Pith. sign in

REVIEW 3 major objections 5 minor 37 references

Uncertainty-Aware Multi-view Arrhythmia Classification from ECG

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Fusing two views of single-lead ECG with an uncertainty-aware Dempster-Shafer rule yields state-of-the-art arrhythmia classification and stronger noise robustness than conventional fusion.

desk verdict Competent engineering, but the DST fusion is not DST and the beat-level split makes the SOTA claims shaky. read the letter →

arxiv 2506.06342 v1 pith:4Z7CZ3WZ submitted 2025-06-01 eess.SP cs.LG

Mohd Ashhad , Sana Rahmani , Mohammed Fayiz , Ali Etemad , Javad Hashemi This is my paper
classification eess.SPcs.LG
keywords ECGarrhythmiaclassificationmulti-viewlearninguncertainty-awarefusionDempster-ShafertheoryGramianAngularFieldVisionTransformerBiLSTMnoiserobustness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

UAMAC classifies each heartbeat twice: a BiLSTM reads the raw time-series while a Vision Transformer reads a Gramian Angular Field image of the same beat, and the two views are merged by an uncertainty-aware rule based on Dempster-Shafer theory. The paper reports that this two-view design reaches 98.8% accuracy on MIT-BIH and 99.5% on INCART, beating the compared baselines in accuracy and precision, and that the uncertainty-aware fusion degrades more gracefully than score-level or feature-level fusion when noise is added. The claim matters because real ECG data are noisy, and a fusion rule that treats disagreement between views as uncertainty rather than averaging it away could make automated arrhythmia screening more trustworthy. The paper positions the fusion mechanism as the core contribution, since its ablations show each view alone performs below the combined model.

What carries the argument

The load-bearing component is the uncertainty-aware fusion block built on Dempster-Shafer theory, a framework for reasoning under uncertainty. Each branch outputs softmax probabilities; for a given class the probability assigned to that class is treated as evidence and the remaining probability mass as conflict. The two branches' evidence terms are multiplied in the numerator and the product of their conflicts is subtracted from one in the denominator, so agreement between views is amplified while conflicting evidence is discounted rather than averaged. Around this block sit the two view encoders: a BiLSTM for morphological time-series features and a Vision Transformer on 16x16 patches of 224x224 Gramian Angular Field images for spatiotemporal features.

What would settle it

Train and test UAMAC under a patient-exclusive split, putting all beats from each patient into either training or testing on both MIT-BIH and INCART; if accuracy on held-out patients drops well below 98.8% and 99.5%, or if the uncertainty-aware fusion no longer beats score-level and feature-level fusion in the noise tests, the claimed advantage is an artifact of heartbeat-level overlap rather than a property of the fusion rule.

Watch

Extended reading notes

Core claim

The central discovery is that converting each view's softmax output into per-class evidence and conflict, then combining them with Dempster's rule of combination, produces a heartbeat classifier that is both more accurate and more noise-resistant than the same two encoders fused by score averaging or feature concatenation. Concretely, for class $y$ the fused probability is $\tilde{p}_y = E^1_y E^2_y / (1 - C^1_y C^2_y)$, where $E^k_y$ is the probability view $k$ assigns to $y$ and $C^k_y$ is the probability it assigns to other classes. On MIT-BIH the model reaches 98.8% accuracy with 95.6% precision and 93.4% recall, and on INCART 99.5% accuracy with 96.3% precision and 98.0% recall, surpassing the listed baselines. Under artificial white noise and three real-world noise types (baseline wander, muscle artifact, electrode motion) at SNRs down to 0 dB, the uncertainty-aware fusion maintains higher accuracy than score-level and feature-level fusion in every tested condition.

Load-bearing premise

The evaluation splits heartbeats randomly into 80% training and 20% test, so beats from the same patient can appear in both sets; the reported accuracy therefore assumes the classifier is only tested on patients whose beats it has already seen during training.

