Signal-Specific and Signal-Independent Features for Real-Time Beat-by-Beat ECG Classification with AI for Cardiac Abnormality Detection
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposed signal-specific and independent features for beat-to-beat ECG classification with artificial intelligence for cardiac abnormality detection. Research is interesting and the obtained results shows that the proposed method reaches high accuracy. Before oublication, I still believe that this paper needs to address several points:
- Update affilliations with address, country, etc.
- I believe there is some mistake in sentence in Lines 16-17.
- Abstract can be more concise, while keeping all relevant information.
- Abbreviations must be revised in the whole manuscript. Many abbreviations are defined mutliple times throughout the paper, while only once is needed.
- Your contributions need to be more highlighted in the Introduction. Elaborate what is new here and why did you make these changes.
- Introduction can have paper outline at the end.
- There are two Materials and Methods sections. Revise that. I believe that the same text is multiplied.
- Did you perform hyperparameter search for deep learning?
- How did you do annotation in training dataset of actual peaks, etc.? Is that annotation correct?
- Add more relevant works from past years in References.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors in the article investigates the problem of classifying cardiac arrhythmias from ECG signals, which is complicated by the high similarity between normal and pathological rhythms, as well as the uneven representation of classes.This study used interplay of feature engineering, signal specific features, model selection and computational resource optimisation in heartbeat classification. Ensemble machine learning methods and deep neural networks were also used, the training of which was optimised using stratified k-fold cross-validation and sample balancing.
Сomments on the publication:
1.Introduction section should be extended using more clearly motivation of this paper.
2.It would be good to add point-by-point the main contributions in the end of the Introduction section
3.The text of the article lacks detail on what features or mechanisms contributed to decision-making in DL models taking into account explanatory artificial intelligence (XAI).
4.A description of MIT-BIH should be added to show how it represents rare types of abbreviations (e.g. merged or undefined) to exclude unbalanced sampling.
5.Add model validation on other open or real ECG databases (e.g. PhysioNet other than MIT-BIH).
6.Conclusion section should be extended using: 1) numerical results obtained in the paper; 2) limitations of the proposed approach; 3) prospects for the future research.
7.A lot of references are outdated. Please fix it using 3-5 years old papers in high-impact journals.
The authors in the article investigates the problem of classifying cardiac arrhythmias from ECG signals, which is complicated by the high similarity between normal and pathological rhythms, as well as the uneven representation of classes.This study used interplay of feature engineering, signal specific features, model selection and computational resource optimisation in heartbeat classification. Ensemble machine learning methods and deep neural networks were also used, the training of which was optimised using stratified k-fold cross-validation and sample balancing.
Сomments on the publication:
1.Introduction section should be extended using more clearly motivation of this paper.
2.It would be good to add point-by-point the main contributions in the end of the Introduction section
3.The text of the article lacks detail on what features or mechanisms contributed to decision-making in DL models taking into account explanatory artificial intelligence (XAI).
4.A description of MIT-BIH should be added to show how it represents rare types of abbreviations (e.g. merged or undefined) to exclude unbalanced sampling.
5.Add model validation on other open or real ECG databases (e.g. PhysioNet other than MIT-BIH).
6.Conclusion section should be extended using: 1) numerical results obtained in the paper; 2) limitations of the proposed approach; 3) prospects for the future research.
7.A lot of references are outdated. Please fix it using 3-5 years old papers in high-impact journals.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis work aims to perform Signal-Specific and Signal-Independent Features for Real-time Beat-by-Beat ECG Classification with AI for Cardiac Abnormality Detection. While the paper is generally well-written, several areas need clarification:
- The study employs the ANOVA algorithm to rank features and selects the top 10 signal-independent features. Could you clarify how this choice compares to other feature ranking methods such as Mutual Information or Relief? Additionally, did you evaluate the stability of the selected features across different datasets or validation splits to ensure their robustness?
- The highest accuracy of 100% is achieved with a 128-layer LSTM trained over 100 epochs. Considering real-time edge device limitations, can you elaborate on the inference latency, power consumption, and memory requirements of such a deep model? Would this approach be practically implementable in wearable or IoT devices?
- Achieving near-perfect accuracy on validation datasets raises potential overfitting concerns. Have you performed any cross-validation or external validation to verify that the models generalize well to unseen data? What measures (e.g., dropout, early stopping) were taken to prevent overfitting?
- While the study shows high accuracy with deep models, how do these results compare with other recent ECG classification approaches, such as CNNs with fewer layers or hybrid models? Including such comparisons could better position your method within the current landscape.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors presented an ECG classification algorithm with AI to identify cardiac abnormality. They considered 99 signal-independent features, also dimensionally reduced 28 signal specific features to further improve classification accuracy.
I have some comments to improve the quality of paper to enhance the clarity:
1) The introduction fails to provide the current AI-based techniques to classify the cardiac abnormality from ECG signals. I recommend to provide a summary in the form of a table format to point the knowledge gap.
2) The algorithm for feature extraction lacs the professionalism (Figure 1 and Algorithm 1).
3) There are some acronyms, not abbreviated during first time citation (ex. TSFEL).
4) There are redundant pages (4 and 6), Section 2.2 is missing, the sequence of tables are wrong (missing tables II and III).
5) The results section should include more statistical performance measures.
6) Please consider extending the conclusion to encompass implications for clinical practice and future research avenues.
Comments on the Quality of English Language
The quality of English language is not appropriate, even readable. There are major grammar errors throughout the manuscript. I recommend having a native English speaker or professional editing service to edit the manuscript thoroughly to refine the quality of overall writing.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe Authors have addressed all my comments. I do not require any further modifications.
Author Response
We sincerely appreciate your time and thoughtful feedback throughout the review process. Thank you for confirming that all issues have been appropriately addressed. We are grateful for your valuable comments, which helped us improve the manuscript, and look forward to further evaluation.
Reviewer 3 Report
Comments and Suggestions for AuthorsThis is an excellent and timely work that proposes a novel framework combining Signal-Specific (SS) and Signal-Independent (SI) features for real-time beat-by-beat ECG classification using AI for cardiac abnormality detection. The paper is generally well-written, and the proposed method addresses an important need for accurate and energy-efficient detection in real-time applications. I appreciate the authors' contributions, particularly the hybrid feature design and thoughtful handling of class imbalance.
To further strengthen the paper, I suggest minor revisions focused on clarifying and expanding a few points:
- In the introduction and Table I, the authors cite several recent works on lightweight ECG classification. It would greatly enhance the paper to include a comparative analysis highlighting key differences in accuracy, power efficiency, interpretability, and robustness. Specifically, how does the hybrid SI + SS feature approach outperform or complement existing deep learning? A table summarizing these comparisons would make the unique contribution of this method clearer to the reader.
- In Table II, the use of class-weighted focal loss is commendable. Please consider adding a brief explanation of how this strategy affected the detection performance of minority classes.
- The mention of power consumption on Page 14 is appreciated but could be expanded. Providing specific power metrics or runtime comparisons between the proposed CNN model and traditional methods would strengthen claims of energy efficiency.
- In the Discussion, the proposed energy-efficient implementation is promising. Could the authors elaborate on how it compares with existing low-power ECG classification solutions in terms of inference latency?
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors addressed all the comments as per requests. I don't have any further comments.
Author Response
We sincerely appreciate your time and thoughtful feedback throughout the review process. Thank you for confirming that all issues have been appropriately addressed. We are grateful for your valuable comments, which helped us improve the manuscript, and look forward to further evaluation.