Blood pressure testing methods and refrigerators
By identifying users and collecting photoplethysmography (PPG) signals on a refrigerator, and utilizing deep learning models and blood pressure correction formulas, the problem of low accuracy in blood pressure detection has been solved, enabling cuffless, convenient blood pressure monitoring and closed-loop management services.
Patent Information
- Application Number
- CN202610246230.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-28
- Publication Date
- 2026-05-26
AI Technical Summary
Existing blood pressure monitoring methods suffer from low accuracy and unclear trend tracking. Traditional methods, such as invasive pressure measurement and cuff pressure measurement, are invasive and costly, while emerging methods, such as millimeter-wave radar and ultrasound measurement, are costly in terms of hardware and have low measurement accuracy.
By identifying user identity information, matching and calibrating photoplethysmography (PPG) signals and cuff pressure standards, collecting actual signals for similarity analysis, using deep learning models for reasoning, and performing precise corrections using blood pressure correction formulas, cuffless detection is achieved by combining the signal acquisition structure on the refrigerator.
It enables cuffless and convenient blood pressure monitoring, improves detection accuracy and trend tracking capabilities, provides real-time blood pressure monitoring services, supports dietary management and exercise recommendations, and forms a closed-loop blood pressure management service.
Smart Images
Figure CN122074930A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of home appliance technology, and in particular to a method for detecting blood pressure and a refrigerator. Background Technology
[0002] In China, over 300 million people suffer from hypertension, with a prevalence rate as high as 27.5%. However, the awareness rate among hypertension patients is only 47%, and the control rate is less than 18%. Hypertension control and prevention are strongly correlated with dietary management. Traditional blood pressure monitoring methods include invasive and cuff-based methods. However, invasive methods cause significant trauma to patients, while cuff-based methods are expensive, inconvenient, and the data cannot be used to scientifically guide dietary decisions. Emerging methods include millimeter-wave radar and ultrasound measurement, but these methods suffer from high hardware costs and low measurement accuracy. Summary of the Invention
[0003] One objective of this invention is to solve the problems of low accuracy and unclear trend tracking in user blood pressure detection, and to achieve accurate blood pressure detection without a cuff.
[0004] A further objective of this invention is to provide users with convenient and real-time blood pressure monitoring services, thereby enhancing the user experience.
[0005] Specifically, the present invention provides a method for detecting blood pressure, comprising: Identify the user's identity information, and match the corresponding calibration photoplethysmography (PPG) signal and cuff gold standard based on the identity information. The calibration PPG signal and the corresponding cuff gold standard under the user's high-voltage and low-voltage states are pre-acquired and saved. Collect the user's actual photoplethysmography (PPG) signal; The actual photoplethysmography (PPG) signal was compared with the calibrated PPG signals under high-voltage and low-voltage conditions, and the corresponding calibration value was selected based on the analysis results. The actual photoplethysmography (PPG) signal is input into a deep learning model corresponding to the calibrated values to obtain the inferred systolic and diastolic blood pressure. The deep learning models for different high-pressure sub-tasks are pre-built. The inferred systolic and diastolic blood pressures are corrected according to the preset blood pressure correction formula to obtain the final systolic and diastolic blood pressures.
[0006] Optionally, the steps of pre-acquiring the calibrated photoplethysmography (PPG) signal and the corresponding cuff gold standard include: The signal acquisition structure set on the refrigerator collects the photoplethysmography (PPG) signals of the user under high and low voltage conditions, and simultaneously obtains the corresponding cuff pressure. The quality of the acquired photoplethysmography (PPG) signals is assessed, and only PPG signals that meet the preset quality requirements and their corresponding cuff pressures are saved as the gold standard for calibrating PPG signals and their corresponding cuff pressures.
[0007] Optionally, the refrigerator door has small holes adapted to fit a user's fingertip, the holes are covered with uniformly transparent glass, and a signal acquisition structure is installed inside the holes. The signal acquisition structure includes a light source and a photodetector, with a light-shielding structure placed between the light source and the photodetector.
[0008] Optionally, the steps for acquiring photoplethysmography (PPG) signals include: When a user presses their finger on the transparent glass, a light source emits light towards the user's finger, which then passes through the skin and reaches the subcutaneous blood vessels; and A photodetector is used to receive light signals reflected from subcutaneous blood vessels and convert them into electrical signals to generate photoplethysmography (PPG) signals.