Editorial extensions

If this is right

  • On MIT-BIH's five-class task, UAMAC's 98.8% accuracy and 95.6% precision exceed every baseline the paper compares against, with recall at 93.4%.
  • On INCART's three-class task, UAMAC's 99.5% accuracy, 96.3% precision, and 98.0% recall beat the closest baseline by 1.4, 2.1, and 3.0 percentage points respectively.
  • Ablation shows both views contribute: removing either the BiLSTM branch or the ViT branch lowers accuracy on both datasets, and BiLSTM and ViT-base are the strongest single-view backbones tested.
  • Noise stress tests with AWGN and NSTDB noise (baseline wander, muscle artifact, electrode motion) show the uncertainty-aware fusion keeping higher accuracy than score-level and feature-level fusion at every SNR level from 15 down to 0 dB.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the paper leaves implicit: because the 80-20 split is made over heartbeats rather than patients, beats from the same patient appear in both training and test sets, so the reported numbers likely overstate performance on unseen patients; a patient-exclusive split would be the stricter clinical test.
  • The two views are both computed from the same single lead, so the multi-view diversity is representational rather than sensor-level; the same Dempster-Shafer fusion could be applied to genuinely independent inputs such as different leads or ECG plus another modality, where disagreements carry different meaning.
  • The fusion rule is parameter-free and does not need a learned gating network, so it could be dropped into any two-branch classifier; a testable prediction is that its benefit grows as the two branches become more complementary and more prone to disagree on hard examples.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes UAMAC, a multi-view ECG heartbeat classifier that combines a BiLSTM on raw single-lead beats with a ViT on GAF images, and fuses the two views through a rule presented as a Dempster-Shafer combination. Experiments on MIT-BIH and INCART report state-of-the-art accuracies of 98.8% and 99.5%, together with ablations and a noise-robustness comparison against score-level and feature-level fusion.

Significance. If the method and evaluation were sound, UAMAC would be a useful contribution to single-lead ECG beat classification, particularly for its stated objective of noise-resistant multi-view fusion. The paper includes a careful ablation of time-series and image-space backbones and compares against several previous systems. However, the central fusion formula is not a valid Dempster-Shafer combination, and the evaluation protocol appears to allow patient leakage; these issues undermine the main technical and empirical claims. The paper does not provide code or statistical significance tests, which further limits verification.

major comments (3)
  1. [Section III-C, Eq. (3)] Equation (3) is not Dempster's rule of combination. For two Bayesian mass functions obtained from softmax outputs, Dempster's rule gives m_{12}({y}) = p^1_y p^2_y / \sum_z p^1_z p^2_z, with a single global normalizer. Equation (3) instead uses the per-class denominator 1 - C^1_y C^2_y, so the fused values do not sum to 1. For example, with p^1 = p^2 = (0.5, 0.3, 0.2), Eq. (3) yields approximately (0.333, 0.207, 0.071), summing to 0.612. Moreover, C^k_y is defined in Eq. (2) as a set, so the product C^1_y C^2_y is undefined as written. Because the paper's central novelty is an uncertainty-aware fusion grounded in Dempster-Shafer theory, this is a load-bearing error: the method is at present an ad hoc per-class product rule, and the theoretical motivation as well as the noise-robustness interpretation drawn from DST are unsupported.
  2. [Section IV-B, Datasets] The description 'we first created an 80-20 split for training and testing' does not specify a patient-disjoint split. On MIT-BIH (47 patients) and INCART (32 patients), a random heartbeat-level split places beats from the same patient in both training and test sets, which is a known form of data leakage and inflates reported accuracy. The claimed state-of-the-art results in Table I (98.8% and 99.5%) are therefore not reliable unless the split is patient-wise. The authors should specify the split explicitly and report results under a patient-independent or inter-patient protocol, for example following AAMI recommendations.
  3. [Table I, Baseline comparison] The comparison in Table I is across prior publications that use different class sets, preprocessing, lead configurations, and train/test protocols. For instance, on INCART the authors use three classes (N, S, V) while some listed baselines use different class definitions, and on MIT-BIH several baselines use different evaluation splits. Without a unified protocol or directly re-run baselines, the claim that UAMAC outperforms all previous methods is not established independently of the split issue raised above.
minor comments (5)
  1. [Section III-C, Eqs. (1)-(3)] The notation C^k_y is introduced as a set in Eq. (2) but used as a scalar in Eq. (3); the authors should define it explicitly as the sum of conflicting probabilities, for example C^k_y = \sum_{j \neq y} p^k_j, or revise the formula accordingly.
  2. [Section III-A and IV-A] The number of ECG time steps n per heartbeat is never specified, despite being a free parameter in the GAF construction and the BiLSTM input; this should be stated in the implementation details.
  3. [Section III-B] It is not stated whether the ViT-base network is pre-trained on ImageNet or trained from scratch for the GAF view, whereas the ResNet-18 and Inception-v3 baselines in Table II are explicitly described as ImageNet-pretrained; this difference should be disclosed.
  4. [Figure 1 and throughout] The figure caption contains typographical errors (e.g., 'consisting1)' instead of 'consisting of:') and the architecture diagram is not referenced in the text; the paper would benefit from a proofreading pass.
  5. [Tables I-III] The numerical differences between some methods are small (e.g., 98.8% vs. 98.6%), but no confidence intervals or significance tests are reported, making it difficult to judge whether the improvements are meaningful.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the fusion rule and experimental claims are self-contained empirical constructs, not derived from or equivalent to their inputs.

