pith. sign in

arxiv: 2506.16494 · v3 · submitted 2025-06-19 · 💻 cs.LG · eess.SP

Manifold Learning for Personalized and Label-Free Detection of Cardiac Arrhythmias

Amir Reza Vazifeh , Jason W. Fleischer This is my paper

Pith reviewed 2026-05-19 08:28 UTC · model grok-4.3

classification 💻 cs.LG eess.SP
keywords ECGarrhythmia detectionnonlinear dimensionality reductiont-SNEUMAPunsupervised learningmanifold learningpersonalized healthcare
0
0 comments X

The pith

Nonlinear dimensionality reduction clusters ECG heartbeats from one person into normal and arrhythmic groups without labels or training.

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

The paper shows that algorithms such as t-SNE and UMAP can map ECG heartbeats into a 2D space where, for any single patient, normal beats form one cluster and arrhythmic beats form separate clusters, all without supervised training or medical annotations. This unsupervised separation works because the methods preserve local neighborhoods in the data, allowing simple nearest-neighbor classification on the embeddings to reach median accuracy above 98 percent for arrhythmia identification. The same approach applied to mixed patient data instead groups signals by individual, highlighting personal morphological differences. Readers may care because the method sidesteps problems of inconsistent labels and population biases that limit many current automated ECG tools.

Core claim

Applying nonlinear dimensionality reduction to heartbeats of a single individual separates normal beats from arrhythmias into distinct clusters identifiable in an unsupervised manner. Both t-SNE and UMAP produce 2D embeddings with trustworthiness scores of at least 0.95, and a k-nearest neighbors classifier on these embeddings discriminates individual recordings with at least 80 percent accuracy while identifying arrhythmias with median accuracy of at least 98 percent and median F1-score of at least 85 percent, outperforming classification in the original high-dimensional space.

What carries the argument

Nonlinear dimensionality reduction via t-SNE and UMAP, which embeds high-dimensional ECG signals into 2D spaces while preserving local neighborhood structure to expose medically relevant clusters.

If this is right

  • NLDR applied to mixed populations groups signals by individual rather than by arrhythmia type.
  • Per-patient embeddings enable label-free arrhythmia detection with high median accuracy using only a simple classifier.
  • The 2D representations outperform the original high-dimensional ECG features for classification tasks.
  • The approach operates without pretraining or population-level labeled datasets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Per-patient manifold learning could support real-time analysis on wearable devices that lack access to large supervised training sets.
  • If clusters map reliably to specific arrhythmia subtypes, the method might reduce dependence on variable expert annotations across clinics.
  • The same per-individual embedding strategy could be tested on other physiological time series such as EEG or blood-pressure waveforms.

Load-bearing premise

The clusters that appear in the 2D space must reflect true differences between normal and abnormal heartbeats rather than noise, lead placement, or other non-medical sources of signal variation.

What would settle it

Re-applying t-SNE or UMAP to the MIT-BIH database and finding that normal and arrhythmic beats from the same patient occupy the same cluster, or that k-NN accuracy on the embeddings falls below 70 percent, would show the unsupervised separation does not hold.

Figures

Figures reproduced from arXiv: 2506.16494 by Amir Reza Vazifeh, Jason W. Fleischer.

Figure 1
Figure 1. Figure 1: ECG signals from multiple leads in a standard 12-lead setup. The data is from recording A6371 of the CPSC2018 dataset [6]. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Different heartbeat profiles. Each subplot is sourced from different individuals in the MIT-BIH dataset, all recorded using the modified limb lead II (MLII). Top row: Normal beats (N), Middle row: Premature ventricular contraction (V), Bottom row: Atrial premature beat (A) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Class distribution of AAMI labels in the reduced MIT-BIH [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic of our approach. 2.2. Signal Segmentation and Filtering The first step in real-world analysis is to segment the ECG signals into isolated heartbeats. This requires identify￾ing the R-peaks, which must be detected from raw signals. Various algorithms have been developed for this purpose, including methods by Christov [40], Pan-Tompkins [41], and the NeuroKit2 framework [42]. A recent study showed … view at source ↗
Figure 5
Figure 5. Figure 5: Fiducial points and method of segmentation selection. Each heartbeat is identified by its corresponding R-peak. The signal for each [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of heartbeat signal populations from 40 recordings. Shown are the 2D latent spaces using PCA, t-SNE, and UMAP from [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: KNN classification results based on the embeddings shown in Figure 6. Left: patient identification, Middle: arrhythmia detection, Right: [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: KNN classification results on 2D embeddings of each patient (Figures A.14–A.17) for arrhythmia detection. A KNN classifier with [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Analysis of Recording 116 with dimensionality reduction methods. Top: 2D visualizations using PCA, t-SNE, and UMAP. A KNN [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Analysis of Recording 231 with dimensionality reduction methods. Top: 2D visualizations using PCA, t-SNE, and UMAP. A KNN [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Analysis of Recording 209 with dimensionality reduction methods. Top: 2D visualizations using PCA, t-SNE, and UMAP. A KNN [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Analysis of Recording 207 with dimensionality reduction methods. Top: 2D visualizations using PCA, t-SNE, and UMAP. A KNN [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
read the original abstract

