2020 19th IEEE International Conference on Machine Learning and Applications (ICMLA) 2020
DOI: 10.1109/icmla51294.2020.00137
|View full text |Cite
|
Sign up to set email alerts
|

Enhance CNN Robustness Against Noises for Classification of 12-Lead ECG with Variable Length

Abstract: Electrocardiogram (ECG) is the most widely used diagnostic tool to monitor the condition of the cardiovascular system. Deep neural networks (DNNs), have been developed in many research labs for automatic interpretation of ECG signals to identify potential abnormalities in patient hearts. Studies have shown that given a sufficiently large amount of data, the classification accuracy of DNNs could reach human-expert cardiologist level. However, despite of the excellent performance in classification accuracy, it h… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

1
4
0

Year Published

2022
2022
2023
2023

Publication Types

Select...
2
1
1
1

Relationship

0
5

Authors

Journals

citations
Cited by 6 publications
(5 citation statements)
references
References 10 publications
(12 reference statements)
1
4
0
Order By: Relevance
“…The classification DNN with CardioDefense at T = 1 still had the best performance, and its accuracy ratio against PGD attacks is still over 80% in two situations, and its decline ratio of accuracy is lower than 1% in situation I. In addition, we can see that adversarial training had good performance against low noise level PGD attacks, and NSR behaved better than JR under PGD attacks, which is consistent with the results of Ma and Liang [2020a].…”
Section: Defense Effects Against Pgd Attackssupporting
confidence: 82%
See 2 more Smart Citations
“…The classification DNN with CardioDefense at T = 1 still had the best performance, and its accuracy ratio against PGD attacks is still over 80% in two situations, and its decline ratio of accuracy is lower than 1% in situation I. In addition, we can see that adversarial training had good performance against low noise level PGD attacks, and NSR behaved better than JR under PGD attacks, which is consistent with the results of Ma and Liang [2020a].…”
Section: Defense Effects Against Pgd Attackssupporting
confidence: 82%
“…Here, we introduce the experimental implementation details. Based on the studies conducted by Ma and Liang Ma and Liang [2020a], the only parameter of JR, λ, is set as 44, due to its outstanding performance, and the two parameters of NSR regularization, ϵ max and β, are set to 1. To make the classification model converge quickly, the regularization term of JR and NSR regularization and the NSR margin loss are not added to the training process until the 11th epoch.…”
Section: Implementation Detailsmentioning
confidence: 99%
See 1 more Smart Citation
“…Instead, it is acceptable, as the trained classifier achieved good accuracy (similar to the literature). Furthermore, it is encouraging in terms of robustness and accuracy dropping in comparison with reference [ 43 ], which used a deep learning classifier with defense methods on an intra-patient paradigm. However, it should be noted that the intra-patient paradigm is mostly focused on beat types, which implies that a certain person’s ECG recording may be found in both data sets (train and test data).…”
Section: Discussionmentioning
confidence: 77%
“…However, once the model trained on clean data was applied to identify the classes from the noisy data sets, F1-scores dropped by (up to) 0.18. In this work [ 43 ], the authors constructed a CNN for the classification of 12-lead ECG signals, and they used three defense systems to increase the robustness of the CNN classifier for ECG classification against a PGD-100 attack and white noise. The evaluation study reveals that the modified CNN achieved an acceptable F1-score and average accuracy, and the defense methods improved their CNN’s robustness against adversarial noise and white noise, with a small decrease in accuracy on clean data.…”
Section: Literature Overviewmentioning
confidence: 99%