[0009] Optionally, the steps of pre-building a deep learning model include: The blood pressure range where the systolic pressure is greater than or equal to a preset threshold is divided into multiple intervals; The system divides multiple intervals into a predetermined number of subtasks, and constructs an independent deep learning model for each subtask within each interval. The deep learning model takes the calibrated photoplethysmography (PPG) signal under high-pressure conditions and the corresponding reference systolic and diastolic blood pressure as input. The reference systolic and diastolic blood pressures are derived from the cuff pressure gold standard. The one-hot encoding is obtained by using a bucket function based on a reference shrinkage pressure, and the output of the corresponding deep learning model is selected as the final output.
[0010] Optionally, the steps of performing similarity analysis between the actual photoplethysmography (PPG) signal and the calibrated PPG signals under high-voltage and low-voltage conditions, and selecting the corresponding calibration value based on the analysis results, include: Similarity analysis was performed using the Pearson correlation coefficient to obtain the high-pressure correlation coefficient and the low-pressure correlation coefficient; Determine whether the high-voltage correlation coefficient is greater than the low-voltage correlation coefficient; and If so, select the calibration photoplethysmography pulse wave signal under high voltage and the cuff pressure gold standard as the calibration values. If not, select the calibration photoplethysmography pulse wave signal and cuff pressure gold standard under low voltage conditions as calibration values.
[0011] Optionally, the steps for presetting the blood pressure correction formula include: Multiple acquisitions of photoplethysmography (PPG) signals and corresponding systolic and diastolic blood pressure were performed under different motion conditions of the user; and Feature parameters of the photoplethysmography (PPG) signal are extracted and fitted to obtain a blood pressure correction formula. The feature parameters include amplitude, frequency, waveform, energy spectrum, information entropy, and heart rate.
[0012] Alternatively, the blood pressure correction formula is: , , ΔSBP sport ΔDBP is the correction value for systolic blood pressure during exercise. sport f is the correction value for diastolic blood pressure during exercise. SBP f is the fitting function for systolic blood pressure. DBP Here, is the fitting function for diastolic blood pressure, where A is the amplitude, ω is the frequency, W is the waveform, ES is the energy spectrum, IE is the information entropy, and HR is the heart rate. The formulas for calculating final systolic blood pressure and final diastolic blood pressure are: , , SBP final For the final systolic pressure, DBP final For final diastolic blood pressure, SBP pre For inference of systolic blood pressure, DBP pre To infer diastolic blood pressure.
[0013] Optionally, the steps for saving the calibrated photoplethysmography (PPG) signal and the corresponding cuff gold standard include: Receive user touch or voice commands; and Wake up the information input interface, input and save the calibrated photoplethysmography (PPG) signal and the corresponding cuff gold standard.
[0014] According to another aspect of the present invention, a refrigerator is also provided, comprising: a controller, the controller including a memory and a processor, wherein the memory stores a machine-executable program, which, when executed by the processor, implements any of the above-described methods for detecting blood pressure.
[0015] The blood pressure detection method and refrigerator of this invention can significantly improve the accuracy and trend tracking ability of users' blood pressure detection. By using a dual calibration method, a multi-task deep learning model and exercise blood pressure tracking correction, it solves the core technical pain point of low accuracy in blood pressure inference from photoplethysmography (PPG) signals, and the detection results are more consistent with the user's real blood pressure status.
[0016] Furthermore, the blood pressure detection method and refrigerator of the present invention enable cuffless and convenient blood pressure monitoring. Relying on the contact acquisition structure of the refrigerator door, no additional special equipment is required. With the assistance of identity recognition and signal quality judgment, users can complete the test in real time and quickly. At the same time, it can provide reliable data support for subsequent dietary management recommendations, forming a closed-loop blood pressure management service.