Electrocardiograms (ECGs) provide non-invasive measurements of heart activity and are established tools for detecting cardiac arrhythmias. Although supervised machine learning has emerged as a promising approach for automated heartbeat classification, substantial variations in ECG signals across individuals and leads, combined with inconsistent labeling standards and dataset biases, make it difficult to develop generalizable models. Dimensionality reduction maps high-dimensional data into a lower-dimensional space while preserving the underlying structure, enabling visualization and pattern discovery. Conventional methods, e.g., principal component analysis, prioritize large variances and typically overlook subtle yet clinically relevant patterns. Here, we show that nonlinear dimensionality reduction (NLDR) algorithms, e.g., t-SNE and UMAP, can identify medically relevant features in ECG signals without pretraining or prior information. Using the MIT-BIH Arrhythmia Database, we show that: a) applying NLDR to a mixed population of heartbeats reveals inter-individual morphological differences, as signals from the same person cluster together in latent spaces; and b) applying NLDR to heartbeats of a single individual separates normal beats from arrhythmias into distinct clusters, identifiable in an unsupervised manner. To our knowledge, this is the first systematic evaluation of NLDR for unsupervised arrhythmia detection. Both UMAP and t-SNE achieved trustworthiness scores >=0.95, indicating that local neighborhoods are well preserved in the embedding. Classification on 2D embeddings outperforms the original high-dimensional space, with a k-NN classifier discriminating individual recordings with >=80% accuracy and identifying arrhythmias with median accuracy >=98% and median F1-score >=85%. These results show that NLDR holds much promise for cardiac monitoring and personalized healthcare.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper claims that nonlinear dimensionality reduction (NLDR) algorithms such as t-SNE and UMAP, when applied to ECG heartbeats from the MIT-BIH Arrhythmia Database, can identify medically relevant features without pretraining or prior information. Specifically, NLDR on mixed-population data reveals inter-individual morphological differences via clustering, while per-patient application separates normal beats from arrhythmias into distinct unsupervised clusters. Both methods achieve trustworthiness scores >=0.95; k-NN classification on the resulting 2D embeddings outperforms the original high-dimensional space, yielding >=80% accuracy for discriminating individual recordings and median accuracy >=98% with median F1-score >=85% for arrhythmia identification.

Significance. If the clusters and performance gains demonstrably isolate arrhythmia-specific morphology rather than artifacts, the approach could enable practical label-free, personalized arrhythmia detection in cardiac monitoring, reducing dependence on large labeled datasets and supervised models. The systematic evaluation on a public benchmark and the reported local neighborhood preservation are strengths that support further investigation.

major comments (3)
  1. [Results (trustworthiness evaluation and cluster visualization)] The trustworthiness score >=0.95 (reported in the abstract and results) only certifies local neighborhood preservation and does not test whether the global structure of the 2D embedding aligns with clinical arrhythmia annotations (N vs. V, S, etc.). Without quantitative alignment metrics such as adjusted Rand index or cluster purity between unsupervised clusters and ground-truth labels, the separation could arise from non-arrhythmia factors like amplitude, baseline wander, or lead-specific shape, weakening the central claim of medically relevant feature identification.
  2. [Methods and Experimental Setup] The manuscript provides insufficient detail on preprocessing steps for the ECG signals, hyperparameter choices for t-SNE (e.g., perplexity) and UMAP (e.g., n_neighbors, min_dist), and the train/test protocol or cross-validation for the k-NN classifier. These omissions make it impossible to assess reproducibility of the reported median accuracy >=98% and F1 >=85%, or to determine whether the outperformance over the original space is statistically significant.
  3. [Abstract and Results (classification experiments)] The label-free detection claim (title and abstract) sits in tension with the use of supervised k-NN classification to quantify arrhythmia identification performance. If detection is intended to be fully unsupervised, evaluation should emphasize unsupervised metrics (e.g., silhouette score on clusters or agreement with annotations via ARI) rather than supervised accuracy, which can succeed even when embeddings capture confounding morphological variation.
minor comments (2)
  1. [Abstract] The abstract states median values without indicating the number of patients or recordings over which the median is taken or the inter-quartile range; adding this information would clarify variability across individuals.
  2. [Figures (embedding visualizations)] Figure captions and axis labels for the 2D embeddings should explicitly note whether points are colored by ground-truth arrhythmia labels or by unsupervised cluster assignment to avoid ambiguity in interpreting the visual separation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the presentation and strengthen the manuscript. We address each major point below and indicate revisions where we agree changes are needed.

