WO2023179516A1 - 一种基于智能移动设备的检测系统、方法及检测模型构建方法 - Google Patents

一种基于智能移动设备的检测系统、方法及检测模型构建方法 Download PDF

Info

Publication number
WO2023179516A1
WO2023179516A1 PCT/CN2023/082388 CN2023082388W WO2023179516A1 WO 2023179516 A1 WO2023179516 A1 WO 2023179516A1 CN 2023082388 W CN2023082388 W CN 2023082388W WO 2023179516 A1 WO2023179516 A1 WO 2023179516A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal
ppg
ppg signal
detection system
unet
Prior art date
Application number
PCT/CN2023/082388
Other languages
English (en)
French (fr)
Inventor
夏超然
亓秀海
Original Assignee
北京华视诺维医疗科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202220656504.9U external-priority patent/CN218500702U/zh
Priority claimed from CN202210393707.8A external-priority patent/CN115054209B/zh
Application filed by 北京华视诺维医疗科技有限公司 filed Critical 北京华视诺维医疗科技有限公司
Publication of WO2023179516A1 publication Critical patent/WO2023179516A1/zh

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/0205—Simultaneously evaluating both cardiovascular conditions and different types of body conditions, e.g. heart and respiratory condition
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/117—Identification of persons
    • A61B5/1171—Identification of persons based on the shapes or appearances of their bodies or parts thereof
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue

