On the Beat Detection Performance in Long-Term ECG Monitoring Scenarios

Figures

• 

  Figure 1. Main stages of the QRS detection algorithms proposed in this work. The yellow boxes are inputs/outputs, the blue boxes represent the preprocessing stage, the green boxes represent the feature signal extraction stage, and the red boxes represent the QRS detection stage.

• 

  Figure 2. Example of the signal preprocessing stage over a segment of 3-lead signal from vPREDICT database.

• 

  Figure 3. Computed feature signals for each proposed method. The upper four panels show the feature signal for the next methods: ICA, PCA, PT, and RMS. In the lower panel, the feature signals from the combinational methods are presented.

• 

  Figure 4. Thresholding process and peak detection for each proposed method. Panel one shows the feature signal, the threshold, and the peaks detected for the next methods, ICA, PCA, PT, and RMS. In panel two, the feature signal, the threshold, and the peaks detected for the combinational methods are presented. The yellow crosses shown the real peak presented in the signal, and the red circles shown the peak detected by each method.

• 

  Figure 5. Mean detection errors and processing times for different segment lengths in the proposed algorithms.

• 

  Figure 6. Estimated computational complexity in terms of the number of operations per recording and of its duration (from 1 to 31 days), and for different number of leads. The dark blue line represents the number of operations of a complete single-lead signal, and the others show different lead configurations when including the segmentation stage.

• 

  Figure 7. Comparison between the non-optimized and the optimized algorithm in record 100. The two upper panels depict leads 1 and 2. The four lower panels depict the feature signals for both leads.

• 

  Figure 8. Evolution of the detection error for each LTM patient across the test duration (up), and examples of noise types present in 7-day Holter recordings (middle and down).

• 

  Figure 9. Example of single beat miss-detection in conventional Holter LTM. The panels in the first row show leads 1 and 2 for an example of Holter miss-detection, the second row shows the feature signal for the leads plotted before. The last four panels follow the same scheme for other Holter miss-detection examples.

• 

  Figure 10. Example of beat miss-detection in proposed algorithm on 7-day Holter recording. The panels in the first row show leads 1 and 2 for an example of algorithm miss-detection, the second row shows the feature signal for the leads plotted before. The last four panels follow the same scheme for other algorithm miss-detection examples.