, which speeds up blood pressure prediction while improving the model's predictive performance.
[0090] Example 2
[0091] This embodiment proposes a non-invasive blood pressure prediction device based on a convolutional neural network. The device includes a data acquisition module, a preprocessing module, and a blood pressure prediction module; wherein:
[0092] The acquisition module is used to acquire PPG signals from the user's fingertips;
[0093] A preprocessing module is used to perform preprocessing on the PPG signal to obtain a first processed signal;
[0094] The blood pressure prediction module is used to input the first processed signal into a blood pressure prediction model based on a convolutional neural network to predict and generate a user's blood pressure signal; the blood pressure prediction model includes a backbone network and an information distillation network based on an attention mechanism.
[0095] Furthermore, the preprocessing module includes a data segmentation module, a bandpass filtering submodule, and a data filtering submodule; wherein:
[0096] The data segmentation module is used to segment the acquired continuous PPG signal into segments of a predetermined length;
[0097] The bandpass filter submodule is used to filter low-frequency and high-frequency noise in the segmented data by bandpass filtering;
[0098] The data filtering submodule is used to remove PPG signals that are damaged by noise by calculating the autocorrelation of the data.
[0099] The backbone network includes a dual-channel encoded Unet network; the Unet network uses a dual-channel encoder, that is, after the first encoder of each layer, encoding is performed again, so that shallow layers can also have deep semantic information; the dual encoders of each layer are concatenated together using Addition.
[0100] In this process, the first encoder of each layer receives data from the first encoder of the previous layer, and the second encoder of each layer receives data from the second encoder of the previous layer and the first encoder of the current layer. The two encoders of each layer are then passed to the decoder of the current layer via Addition.
[0101] The encoder includes an information distillation network based on an attention mechanism;
[0102] The information distillation network based on the attention mechanism includes:
[0103] The input data is sequentially processed through one-dimensional convolution operations with kernel sizes of 3, 5, and 7, and the semantic information at a deeper level is extracted through information distillation.
[0104] The data from the three outputs are concatenated in parallel on the channels using the concatenation operation, so that the output contains semantic information at different scales. At the same time, the original input is compressed by using the GlobalAveragePooling1D operation in parallel. After that, it goes through two fully connected layers, which triples the original number of channels. Finally, the output is multiplied by the information distillation output to get the result.
[0105] The device also includes a training module for pre-training a blood pressure prediction model based on a convolutional neural network;
[0106] The training module is used to: acquire a dataset consisting of PPG signals and corresponding ABP signals; sample and segment the signals in the dataset; denoise the PPG and ABP signal data in the dataset; filter the denoised PPG and ABP signal data; divide the filtered PPG and ABP signal data into training, validation, and test sets according to a predetermined ratio; and train the blood pressure prediction model based on the training set data to generate the optimal prediction model.
[0107] Example 3
[0108] This embodiment proposes a non-invasive real-time blood pressure monitoring system based on convolutional neural networks, comprising two parts: a hardware subsystem and a platform subsystem. The hardware subsystem can also be implemented using the non-invasive blood pressure prediction device based on convolutional neural networks described in Embodiment 2. The non-invasive blood pressure prediction device based on convolutional neural networks described in Embodiment 2 performs real-time acquisition of PPG signals and real-time prediction of the user's blood pressure signal; and includes the platform subsystem.
[0109] (1) The hardware subsystem may include a data acquisition module, a data transmission module, a data preprocessing module, and a blood pressure prediction module.
[0110] The data acquisition module mainly consists of a PPG sensor, a temperature sensor, and a humidity sensor. Unlike traditional cuff-based blood pressure monitoring, using a PPG sensor only requires pressing the fingertip into the sensor to collect the user's PPG signal. Compared to the cuff method, it is simpler and more convenient to operate.
[0111] The data transmission module refers to the transmission of PPG signals collected by the PPG sensor to a nearby embedded development board. Currently, this transmission can be achieved via Bluetooth or serial port protocols.
[0112] The data preprocessing module mainly uses the preprocessing method described in Example 1. Its implementation is carried out on an embedded development board.
[0113] The blood pressure prediction module refers to running the convolutional neural network from Example 1 on an embedded development board. It uses the previously predicted PPG signal as input and the ABP signal as output. The embedded development board is based on an ARM framework, which has a lower computational load compared to a GPU. Models trained on GPUs cannot be directly run on the embedded development board. TensorRT is a C++ library that facilitates high-performance inference on NVIDIA GPUs, specifically designed for fast and efficient network inference on GPUs. Installing TensorRT on the embedded development board allows models trained on GPUs to run on the ARM framework. Therefore, models trained on GPUs need to be converted to a type recognized by TensorRT, i.e., converting h5 files to pb files, before running them using a machine learning framework in Python.