read point-by-point responses

  1. Referee: The trustworthiness score >=0.95 only certifies local neighborhood preservation and does not test whether the global structure of the 2D embedding aligns with clinical arrhythmia annotations (N vs. V, S, etc.). Without quantitative alignment metrics such as adjusted Rand index or cluster purity, the separation could arise from non-arrhythmia factors.

    Authors: We agree that trustworthiness primarily assesses local neighborhood preservation and does not directly quantify global alignment with clinical labels. Our visualizations show patient-specific and arrhythmia-specific clustering, but to better support the claim of medically relevant features, we will add adjusted Rand index (ARI) and cluster purity metrics comparing unsupervised clusters to ground-truth annotations in the revised results section. revision: yes

  2. Referee: The manuscript provides insufficient detail on preprocessing steps for the ECG signals, hyperparameter choices for t-SNE (e.g., perplexity) and UMAP (e.g., n_neighbors, min_dist), and the train/test protocol or cross-validation for the k-NN classifier.

    Authors: We acknowledge this omission limits reproducibility. In the revised Methods section we will provide: full preprocessing details (filtering, normalization, R-peak detection and segmentation from MIT-BIH), exact hyperparameter settings for t-SNE and UMAP, and the train/test splits plus cross-validation scheme used for k-NN, including any statistical tests for performance differences. revision: yes

  3. Referee: The label-free detection claim (title and abstract) sits in tension with the use of supervised k-NN classification to quantify arrhythmia identification performance. If detection is intended to be fully unsupervised, evaluation should emphasize unsupervised metrics (e.g., silhouette score on clusters or agreement with annotations via ARI) rather than supervised accuracy.

    Authors: The embedding step itself is unsupervised; k-NN serves only as a post-hoc quantitative measure of separation quality, which is standard for validating embeddings. We will revise the abstract and discussion to reduce ambiguity around 'label-free' and will add unsupervised metrics such as silhouette scores alongside the existing k-NN results. revision: partial

Circularity Check

0 steps flagged

No circularity detected; standard NLDR application remains self-contained

full rationale

The paper applies established NLDR methods (t-SNE and UMAP) directly to raw or lightly preprocessed ECG segments from the public MIT-BIH database. Cluster separation and k-NN accuracies on the resulting 2D embeddings are reported as empirical outcomes evaluated against external clinical annotations; no equations in the manuscript define the embedding coordinates in terms of the target arrhythmia labels or vice versa. No self-citations, uniqueness theorems, or ansatzes from prior author work are invoked to justify the core pipeline. The trustworthiness metric certifies neighborhood preservation independently of the arrhythmia classification results. This is a conventional experimental workflow with no load-bearing reductions to fitted inputs or self-referential definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions of manifold learning that ECG data lies on a lower-dimensional manifold and that local neighborhood preservation in embeddings captures clinically relevant structure, plus properties of the MIT-BIH Arrhythmia Database.

axioms (1)
  • domain assumption High-dimensional ECG heartbeat data lies on a lower-dimensional manifold that can be recovered by nonlinear dimensionality reduction
    This is the core premise invoked when applying t-SNE and UMAP to reveal clusters corresponding to individuals or arrhythmia types.

pith-pipeline@v0.9.0 · 5833 in / 1374 out tokens · 55303 ms · 2026-05-19T08:28:45.957452+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages · 1 internal anchor

  1. [1]

    The Heart of the World,

    M. Di Cesare, P. Perel, S. Taylor, C. Kabudula, H. Bixby, T. A. Gaziano, D. V . McGhie, J. Mwangi, B. Pervan, J. Narula, D. Pineiro, and F. J. Pinto, “The Heart of the World,”Global Heart, vol. 19, no. 1, p. 11, Jan. 2024, doi: https://doi.org/10.5334/gh.1288

    work page doi:10.5334/gh.1288 2024

  2. [2]

