Dynamic Bayesian Model for Detecting Obstructive Respiratory Events by Using an Experimental Model

Figures

• 

  Figure 1. The RR time series (in blue) before (top) and after (bottom) removing the trends and slow oscillations (in black). Red vertical lines mark the time instant of apneas onset and end.

• 

  Figure 2. Examples of dynamic Bayesian networks obtained using four time slices (L = 3). (a) Only the RR time series. (b) The combination of Ra and RR time series. (c) All the ECG parameters (Ra, Sa, RSd, Us, Ds) combined with the RR time series.

• 

  Figure 3. (Top) Normalized RR and QRSpca time series (blue color) and apnea occurrence (red color) for a particular segment of four apnea episodes. (Bottom) Apnea probability estimates using a DBN with L = 10, before (gray color) and after smoothing (black color) using only the RR marker, and when adding the QRSpca marker.

• 

  Figure 4. (Top) Original and normalized RR (blue color) and apnea occurrence (red color) for a particular sequence of 15 episodes. (Bottom) Apnea probability estimates (gray, raw values; black, average values) before and after the normalization process.

• 

  Figure 5. RMSE values as a function of the Markovian order when training DBNs in three different recordings for RR (top) and Ra (bottom) markers. Each color represents one particular sequence/recording of 15 min within the same rat having a different number of apnea episodes.

• 

  Figure 6. Complexity of the dynamic Bayesian networks as a function of lag (L) in three different recordings for the RR + Ra combination. The blue line represents the number of nodes involved while the other colors indicate the total number of arcs obtained from the networks’ structure in each sequence/recording.

• 

  Figure 7. Performance metrics as a function of threshold obtained for Bayesian networks trained with individual (left RR, middle Ra) and combined (right RR + Ra) markers. Color intensity represents the lag (L = 1:15), where the lighter tones refer to smaller values.