[0114] The intermediate message module refers to the process of transmitting the blood pressure signal predicted by the embedded development board to the MQTT server via the MQTT intermediate message protocol, and then transmitting it from the MQTT server to the service subsystem, thus enabling asynchronous information transmission.
[0115] (2) The platform subsystem includes a database module, a blood pressure analysis module and a message module.
[0116] The database module is used to store uploaded blood pressure signals based on the MQTT server;
[0117] The blood pressure analysis module is used to calculate and obtain the user's blood pressure analysis information based on the blood pressure signal; the blood pressure analysis information includes diastolic blood pressure, systolic blood pressure, and mean arterial pressure information;
[0118] The message module is used to transmit blood pressure signals to the MQTT server of the database module via the MQTT intermediate message protocol; and in response to user requests, to send the blood pressure signals and blood pressure analysis information to the client.
[0119] Specifically, the platform subsystem includes a database module, a blood pressure analysis module, and a messaging module. The database module primarily stores all blood pressure signals obtained from the MQTT server by all users, as well as users' personal information. The blood pressure analysis module uses data mining programs to extract information such as diastolic blood pressure, systolic blood pressure, and mean arterial pressure from users' blood pressure signals. The messaging module mainly processes health-related information pushed by the system and also enables online consultations, allowing individuals to better understand their own health status.
[0120] To more clearly illustrate the system provided by the present invention, embodiments of this disclosure will be discussed. Figure 5 The serial numbers in the overall framework diagram of the non-invasive real-time blood pressure monitoring system are explained. Figure 5 ①, ②, ③, ④, ⑤, ⑥, and ⑦ represent: PPG signal sensor, microcontroller, embedded development board, wired network, server, individual user, and individual doctor, respectively. Sensors include PPG sensors, temperature sensors, and humidity sensors, etc.
[0121] The detection steps of the real-time blood pressure monitoring system based on convolutional neural networks are as follows:
[0122] Steps (1)-(4) constitute the data acquisition phase. In step (1), ordinary users collect humidity and temperature information using a fingertip PPG sensor, along with a temperature sensor and a humidity sensor. In step (2), the sensors read the data and send it to the microcontroller via serial port. In step (3), the microcontroller transmits the data to an embedded development board via wired or Bluetooth connection. The embedded development board preprocesses the PPG data and performs blood pressure prediction, then sends the results to an MQTT server via the network. In step (4), the user logs into the server to query the data, and the server retrieves the data from the data queue in the MQTT server.
[0123] Steps 5-6 constitute the data query phase for ordinary users. In step 5, ordinary users log in to the software to query their blood pressure, as well as information such as air humidity and temperature. In step 6, the server analyzes the transmitted blood pressure signal, calculates information such as diastolic blood pressure, systolic blood pressure, and mean arterial pressure, and displays it as an image, while simultaneously generating a health report for the user.
[0124] Steps 7-8 constitute the patient's consultation with the doctor stage. In step 7, a regular user logs in, selects a doctor, and requests a consultation by initiating an HTTP request to establish a connection. In step 8, the doctor reviews the message and confirms acceptance, establishing a messaging channel. The doctor and patient then exchange information.
[0125] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0126] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.
[0127] The units described as separate components may or may not be physically separate. As will be appreciated by those skilled in the art, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0128] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0129] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or grid device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0130] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A non-invasive blood pressure prediction method based on a convolutional neural network, characterized in that, The method includes: Collect PPG signals from the user's fingertips; The PPG signal is preprocessed to obtain a first processed signal; The first processed signal is input into a blood pressure prediction model based on a convolutional neural network to predict and generate the user's blood pressure signal; the blood pressure prediction model includes a backbone network composed of a dual-channel encoded Unet network and an information distillation network based on an attention mechanism. The backbone network includes a dual-channel encoded Unet network; the Unet network uses a dual-channel encoder, that is, after the first encoder of each layer, encoding is performed again, so that shallow layers can also have deep semantic information; the dual encoders of each layer are concatenated together using an append operation; In this process, the first encoder of each layer receives data from the first encoder of the previous layer, and the second encoder of each layer receives data from the second encoder of the previous layer and the first encoder of the current layer. The two encoders of each layer are then fed into the decoder of the current layer through an add operation.
