The Relevance of Calibration in Machine Learning-Based Hypertension Risk Assessment Combining Photoplethysmography and Electrocardiography

Figures

• 

  Figure 1. Graphical definition of fiducial points detected from photoplethysmogram (PPG), velocity plethysmogram (VPG) and acceleration plethysmogram (APG) signals.

• 

  Figure 2. Representation of PATfoot, PATderivate and PATpeak features obtained by the time interval between ECG R-peak and fiducial points of PPG signals as well as PPG morphological parameters: Systolic peak amplitude, TPP, rise time, areas under the pulse, width and TPI.

• 

  Figure 3. Correlation matrix of the 23 initial discriminatory features used in the study. Dark red values represent higher correlation coefficients and dark blue values represent lower correlation coefficients.

• 

  Figure 4. Classification performance provided by the KNN classifier in the discrimination between NTS and HTS individuals. Results obtained for sequential validation of consecutive sub-segments for each of the selected subject segments. In each box, the red line indicates the median, and the bottom and top edges indicate the 15th and 85th percentiles, respectively. The whiskers cover the most extreme data points not considered outliers, and the red symbol (+) stands for outliers. Black squares inside each box indicate mean accuracies.

• 

  Figure 5. Classification accuracy to distinguish between NTS and HTS individuals using KNN sequential validation of segments with calibration distant from test measurements. (a) Distance below 1 h. (b) Distance between 1 and 6 h. (c) Distance between 6 and 24 h. (d) Distance above 24 h. In each box, the red line indicates the median, and the bottom and top edges indicate the 15th and 85th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers, and the red symbol (+) stands for outliers. Black squares inside each box indicate mean accuracies.