full rationale

The paper's central methodological claim is a multi-view architecture with an uncertainty-aware fusion module. The fusion formula in Eq. (3) is defined directly from the softmax outputs of the two views, and it is not fitted to any target-derived quantity. The reported accuracies are measured on external public datasets (MIT-BIH and INCART) against published baselines, so the performance claims are empirically anchored rather than derived from the method's own assumptions. The only self-citations, Refs. [14] and [15], appear in the related-work discussion of contrastive ECG representation learning and are not load-bearing for the proposed UAMAC framework. Concerns raised by a careful reader, such as the departure of Eq. (3) from textbook Dempster's rule of combination and the random heartbeat-level 80-20 split that may allow patient leakage, are correctness and evaluation-protocol threats, not circularity: they do not make any prediction equivalent to its input by construction. Accordingly, no circular step can be exhibited under the required standard, and the correct circularity score is 0.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on hand-picked design choices (GAF constants, input length, ViT pretraining, SMOTE parameters) that are not specified, and on two domain assumptions that are questionable: that softmax outputs can serve as DST mass functions and that the two views are independent. The evaluation additionally assumes that a random heartbeat split is valid, which is not the clinically standard protocol.

free parameters (4)
  • GAF polar radius constant t
    Controls the radius scaling in the polar mapping (Section III-B); the value is not specified, which affects the GAF image.
  • Number of ECG time steps n per heartbeat
    Input length for both BiLSTM and GAF generation; not stated in the paper, though it determines the input dimensions.
  • Vision Transformer pretraining strategy
    The paper adopts ViT-base [19] but does not state whether ImageNet-pretrained weights are used or if it is trained from scratch, which heavily affects performance.
  • SMOTE nearest neighbors
    SMOTE is applied to the training split (Section IV-B) but the number of neighbors and other parameters are not specified.
assumptions (3)
  • domain assumption Softmax outputs of each view can be treated as Dempster-Shafer mass functions over singleton classes.
    Section III-C assumes the conditions for DST hold, but standard DST allows masses over the power set; reducing to singletons makes the combination equivalent to a simple product rule, not true DST.
  • domain assumption The two views are independent evidence sources for Dempster's rule.
    Equation 3 multiplies the evidence terms as if the views are independent, but the GAF image is derived from the same raw ECG signal, so the views are strongly dependent.
  • ad hoc to paper A random 80-20 heartbeat split provides an unbiased evaluation.
    Section IV-B describes an 80-20 split without holding out patients, so the same patient appears in both training and test sets, violating the standard AAMI inter-patient protocol.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uncertainty-Aware Multi-view Arrhythmia Classification from ECG." pith.science (2026). https://pith.science/paper/4Z7CZ3WZ

@misc{pith2026250606342,
  author       = {Pith},
  title        = {Pith review of: Uncertainty-Aware Multi-view Arrhythmia Classification from ECG},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4Z7CZ3WZ}},
  note         = {Machine review of arXiv:2506.06342}
}
read the original abstract

We propose a deep neural architecture that performs uncertainty-aware multi-view classification of arrhythmia from ECG. Our method learns two different views (1D and 2D) of single-lead ECG to capture different types of information. We use a fusion technique to reduce the conflict between the different views caused by noise and artifacts in ECG data, thus incorporating uncertainty to obtain stronger final predictions. Our framework contains the following three modules (1) a time-series module to learn the morphological features from ECG; (2) an image-space learning module to learn the spatiotemporal features; and (3) the uncertainty-aware fusion module to fuse the information from the two different views. Experimental results on two real-world datasets demonstrate that our framework not only improves the performance on arrhythmia classification compared to the state-of-the-art but also shows better robustness to noise and artifacts present in ECG.