[0017] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description
[0018] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings: Figure 1 This is a schematic diagram of a blood pressure detection method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall structure of a refrigerator according to an embodiment of the present invention; Figure 3 yes Figure 2 Enlarged view of region A in the middle; Figure 4 This is a schematic block diagram of a refrigerator controller according to an embodiment of the present invention; and Figure 5 This is a schematic diagram of a machine-readable storage medium according to an embodiment of the present invention. Detailed Implementation
[0019] This embodiment first provides a blood pressure detection method that can significantly improve the accuracy and trend tracking ability of users' blood pressure detection. By using a dual calibration method, a multi-task deep learning model, and exercise blood pressure following correction, it solves the core technical pain point of low accuracy in blood pressure inference from PPG signals, and the detection results are more consistent with the user's real blood pressure status. Figure 1 This is a schematic diagram of a blood pressure detection method according to an embodiment of the present invention. Figure 1 As shown, the blood pressure detection method may include the following steps: Step S102: Identify the user's identity information and match the corresponding calibrated photoplethysmography (PPG) signal and cuff gold standard based on the identity information. Step S104: Acquire the user's actual photoplethysmography (PPG) signal; Step S106: Perform similarity analysis between the actual photoplethysmography (PPG) signal and the calibrated PPG signals under high voltage and low voltage conditions, respectively, and select the corresponding calibration value based on the analysis results. Step S108: Input the actual photoplethysmography (PPG) signal into the deep learning model corresponding to the calibration value to obtain the inferred systolic blood pressure and inferred diastolic blood pressure. Step S110: Correct the inferred systolic blood pressure and inferred diastolic blood pressure according to the preset blood pressure correction formula to obtain the final systolic blood pressure and final diastolic blood pressure.
[0020] In the above steps, the calibration photoplethysmography (PPG) signals and corresponding cuff pressure standards under the user's high-voltage and low-voltage states in step S102 are pre-acquired and saved. Furthermore, in a preferred embodiment, the step of pre-acquiring the calibration PPG signals and corresponding cuff pressure standards may include: acquiring PPG signals under the user's high-voltage and low-voltage states through a signal acquisition structure set on the refrigerator, and simultaneously acquiring the corresponding cuff pressure; and performing quality judgment on the acquired PPG signals, saving only PPG signals and corresponding cuff pressures that meet preset quality requirements as calibration PPG signals and corresponding cuff pressure standards.
[0021] This embodiment also provides a refrigerator. Figure 2 This is a schematic diagram of the overall structure of a refrigerator 100 according to an embodiment of the present invention. Figure 3 yes Figure 2 An enlarged view of region A in the diagram. (See diagram below.) Figure 2 As shown, the refrigerator 100 may include a cabinet 110 and a door 120. The cabinet 110 defines a storage space inside. The door 120 is disposed on the front surface of the cabinet 110 to close the storage space.
[0022] It should be noted that the quantity, structure, and function of the storage spaces and doors 120 can be selected according to specific circumstances. Specifically, storage spaces can be configured as refrigeration spaces, freezing spaces, variable temperature spaces, or preservation spaces, depending on their intended use. Doors 120 can be correspondingly configured with storage spaces, meaning each storage space can have one or more doors 120. Figure 2 The refrigerator 100 shown has two doors 120 that open relative to each other.
[0023] In one specific embodiment, such as Figure 3 As shown, the refrigerator 100 has a small hole 135 on its door 120, adapted to fit a user's fingertip. The hole 135 is covered by a uniformly transparent glass 136, and a signal acquisition structure 130 is installed inside the hole 135. The diameter of the hole 135 is adapted to fit a user's fingertip, allowing the fingertip to completely cover the hole 135 and preventing ambient light from entering. The uniformly transparent glass 136 covering the hole 135 further protects the signal acquisition structure 130 inside from wear and tear.
[0024] In a preferred embodiment, the signal acquisition structure 130 includes a light source 131 and a photodetector 132. The light source 131 can be a light-emitting diode (LED). The photodetector 132 can be a photodiode (PD). It should be noted that... Figure 3 The illustration shows a reflective blood pressure detection method. A light-shielding structure 133 can be provided between the light source 131 and the photodetector 132 to ensure that the light source 131 and the photodetector 132 on the left and right sides work independently, and to prevent the light signal received by the photodetector 132 from being affected by the light emitted by the light source 131. The light-shielding structure 133 can be embedded in the transparent glass 136.
[0025] Furthermore, the steps for acquiring photoplethysmography (PPG) signals may include: when the user's finger 200 presses against the transparent glass 136, a light source 131 emits light towards the user's finger 200, the light passing through the skin and illuminating subcutaneous blood vessels; and a photodetector 132 receives the light signal reflected by the subcutaneous blood vessels and converts it into an electrical signal to generate a PPG signal. It should be noted that a pressure sensor 134 may also be installed at the transparent glass 136. When the user's finger 200 presses against the transparent glass 136, the pressure sensor 134 detects the pressure signal, thereby triggering the signal acquisition structure 130 to operate and acquire the PPG signal.