Definitions

  • the invention relates to the field of medical instruments, and in particular to a detection system, method and detection model construction method based on intelligent mobile devices.
  • Blood pressure commonly known as systolic blood pressure and diastolic blood pressure
  • systolic blood pressure and diastolic blood pressure reflects the pressure exerted by aortic blood on the blood vessel wall when the human heart contracts and relaxes.
  • blood pressure can reflect the human cardiovascular system. Function, the greatest harm of abnormal blood pressure values to the human body is that it may cause damage to the heart, brain, kidneys and other important organs, such as cerebral infarction, renal artery stenosis, myocardial infarction, etc. Therefore, daily monitoring of blood pressure and fully understanding the real-time status and dynamic changes of an individual's own blood pressure are of great significance for individual health monitoring.
  • Mechanical sphygmomanometer refers to a mercury sphygmomanometer, which uses a cuff to pressurize and a stethoscope to distinguish pulse waves to measure blood pressure.
  • Using a mechanical sphygmomanometer to measure blood pressure requires professional operation. This measurement method may cause the white coat effect, that is, because the doctor's blood pressure is different from normal conditions during measurement, the pressure of the cuff will make the subject uncomfortable.
  • Electronic blood pressure monitors can be divided into electronic blood pressure monitors and smart watches with blood pressure measurement functions.
  • the electronic sphygmomanometer is opposite to the mercury sphygmomanometer, but it uses electronic listening instead of human listening, without the influence of the doctor's white coat, and is more objective and stable.
  • the cuff-based pressurization method has not changed, especially its size and inconvenience to carry.
  • blood pressure measurement watches with integrated micro air pumps and double-layered narrow air bags their overall weight is heavy.
  • the air bag design under the watch strap will not be breathable, and they are not suitable for daily long-term wear.
  • the present invention hopes to provide a multi-parameter physiological information detection system, method and model construction method that neither require conventional pressure equipment such as cuffs nor rely on hardware equipment such as separate photoelectric sensors, that is, the present invention
  • the invention proposes a detection system, method and model construction method based on smart mobile devices (non-wearable).
  • the present invention provides a detection system based on a smart mobile device.
  • the smart mobile device includes a lighting device and a sensing device located on the same side of the smart mobile device.
  • the detection system includes: a signal acquisition module and a signal processing module,
  • the lighting equipment is used to illuminate specific parts of the body of the individual to be tested
  • the sensing equipment is used to collect images of the illuminated parts or reflected signal images
  • the signal acquisition module is used to obtain multiple frames of images or reflection signals within a predetermined time, and extract PPG signals based on the images or reflection signals;
  • the signal processing module includes a UNet-ANN multi-branch feedback enhancement model, which performs multi-parameter physiological information detection based on the PPG signal.
  • the UNet-ANN multi-branch feedback enhancement model The enhanced model includes an asymmetric UNet backbone network and a multi-branch ANN network. Each branch in the multi-branch ANN network is used to detect a physiological information parameter.
  • the detection system also includes a data processing module, which is used to process the collected images into frames, weight and sum the R channel pixel values of each frame of image, and finally combine them into a continuous segment.
  • a data processing module which is used to process the collected images into frames, weight and sum the R channel pixel values of each frame of image, and finally combine them into a continuous segment. "PPG signal”.
  • the number of channels of input signals in the UNet backbone network is less than the number of sampling points, and the number of channels of output signals is greater than the number of sampling points.
  • the physiological information parameters include one or more of heart rate, blood oxygen, systolic blood pressure and diastolic blood pressure.
  • the UNet-ANN multi-branch feedback enhancement model adjusts the parameters of the UNet backbone network based on the difference between the true value and the detected value of the individual's heart rate and blood oxygen as a feedback condition.
  • the sensing device is a camera device, and the image is a video image.
  • the camera device is also used to capture a face video of the individual to be tested, and determine the gender of the individual to be tested based on the face video.
  • the data processing module is used to perform 0.5-8HZ elliptical bandpass filtering on the fingertip "PPG signal” to detect abnormal points and clean them; to divide the 30-second "PPG signal” into windows with a window width of 256 samples. points, with a step size of 150 sampling points; determine whether the data after cleaning and windowing contains data segments that meet the data quality requirements, and extract the corresponding data segments.
  • the lighting device is a display screen with a light-emitting function
  • the specific part of the body is a finger
  • the display screen is used to contact the finger to generate reflected or scattered light of the finger
  • the sensing device is a device with a light emitting function.
  • the photosensitive component is arranged below the active light-emitting display screen and is used to convert the reflected or scattered light into an electrical signal;
  • the processor is connected to the photosensitive component and is used to synthesize the electrical signal into a PPG signal.
  • the signal acquisition module includes an analog front-end amplifier
  • the analog front-end amplifier includes a signal amplifier, an analog-to-digital converter and a digital filter;
  • the input end of the signal amplifier is connected to the photosensitive component and is used to amplify the unit current value generated by the photosensitive component;
  • the analog-to-digital converter is connected to the signal amplifier and is used to convert the unit current value from an analog signal into a digital signal;
  • the input terminal of the digital filter is connected to the analog-to-digital converter, and the output terminal of the digital filter is connected to the processor, and is used to filter out noise mixed in the current value of each unit.
  • the measurement module further includes a timing control circuit
  • the timing control circuit includes a timing controller, a controller and a driver; the timing controller has three output terminals, the first output terminal is connected to the processor, the second output terminal is connected to the analog front-end amplifier, and the third output terminal Connect the controller to control the working sequence of the photosensitive array; the controller is connected to the timing controller to control the working frequency of the photosensitive array; the input end of the driver is connected to the controller, and the output end is connected to the controller.
  • the photosensitive component is connected to drive the photosensitive component to work.
  • the photosensitive component is a photosensitive element or a photosensitive array composed of multiple photosensitive elements.
  • the measurement module further includes a memory and a voltage stabilizing circuit
  • the memory is connected to the processor and used to store preset instructions and blood pressure data
  • the voltage stabilizing circuit is respectively connected to the processor and the analog front-end amplifier to stabilize the power supply and reduce the impact of the external circuit environment.
  • the present invention provides a physiological information detection model construction method, which method includes:
  • Step 1) use the lighting equipment of the smart mobile device to illuminate the end of the body part of the individual to be tested;
  • Step 2) use the sensing device of the smart mobile device to collect multiple frames of continuous images or reflection signals of the individual to be tested;
  • Step 3 extract the terminal PPG signal based on the image or reflection signal and process the terminal PPG signal;
  • Step 4 construct a UNet-ANN multi-branch feedback enhancement model.
  • the UNet-ANN multi-branch feedback enhancement model includes an asymmetric UNet backbone network and a multi-branch ANN network. Each branch in the multi-branch ANN network Each is used to detect a physiological information parameter;
  • Step 5 Use the labeled fingertip PPG signals to train the enhanced model.
  • the step of processing the terminal PPG signal includes:
  • Step 2.1 Divide the obtained video into frames, weight and sum the R channel pixel values of each frame of image, and combine them into a continuous "PPG signal";
  • Step 2.2 Perform 0.5-8HZ elliptical bandpass filtering on the PPG signal to detect abnormal points and clean them out;
  • Step 2.3 Divide the PPG signal into windows
  • Step 2.4 Extract and inspect the data segments of the PPG data after cleaning and windowing. If the verification is qualified, proceed to the next step; otherwise, collect the data again;
  • the extraction and verification steps of the PPG signal include:
  • Step 3.1 Find the peak point of the processed PPG signal and find the peak point and valley point in the PPG signal;
  • d_valley in fluctuating conditions, if d_peak ⁇ 0.05 and d_valley ⁇ 0.08, this segment of PPG data will be retained;
  • Step 3.3 Based on the number of peak points and valley points, determine whether the number of cycles contained in the current signal window is less than the predetermined value. If it is less than the predetermined value, use the second extraction plan;
  • Step 3.4 Divide the processed PPG signal into windows of a single period
  • Step 3.5 Calculate the correlation coefficient r of adjacent periodic waveforms. If r > the correlation coefficient threshold, retain the PPG signal, otherwise discard it;
  • Step 3.6 If the PPG signal processed in step 2 does not meet both extraction standards, delete the PPG signal directly.
  • the present invention provides a detection method based on smart mobile devices, which method is used to detect physiological parameter information.
  • the method includes:
  • Step 1) use the lighting equipment of the smart mobile device to illuminate the end of the body part of the individual to be tested;
  • Step 2) use the sensing device of the smart mobile device to collect multiple frames of continuous images or reflection signals of the individual to be tested;
  • Step 3 extract the terminal PPG signal based on the image or reflection signal and process the terminal PPG signal;
  • Step 4 input the processed signal into the UNet-ANN multi-branch feedback enhancement model constructed in claim 13 or 14;
  • Step 5 use the output of the UNet-ANN multi-branch feedback enhancement model as the result of corresponding physiological parameter information detection.
  • the "specific parts of the body” and “ends of the body” mentioned in the present invention refer to parts of the body that can collect blood-related information, such as finger tips, earlobes or toe tips.
  • the “smart mobile devices” mentioned in the present invention refer to non-wristband pressurized mobile devices such as mobile phones, tablets, and Ipods, which have lighting sources and camera equipment.
  • This invention technologically proposes to use a combination of cameras and lighting equipment of smart mobile devices such as mobile phones and tablets to obtain video images of the end of the individual to be tested (such as fingertips and ear ends), and extract photoplethysmography (PPG) signals therefrom. , perform a weighted summation of the acquired video frames and convert them into PPG signal sampling points.
  • the present invention uses the acquired PPG signal to use multi-branch feedback to strengthen the model to obtain multiple physiological parameter indicators, and can simultaneously realize synchronous measurement of multiple physiological parameters based on mobile phones without the need for professional equipment.
  • the finger-clip pulse oximeter which is fixed on the fingertip, it emits light in a fixed wavelength band (usually green light or red light and near-infrared light) and the PPG signal obtained by the corresponding photoelectric sensor, which is obtained from smart devices such as mobile phones.
  • the PPG signal obtained from the camera is relatively unstable. Therefore, currently existing equipment and methods cannot use the PPG extracted from the fingertip continuous video frames obtained from the camera of the smart device to effectively measure blood oxygen, blood pressure and other parameters. .
