ECGAssess: A Python-Based Toolbox to Assess ECG Lead Signal Quality

Figure 1. This recording (Recording 1609343) was annotated as being of unacceptable quality in the PhysioNet/Computing in Cardiology Challenge. However, according to the assessment in this study, only Lead V1 was unacceptable because the heartbeats provided by this lead were not clearly recognizable. The voltage (y axis) is given in 100 micro volts, the time (x axis) is given in seconds. Note that the 12-lead ECG signals are visible in the standard ECG grid to improve the clinical interpretation of the ECG waveforms.

Figure 2. Six examples of ECG signals that were classified and annotated into acceptable and unacceptable with respect to their quality. The filtered signals were visually classified, with a focus on the heartbeats. If all heartbeats were visible, the quality was considered acceptable.

Figure 3. Flowcharts of the three signal quality classification algorithms. Algorithm I is a stationary check, Algorithm II is a heart rate check, and Algorithm III is an SNR check. A passed status on an algorithm means that no characteristics of a low-quality signal were detected. A not-passed status means that characteristics of poor quality were detected for the ECG signal.

Figure 4. Stationary signal check. The window, represented by the red rectangle, moved through the signal (Recording 1072620) in the direction of the arrows, and the signal was viewed through the window in this algorithm. If the signal was stationary in the window at a specific time, it was declared not passed. A not-passed signal is shown on the left and a signal that has passed is shown on the right.

Figure 5. Heart rate check. The filtered signal (Recording 2167341) was analyzed using the Pan-Tompkins algorithm. The detected heartbeats are visible as dots. If the number of detected heartbeats was not within the accepted range, the signal was marked not passed. A signal that has not passed is shown on the left; a signal that has passed is shown on the right.

Figure 6. SNRs of the signals from the testing data set. The signals are shown as diamonds if they were annotated as acceptable and as stars if they were annotated as unacceptable. The selected SNR threshold of 0.5 dB is shown as a black line.

Figure 7. An SNR example. With respect to Algorithm III, an example of a not-passed signal is on the left and an example of a passed signal is on the right (Recording 1896934). The SNRs of the signals are 0.4384 dB for the signal on the left and 8.1156 dB for the signal on the right.

Figure 8. ECGAssess toolbox output. The toolbox graphical user interface was developed to enable people without programming knowledge to use the application. Pressing the Import Data button opens an explorer window where the .txt file of interest can be selected. The signal is also visible in the plot window with the standard ECG grid, to improve the interpretation to cardiologists. The visible lead can be selected using the slider. When the “Process” button is pressed, the signal quality assessment is executed, and the results are displayed in a table format.