[0026] The aforementioned signal acquisition structure 130, its power supply circuit unit, and auxiliary protection and adaptation units, such as the pinhole 135, transparent glass 136, light-shielding structure 133, and pressure sensor 134, can together constitute a vital signs detection module to jointly acquire photoplethysmography (PPG) signals. Figure 3 The illustration shows a reflective blood pressure detection method. In some other embodiments, a light source 131 and a photodetector 132 can be respectively set on the front and back sides of the finger 200. The photodetector 132 receives the light signal transmitted through the subcutaneous blood vessels and converts it into an electrical signal to generate a photoplethysmography (PPG) wave signal, thereby realizing a transmissive blood pressure detection method.
[0027] In a preferred embodiment, the step of saving the calibrated photoplethysmography (PPG) signal and the corresponding cuff gold standard may include: receiving a user's touch operation command or voice operation command; and waking up the information input interface to input and save the calibrated PPG signal and the corresponding cuff gold standard. Specifically, the information input interface is a dedicated interactive operation interface set up by the refrigerator 100. It is a visual operation entry designed to realize the input of the calibrated PPG signal and the cuff gold standard under high / low pressure conditions, and is integrated into the smart interactive terminal of the refrigerator 100, for example... Figure 2The refrigerator 100 shown has a display panel 121 installed on the door 120. The display panel 121 can also prompt the user on the correct way and timing to place their finger 200 on the small hole 135, and can also remind the user to re-enter the signal if the acquired photoplethysmography signal does not meet the preset quality requirements.
[0028] Step S106 involves performing similarity analysis between the actual photoplethysmography (PPG) signal and the calibrated PPG signals under high-voltage and low-voltage conditions, respectively. Based on the analysis results, the corresponding calibration value is selected. Specifically, this may include: using the Pearson correlation coefficient to perform similarity analysis, obtaining the high-voltage correlation coefficient and the low-voltage correlation coefficient; determining whether the high-voltage correlation coefficient is greater than the low-voltage correlation coefficient; and if so, selecting the calibrated PPG signal under high-voltage conditions and the cuff pressure standard as the calibration value; otherwise, selecting the calibrated PPG signal under low-voltage conditions and the cuff pressure standard as the calibration value.
[0029] In step S108, deep learning models are pre-built for different systolic blood pressure range sub-tasks. Specifically, the pre-building of deep learning models may include: dividing the blood pressure range (systolic blood pressure greater than or equal to a preset threshold) into multiple ranges; dividing these ranges into a preset number of sub-tasks, and building an independent deep learning model for each sub-task corresponding to each range. The deep learning model takes the calibrated photoplethysmography (PPG) signal of the user under systolic blood pressure and the corresponding reference systolic and diastolic blood pressures as input, with the reference systolic and diastolic blood pressures derived from the cuff pressure gold standard; and obtaining the one-hot encoding based on the reference systolic blood pressure using a bucket function, and selecting the output of the corresponding deep learning model as the final output.
[0030] In one specific embodiment, the deep learning model can be the InceptionTime model. The InceptionTime model, which is designed independently for multiple tasks, is the core algorithm carrier for achieving accurate blood pressure inference from PPG signals in this embodiment. Its adaptability and technical advantages are perfectly suited to the application scenarios of refrigerators and the technical requirements for blood pressure detection in hypertensive populations. It not only leverages the algorithmic characteristics of the model itself, but also amplifies its advantages through the customized design of interval and molecular tasks, thus solving the core problems of low accuracy and weak trend tracking in traditional PPG signal blood pressure inference.
[0031] The advantage of the InceptionTime model in this embodiment is not simply its algorithmic characteristics, but rather the deep integration of the model's temporal adaptability with the customized design of molecular tasks for high-pressure populations, the hardware / usage requirements of the refrigerator scenario, and the full-process technical requirements for exercise blood pressure tracking.
[0032] The InceptionTime model ultimately achieves a full-process accuracy improvement from PPG signal feature extraction to precise inference across intervals and dynamic correction of motion state. It perfectly solves the core technical problems that this embodiment aims to address in the refrigerator scenario, namely, the low accuracy of PPG signal inference for high blood pressure in hypertensive populations and the lack of obvious trend tracking. It is a core manifestation of the high degree of adaptation between the algorithm and hardware scenario and business requirements.