Figures

Figures reproduced from arXiv: 2506.06342 by the authors.

Figure 1
Figure 1. Proposed architecture, consisting 1) BiLSTM module for time-series learning 2) Vision Transformer module for image￾space learning on ECG GAF 3) The Uncertainty-Aware Fusion Block. data, an S shape reconstruction method was utilized in [6] within 19 layer deep residual network for automatic heartbeat classification. In contrast, 2D representations of ECG signals, including spectrograms and scalograms, were incorporat… view at source ↗
Figure 2
Figure 2. ECG signals with different levels of noise. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Performance of our method and different fusion techniques with both artificial and real-world noise from NSTDB. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 34 canonical work pages

  1. [1]

    Screening for cardiac contractile dysfunction using an artificial intelligence–enabled electrocardiogram,

    Z. I. Attia, S. Kapa, F. Lopez-Jimenez, P. M. McKie, D. J. Ladewig, G. Satam, P. A. Pellikka, M. Enriquez-Sarano, P. A. Noseworthy, and T. M. Munger, “Screening for cardiac contractile dysfunction using an artificial intelligence–enabled electrocardiogram,”Nature Medicine, vol. 25, no. 1, pp. 70–74, 2019

    work page 2019

  2. [2]

    Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network,

    A. Y . Hannun, P. Rajpurkar, M. Haghpanahi, G. H. Tison, C. Bourn, M. P. Turakhia, and A. Y . Ng, “Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network,”Nature Medicine, vol. 25, pp. 65–69, 2019

    2019

  3. [3]

    Ecg heart-beat classification using multimodal image fusion,

    Z. Ahmad, A. Tabassum, L. Guan, and N. Khan, “Ecg heart-beat classification using multimodal image fusion,” inIEEE Int. Conf. on Acoustics, Speech and Signal Processing, 2021, pp. 1330–1334

    work page 2021

  4. [4]

    A dilated residual vision transformer for atrial fibrillation detection from stacked time-frequency ecg representations,

    S. Pratiher, A. Srivastava, Y . B. Priyatha, N. Ghosh, and A. Patra, “A dilated residual vision transformer for atrial fibrillation detection from stacked time-frequency ecg representations,” inIEEE Int. Conf. on Acoustics, Speech and Signal Processing, 2022, pp. 1121–1125

    work page 2022

  5. [5]

    A survey on multi-view learning,

    C. Xu, D. Tao, and C. Xu, “A survey on multi-view learning,”arXiv preprint arXiv:1304.5634, 2013

    arXiv 2013

  6. [6]

    Automatic heartbeat classification using s-shaped reconstruction and a squeeze-and- excitation residual network,

    X. Li, F. Zhang, Z. Sun, D. Li, X. Kong, and Y . Zhang, “Automatic heartbeat classification using s-shaped reconstruction and a squeeze-and- excitation residual network,”Computers in Biology and Medicine, vol. 140, pp. 105108, 2022

    work page 2022

  7. [7]

    Atrial fibrillation detection using convolutional neural networks on 2-dimensional representation of ecg signal,

    B. Król-Józaga, “Atrial fibrillation detection using convolutional neural networks on 2-dimensional representation of ecg signal,”Biomedical Signal Processing and Control, vol. 74, pp. 103470, 2022

    work page 2022

  8. [8]

    Spectro-temporal feature based multi-channel convolutional neural network for ecg beat classification,

    C. Hao, S. Wibowo, M. Majmudar, and K. S. Rajput, “Spectro-temporal feature based multi-channel convolutional neural network for ecg beat classification,” inInt. Conf. of the IEEE Engineering in Medicine and Biology Society, 2019, pp. 5642–5645

    work page 2019

Show all 37 references
  1. [9]

    Multi-class arrhythmia detection based on neural network with multi-stage features fusion,

    R. Wang, Q. Yao, X. Fan, and Y . Li, “Multi-class arrhythmia detection based on neural network with multi-stage features fusion,” inIEEE Int. Conf. on Systems, Man and Cybernetics, 2019, pp. 4082–4087

    2019

  2. [10]

    On merging feature engineering and deep learning for diagnosis, risk prediction and age estimation based on the 12-lead ecg,

