Monitoring blood pressure and cardiac function without positioning via a deep learning–assisted strain sensor array

Figures

• 

  Fig. 1.. Illustration showing the structure and working mechanism of the deep learning–assisted strain sensor array for monitoring blood pressure and cardiac function. (A) Schematic and photograph of the monitoring system attached on the wrist. (B) Schematic showing the structure of the sensor array and images of the carbonized silk. (C) Schematic showing the deformation of the strain sensor induced by the radial artery pulse and its working mechanism. (D) Typical radial artery pulse of a healthy person, showing plenty of features, including heartbeat total time (TT), systolic upstroke time (UT), total time (TT), reflected wave transit time (RWTT), systolic-diastolic time (PPT), left ventricular ejection time (LVET), derivative-systolic upstroke time (DPT), amplitude of the advancing wave (HP), amplitude of the reflected wave (HT), and amplitude of the dicrotic wave (HD). (E) Architecture of the deep learning model. CNN, convolutional neural network.

• 

  Fig. 2.. Experimental electromechanical response and simulation results of the strain sensor array. (A) Relative change in resistance of the strain sensor versus the applied strain, showing high linearity. The experiment was repeated three times independently with similar results. (B and C) Relative change in resistance of the strain sensor array while being pressed at different sites. The experiment was repeated three times independently with similar results. (D) Schematic illustration of the sensor array collecting pulse. (E) Deformation distribution in the active layer from the bottom view while the center of the array is pressed. (F) Deformation of different sensors while the center of the array is pressed.

• 

  Fig. 3.. Performance of the strain sensor array in acquiring pulse waves and the extracted key features. (A) Typical pulse waveform of radial artery collected by one strain sensor, showing high precision and rich features. (B and C) Radial artery pulse collected before (B) and after (C) 30 days of service, showing long-term stability. The experiment was repeated three times independently with similar results. (D) Pulse waves collected with the sensor array from different locations, indicating its pulse acquisition ability without precise positioning. (E) UT, TT, RWTT, and PPT from the pulse signal. (F) LVET, DPT, HT/HP (amplitude ratio of the reflected wave to advancing wave), and HD/HP (amplitude ratio of the dicrotic wave to advancing wave). (G) Extracted systolic time @ 25% of the pulse amplitude (Up_25), diastolic time @ 25% of the pulse amplitude (Down_25), width @ 50% of the pulse amplitude (Width_50), and width @ 70% of the pulse amplitude (Width_70).

• 

  Fig. 4.. The mechanism to monitor blood pressure and cardiac function based on the pulse waves obtained by the sensor array and deep learning neural networks. (A) Schematic illustration of the system, which contains pulse acquisition, preprocessing, and deep learning modules. ADC, analog-to-digital converter; BP, blood pressure; CFP, cardiac function parameters. (B) Violin plots of the predicted and measured diastolic blood pressure (DBP), systolic blood pressure (SBP), and mean blood pressure (MBP), indicating the high accuracy of the blood pressure monitoring system. The violin plots include box plots (defined as Q1 and Q3 quartiles, and median) with a kernel density estimation over the points. (C) Bland-Altman plot of the differences between the predicted systolic blood pressure using our system (SBP_Pre) and the measured systolic blood pressure using the CNAP system (SBP_Mea). (D) Bland-Altman plot of the differences between the predicted diastolic blood pressure using our system (DBP_Pre) and measured diastolic blood pressure using the CNAP system (DBP_Mea). (E to H) Violin plots of the predicted and measured cardiac function parameters, including PPV, CO, SV, and SVR. The violin…

• 

  Fig. 5.. Demonstration of the intelligent blood pressure and cardiac function monitoring system in practical application. (A) Schematic diagram of the application scenario of the system. (B) Photograph of the health management system, which enabled real-time blood pressure and cardiac function monitoring using a mobile phone. (C) Predicted and measured blood pressure in a continuous 5-min test. (D) Predicted and measured blood pressure when the participant performed the Valsalva maneuver. (E and F) Predicted and measured cardiac function parameters in a continuous 5-min test, showing good consistency and indicating the reliable performance of the system.