Automated Signal Quality Assessment of Single-Lead ECG Recordings for Early Detection of Silent Atrial Fibrillation

Figures

• 

  Figure 1. The MyDiagnostick device used in the our study. The device incorporates two metal handle electrodes and performs an Einthoven I-lead ECG when held by the user.

• 

  Figure 2. Capacitance of the stratum corneum over the layer thickness. The relative permittivity is additionally dependent on the skin humidity.

• 

  Figure 3. Simplified equivalent circuit diagram of the ECG measurement chain provided by the device used in this study. The circuit consists of the internal voltage source Uecg, which equals a standard ECG signal measurement, the skin-electrode interface, Csc and Rsc, of the metal electrodes, the input high pass-filter characteristic formed by the coupling capacity Ccpl and the input impedance Rin of the analog amplifiers.

• 

  Figure 4. Results of the measurement chain simulations. (a) Resulting changes in ECG signal morphology due to input high-pass filter characteristic over varying coupling capacity Ccpl. Different coupling frequencies are represented by different shades of blue from 0.1pF to 0.1nF, the original ECG signal is shown in orange. (b) Comparison between original ECG signal morphology and amplitude-aligned filtered signal for the input high-pass characteristic with a cut-off frequency of fc=10.6Hz.

• 

  Figure 5. Typical measurement scenario of a potential study participant in the pharmacy. The measurement with the ECG device was performed under the guidance of trained pharmacy staff.

• 

  Figure 6. Signal processing chain for preprocessing the ECG recordings prior to feature extraction.

• 

  Figure 7. Two exemplary ECG recordings including motion artifact detection and R-peaks identified by the peak detection algorithm. (a) Recording of sinus rhythm affected by a motion artifact at t=0–3s. The segment affected by the motion artifact is discarded from further analysis. (b) Recording of a patient suffering from AF. Motion artifact detection is not affected by irregular cardiac rhythm.

• 

  Figure 8. ECG heart cycle prototypes extracted from the excellent´signal quality category. The signal prototypes were clustered into four separate groups according to maximum signal similarity using the DTW distance. The mean signal morphology is shown in dark blue, the standard deviation is represented by the light blue area.

• 

  Figure 9. Violin plot presenting the results of the training process of mapping the input feature space to the clinical expert rating.

• 

  Figure 10. Best-rated ECG recordings according to the AISQA index from each signal quality category assessed by the clinical experts.

• 

  Figure 11. Worst-rated ECG recordings according to the AISQA index from each signal quality category as assessed by the clinical experts.

• 

  Figure 12. Percentages of repeated measurements and remaining false AF classification over AISQA threshold if a measurement is assumed to be repeated when the signal quality exceeds the threshold.