    E. Zvuloni, J. Read, A. H. Ribeiro, A. L. P. Ribeiro, and J. A. Behar, “On merging feature engineering and deep learning for diagnosis, risk prediction and age estimation based on the 12-lead ecg,”IEEE Trans. on Biomedical Engineering, 2023

    2023

  3. [11]

    Multi-class arrhythmia detection from 12-lead varied-length ecg using attention-based time- incremental convolutional neural network,

    Q. Yao, R. Wang, X. Fan, J. Liu, and Y . Li, “Multi-class arrhythmia detection from 12-lead varied-length ecg using attention-based time- incremental convolutional neural network,”Information Fusion, vol. 53, pp. 174–182, 2020

    2020

  4. [12]

    Multi-label correlation guided feature fusion network for abnormal ecg diagnosis,

    Z. Ge, X. Jiang, Z. Tong, P. Feng, B. Zhou, M. Xu, Z. Wang, and Y . Pang, “Multi-label correlation guided feature fusion network for abnormal ecg diagnosis,”Knowledge-Based Systems, vol. 233, pp. 107508, 2021

    2021

  5. [13]

    A multi- view multi-scale neural network for multi-label ecg classification,

    S. Yang, C. Lian, Z. Zeng, B. Xu, J. Zang, and Z. Zhang, “A multi- view multi-scale neural network for multi-label ecg classification,”IEEE Trans. on Emerging Topics in Computational Intelligence, 2023

    2023

  6. [14]

    Analysis of augmentations for contrastive ecg representation learning,

    S. Soltanieh, A. Etemad, and J. Hashemi, “Analysis of augmentations for contrastive ecg representation learning,” inIEEE Int. Joint Conf. on Neural Networks, 2022, pp. 1–10

    2022

  7. [15]

    In-distribution and out-of- distribution self-supervised ecg representation learning for arrhythmia detection,

    S. Soltanieh, J. Hashemi, and A. Etemad, “In-distribution and out-of- distribution self-supervised ecg representation learning for arrhythmia detection,”IEEE Journal of Biomedical and Health Informatics, 2023

    2023

  8. [16]

    Encoding time series as images for visual inspection and classification using tiled convolutional neural networks,

    Z. Wang and T. Oates, “Encoding time series as images for visual inspection and classification using tiled convolutional neural networks,” inAAAI Conf. on Artificial Intelligence, 2015, vol. 1

    2015

  9. [17]

    Ha-resnet: Residual neural network with hidden attention for ecg arrhythmia detection using two- dimensional signal,

    Y . Guan, Y . An, J. Xu, N. Liu, and J. Wang, “Ha-resnet: Residual neural network with hidden attention for ecg arrhythmia detection using two- dimensional signal,”IEEE/ACM Trans. on Computational Biology and Bioinformatics, 2022

    2022

  10. [18]

    A survey on vision transformer,

    K. Han, Y . Wang, H. Chen, X. Chen, J. Guo, Z. Liu, Y . Tang, A. Xiao, C. Xu, and Y . Xu, “A survey on vision transformer,”IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 87–110, 2022

    2022

  11. [19]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, and S. Gelly, “An image is worth 16x16 words: Transformers for image recognition at scale,”Int. Conf. on Learning Representations, 2020

    2020

  12. [20]

    Ecg heartbeat classi- fication: A deep transferable representation,

    M. Kachuee, S. Fazeli, and M. Sarrafzadeh, “Ecg heartbeat classi- fication: A deep transferable representation,” inIEEE Int. Conf. on Healthcare Informatics, 2018, pp. 443–444

    2018

  13. [21]

    Combining the taguchi method and convolutional neural networks for arrhythmia classification by using ecg images with single heartbeats,

    S. Li, M. Huang, and Y . Wu, “Combining the taguchi method and convolutional neural networks for arrhythmia classification by using ecg images with single heartbeats,”Mathematics, vol. 11, no. 13, pp. 2841, 2023

    2023

  14. [22]

    A new approach of transparent and explainable artificial intelligence technique for patient-specific ecg beat classification,

    A. J. Prakash, K. K. Patro, S. Saunak, P. Sasmal, P. L. Kumari, and T. Geetamma, “A new approach of transparent and explainable artificial intelligence technique for patient-specific ecg beat classification,”IEEE Sensors Letters, 2023