[0033] The following describes a specific embodiment, with a preset number of three subtasks. Based on the statistical characteristics of the normal distribution of high-pressure data samples, the blood pressure range of systolic blood pressure (SBP) ≥ 140 mmHg is categorized according to... ...The interval is divided into n segments and further divided into three sub-tasks. For each sub-task, an independent InceptionTime model is developed and designed. The model takes the PPG signal and the reference systolic pressure and reference diastolic pressure as inputs. Then, the bucket function obtains the corresponding one-hot encoding based on the reference systolic pressure, and selects the output of the corresponding InceptionTime model as the final output of the current input.
[0034] The blood pressure correction formula in step S110 is preset, and the specific steps of preset blood pressure correction formula may include: repeatedly collecting photoplethysmography (PPG) signals and corresponding systolic and diastolic blood pressures under different exercise states of the user; and extracting the feature parameters of the PPG signals and fitting them to obtain the blood pressure correction formula, wherein the feature parameters include: amplitude, frequency, waveform, energy spectrum, information entropy and heart rate.
[0035] In one specific embodiment, the blood pressure correction formula is: , , ΔSBP sport ΔDBP is the correction value for systolic blood pressure during exercise. sport f is the correction value for diastolic blood pressure during exercise. SBP f is the fitting function for systolic blood pressure. DBP Let A be the fitting function for diastolic blood pressure, ω be the frequency, W be the waveform, ES be the energy spectrum, IE be the information entropy, and HR be the heart rate.
[0036] The formulas for calculating final systolic blood pressure and final diastolic blood pressure are: , , SBP final For the final systolic pressure, DBP final For final diastolic blood pressure, SBP pre For inference of systolic blood pressure, DBPpre The inferred systolic and diastolic blood pressures are corrected using a preset blood pressure correction formula, resulting in final systolic and diastolic blood pressures that more closely reflect the user's actual blood pressure status.
[0037] In summary, the blood pressure detection method in this embodiment can significantly improve the accuracy and trend tracking ability of users' blood pressure detection. By using a dual calibration method, a multi-task deep learning model, and exercise blood pressure tracking correction, it solves the core technical pain point of low accuracy in blood pressure inference from photoplethysmography (PPG) signals, and the detection results are more consistent with the user's real blood pressure status.
[0038] Furthermore, the blood pressure detection method in this embodiment enables cuffless and convenient blood pressure monitoring. Relying on the contact acquisition structure of the refrigerator door 120, no additional special equipment is required. With the assistance of identity recognition and signal quality judgment, users can complete the test in real time and quickly. At the same time, it can provide reliable data support for subsequent dietary management recommendations, forming a closed-loop blood pressure management service.
[0039] Based on precise blood pressure data collected by the Refrigerator 100 from individuals with high blood pressure, it can be deeply integrated with the Refrigerator 100's core scenarios of food storage and dietary management, creating a closed-loop health service system encompassing blood pressure monitoring, data analysis, dietary guidance, and exercise recommendations. The Refrigerator 100 can intelligently match a user's real-time blood pressure values and blood pressure trends during exercise with the nutritional properties of food. If it detects that a user's systolic blood pressure is consistently high, it will automatically identify high-sodium and high-fat foods in the Refrigerator 100 and issue consumption reminders. Simultaneously, it will recommend low-salt, high-potassium, and fiber-rich food combinations and light recipes based on blood pressure data, such as celery and lily bulbs, multigrain porridge, and steamed fish. Furthermore, it can generate personalized blood pressure-lowering dietary plans based on the existing food in the Refrigerator 100, ensuring that dietary recommendations align with the user's actual dietary reserves and improving feasibility.
[0040] Meanwhile, the Refrigerator 100 can customize suitable light exercise methods for users based on blood pressure test results and exercise-based blood pressure tracking data. For people with high blood pressure who experience large fluctuations, it recommends low-intensity aerobic exercises such as walking, Tai Chi, and Baduanjin, clearly defining the reasonable range of exercise duration and frequency. For people with relatively stable blood pressure control, it can suggest moderate brisk walking, jogging, and other exercises, and combine exercise recommendations with dietary plans. For example, it recommends supplementing with high-quality protein and vitamins after exercise. The Refrigerator 100 can also push exercise reminders in real time through its smart interactive interface, and simultaneously record the correlation data between blood pressure, exercise, and diet to continuously optimize the recommendation strategy. This allows people with high blood pressure to achieve a scientific combination of blood pressure management and diet and exercise in their daily home life, forming a sustainable health management model.