  • the present invention realizes the extraction and utilization of PPG signals collected based on intelligent devices, and has a high accuracy after experimental verification.
  • the multi-branch feedback enhancement model of the present invention uses the Unet backbone network and multiple ANN branches to jointly predict heart rate, blood oxygen, and blood pressure values.
  • the feedback from the heart rate and blood oxygen branches is first used to optimize the Unet backbone network. Then the backbone network is fixed, and the blood pressure prediction branch is trained on this basis.
  • the present invention uses the predicted gender and age signals of the portrait obtained by the front or rear camera to correct the model prediction results.
  • blood pressure is a physiological indicator with large individual differences, it still has certain statistical significance with individual age, gender and other signals in a large sample. Combined with the individual information obtained from the portrait, if the physiological indicators predicted by the model are too different from the statistical prediction results, the subject can be prompted to perform secondary measurements. The results of the three measurements still do not fluctuate much.
  • the system and method of the present invention avoid some accidental errors and extreme results to a certain extent.
  • Figure 1 is a flow chart of parameter detection using the detection system of Embodiment 1 of the present invention.
  • Figure 2 is a schematic structural diagram of a mobile phone used to perform the method of the present invention
  • Figure 3 is a schematic diagram of the process of reconstructing finger-end PPG signals in Embodiment 1 of the present invention.
  • Figure 4 is a schematic diagram of the overall framework of the model used in two embodiments of the present invention.
  • FIG. 5 is a detailed framework diagram of the backbone network Unet of the model used in the two embodiments of the present invention.
  • Figure 6 is a detailed framework diagram of a branch network of the model used in two embodiments of the present invention.
  • Figure 7 is a schematic structural diagram of the detection system in Embodiment 2 of the present invention.
  • Figure 8 is a schematic structural diagram of the measurement module in Embodiment 2 of the present invention.
  • Figure 9 is a schematic structural diagram of an analog front-end amplifier in Embodiment 2 of the present invention.
  • FIG. 10 is a schematic structural diagram of a timing control circuit in Embodiment 2 of the present invention.
  • 1-measurement module 2-active light-emitting display screen, 11-finger, 121-incident light, 122-reflected or scattered light, 21-photosensitive component, 22-processor, 23-memory, 24-input and output interface, 32-analog front-end amplifier, 36-timing control circuit, 37-voltage stabilizing circuit, 41-signal amplifier, 42-analog-to-digital converter, 43-digital filter, 51-timing controller, 52-controller, 53-driver .
  • the detection system based on smart mobile devices in this embodiment is described using a smart phone as an example.
  • other smart mobile devices such as tablet computers can also be used to implement the system of the present invention.
  • Smart mobile devices need to have lighting equipment and camera equipment on the same side. Currently, their rear cameras are mainly used, so that both lighting and image collection can be achieved on the same side.
  • auxiliary light sources can also be used for illumination, and then the camera of the mobile device can be used for image collection.
  • the lighting device uses the flash 200 of the mobile phone
  • the camera device uses the rear camera 100 of the mobile phone.
  • the detection system in this embodiment includes: a signal acquisition module 300 and a signal processing module 400.
  • the signal acquisition module 300 and the signal processing module 400 can both be implemented by the processor in the mobile phone.
  • the flash 200 is used to illuminate the fingertips or ears of the human body; the camera 100 is used to collect video images of the illuminated parts.
  • the signal acquisition module is used to obtain multi-frame images within 20-40s, such as 30s, obtained by the camera equipment, and extract PPG signals based on the images; the signal processing module contains the UNet-ANN multi-branch feedback enhancement model , the UNet-ANN multi-branch feedback enhancement model performs multi-parameter physiological information detection based on the PPG signal.
  • the signal processing module carries the application of a multi-branch feedback enhancement algorithm network and runs within the processor of the mobile device.
  • the invention can use mobile phone cameras, lights and applications to realize the measurement, storage and analysis system of human physiological indicators (heart rate, blood oxygen, blood pressure).
  • human physiological indicators heart rate, blood oxygen, blood pressure.
  • Step 1.1 Let the person being tested take a short rest to ensure that their various physiological indicators are in a stable state
  • Step 1.2 Set mobile video recording to 60fps or adjust according to accuracy.
  • Step 1.3 Use the front camera of the mobile phone to collect the face video and save it (this step only needs to be performed once when any subject uses this function for the first time). It should be noted that steps 1.1-1.3 can be implemented by setting up a separate login setting module in the system. It is an optional module and is mainly used for personnel information login confirmation and mobile phone parameter adjustment.
  • Step 1.4 Press your finger against the light of the mobile phone and the rear camera connected to it, maintain a quiet and stable state, and collect continuous image frames from the finger tip for 30 seconds, as shown in Figure 3.
  • Step 2.1 Divide the fingertip video obtained in step 1.3 into frames, weight and sum the R channel pixel values of each frame image, and finally combine them into a continuous fingertip "PPG signal", as shown in Figure 3.
  • Step 2.2 Perform 0.5-8Hz elliptical bandpass filtering on the fingertip "PPG signal" to detect abnormal points and clean them to remove them.
  • Step 2.3 Divide the 30-second "PPG signal" into windows with a window width of 256 sampling points and a step length of 150 sampling points.
  • Step 2.4 Determine the data after cleaning and windowing, extract the data segments and check according to step S3. If the extraction criteria are met, proceed to the next step, otherwise return to 1.3 to collect data again.
  • Step 3.1 Use the processed PPG signal to find the peak point and valley point algorithm to find the peak point and valley point in the PPG signal.
  • Step 3.3 Based on the number of peak points and valley points, determine whether the number of cycles contained in the current signal window is less than a predetermined value (no more than 2 peak points or valuation points), then use the second extraction scheme.
  • Step 3.5 Calculate the correlation coefficient r of adjacent periodic waveforms (just use the existing correlation algorithm). If r>0.85 (0.85 is an empirical value), retain the PPG signal in this segment.
  • Step 3.6 If the PPG signal processed in step 2 does not meet the above two extraction standards, the PPG signal is directly deleted.
  • Step 4.1 Call the openCV library to determine the gender and age information of the subject from the face video obtained in step 1.2.
  • Step 4.2 Input the pre-processed fingertip "PPG signal" into the pre-trained UNet-ANN multi-branch feedback enhancement model to obtain predicted heart rate, blood oxygen, and blood pressure values.
  • PPG signal the pre-processed fingertip "PPG signal” into the pre-trained UNet-ANN multi-branch feedback enhancement model to obtain predicted heart rate, blood oxygen, and blood pressure values.
  • Step 4.3 Adjust the 3.2 model based on the subject information obtained in 4.1.
  • Step 5.1 Choose whether to save the measured results
  • Step 5.2 If you choose to save in 5.1, the results will be saved to personal data and a statistical analysis will be performed.
  • the UNet-ANN multi-branch feedback enhancement model in this embodiment is divided into two parts: the backbone network and the branch network.
  • the backbone network adopts the Unet network
  • the branch network adopts the ANN branch network.
  • the structural configuration of a single Unet network and four ANN branch networks is adopted.
  • the original Unet is a symmetric structure, and the Unet in the present invention is adjusted as a feature extraction module and adopts an asymmetric structure.
  • the Unet network was originally created to solve the problem of medical image segmentation.
  • the Unet network utilizes a U-shaped network structure to obtain contextual information and location information.
  • the asymmetric Unet backbone network is mainly divided into two modules: left and right.
  • the module on the left includes four layers. Each layer is mainly composed of four two-level 3*1 convolution kernels plus a 2*1 downsampling sub-module (the number of convolution kernels and sub-modules in each layer is adjustable).
  • the right module also includes four layers.
  • Each layer is composed of four two-level 3*1 convolution kernels plus a 2*1 upsampling sub-module.
  • Each layer between the left and right modules is connected through splicing, and the output is from the rightmost side. The asymmetry is reflected in the difference between the input layer and the output layer.
  • the input layer is a PPG signal with a length of 1 channel and 256 sampling points, while the output layer is a multi-feature signal with a length of 1024 channels and 2 sampling points.
  • the improved Unet structure in the present invention uses one-dimensional convolution to replace the two-dimensional convolution of the image. In the last layer (the top layer of the right module), multiple output channels are used to replace the single channel in the original Unet to obtain multiple features. The values are sent to each ANN branch network respectively.
  • the PPG signal with a length of 1 channel and 256 sampling points is used as input, and it is input into the backbone network Unet, and becomes a signal with a length of 64 channels and 256 sampling points under the two-stage 3*1 convolution network;
  • the signal is transformed into a signal with a length of 64 channels and 128 sampling points through the 2*1 down-sampling network; it is transformed into a signal with a length of 128 channels and 128 sampling points under the two-stage 3*1 convolution network; it is transformed into a signal with a length of 128 sampling points through the 2*1 down-sampling network.
  • the signal is spliced with the signal of 64 channels and 256 sampling points on the left to form a signal of 128 channels and 256 sampling points; it becomes a signal of 128 channels and 64 sampling points through the 4*1 down-sampling network; at 5*1, the step size is Under the convolution network of 3, it becomes a signal with a length
  • the processing of the branch network is mainly based on the feature combination processing of multi-feature signals with a length of 1024 channels and 2 sampling points in the output layer of the backbone network.
  • the predicted values are obtained through multi-layer convolution feature processing.
  • the high-pressure and low-pressure values of the blood pressure value are related data. They are each convolved to obtain the predicted value from one dimension of the characteristic signal with a length of 1024 channel 2 sampling points.
  • the signal with a length of 1024 channel 2 sampling points passes through a 2*1 convolution network and becomes a signal with a length of 1024 channel 1 sampling points; the signal with a length of 1024 channels 1 sampling point passes through 1
  • the *1 convolution network becomes a signal with a length of 64 channels and 1 sampling point; the signal with a length of 64 channels and 1 sampling point passes through the 1*1 convolution network and becomes a signal with a length of 1 channel and 1 sampling point, that is, the predicted heart rate value.
  • the signal with a length of 1024 channels and 2 sampling points passes through a 2*1 convolution network and becomes a signal with a length of 1024 channels and 1 sampling points; the signal with a length of 1024 channels and 1 sampling points passes through a 1*1 convolution network and becomes a signal with a length of 1024 channels and 1 sampling points.
  • the signal with length of 256 channels and 1 sampling point passes through a 1*1 convolution network and becomes a signal with a length of 64 channels and 1 sampling point; the signal with a length of 64 channels and 1 sampling point passes through a 1*1 convolution network and becomes 1
  • the signal of the length of the sampling point of channel 1 is the predicted blood oxygen value.
  • the last output signal of the Unet backbone network with the length of 1024 channel 2 sampling points is used to predict blood pressure.