    2023

  15. [23]

    Gen- eralization of convolutional neural networks for ecg classification using generative adversarial networks,

    A. M. Shaker, Ma. Tantawi, H. A. Shedeed, and M. F. Tolba, “Gen- eralization of convolutional neural networks for ecg classification using generative adversarial networks,”IEEE Access, vol. 8, pp. 35592–35605, 2020

    2020

  16. [24]

    Ecg classification using 1-d convolutional deep residual neural network,

    F. Khan, X. Yu, Z. Yuan, and A. U. Rehman, “Ecg classification using 1-d convolutional deep residual neural network,”PLoS One, vol. 18, no. 4, pp. e0284791, 2023

    2023

  17. [25]

    A pyramid-like model for heartbeat classification from ecg recordings,

    J. He, L. Sun, J. Rong, H. Wang, and Y . Zhang, “A pyramid-like model for heartbeat classification from ecg recordings,”PLoS One, vol. 13, no. 11, pp. e0206593, 2018

    2018

  18. [26]

    Three-class ecg beat classification by ordinal entropies,

    Jean B. B. à Mougoufan, J.S. A. E. Fouda, M. Tchuente, and W. Koepf, “Three-class ecg beat classification by ordinal entropies,”Biomedical Signal Processing and Control, vol. 67, pp. 102506, 2021

    2021

  19. [27]

    Cross-database generalization of deep learning models for arrhythmia classification,

    E. Merdjanovska and A. Rashkovska, “Cross-database generalization of deep learning models for arrhythmia classification,” inInt. Convention on Information, Communication and Electronic Technology, 2021, pp. 346–351

    2021

  20. [28]

    A micro neural network for healthcare sensor data stream classification in sustainable and smart cities,

    J. Wu, L. Sun, D. Peng, and S. Siuly, “A micro neural network for healthcare sensor data stream classification in sustainable and smart cities,”Computational Intelligence and Neuroscience, vol. 2022, 2022

    2022

  21. [29]

    A generalization of bayesian inference,

    A. P. Dempster, “A generalization of bayesian inference,”Journal of the Royal Statistical Society: Series B (Methodological), vol. 30, no. 2, pp. 205–232, 1968

    1968

  22. [30]

    A mathematical theory of evidence for gls shackle,

    G. Fioretti, “A mathematical theory of evidence for gls shackle,”Mind & Society, vol. 2, pp. 77–98, 2001

    2001

  23. [31]

    Motor imagery eeg classification based on riemannian sparse optimization and dempster- shafer fusion of multi-time-frequency patterns,

    J. Jin, T. Qu, R. Xu, X. Wang, and A. Cichocki, “Motor imagery eeg classification based on riemannian sparse optimization and dempster- shafer fusion of multi-time-frequency patterns,”IEEE Trans. on Neural Systems and Rehabilitation Engineering, vol. 31, pp. 58–67, 2022

    2022

  24. [32]

    Hierarchical filtering with online learned priors for ecg denoising,

    T. Locher, G. Revach, N. Shlezinger, R. J. G. van Sloun, and R. Vullings, “Hierarchical filtering with online learned priors for ecg denoising,” in IEEE Int. Conf. on Acoustics, Speech and Signal Processing, 2023, pp. 1–5

    2023

  25. [33]

    The impact of the mit-bih arrhythmia database,

    G. B. Moody and R. G. Mark, “The impact of the mit-bih arrhythmia database,”IEEE Engineering in Medicine and Biology Magazine, vol. 20, no. 3, pp. 45–50, 2001

    2001

  26. [34]

    Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals,

    A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C. Peng, and H. E. Stanley, “Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals,”Circulation, vol. 10...

    2000

  27. [35]

    Smote: synthetic minority over-sampling technique,

    N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,”Journal of Artificial Intelligence Research, vol. 16, pp. 321–357, 2002

    2002

  28. [36]

    Imagenet: A large-scale hierarchical image database,

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei- Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE conference on computer vision and pattern recognition. 2009, pp. 248–255, Ieee

    2009

  29. [37]

    A noise stress test for arrhythmia detectors,

    George B. Moody, W. Muldrow, and Roger G. Mark, “A noise stress test for arrhythmia detectors,”Computers in Cardiology, vol. 11, no. 3, pp. 381–384, 1984

    1984

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.