[0041] The refrigerator 100 in this embodiment may include a controller 300. Figure 4 This is a schematic block diagram of the controller 300 of a refrigerator 100 according to an embodiment of the present invention. Figure 4 As shown, the controller 300 may include a processor 310 and a memory 320. The memory 320 stores a machine-executable program 321. When the machine-executable program 321 is executed by the processor 310, it is used to implement any of the above-mentioned blood pressure detection methods.
[0042] Processor 310 can be a central processing unit (CPU), a digital processing unit, etc. Processor 310 sends and receives data via a communication interface. Memory 320 is used to store the machine-executable program 321 executed by processor 310. Memory 320 can be any medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer; it can also be a combination of multiple memories 320.
[0043] The aforementioned machine-executable program 321 can be downloaded from a computer-readable storage medium to a corresponding computing / processing device or downloaded and installed to the controller 300 via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). The controller 300 can be used to control the operation of the refrigerator 100 itself, as well as to receive signals from other devices and send signals to other devices, for example, it can send signals to and receive signals sent by a mobile terminal.
[0044] It should be noted that the blood pressure detection method implemented by the refrigerator 100 in this embodiment identifies the user's identity information and matches the corresponding calibrated photoplethysmography (PPG) signal and cuff gold standard based on the identity information. The calibrated PPG signals and corresponding cuff gold standards for the user's systolic and diastolic states are pre-acquired and saved. The user's actual PPG signal is collected. Similarity analysis is performed between the actual PPG signal and the calibrated PPG signals for systolic and diastolic states, respectively, and the corresponding calibration value is selected based on the analysis results. The actual PPG signal is input into a deep learning model corresponding to the calibration value to obtain the inferred systolic and diastolic blood pressure. Deep learning models for different systolic interval subtasks are pre-constructed. Finally, the inferred systolic and diastolic blood pressures are corrected according to a preset blood pressure correction formula to obtain the final systolic and diastolic blood pressures.
[0045] This approach not only significantly improves the accuracy and trend tracking capabilities of blood pressure monitoring, but also addresses the core technical challenge of low accuracy in PPG signal-based blood pressure inference through dual calibration, multi-task deep learning models, and exercise-based blood pressure correction. The resulting test results more closely reflect the user's actual blood pressure status. Furthermore, it enables cuffless and convenient blood pressure monitoring. Utilizing the refrigerator's 120-degree contact-type acquisition structure on the door, no additional specialized equipment is required. Combined with identity recognition and signal quality assessment, users can complete the test quickly and in real-time. This also provides reliable data support for subsequent dietary management recommendations, forming a closed-loop blood pressure management service.
[0046] This embodiment also provides a machine-readable storage medium 400. Figure 5 This is a schematic diagram of a machine-readable storage medium 400 according to an embodiment of the present invention. The machine-readable storage medium 400 stores a machine-executable program 321, which, when executed by a processor 310, implements the blood pressure detection method of any of the above embodiments.
[0047] The machine-readable storage medium 400 of this embodiment may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. The machine-readable storage medium 400 has storage space for a machine-executable program 321 for performing any of the method steps described above. These machine-executable programs 321 can be read from or written to one or more computer program products.
[0048] These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. When the device containing the machine-readable storage medium 400 runs the machine-executable program 321 described above, it can perform the various steps of the methods described above.
[0049] In the description of this embodiment, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0050] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.
Claims
1. A method for detecting blood pressure, comprising: Identify the user's identity information, and match the corresponding calibrated photoplethysmography (PPG) signal and cuff pressure standard based on the identity information. The calibrated PPG signal and the corresponding cuff pressure standard under the user's high-voltage and low-voltage states are pre-acquired and stored. Collect the actual photoplethysmography (PPG) signal of the user; The actual photoplethysmography (PPG) signal is compared with the calibrated PPG signal under high voltage and low voltage conditions, respectively, and the corresponding calibration value is selected based on the analysis results. The actual photoplethysmography (PPG) signal is input into a deep learning model corresponding to the calibration value to obtain inferred systolic blood pressure and inferred diastolic blood pressure, wherein the deep learning model is pre-built for different high-pressure interval sub-tasks. as well as The inferred systolic blood pressure and the inferred diastolic blood pressure are corrected according to a preset blood pressure correction formula to obtain the final systolic blood pressure and the final diastolic blood pressure.