  • the signal with a length of 1024 channels and 1 sampling point passes through a 1*1 convolution network and becomes a signal with a length of 256 channels and 1 sampling point; the signal with a length of 256 channels and 1 sampling point passes through a 1*1 convolution network and becomes A signal with a length of 64 channels and 1 sampling point; a signal with a length of 64 channels and 1 sampling point passes through a 1*1 convolution network and becomes a signal with a length of 1 channel and 1 sampling point, which is the predicted systolic blood pressure.
  • the signal with a length of 1024 channels and 1 sampling point passes through a 1*1 convolution network and becomes a signal with a length of 256 channels and 1 sampling point; the signal with a length of 256 channels and 1 sampling point passes through a 1*1 convolution network and becomes The signal with the length of 64 channels and 1 sampling point; the signal with the length of 64 channels and 1 sampling point passes through the 1*1 convolution network and becomes the signal with the length of 1 channel and 1 sampling point, which is the predicted diastolic blood pressure.
  • Heart rate accuracy more than 90% probability of falling within +/-3bpm
  • Blood pressure accuracy More than 85% probability of high pressure SBP falling within the range of +/-10bpm
  • the active light-emitting display screen of the smart device is used as a lighting device as an example to describe the process of detecting physiological information using the front panel of the smart device.
  • the detection system in this embodiment can be implemented using an intelligent device with a display screen.
  • the display screen can be an organic light emitting diode (Organic Light Emitting Diode) display screen or a micro-light emitting diode (Micro-LED) display screen.
  • the intelligent device can be a smartphone. , smart watches, tablets, etc., but are not limited to these.
  • this embodiment provides a schematic diagram of the data collection process of another smart device.
  • the rear camera of the mobile phone of Embodiment 1 is not used for image collection.
  • the measurement module 1 provided below the active light-emitting display screen 2 is used to detect the reflected or scattered light 122 on the finger 11 .
  • the lighting device uses an active light-emitting display screen 2 .
  • the active light-emitting display screen 2 When the person to be tested places the finger 11 on a designated local area on the active light-emitting display screen 2, the active light-emitting display screen 2 emits incident light 121 to the finger as a lighting device.
  • the incident light 121 irradiates the finger 11 and is reflected or scattered. or scattered light122.
  • the measurement module 1 is arranged below the active light-emitting display screen 12, and is used to obtain reflected or scattered light 122 to obtain multiple current values in the time domain within a certain period of time, and generate the multiple current values into PPG signals, and finally use all the current values.
  • the PPG signal is calculated to generate a blood pressure value.
  • the measurement module 1 includes a photosensitive component 21, a processor 22, a memory 23 and an input and output interface 24.
  • the photosensitive component 21 is disposed below the active light-emitting display screen 2 and is used to acquire light signals with physiological characteristics of the human body and convert them into electrical signals; the photosensitive component 21 can be a photosensitive array composed of multiple photosensitive elements and is disposed below the active light-emitting display screen 2 , to achieve multi-position blood pressure measurement or half-screen or even full-screen blood pressure measurement.
  • the processor 22 is connected to the photosensitive component 21.
  • the processor 22 has a built-in PPG algorithm or model for synthesizing the electrical signal output by the photosensitive component 21 into a PPG signal through an algorithm to calculate the blood pressure value.
  • the memory 23 is connected to the processor 22 and is used to store designed algorithm instructions and user physiological data.
  • the input and output interface 24 is connected to the processor 22 and is used for receiving work instructions and transmitting blood pressure data to external devices when blood pressure measurement is completed.
  • the PPG detection model built into the processor 22 can be the UNet-ANN multi-branch feedback enhancement model constructed in Embodiment 1, but this model needs to be trained and used based on the signals collected by the device in this embodiment.
  • the measurement module 1 also includes an analog front-end amplifier 32 , a timing control circuit 36 and a voltage stabilizing circuit 37 .
  • the input end of the analog front-end amplifier 32 is connected to the photosensitive component 21, and is used to amplify the unit current value generated by each photosensitive array unit, perform analog-to-digital conversion and circuit denoising on each unit current value, and generate each preprocessing unit current value; the output end is with Processor 22 is connected.
  • the timing control circuit 36 is respectively connected to the photosensitive component 21, the processor 22 and the analog front-end amplifier 32, and is used to control the working timing of the devices in the measurement module 1 and the working frequency of the photosensitive array.
  • the voltage stabilizing circuit 37 is connected to the processor 22 and the analog front-end amplifier 32 respectively, and is used to stabilize the power supply of the devices in the measurement module 1 and reduce the influence of the external circuit environment.
  • the analog front-end amplifier 32 has a data input terminal and a data output terminal.
  • the data input terminal receives the unit current value generated by each photosensitive array unit from the photosensitive component 21 at each frequency time point within the measurement time according to the operating frequency.
  • the data output terminal is connected to the processor 22 and outputs the current value of each preprocessing unit at each frequency time point.
  • the internal circuit of the analog front-end amplifier 32 includes a signal amplifier 41 connected in sequence, which is used to amplify the weak unit current value generated by each photosensitive array unit to facilitate subsequent signal processing and increase the value of the unit current value of the same photosensitive array unit at different frequency time points.
  • the analog-to-digital converter 42 is used to convert the current value of each unit from an analog signal to a digital signal to facilitate calculation and analysis by the processor;
  • the digital filter 43 is used to filter out the noise mixed in the current value of each unit.
  • the signal amplifier 41 uses an operational amplifier to linearly amplify the signal, and at the same time, the analog front-end amplifier 32 sends the current value of each preprocessing unit to the processor 22 once at a frequency time point.
  • the signal amplifier 41 can also use a transistor amplifier and a power amplifier, and the analog front-end amplifier 32 can uniformly send the current values of each preprocessing unit at each frequency time point to the processor 22 at the end of the measurement time.
  • the timing control circuit 36 includes: a timing controller 51, which is used to control the working timing of components in the measurement module, and has three output terminals.
  • the first output terminal is connected to the processor 22 and is used to control the working timing of the processor 22;
  • the second output terminal The analog front-end amplifier 32 is connected to control the working sequence of the analog front-end amplifier 32;
  • the third output terminal is connected to the photosensitive component 21 through the controller 52 and the driver 53, and is used to control the working sequence of the photosensitive array;
  • the controller 52 is connected to the timing controller 51 , used to control the operating frequency of the photosensitive array;
  • the driver 53 is connected to the controller 52 and used to receive control instructions from the controller 52 and drive the photosensitive array to work.
  • the photosensitive component 21 uses the image CMOS sensor OV9782 of OmniVision Technology; the processor 22 uses the DM505 SoC of TI Texas Instruments, and the input and output interface 24 is integrated and set on the DM505 SoC of the processor 22; the memory 23 Using Winbond Electronics W25X05C LFLASH memory.
  • the analog front-end amplifier 32 adopts Texas Electronics' single-chip biosensing signal acquisition analog front-end AFE4900.
  • the AFE4900 chip integrates a signal amplifier 41, an analog-to-digital converter 42 and a digital filter 43.
  • the timing control circuit 36 adopts the ADM1066 digital timing controller of Analog Devices.
  • the ADM1066 digital timing controller integrates the timing controller 51 , the controller 52 and the driver 53 .
  • the voltage stabilizing circuit 37 uses the synchronous boost converter TPS61099 and the non-inverting buck-boost converter TPS63036; among them, the AFE4900 chip has three power inputs, the TPS61099 is used to provide voltage stabilization of one power supply, and the TOS6303 provides the stabilization of the other two power supplies. pressure.
  • the blood pressure measurement system also includes a Bluetooth module and a WiFi module for wireless user blood pressure data transmission.
  • the operating frequency of the photosensitive array is greater than or equal to 180 Hz, and each photosensitive array unit generates a unit current value at one frequency time point of the photosensitive array.
  • the processor 22 obtains the current value data of each preprocessing unit at each frequency time point.
  • the photosensitive array includes multiple photosensitive array units.
  • the processor 22 performs a weighted average of the current values of each preprocessing unit at a frequency time point. Or an arithmetic average operation is performed to obtain a reference current value, which is used to generate a value of the PPG signal.
  • the processor 22 combines the current values of each preprocessing unit at each frequency time point to find the photosensitive array unit with the most obvious value change, and uses the preprocessing unit current value of the photosensitive array unit at one frequency time point as a reference.
  • the processor performs a weighted average or arithmetic average operation on the current values of the preprocessing units of the several photosensitive array units at a frequency time point to obtain the reference current value,
  • the reference current value is used to generate a value of PPG.
  • the photosensitive array only contains one photosensitive array unit, and the processor 22 uses the preprocessing unit current value of the only photosensitive array unit at a frequency time point as a reference current value, and the reference current value is used to generate a PPG signal. value.
  • the processor 22 obtains the reference current value at each frequency time point and removes the DC part of each reference current value to obtain each reference AC current value, generates a PPG signal according to each reference AC current value, and filters the generated PPG signal. Noise is reduced to reduce the interference caused by noise on the calculation of blood pressure values. Finally, the user's blood pressure value is calculated based on the correlation between the PPG signal waveform characteristics and blood pressure.
  • the workflow of blood pressure measurement includes: when not working, the system is in a standby energy-saving state, that is, the photosensitive array is not started and the system continues to wait for a work signal.
  • the user selects blood pressure measurement and presses his finger on the designated area of the display for more than 8 seconds.
  • the blood pressure measurement system receives the working signal and receives the light signal emitted by the smart device display and reflected or scattered by the finger to obtain the working frequency of each photosensitive array unit.
  • the unit current value generated at each frequency time point within the measurement time, the unit current value generated by each photosensitive array unit at each frequency time point is used to form the photoplethysmography PPG signal, and the correlation between the PPG signal waveform shape and blood pressure Used to get the user's blood pressure information.
  • the measured PPG signal is used to input it into the already trained UNet-ANN multi-branch feedback enhancement model, or into other blood pressure monitoring models based on PPG signal to calculate blood pressure data.
  • this embodiment is the same as the processing method in Embodiment 1, and will not be described in detail here.
  • the system automatically stores and updates the user's blood pressure data, and promptly transmits the current user's blood pressure status to the external device. After the measurement is completed, it enters the standby state, that is, waiting for the next work signal.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Cardiology (AREA)
  • Physiology (AREA)
  • Pulmonology (AREA)
  • Optics & Photonics (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Abstract

