The identification of blood pressure variation with hypovolemia based on the volume compensation method

FIGURE 1. Schematic diagram of animal blood loss experiment. PPG signal (human finger probe used) under constant volume and invasive blood pressure were collected based on synchronization level signal sent by the finger cuff to achieve strict time alignment.

FIGURE 2. Relationship between PPG pulsation and external pressure at the detection site. The left y-label represents the externally applied pressure value, the black diagonal line in the figure represents the pressure change, the right y-label represents the normalized PPG value, the blue line in the figure represents the PPG waveform, and the red curve represents the upper and lower envelopes of the PPG waveform.

FIGURE 3. Data processing result graph. Figures (A), (B), (C) are the pulse wave waveform after single-cycle preprocessing, the pulse wave waveform after single-cycle normalization, and the pulse wave waveform after single-cycle template, respectively; Figures (D), (E) and (F) are the ten consecutive cycles pulse waveforms after of preprocessing, ten consecutive cycles pulse waveforms after normalizing, and ten consecutive pulse cycles waveforms after templating, respectively.

FIGURE 4. The change of PPG waveform shape during blood pressure change: Blood pressure drops and PPG waveform become shorter and wider. Blood pressure increased, PPG waveform became high and narrow. The red line represents the blood pressure and the blue line represents the PPG waveform at that blood pressure.

FIGURE 5. Schematic diagram of feature extraction.

FIGURE 6. The overall flow chart of blood pressure identification. ①Firstly, the animal experiment was carried out. IBP was collected from the left femoral artery, PPG and Finger pressure were collected from the pig tail by self-developed device, and a total of three signals were collected; ②The collected PPG signals were preprocessed, the baseline drift and noise were removed by filtering, the amplitude and length of the signals were normalized, and finally the PPG signals were templated; ③Feature extraction was carried out for the preprocessed PPG signals, and the integral area of the rising edge of the waveform, the integral area of a single cycle of the waveform and the difference between the cross-correlation of the sample waveform and the template waveform and the autocorrelation of the template waveform; ④After feature extraction, the ten-fold cross-validation idea was used to construct the model. The magnitude of a single feature value is used to determine changes in blood pressure and a machine learning model with multiple features is built to identify changes in blood pressure, respectively. Feature 1 represents the integral area of the rising edge of the waveform, Feature 2…

FIGURE 7. Bland-Altman plot comparing non-invasive mean blood pressure and invasive mean blood pressure. There are 214 groups of samples, with each circle representing the mean blood pressure at the beginning of identification of blood pressure.

FIGURE 8. (A–C) are the correlation analysis of characteristic parameters with systolic blood pressure, diastolic blood pressure and mean blood pressure, respectively.

FIGURE 9. (A–C) are the analysis of identification results under different blood pressure variability under diastolic blood pressure, mean blood pressure and systolic blood pressure, respectively.

FIGURE 10. (A) is the mean values of ACC, MCC, F1 score, AUC, KAPPA and AUPRC for systolic blood pressure at different sliding window sizes, in models established for blood pressure changes spanning 5–15 mmHg, respectively. (B) is the mean values of ACC, MCC, F1 score, AUC, KAPPA and AUPRC for diastolic blood pressure at different sliding window sizes, in models established for blood pressure changes spanning 5–15 mmHg, respectively. (C) is the mean values of ACC, MCC, F1 score, AUC, KAPPA and AUPRC for mean blood pressure at different sliding window sizes, in models established for blood pressure changes spanning 5–15 mmHg, respectively. The X label of each graph represents the size of the sliding window and the Y label represents the different evaluation metrics for blood pressure changes across 5–15 mmHg under the current sliding window.

FIGURE 11. (A) is the standard deviations of ACC, MCC, F1 score, AUC, KAPPA and AUPRC for systolic blood pressure at different sliding window sizes, in models established for blood pressure changes spanning 5–15 mmHg, respectively. (B) is the standard deviations of ACC, MCC, F1 score, AUC, KAPPA and AUPRC for diastolic blood pressure at different sliding window sizes, in models established for blood pressure changes spanning 5–15 mmHg, respectively. (C) is the standard deviations of ACC, MCC, F1 score, AUC, KAPPA and AUPRC for mean blood pressure at different sliding window sizes, in models established for blood pressure changes spanning 5–15 mmHg, respectively. The X label of each graph represents the size of the sliding window and the Y label represents the different evaluation metrics for blood pressure changes across 5–15 mmHg under the current sliding window.