2. The non-invasive blood pressure prediction method based on convolutional neural networks according to claim 1, characterized in that, The preprocessing of the PPG signal to obtain a first processed signal includes: The preprocessing includes data segmentation, bandpass filtering, and data filtering operations. The data segmentation operation includes segmenting the acquired continuous PPG signal into segments of a predetermined length. The bandpass filtering operation includes filtering low-frequency and high-frequency noise from the segmented data using bandpass filtering; The data filtering operation includes removing PPG signals that are damaged by noise by calculating the autocorrelation of the data.
3. The non-invasive blood pressure prediction method based on convolutional neural networks according to claim 2, characterized in that, The encoder includes an information distillation network based on an attention mechanism; The information distillation network based on the attention mechanism includes: The input data is sequentially processed through one-dimensional convolution operations with kernel sizes of 3, 5, and 7, and the semantic information at a deeper level is extracted through information distillation. The data from the three outputs are concatenated in parallel on the channels through a concatenation operation, so that the output contains semantic information at different scales. At the same time, global average pooling is applied in parallel to the original input to compress each channel. After that, the data passes through two fully connected layers, which triples the original number of channels. Finally, the output is multiplied by the information distillation output to obtain the final result.
4. The non-invasive blood pressure prediction method based on convolutional neural networks according to claim 3, characterized in that, Before collecting the PPG signal from the user's fingertip, the method further includes: Pre-train a blood pressure prediction model based on a convolutional neural network; The pre-trained blood pressure prediction model based on a convolutional neural network includes: Step 1: Obtain a dataset consisting of PPG signals and corresponding ABP signals, and perform data sampling and segmentation on the signals in the dataset; Step 2: Denoise the PPG and ABP signal data in the dataset; Step 3: Filter the denoised PPG and ABP signal data. Step 4: Divide the selected PG signal data and ABP signal data into training set, validation set and test set according to a predetermined ratio; Step 5: Train the blood pressure prediction model based on the training set data into a convolutional neural network to generate the optimal prediction model.
5. A non-invasive blood pressure prediction device based on a convolutional neural network, characterized in that, The device includes a data acquisition module, a preprocessing module, and a blood pressure prediction module; wherein: The acquisition module is used to acquire PPG signals from the user's fingertips; A preprocessing module is used to perform preprocessing on the PPG signal to obtain a first processed signal; The blood pressure prediction module is used to input the first processed signal into a blood pressure prediction model based on a convolutional neural network to predict and generate a user's blood pressure signal; the blood pressure prediction model includes a backbone network composed of a dual-channel encoded Unet network and an information distillation network based on an attention mechanism. The backbone network includes a dual-channel encoded Unet network; the Unet network uses a dual-channel encoder, that is, after the first encoder of each layer, encoding is performed again, so that shallow layers can also have deep semantic information; the dual encoders of each layer are concatenated together using an append operation; In this process, the first encoder of each layer receives data from the first encoder of the previous layer, and the second encoder of each layer receives data from the second encoder of the previous layer and the first encoder of the current layer. The two encoders of each layer are then fed into the decoder of the current layer through an add operation.
6. The non-invasive blood pressure prediction device based on a convolutional neural network according to claim 5, characterized in that, The preprocessing module includes a data segmentation module, a bandpass filtering submodule, and a data filtering submodule; wherein: The data segmentation module is used to segment the acquired continuous PPG signal into segments of a predetermined length; The bandpass filter submodule is used to filter low-frequency and high-frequency noise in the segmented data by bandpass filtering; The data filtering submodule is used to remove PPG signals that are damaged by noise by calculating the autocorrelation of the data.
7. A non-invasive real-time blood pressure monitoring system based on convolutional neural networks, characterized in that, The device includes a non-invasive blood pressure prediction device based on a convolutional neural network as described in any one of claims 5-6, which performs real-time acquisition of PPG signals and real-time prediction and generation of user blood pressure signals; and a platform subsystem. After the non-invasive blood pressure prediction device based on convolutional neural network predicts and generates the user's blood pressure signal, it uploads it to the platform subsystem; the platform subsystem includes a database module, a blood pressure analysis module, and a message module; The database module is used to store uploaded blood pressure signals based on the MQTT server; The blood pressure analysis module is used to calculate and obtain the user's blood pressure analysis information based on the blood pressure signal; the blood pressure analysis information includes diastolic blood pressure, systolic blood pressure, and mean arterial pressure information; The message module is used to transmit blood pressure signals to the MQTT server of the database module via the MQTT intermediate message protocol; In response to user requests, the system sends the blood pressure signal and blood pressure analysis information to the client.