一种基于智能移动设备的检测系统、方法及检测模型构建方法。智能移动设备包括位于智能移动设备同一侧的照明设备和摄像设备，检测系统包括：信号采集模块（300）和信号处理模块（400）。利用智能终端照明设备和信号检测设备进行指端信号采集，进而视频预测心率，血氧，血压。该系统利用视频取帧，图像分通道，像素点加权求和从指端连续视频帧中合成PPG信号或利用感光设备进行信号采集获得PPG信号；在PPG基础上利用Unet网络提取特征，由多层卷积神经网络组成的多个支路的分支网络分别预测心率，血氧，血压值。

Description

一种基于智能移动设备的检测系统、方法及检测模型构建方法 技术领域
本申请主张于2022年4月14日提交的、名称为“一种基于智能移动设备的多参数生理信息检测系统和方法”的中国发明专利申请：202210393707.8的优先权以及2022年3月23日提交的、名称为“一种用于智能设备的屏下血压测量系统”的中国实用新型专利申请：202220656504.9的优先权。
本发明涉及医疗仪器领域，具体涉及一种基于智能移动设备的检测系统、方法及检测模型构建方法。
 背景技术
面对人口老龄化及上班族日常工作高强度趋势，常规生理指标（心率，血氧，血压等）的随时随地监测，成为迫切需求。
以上需求中，血压的精确测量是一个难点。血压，即通常人们所说的收缩压和舒张压，其反应的是人体心脏收缩与舒张时主动脉血液对血管壁造成的压强的大小，血压作为人体生理信号的重要指标，可以反应人体心血管功能，异常的血压值对人体最大的危害在于可能引起心，脑，肾等重要器官的损害，如脑梗，肾动脉狭窄，心肌梗死等。所以日常对血压的监测，充分了解个体自身的血压实时情况及动态变化情况对于个体的健康监护有十分重要的意义。
1.目前市面上用于生理指标检测的设备大多功能独立，比如用于血氧检测的指夹式脉搏血氧仪，用于血压检测的血压计。很少有集成的一站式的多生理参数检测的综合设备。
2.目前市面上对血压的检测设备分机械式和电子式两大类。
机械式血压计指水银血压计，通过袖带加压，听诊器分辨脉波音测量血压值。利用机械式血压计进行血压测量需要专业人员操作，这种测量方式可能会引发白大褂效应，即测量时因为面对医生血压和正常情况有差异，袖带的加压会使得受试者不适。
电子式血压计又可以分为电子血压计和带有血压测量功能的智能手表。电子血压计与水银血压计相对，只是用电子听音代替的人的听音，没有了医生白大褂影响，也更加客观稳定。但是基于袖带加压的方式还是没有改变，尤其其体积问题也不方便携带。虽说现在有集成微型气泵、双层窄气囊的血压测量手表，但其整体重量偏重，加上表带下气囊设计会不透气，也不适合日常长期佩戴。
而其他带有血压测量功能的智能手表，不论是基于两路PPG信号或是PPG与ECG信号的结合方案测量血压对于硬件要求都比较高，必须搭载在智能手表上。
 技术问题
综上所述，现有血压测量方案中以部分是基于袖带加压的测量方式，此种方式一方面不够便捷，另一方面会给被测者带来一定的不适感；另一部分采用基于PPG光电信号的方式，这种方式依赖于光电传感器等硬件设备。
 技术解决方案
针对上述问题，本发明希望提供一种既不需要袖带等常规的加压设备，也不依赖于单独的光电传感器等硬件设备的多参数生理信息检测系统、方法及模型构建方法，即，本发明提出了基于智能移动设备（非佩戴式）的检测系统、方法及模型构建方法。
具体而言，本发明提供一种基于智能移动设备的检测系统，所述智能移动设备包括位于所述智能移动设备同一侧的照明设备和传感设备，所述检测系统包括：信号采集模块和信号处理模块，
  所述照明设备用于对待测个体的身体特定部位进行照射；
  所述传感设备用于采集被照射部位的图或反射信号像；
  所述信号采集模块用于获取预定时间内的多帧图像或反射信号，并基于所述图像或反射信号提取PPG信号；
  所述信号处理模块中包含有UNet-ANN多支路反馈加强模型，所述UNet-ANN多支路反馈加强模型基于所述PPG信号进行多参数生理信息检测，所述UNet-ANN多支路反馈加强模型包括非对称的UNet主干网络和多分支ANN网络，所述多分支ANN网络中的每个分支分别用于进行一种生理信息参数的检测。
优选地，所述检测系统还包括数据处理模块，所述数据处理模块用于将所采集的图像进行分帧处理，将每一帧图像的R通道像素值加权求和，最后组合成一段连续的“PPG信号”。
优选地，所述UNet主干网络中输入信号的通道数小于采样点数，输出信号的通道数大于采样点数。
优选地，所述生理信息参数包括心率、血氧、收缩压以及舒张压中的一种或多种。
优选地，所述UNet-ANN多支路反馈加强模型基于待测个体的心率和血氧的真实值与检测值之差作为反馈条件进行UNet主干网络的参数调整。
优选地，所述传感设备为摄像设备，所述图像为视频图像，所述摄像设备还用于拍摄待测个体的人脸视频，并且基于所述人脸视频进行待测个体的性别判定。
优选地，所述数据处理模块用于将指端“PPG信号”进行0.5-8HZ椭圆带通滤波处理，检测异常点并清洗；将30秒的“PPG信号”中进行分窗，窗宽256采样点，步长150采样点；判断清洗分窗过后的数据是否包含符合数据质量要求的数据段，并提取相应数据段。
优选地，所述照明设备为具有发光功能的显示屏，所述身体特定部位为手指，所述显示屏用于与手指接触，产生所述手指的反射或散射光，所述传感设备为具有感光组件的测量模块；
所述感光组件设置在所述主动发光显示屏下方，用于将所述反射或散射光转换为电信号；
所述处理器与所述感光组件连接，用于将所述电信号合成为PPG信号。
优选地，所述信号采集模块包括模拟前端放大器；
所述模拟前端放大器包括信号放大器、模数转换器和数字滤波器；
所述信号放大器输入端与所述感光组件连接，用于放大所述感光组件生成的单元电流值；
所述模数转换器与所述信号放大器连接，用于将所述单元电流值从模拟信号转换成数字信号；
所述数字滤波器输入端与所述模数转换器连接，所述数字滤波器输出端与所述处理器连接，用于滤除混杂在各个单元电流值中的噪声。
优选地，所述测量模块还包括时序控制电路；
所述时序控制电路包括时序控制器、控制器和驱动器；所述时序控制器具有三个输出端，第一输出端连接所述处理器，第二输出端连接所述模拟前端放大器，第三输出端连接所述控制器，用于控制感光阵列工作时序；所述控制器与所述时序控制器连接，用于控制感光阵列的工作频率；所述驱动器输入端与所述控制器连接，输出端与所述感光组件连接，用于驱动所述感光组件工作。
优选地，所述感光组件是一个感光元件或多个感光元件构成的感光阵列。
优选地，所述测量模块还包括存储器和稳压电路；
所述存储器与所述处理器连接，用于存储预设指令和血压数据；
所述稳压电路分别连接所述处理器和模拟前端放大器，用于稳定电源，减小外部电路环境带来的影响。
另一方面，本发明提供一种生理信息检测模型构建方法，所述方法包括：
    步骤1）、利用智能移动设备的照明设备对待测个体的身体部位末端进行照射；
    步骤2）、利用所述智能移动设备的传感设备采集待测个体的多帧连续图像或反射信号；
    步骤3）、基于所述图像或反射信号提取末端PPG信号并对所述末端PPG信号进行处理；
    步骤4）、构建UNet-ANN多支路反馈加强模型，所述UNet-ANN多支路反馈加强模型包括非对称的UNet主干网络和多分支ANN网络，所述多分支ANN网络中的每个分支分别用于进行一种生理信息参数的检测；
    步骤5）、利用经标注的指端PPG信号对所述加强模型进行训练。
优选地，对所述末端PPG信号进行处理的步骤包括：
    步骤2.1:将所获得的视频分帧，将每一帧图像的R通道像素值加权求和，组合成一段连续的 “PPG信号”；
    步骤2.2:将PPG信号进行0.5-8HZ椭圆带通滤波处理，检测异常点并清洗掉异常点；
    步骤2.3:将 PPG信号进行分窗；
    步骤2.4:对清洗分窗过后的PPG数据进行数据段提取和检验，若校验合格则进行下一步，否则重新采集数据；
    其中PPG信号的提取和检验步骤包括：
    步骤3.1:对处理后的PPG信号查找峰值点，找到PPG信号中的峰值点和谷值点；
    步骤3.2：由步骤3.1找出的峰值点、谷值点的最大值max，最小值min通过公式d=（max-min）/min分别计算表征峰值点的波动情况的d_peak和表征谷值点的波动情况的d_valley，若d_peak<0.05且d_valley<0.08，则保留该段PPG数据；
    步骤3.3：基于峰值点、谷值点的数量，判断当前信号窗口内包含的周期数是否少于预定值，若少于则采用第二提取方案；
第二提取方案：
    步骤3.4将处理后的PPG信号划分成单个周期的窗口
    步骤3.5：计算相邻周期波形的相关系数r ，若r>相关系数阈值，则保留该段PPG信号，否则舍弃；
    步骤3.6：若步骤2处理后的PPG信号对于两种提取标准皆不满足，则直接删除该段PPG信号。
另一方面，本发明提供一种基于智能移动设备的检测方法，所述方法用于进行生理参数信息检测，所述方法包括：
步骤1）、利用智能移动设备的照明设备对待测个体的身体部位末端进行照射；
    步骤2）、利用所述智能移动设备的传感设备采集待测个体的多帧连续图像或反射信号；
    步骤3）、基于所述图像或反射信号提取末端PPG信号并对所述末端PPG信号进行处理；
    步骤4）、将经处理的信号输入到权利要求13或14所构建的UNet-ANN多支路反馈加强模型中；
    步骤5）、将所述UNet-ANN多支路反馈加强模型的输出作为相应生理参数信息检测的结果。
 需要说明的是，本发明中所提到的“身体特定部位”“身体末端”指的是能够进行血液相关信息采集的身体部位，比如，手指尖、耳垂或者脚趾尖等。
 需要说明的是，本发明中所提到的“智能移动设备”指的是手机、平板电脑、Ipod等非腕带加压式的移动设备，其具有照明光源和摄像设备。
 有益效果
本发明开创性的提出了利用手机、平板等智能移动设备的摄像头和照明设备的组合获取待测个体末端（比如，指端、耳端）的视频图像，并从中提取光电容积描记（PPG）信号，对所获取的视频帧进行加权求和转换为PPG信号采样点。本发明利用获取的PPG信号利用多支路反馈加强模型获取多生理参数指标，可以同时实现基于手机的、无需专业设备的多种生理参数的同步测量。