2. The method according to claim 1, wherein the step of pre-acquiring the calibrated photoplethysmography signal and the corresponding cuff pressure gold standard includes: The photoplethysmography (PPG) signals of the user under high and low voltage conditions are acquired by the signal acquisition structure set on the refrigerator, and the corresponding cuff pressure is acquired simultaneously. The quality of the acquired photoplethysmography (PPG) signals is assessed, and only PPG signals that meet the preset quality requirements and their corresponding cuff pressures are saved as the gold standard for calibrating PPG signals and their corresponding cuff pressures.
3. The method according to claim 2, wherein, The refrigerator door has a small hole adapted to fit the user's fingertip. The hole is covered with uniformly transparent glass, and the signal acquisition structure is installed inside the hole. The signal acquisition structure includes a light source and a photodetector, with a light-shielding structure disposed between the light source and the photodetector.
4. The method according to claim 3, wherein the step of acquiring the photoplethysmography signal includes: When the user's finger presses the transparent glass, light is emitted from the light source toward the user's finger, and the light passes through the skin and irradiates the subcutaneous blood vessels. as well as The photodetector receives the light signal reflected by the subcutaneous blood vessel and converts it into an electrical signal to generate the photoplethysmography (PPG) signal.
5. The method of claim 1, wherein the step of pre-constructing the deep learning model comprises: The blood pressure range where the systolic pressure is greater than or equal to a preset threshold is divided into multiple intervals; The multiple intervals are divided into a preset number of sub-tasks, and an independent deep learning model is constructed for each sub-task corresponding to each interval. The deep learning model takes the calibrated photoplethysmography pulse wave signal and the corresponding reference systolic pressure and reference diastolic pressure as inputs. The reference systolic pressure and reference diastolic pressure are derived from the cuff pressure gold standard. as well as The one-hot encoding is obtained by using the bucket function based on the reference shrinkage pressure, and the output of the corresponding deep learning model is selected as the final output.
6. The method according to claim 1, wherein the step of performing similarity analysis between the actual photoplethysmography (PPG) signal and the calibrated PPG signals under high-voltage and low-voltage conditions, respectively, and selecting the corresponding calibration value based on the analysis results, includes: Similarity analysis was performed using the Pearson correlation coefficient to obtain the high-pressure correlation coefficient and the low-pressure correlation coefficient; Determine whether the high-pressure correlation coefficient is greater than the low-pressure correlation coefficient; as well as If so, the calibration photoplethysmography (PPG) signal under high voltage and the cuff pressure standard are selected as the calibration values. If not, select the calibration photoplethysmography pulse wave signal under low pressure and the cuff pressure gold standard as the calibration value.
7. The method according to claim 1, wherein the step of presetting the blood pressure correction formula includes: The photoplethysmography (PPG) signals and corresponding systolic and diastolic blood pressure were collected multiple times under different motion states of the user. as well as The characteristic parameters of the photoplethysmography (PPG) signal are extracted and fitted to obtain the blood pressure correction formula. The characteristic parameters include amplitude, frequency, waveform, energy spectrum, information entropy, and heart rate.
8. The method according to claim 7, wherein, The blood pressure correction formula is as follows: ， ， ΔSBP sport ΔDBP is the correction value for systolic blood pressure during exercise. sport f is the correction value for diastolic blood pressure during exercise. SBP f is the fitting function for systolic blood pressure. DBP Here, A is the diastolic blood pressure fitting function, ω is the frequency, W is the waveform, ES is the energy spectrum, IE is the information entropy, and HR is the heart rate. The formulas for calculating the final systolic blood pressure and the final diastolic blood pressure are as follows: ， ， SBP final For the final systolic pressure, DBP final SBP is the final diastolic pressure. pre For the inference systolic pressure, DBP pre The inference is for diastolic pressure.
9. The method according to claim 2, wherein the step of saving the calibrated photoplethysmography signal and the corresponding cuff pressure gold standard includes: Receive user touch or voice commands; as well as The wake-up information input interface is used to input and save the calibration photoplethysmography pulse wave signal and the corresponding cuff gold standard.
10. A refrigerator, comprising: A controller, comprising a memory and a processor, wherein the memory stores a machine-executable program that, when executed by the processor, implements the method for detecting blood pressure according to any one of claims 1 to 9.