An Open-Source Graphical User Interface-Embedded Automated Electrocardiogram Quality Assessment: A Balanced Class Representation Approach

Figure 1. Schematic overview of the proposed CNN-LSTM network. The Feature extraction box outlines the CNN part of the model. The Classification box outlines the LSTM part of the model. The values under the input layer represent the input shape. The values under the Conv1D layers represent the filter and kernel size. The value under the dropout layers represents the dropout rate. The value under the LSTM and Dense layers represents the dimensionality of the output. The Sigmoid under the last Dense layer represents the activation function of the output.

Figure 2. Examples of true positives and true negatives yielded by the proposed CNN-LSTM model, utilizing the BUT QDB dataset.

Figure 3. Showcase of the new toolbox with an instance of an ECG signal with acceptable quality. The GUI marks it as bad quality due to automatic labeling based on the proposed CNN-LSTM algorithm. Other ECG signal quality indices in agreement. The displayed ECG corresponds to ‘Lead 1’, extracted from the CinC11 dataset, which includes 12-lead ECG data. By utilizing the scroll bar situated in the top left corner of the graphical user interface, we can load and examine each ECG lead within the dataset. Remarkably, all 12 ECG leads exhibit acceptable quality.

Figure 4. Showcase of the new toolbox with an instance of an ECG signal with unacceptable quality. The GUI marks it as bad quality due to automatic labeling based on the proposed CNN-LSTM algorithm. Other ECG signal quality indices in disagreement. The displayed ECG corresponds to ‘Lead 8,’ extracted from the CinC11 dataset, which includes 12-lead ECG data. By utilizing the scroll bar situated in the top left corner of the graphical user interface, we can load and examine each ECG lead within the dataset. Remarkably, all 12 ECG leads, excluding ‘Lead 8,’ exhibit acceptable quality.