相比于指夹式脉搏血氧仪这种固定在指端，有定波长段发射光（通常是绿光或者红光与近红外光）和对应光电传感器获取的PPG信号，从手机等智能设备的摄像头中获取的PPG信号较为不稳定，所以，目前现有设备和方法中都无法利用从智能设备摄像头中获取的指端连续视频帧中提取的PPG进行有效的血氧、血压等参数进行测定。但是，本发明通过对数据的处理和模型的设计调整，实现了基于智能设备采集的PPG信号的提取和利用，并且经实验验证，具有较高的准确率。
本发明的多支路反馈加强模型由Unet主干网络配合多个ANN支路共同预测心率，血氧，血压值。在模型的预训练阶段，先利用心率，血氧两支路的反馈优化Unet主干网络。然后固定主干网络，在此基础上训练血压预测支路。
此外，本发明利用前置或后置摄像头获取的人像预测的性别，年龄信号修正模型预测结果。
注：虽然血压是一个个体差异较大的生理指标，但在大样本下其与个体年龄，性别等信号还是有一定统计学意义的。结合从人像中获取的个体信息，如果模型预测生理指标与统计学预测结果差异过大，可以提示受试者进行二次测量。三次测量结果依旧没有太大波动即可。本发明的系统和方法一定程度上避免了一些偶然误差和极端结果。
 附图说明
图1是利用本发明实施例1的检测系统进行参数检测的流程图；
图2是用于执行本发明方法的手机的结构示意图；
图3是本发明实施例1中重构指端PPG信号的过程示意图；
图4是本发明两个实施例所采用的模型的整体框架的示意图；
图5是本发明两个实施例所采用的模型的主干网络Unet详细框架图；
图6是本发明两个实施例所采用的模型的分支网络详细框架图；
图7为本发明实施例2中检测系统的结构示意图；
图8为本发明实施例2中测量模块的结构示意图；
图9为本发明实施例2中模拟前端放大器的结构示意图；
图10为本发明实施例2中时序控制电路的结构示意图。
其中，1-测量模块、2-主动发光显示屏、11-手指、121-入射光、122-反射或散射光、21-感光组件、22-处理器、23-存储器、24-输入输出接口、32-模拟前端放大器、36-时序控制电路、37-稳压电路、41-信号放大器、42-模数转换器、43-数字滤波器、51-时序控制器、52-控制器、53-驱动器。
 本发明的最佳实施方式
为了使本领域的技术人员更好地理解本申请的技术方案，以下将结合附图及实施例对本实用新型做进一步详细说明。
实施例1
本实施例的基于智能移动设备的检测系统以智能手机为例进行描述，当然，也可以采用平板电脑等其他智能移动设备来实现本发明的系统。智能移动设备需要具有位于同一侧的照明设备和摄像设备，目前主要采用其后置摄像头，这样才能够在同一侧既实现照明，又实现图像的采集。当然，对于一些特殊情况，也可以采用辅助光源进行照明，然后利用移动设备的摄像头进行图像采集。
如图2所示，本实施例中，照明设备采用手机的闪光灯200，摄像设备采用手机的后置摄像头100。本实施例的检测系统包括：信号采集模块300和信号处理模块400。本领域技术人员应该理解，信号采集模块300和信号处理模块400可以均由手机中的处理器实现。
闪光灯200用于对人体的指端或耳端进行照射；摄像头100用于采集被照射部位的视频图像。信号采集模块用于获取所述摄像设备获得的20-40s，比如30s内的多帧图像，并基于所述图像提取PPG信号；所述信号处理模块中包含有UNet-ANN多支路反馈加强模型，所述UNet-ANN多支路反馈加强模型基于所述PPG信号进行多参数生理信息检测。
信号处理模块搭载多支路反馈加强算法网络的应用程序，在移动设备的处理器内运行。本发明可以利用手机摄像头，灯光和应用程序实现人体生理指标（心率，血氧，血压）的测量，保存与分析系统。参照图1，具体模型训练及参数测量如下：
S1：数据采集
步骤1.1:令被测人员稍作休息，确保自身各项生理指标处于平稳状态；
步骤1.2:将手机视频录制设置为60fps或根据精度进行调整。
步骤1.3:用手机前置摄像头采集人脸视频并保存（该步骤只需在任意受试者第一次使用该功能时执行一次）。需要说明的是，步骤1.1-1.3可以在系统中单独设置登录设置模块来实现，为可选模块，主要用于进行人员信息登录确认以及手机参数调整。
步骤1.4:将手指按于手机灯光及与其相连的后置摄像头，保持安静平稳的状态，采集指端连续图像帧30秒，如图3所示。
S2：数据处理
步骤2.1:将步骤1.3中得到的指端视频分帧，将每一帧图像的R通道像素值加权求和，最后组合成一段连续的指端“PPG信号”，如图3所示。
步骤2.2:将指端“PPG信号”进行0.5-8Hz椭圆带通滤波处理，检测异常点并清洗，去除异常点。
步骤2.3:将30秒的“PPG信号”进行分窗，窗宽256个采样点，步长150个采样点。
步骤2.4:判断清洗分窗过后的数据进行数据段提取并按照步骤S3进行检验，若满足提取标准则进行下一步，否则返回1.3重新采集数据
S3：PPG信号的提取：
第一提取方案
步骤3.1:将处理后的PPG信号通过查找峰值点，谷值点算法找到PPG信号中的峰值点和谷值点。
步骤3.2：由步骤3.1找出的峰值点、谷值点的最大值（max），最小值（min）通过公式d=（max-min）/min分别计算d_peak(表征峰值点的波动情况)和d_valley(表征谷值点的波动情况)，若d_peak<0.05且d_valley<0.08（其中0.05，0.08是经验值，会随着设备变化而不同），则保留该段PPG数据；
步骤3.3：基于峰值点、谷值点的数量，判断当前信号窗口内包含的周期数是否少于预定值（峰值点或估值点不超过2个），则采用第二提取方案。  
第二提取方案：
3.4将预处理后的PPG信号划分成多个单周期的窗口（每一段数据长度相同都是256个采样点，每个周期窗口对应于数据测量者的一个心跳周期）。
步骤3.5：计算相邻周期波形的相关系数r（采用现有的相关算法即可），若r>0.85(0.85是经验值)则保留该段PPG信号。
步骤3.6：若步骤2处理后的PPG信号对于以上两种提取标准皆不满足，则直接删除该段PPG信号。
S4：模型预测
步骤4.1:调用openCV库从步骤1.2中获取的人脸视频中判断受试者的性别、年龄信息。
步骤4.2: 将预处理后的指端“PPG信号”输入预训练好的UNet-ANN多支路反馈加强模型得到预测心率，血氧，血压值。该模型的具体结构在下面进一步详细论述。
步骤4.3: 结合4.1获取的受试者信息调整3.2模型。
S5：结果保存
步骤5.1:选择是否保存已测量结果
步骤5.2:若5.1选择保存，则将结果保存至个人数据中并进行一个统计分析。
注：心率，血氧，血压预测模型详细的介绍
本实施例中的UNet-ANN多支路反馈加强模型分主干网络和分支网络两大部分。
主干网络采用Unet网络，分支网络采用ANN分支网络，本实施例中采用单Unet网络，四ANN分支网络的结构构型。
Unet网络结构：
原始的Unet是一个对称结构，本发明中的Unet作为特征提取模块对其进行了调整，采用了非对称结构。Unet网络最初是为了解决医学图像分割问题而创建的，Unet网络利用U型网络结构获取上下文信息和位置信息。本实施例中，该非对称Unet主干网络主要分左右两个模块。左边模块包括四层，每层主要由4个两级3*1的卷积核加2*1下采样子模块组成（各层的卷积核和子模块数量可调）。右边模块也包括四层，每层由4个两级3*1的卷积核加2*1上采样子模块组成，左右模块之间每层通过拼接相连，从最右侧进行输出。而非对称性体现在输入层与输出层的差别，输入层是1通道256采样点长度的PPG信号，而输出层是1024通道2采样点长度的多特征信号。本发明中的改进型Unet结构用一维卷积代替了图像的二维卷积，在最后一层（右侧模块的最上层）用多个输出通道代替原Unet中单通道以获取多个特征值，分别送入到各个ANN分支网络。
如图5-6所示，以1通道256采样点长度的PPG信号作为输入，将其输入主干网络Unet，在两级3*1的卷积网络下变为64通道256采样点长度的信号；该信号经2*1下采样网络变为64通道128采样点长度的信号；在两级3*1的卷积网络下变为128通道128采样点长度的信号；经2*1下采样网络变为128通道64采样点长度的信号；在两级3*1的卷积网络下变为256通道64采样点长度的信号；经2*1下采样网络变为256通道32采样点长度的信号；在两级3*1的卷积网络下变为512通道32采样点长度的信号；经2*1下采样网络变为512通道16采样点长度的信号；在两级3*1的卷积网络下变为1024通道16采样点长度的信号；经2*1上采样网络变为512通道32采样点长度的信号；该512通道32采样点长度的信号与左侧512通道32采样点长度的信号拼接形成1024通道32采样点长度的信号；在两级3*1的卷积网络下变为512通道32采样点长度的信号；经2*1上采样网络变为256通道64采样点长度的信号；该256通道64采样点长度的信号与左侧256通道64采样点长度的信号拼接形成512通道64采样点长度的信号；在两级3*1的卷积网络下变为256通道64采样点长度的信号；经2*1上采样网络变为128通道128采样点长度的信号；该128通道128采样点长度的信号与左侧128通道128采样点长度的信号拼接形成256通道128采样点长度的信号；在两级3*1的卷积网络下变为128通道128采样点长度的信号；经2*1上采样网络变为64通道256采样点长度的信号；该64通道256采样点长度的信号与左侧64通道256采样点长度的信号拼接形成128通道256采样点长度的信号；经4*1下采样网络变为128通道64采样点长度的信号；在5*1，步长为3的卷积网络下变为256通道20采样点长度的信号；又在5*1，步长为3的卷积网络下变为512通道6采样点长度的信号；最后在3*1，步长为2的卷积网络下变为1024通道2采样点长度的信号。
ANN分支网络：
