Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

Flowchart of the proposed polar spectrogram visualization method.

The polar spectrogram images and their corresponding labels are used to train and validate a deep CNN classification model. Abbreviations: ECG, electrocardiogram; P-T, Pan-Tompkins; STFT, short-time Fourier transform; CNN, convolutional neural network.

More »

Fig 1 Expand

Fig 2.

Processed results from non-filtered ECG data.

(a) ECG time series, (b) spectrogram, (c) polar spectrogram, and (d) reverse polar spectrogram. Low frequency data is visualized in red, while high frequency data is visualized in blue. We used the ‘jet’ colormap for color display. Reverse polar transformed spectrogram images such as (d) were used to train, validate, and test the deep CNN models in our study.

More »

Fig 2 Expand

Fig 3.

Processed results from ECG data with the P-T algorithm.

(a) ECG time series after applying the P-T algorithm, (b) spectrogram, (c) polar spectrogram, and (d) reverse polar spectrogram. Reverse polar transformed spectrogram images such as (d) were used to train, validate, and test the deep CNN models in our study.

More »

Fig 3 Expand

Table 1.

The number of records for the model development and test data.

More »

Table 1 Expand

Table 2.

Prediction results on test data.

More »

Table 2 Expand

Table 3.

Comparison with existing methods.

More »

Table 3 Expand

Fig 4.

Confusion matrices for (a) MobileNet, (b) ResNet50, (c) DenseNet121, and (d) Voting classifiers when the P-T algorithm was used for ECG data preprocessing. A: Afib, N: normal sinus rhythm, O: other rhythm, and ~: noise.

More »

Fig 4 Expand

Fig 5.

tSNE visualization after dimensionality reduction of the penultimate features.

The test samples are shown in colors for difference classes. The left and right columns show results for without and with P-T preprocessing, respectively. The intra-class samples in the good-performance models (d-f) tend to be more clustered than those in the poor-performance models (a-c). Afib: atrial fibrillation, Normal: normal sinus rhythm, Other: other rhythm, Noise: noisy signal.

More »

Fig 5 Expand

Fig 6.

Representative examples of correct predictions in the reverse polar transformed spectrograms.

Afib: atrial fibrillation, Normal: normal sinus rhythm, Other: other rhythm, Noise: noisy signal.

More »

Fig 6 Expand

Fig 7.

Representative examples of incorrect predictions in the reverse polar transformed spectrograms.

Afib: atrial fibrillation, Normal: normal sinus rhythm, Other: other rhythm, Noise: noisy signal.

More »

Fig 7 Expand