分支网络的处理主要是基于主干网络输出层1024通道2采样点长度的多特征信号进行特征组合处理。对于心率和血氧饱和度而言，经过多层卷积的特征处理得到预测值。血压值的高压低压值是一个相关联的数据，他们分别由1024通道2采样点长度的特征信号的一个维度经过卷积处理得到预测值。
2.1如图6所示，最上方支路中，1024通道2采样点长度的信号经过2*1的卷积网络变为1024通道1采样点长度的信号；1024通道1采样点长度的信号经过1*1的卷积网络变为64通道1采样点长度的信号；64通道1采样点长度的信号经过1*1的卷积网络变为1通道1采样点长度的信号，即预测的心率值。
2.2 在第二条支路中，利用Unet主干网络最后输出的1024通道2采样点长度的信号来预测血氧
1024通道2采样点长度的信号经过2*1的卷积网络变为1024通道1采样点长度的信号；1024通道1采样点长度的信号经过1*1的卷积网络变为256通道1采样点长度的信号；256通道1采样点长度的信号经过1*1的卷积网络变为64通道1采样点长度的信号；64通道1采样点长度的信号经过1*1的卷积网络变为1通道1采样点长度的信号，即预测的血氧值。
2.3第三和第四之路中，利用Unet主干网络最后输出的1024通道2采样点长度的信号来预测血压
取第一维数据1024通道1采样点长度的信号经过1*1的卷积网络变为256通道1采样点长度的信号；256通道1采样点长度的信号经过1*1的卷积网络变为64通道1采样点长度的信号；64通道1采样点长度的信号经过1*1的卷积网络变为1通道1采样点长度的信号，即预测的收缩压。
取第二维数据1024通道1采样点长度的信号经过1*1的卷积网络变为256通道1采样点长度的信号；256通道1采样点长度的信号经过1*1的卷积网络变为64通道1采样点长度的信号；64通道1采样点长度的信号经过1*1的卷积网络变为1通道1采样点长度的信号，即预测的舒张压。
需要说明的是，虽然上述给出了通道、采样点的具体数目，但是本领域技术人员在实际使用中，可以根据需要调整通道数目和采样点数目，这里不作限制。
实验测试，通过本发明所构建的模型，发明人实际测试，选取训练集100人，验证集50人，测试集50人，最终测试效果证明可以应用于实际应用中。
心率准确率：90%以上的概率落在+/-3bpm的范围内
血氧饱和度准确率：90%以上的概率落在+/-2%的范围内
血压准确率：高压SBP85%以上的概率落在+/-10bpm的范围内
低压DBP95%以上的概率落在+/-10bpm的范围内。
实施例2
本实施例中，以通过智能设备的主动发光显示屏作为照明设备为例来描述采用智能设备的前面板进行生理信息检测的过程。
本实施例的检测系统可以采用带显示屏的智能设备来实现，显示屏可以是有机发光二极管(Organic Light Emitting Diode)显示屏和微型发光二极管（Micro-LED）显示屏，智能设备可以是智能手机、智能手表、平板电脑等，但不限定于此。
如图7-10所示，本实施例给出了另一种智能设备采集数据过程的示意。本实施例中没有采用实施例1的手机后置摄像头进行图像采集，而是采用设置于主动发光显示屏2下方的测量模块1进行手指11上的反射或散射光122检测。本实施例中，照明设备采用的是主动发光显示屏2。
当待测人员将手指 11放在主动发光显示屏2上的指定的局部区域之后，主动发光显示屏2作为照明设备向手指发射入射光121，入射光121照射手指11后经反射或散射得到反射或散射光122。测量模块1设置在主动发光显示屏12的下方，用于获取反射或散射光122以获取一定时间内的时域上的多个电流值，并将多个电流值生成为PPG信号，最后利用所述PPG信号通过计算生成血压值。
测量模块1包括感光组件21、处理器22、存储器23和输入输出接口24。感光组件21设置在主动发光显示屏2下方，用于获取具有人体生理特征的光信号并转换为电信号；感光组件21可以为具有多个感光元件构成的感光阵列设置于主动发光显示屏2下方，实现多位置血压测量或者半屏乃至全屏血压测量。处理器22与感光组件21连接，处理器22内置有PPG算法或模型，用于将感光组件21输出的电信号通过算法合成为PPG信号，计算得到血压值。存储器23与处理器22连接，用于存储所设计的算法指令和用户生理数据。输入输出接口24与处理器22连接，用于接收工作指令以及血压测量完成时向外部设备传输血压数据。
处理器22内置的PPG检测模型可以为实施例1中所构建的UNet-ANN多支路反馈加强模型，只是该模型需要基于本实施例中设备采集的信号进行训练并使用。
在一种优选实现方式中，测量模块1还包括模拟前端放大器32、时序控制电路36和稳压电路37。模拟前端放大器32输入端与感光组件21连接，用于放大各个感光阵列单元生成的单元电流值，对各个单元电流值进行模数转换和电路去噪，生成各个预处理单元电流值；输出端与处理器22连接。时序控制电路36分别连接感光组件21、处理器22和模拟前端放大器32，用于控制测量模块1中器件的工作时序和感光阵列的工作频率。稳压电路37分别连接处理器22和模拟前端放大器32，用于稳定测量模块1中器件的电源，减小外部电路环境带来的影响。
具体的，模拟前端放大器32具有一个数据输入端和一个数据输出端，数据输入端接收来自感光组件21的各个感光阵列单元按工作频率在测量时间内的各个频率时间点上生成的单元电流值，数据输出端连接至处理器22，输出各个频率时间点上的各个预处理单元电流值。模拟前端放大器32内部电路包括顺次连接的信号放大器41，用于放大各个感光阵列单元生成的微弱的单元电流值，便于后续信号处理和增大同一感光阵列单元不同频率时间点单元电流值的数值差别；模数转换器42，用于将各个单元电流值从模拟信号转换成数字信号，便于处理器计算分析；数字滤波器43，用于滤除混杂在各个单元电流值中的噪声。
进一步的，信号放大器41采用运算放大器，线性放大信号，同时在一个频率时间点模拟前端放大器32向处理器22发送一次各个预处理单元电流值。
信号放大器41也可以采用晶体管放大器和功率放大器，模拟前端放大器32可以在测量时间结束时将各个频率时间点的各个预处理单元电流值统一发送给处理器22。
时序控制电路36包括：时序控制器51，用于控制测量模块中部件的工作时序，具有三个输出端，第一输出端连接处理器22，用于控制处理器22工作时序；第二输出端连接模拟前端放大器32，用于控制模拟前端放大器32工作时序；第三输出端通过控制器52和驱动器53与感光组件21连接，用于控制感光阵列工作时序；控制器52与时序控制器51连接，用于控制感光阵列的工作频率；驱动器53与控制器52连接，用于接收控制器52的控制指令，驱动感光阵列工作。
在一种优选实现方式中，感光组件21采用豪威科技的图像CMOS传感器OV9782；处理器22采用TI德州仪器公司的DM505  SoC，输入输出接口24集成设置在处理器22的DM505  SoC上；存储器23采用华邦电子W25X05C LFLASH存储器。模拟前端放大器32采用德州电子的单芯片生物传感信号采集模拟前端AFE4900，AFE4900芯片内部集成了信号放大器41、模数转换器42和数字滤波器43。时序控制电路36采用亚德诺半导体的ADM1066数字时序控制器，ADM1066数字时序控制器内部集成了时序控制器51、控制器52和驱动器53。稳压电路37采用同步升压转换器TPS61099与同相降压-升压转换器TPS63036；其中，AFE4900芯片具有三个电源输入，TPS61099用于提供一个电源的稳压，TOS6303提供另外两个电源的稳压。
在一种优选实现方式中，血压测量系统还包括蓝牙模块和WiFi模块，用于无线式用户血压数据传输。
感光阵列的工作频率大于等于180赫兹，感光阵列一个频率时间点各个感光阵列单元生成单元电流值。
处理器22获取各个频率时间点的各个预处理单元电流值数据，在本实施例中感光阵列包含多个感光阵列单元，处理器22对一个频率时间点上的各个预处理单元电流值进行加权平均或算数平均运算，以获取参考电流值，所述参考电流值用于生成PPG信号的一个值。在本实施例中，处理器22结合各个频率时间点上的各个预处理单元电流值，找到数值变化最明显的感光阵列单元，将感光阵列单元一个频率时间点的预处理单元电流值为一个参考电流值；或者找到数值变化最明显的几个感光阵列单元，处理器对所述几个感光阵列单元一个频率时间点的预处理单元电流值进行加权平均或算数平均运算，以获取参考电流值，所述参考电流值用于生成PPG的一个值。
或者，感光阵列仅包含一个感光阵列单元，处理器22将唯一的感光阵列单元在一个频率时间点上的预处理单元电流值为一个参考电流值，所述参考电流值用于生成PPG信号的一个值。
处理器22得到各个频率时间点的参考电流值后去除各个参考电流值中的直流部分，以获取各个参考交流电流值，根据各个参考交流电流值生成PPG信号，并对生成的PPG信号进行滤波去噪，减少噪声对计算血压值带来的干扰，最后根据PPG信号波形特征与血压的相关性通过计算得到用户血压值。
以血压测量为例，血压测量的工作流程包括：未工作时系统处于待机节能状态，即感光阵列未启动、系统持续等待工作信号的状态。用户选择血压测量，将手指按在显示屏指定区域8秒以上，血压测量系统收到工作信号，接收智能设备显示屏发出并通过手指反射或散射的光信号，以获取各个感光阵列单元按工作频率在测量时间内的各个频率时间点上生成的单元电流值，各个频率时间点上各个感光阵列单元生成的单元电流值用于形成光电容积脉搏波描记PPG信号，PPG信号波形形态与血压的相关性用于得到用户的血压信息。利用所测得的PPG信号，输入到已经训练好的的UNet-ANN多支路反馈加强模型，或者送入到其他基于PPG信号的血压监测模型中，进行血压数据计算。
对于PPG信号的处理方法，本实施例与实施例1的处理方法相同，这里不再详细介绍。
测量结束后，本系统自动存储和更新用户血压数据，并及时向外部设备传输当前用户血压状态，结束后进入待机状态，即等待下一个工作信号。
虽然上面结合实施例对本发明进行了详细描述，本发明不仅局限于上述具体实施方式，本领域一般技术人员根据实施例和附图公开内容，可以采用其它多种具体实施方式实施本发明的方法和系统，因此，凡是采用本发明的设计结构和思路，进行简单的变换或更改的设计，都落入本发明保护的范围。

Claims (15)

  1. 一种基于智能移动设备的检测系统，其特征在于，所述智能移动设备包括位于所述智能移动设备同一侧的照明设备和传感设备，所述检测系统包括：信号采集模块和信号处理模块，
        所述照明设备用于对待测个体的身体特定部位进行照射；
        所述传感设备用于采集被照射部位的图像或反射信号；
        所述信号采集模块用于获取预定时间内的多帧图像或反射信号，并基于所述图像或反射信号提取PPG信号；
        所述信号处理模块中包含有UNet-ANN多支路反馈加强模型，所述UNet-ANN多支路反馈加强模型基于所述PPG信号进行多参数生理信息检测，所述UNet-ANN多支路反馈加强模型包括非对称的UNet主干网络和多分支ANN网络，所述多分支ANN网络中的每个分支分别用于进行一种生理信息参数的检测。
  2. 根据权利要求1所述的多参数生理信息检测系统，其特征在于，还包括数据处理模块，所述数据处理模块用于将所采集的图像进行分帧处理，将每一帧图像的R通道像素值加权求和，最后组合成一段连续的“PPG信号”。
  3. 根据权利要求1所述的多参数生理信息检测系统，其特征在于，所述UNet主干网络中输入信号的通道数小于采样点数，输出信号的通道数大于采样点数。
  4. 根据权利要求1所述的多参数生理信息检测系统，其特征在于，所述生理信息参数包括心率、血氧、收缩压以及舒张压中的一种或多种。
  5. 根据权利要求4所述的多参数生理信息检测系统，其特征在于，所述UNet-ANN多支路反馈加强模型基于待测个体的心率和血氧的真实值与检测值之差作为反馈条件进行UNet主干网络的参数调整。
  6. 根据权利要求1所述的多参数生理信息检测系统，其特征在于，所述传感设备为摄像设备，所述图像为视频图像，所述摄像设备还用于拍摄待测个体的人脸视频，并且基于所述人脸视频进行待测个体的性别判定。
  7. 根据权利要求1所述的多参数生理信息检测系统，其特征在于，所述数据处理模块用于将指端“PPG信号”进行0.5-8HZ椭圆带通滤波处理，检测异常点并清洗；将30秒的“PPG信号”中进行分窗，窗宽256采样点，步长150采样点；判断清洗分窗过后的数据是否有符合数据质量要求的数据段，并提取相应数据段。
  8. 根据权利要求1所述的检测系统，其特征在于，所述照明设备为具有发光功能的显示屏，所述身体特定部位为手指，所述显示屏用于与手指接触，产生所述手指的反射或散射光，所述传感设备为具有感光组件的测量模块；
          所述感光组件设置在所述主动发光显示屏下方，用于将所述反射或散射光转换为电信号；
          所述处理器与所述感光组件连接，用于将所述电信号合成为PPG信号。
  9. 根据权利要求8所述的检测系统，其特征在于，所述信号采集模块包括模拟前端放大器；
          所述模拟前端放大器包括信号放大器、模数转换器和数字滤波器；
          所述信号放大器输入端与所述感光组件连接，用于放大所述感光组件生成的单元电流值；
          所述模数转换器与所述信号放大器连接，用于将所述单元电流值从模拟信号转换成数字信号；
          所述数字滤波器输入端与所述模数转换器连接，所述数字滤波器输出端与所述处理器连接，用于滤除混杂在各个单元电流值中的噪声。
  10. 根据权利要求9所述的检测系统，其特征在于，所述测量模块还包括时序控制电路；
    所述时序控制电路包括时序控制器、控制器和驱动器；所述时序控制器具有三个输出端，第一输出端连接所述处理器，第二输出端连接所述模拟前端放大器，第三输出端连接所述控制器，用于控制感光阵列工作时序；所述控制器与所述时序控制器连接，用于控制感光阵列的工作频率；所述驱动器输入端与所述控制器连接，输出端与所述感光组件连接，用于驱动所述感光组件工作。
  11. 根据权利要求9所述的检测系统，其特征在于，所述感光组件是一个感光元件或多个感光元件构成的感光阵列。
  12. 根据权利要求9所述的检测系统，其特征在于，所述测量模块还包括存储器和稳压电路；
          所述存储器与所述处理器连接，用于存储预设指令和血压数据；
          所述稳压电路分别连接所述处理器和模拟前端放大器，用于稳定电源，减小外部电路环境带来的影响。
  13. 一种生理信息检测模型构建方法，其特征在于，所述方法包括：
        步骤1）、利用智能移动设备的照明设备对待测个体的身体部位末端进行照射；
        步骤2）、利用所述智能移动设备的传感设备采集待测个体的多帧连续图像或反射信号；
        步骤3）、基于所述图像或反射信号提取末端PPG信号并对所述末端PPG信号进行处理；
        步骤4）、构建UNet-ANN多支路反馈加强模型，所述UNet-ANN多支路反馈加强模型包括非对称的UNet主干网络和多分支ANN网络，所述多分支ANN网络中的每个分支分别用于进行一种生理信息参数的检测；
        步骤5）、利用经标注的指端PPG信号对所述加强模型进行训练。
  14. 根据权利要求13所述的方法，其特征在于，
        对所述末端PPG信号进行处理的步骤包括：
        步骤2.1:将所获得的视频分帧，将每一帧图像的R通道像素值加权求和，组合成一段连续的 “PPG信号”；
        步骤2.2:将PPG信号进行0.5-8HZ椭圆带通滤波处理，检测异常点并清洗掉异常点；
        步骤2.3:将 PPG信号进行分窗；
        步骤2.4:对清洗分窗过后的PPG数据进行数据段提取和检验，若校验合格则进行下一步，否则重新采集数据；
        其中PPG信号的提取和检验步骤包括：
        步骤3.1:对处理后的PPG信号查找峰值点，找到PPG信号中的峰值点和谷值点；
        步骤3.2：由步骤3.1找出的峰值点、谷值点的最大值max，最小值min通过公式d=（max-min）/min分别计算表征峰值点的波动情况的d_peak和表征谷值点的波动情况的d_valley，若d_peak<0.05且d_valley<0.08，则保留该段PPG数据；
        步骤3.3：基于峰值点、谷值点的数量，判断当前信号窗口内包含的周期数是否少于预定值，若少于则采用第二提取方案；
        第二提取方案：
        步骤3.4将处理后的PPG信号划分成单个周期的窗口
        步骤3.5：计算相邻周期波形的相关系数r ，若r>相关系数阈值，则保留该段PPG信号，否则舍弃；
        步骤3.6：若步骤2处理后的PPG信号对于两种提取标准皆不满足，则直接删除该段PPG信号。
  15. 一种基于智能移动设备的检测方法，所述方法用于进行生理参数信息检测，其特征在于，所述方法包括：
        步骤1）、利用智能移动设备的照明设备对待测个体的身体部位末端进行照射；
        步骤2）、利用所述智能移动设备的传感设备采集待测个体的多帧连续图像或反射信号；
        步骤3）、基于所述图像或反射信号提取末端PPG信号并对所述末端PPG信号进行处理；
        步骤4）、将经处理的信号输入到权利要求13或14所构建的UNet-ANN多支路反馈加强模型中；
        步骤5）、将所述UNet-ANN多支路反馈加强模型的输出作为相应生理参数信息检测的结果。
PCT/CN2023/082388 2022-03-23 2023-03-20 一种基于智能移动设备的检测系统、方法及检测模型构建方法 WO2023179516A1 (zh)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202220656504.9U CN218500702U (zh) 2022-03-23 2022-03-23 一种用于智能设备的屏下血压测量系统
CN202220656504.9 2022-03-23
CN202210393707.8 2022-04-14
CN202210393707.8A CN115054209B (zh) 2022-04-14 2022-04-14 一种基于智能移动设备的多参数生理信息检测系统和方法

Publications (1)

Publication Number Publication Date
WO2023179516A1 true WO2023179516A1 (zh) 2023-09-28

Family

ID=88099945

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/082388 WO2023179516A1 (zh) 2022-03-23 2023-03-20 一种基于智能移动设备的检测系统、方法及检测模型构建方法

Country Status (1)

Country Link
WO (1) WO2023179516A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111870237A (zh) * 2020-09-04 2020-11-03 平安科技（深圳）有限公司 血压检测方法、装置、设备及介质
CN112836546A (zh) * 2019-11-22 2021-05-25 深圳市理邦精密仪器股份有限公司 检测生理信号质量的方法、装置及电子设备
CN113375676A (zh) * 2021-05-26 2021-09-10 南京航空航天大学 一种基于脉冲神经网络的探测器着陆点定位方法
WO2021259604A1 (en) * 2020-06-22 2021-12-30 Agfa Healthcare Nv Domain aware medical image classifier interpretation by counterfactual impact analysis
CN115054209A (zh) * 2022-04-14 2022-09-16 杭州华视诺维医疗科技有限公司 一种基于智能移动设备的多参数生理信息检测系统和方法
CN218500702U (zh) * 2022-03-23 2023-02-21 杭州华视诺维医疗科技有限公司 一种用于智能设备的屏下血压测量系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112836546A (zh) * 2019-11-22 2021-05-25 深圳市理邦精密仪器股份有限公司 检测生理信号质量的方法、装置及电子设备
WO2021259604A1 (en) * 2020-06-22 2021-12-30 Agfa Healthcare Nv Domain aware medical image classifier interpretation by counterfactual impact analysis
CN111870237A (zh) * 2020-09-04 2020-11-03 平安科技（深圳）有限公司 血压检测方法、装置、设备及介质
CN113375676A (zh) * 2021-05-26 2021-09-10 南京航空航天大学 一种基于脉冲神经网络的探测器着陆点定位方法
CN218500702U (zh) * 2022-03-23 2023-02-21 杭州华视诺维医疗科技有限公司 一种用于智能设备的屏下血压测量系统
CN115054209A (zh) * 2022-04-14 2022-09-16 杭州华视诺维医疗科技有限公司 一种基于智能移动设备的多参数生理信息检测系统和方法

Similar Documents

Publication Publication Date Title
Nemcova et al. Monitoring of heart rate, blood oxygen saturation, and blood pressure using a smartphone
Sun et al. Photoplethysmography revisited: from contact to noncontact, from point to imaging
Wang et al. Multichannel reflective PPG earpiece sensor with passive motion cancellation
CN104640498A (zh) 移动心脏健康监视
Siddiqui et al. A pulse rate estimation algorithm using PPG and smartphone camera
CN105813564A (zh) 用于确定对象的生命体征的设备和方法
CN109512409B (zh) 一种心血管健康评估装置及其使用方法
Lamonaca et al. Application of the artificial neural network for blood pressure evaluation with smartphones
CN106419880A (zh) 穿戴式血压监测装置
US20180008200A1 (en) Motion-dependent averaging for physiological metric estimating systems and methods
US20190246917A1 (en) Image Based Blood Pressure Monitoring Method
CN112890792A (zh) 一种基于网络摄像头的云计算心血管健康监护系统及方法
JPWO2017188099A1 (ja) 装置、端末および生体情報システム
Ayesha et al. Heart rate monitoring using PPG with smartphone camera
CN115770027A (zh) 一种基于人工智能的血压检测方法及装置
CN115054209B (zh) 一种基于智能移动设备的多参数生理信息检测系统和方法
JP6615970B1 (ja) 血圧推定装置および血圧推定プログラム
WO2023179516A1 (zh) 一种基于智能移动设备的检测系统、方法及检测模型构建方法
TWI563969B (zh) 生理資訊之計算機分析方法及裝置
Phan et al. Noninvasive, wearable multi biosensors for continuous, long-term monitoring of blood pressure via internet of things applications
US11653847B2 (en) Method and apparatus for hypertension classification
CN114557684A (zh) 一种腕表式脉搏心电同步采集显示装置
CN212165772U (zh) 一种血压测量装置
Khong et al. Contact and non-contact heart beat rate measurement techniques: Challenges and issues
Itagi et al. Analysis of blood pressure through video Techniques imaging

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23773764

Country of ref document: EP

Kind code of ref document: A1