    R. G. Carbone and A.-M. Russell, ”Smoking cessation in heart and chronic respiratory disease: A healthy global strategy,” International Journal of Cardiology, vol. 418, Art no. 132584, 2025. doi: 10.1016/j.ijcard.2024.132584

    work page doi:10.1016/j.ijcard.2024.132584 2025

  3. [3]

    Desai and S

    D. Desai and S. Hajouli, ”Arrhythmias,” StatPearls [Internet], Treasure Island (FL): StatPearls Publishing, updated June 5, 2023. Available: https://www.ncbi.nlm.nih.gov/books/NBK558923/

    work page 2023

  4. [4]

    Kligfield et al., ”Recommendations for the Standardization and Interpretation of the Electrocardiogram,”Circulation, vol

    P. Kligfield et al., ”Recommendations for the Standardization and Interpretation of the Electrocardiogram,”Circulation, vol. 115, no. 10, pp. 1306–1324, 2007, doi: 10.1161/CIRCULATIONAHA.106.180200

    work page doi:10.1161/circulationaha.106.180200 2007

  5. [5]

    Rajoub, ”Machine learning in biomedical signal processing with ECG applications,” inBiomedical Signal Processing and Artificial Intelli- gence in Healthcare, W

    B. Rajoub, ”Machine learning in biomedical signal processing with ECG applications,” inBiomedical Signal Processing and Artificial Intelli- gence in Healthcare, W. Zgallai, Ed., Developments in Biomedical Engineering and Bioelectronics, ch. 4, pp. 91-112. Academic Press, 2020. ISBN: 9780128189467. doi: 10.1016/B978-0-12-818946-7.00004-4. [Online]. Availab...

    work page doi:10.1016/b978-0-12-818946-7.00004-4 2020

  6. [6]

    F. Liu, C. Liu, L. Zhao, X. Zhang, X. Wu, X. Xu, Y . Liu, C. Ma, S. Wei, Z. He, J. Li, and E. N. Kwee, ”An Open Access Database for Evaluating the Algorithms of Electrocardiogram Rhythm and Morphology Abnormality Detection,” Journal of Medical Imaging and Health Informatics, vol. 8, no. 7, pp. 1368-1373, Sep. 2018, doi: https://doi.org/10.1166/jmihi.2018.2442

    work page doi:10.1166/jmihi.2018.2442 2018

  7. [7]

    Accuracy of Physicians’ Electrocardiogram Interpretations: A Systematic Review and Meta-analysis,

    D. A. Cook, S. Y . Oh, and M. V . Pusic, “Accuracy of Physicians’ Electrocardiogram Interpretations: A Systematic Review and Meta-analysis,” JAMA Internal Medicine, vol. 180, no. 11, pp. 1461–1471, Nov. 2020, doi: https://doi.org/10.1001/jamainternmed.2020.3989

    work page doi:10.1001/jamainternmed.2020.3989 2020

  8. [8]

    ECG interpretation skill acquisition: A review of learning, teaching and assessment,

    C. J. Breen, G. P. Kelly, and W. G. Kernohan, “ECG interpretation skill acquisition: A review of learning, teaching and assessment,” Journal of Electrocardiology, vol. 73, pp. 125–128, 2022, doi: https://doi.org/10.1016/j.jelectrocard.2019.03.010

    work page doi:10.1016/j.jelectrocard.2019.03.010 2022

  9. [9]

    Usha Kumari, A

    Ch. Usha Kumari, A. Sampath Dakshina Murthy, B. Lakshmi Prasanna, M. Pala Prasad Reddy, and Asisa Kumar Panigrahy, ”An automated detection of heart arrhythmias using machine learning technique: SVM,”Materials Today: Proceedings, vol. 45, part 2, pp. 1393–1398, 2021, doi: 10.1016/j.matpr.2020.07.088

    work page doi:10.1016/j.matpr.2020.07.088 2021

  10. [10]

    Baraeinejad et al., ”Design and Implementation of an Ultralow-Power ECG Patch and Smart Cloud-Based Platform,” IEEE Transactions on Instrumentation and Measurement, vol

    B. Baraeinejad et al., ”Design and Implementation of an Ultralow-Power ECG Patch and Smart Cloud-Based Platform,” IEEE Transactions on Instrumentation and Measurement, vol. 71, pp. 1-11, 2022, Art no. 2506811. doi: 10.1109/TIM.2022.3164151

    work page doi:10.1109/tim.2022.3164151 2022

  11. [11]

    Ebrahimzadeh and A

    A. Ebrahimzadeh and A. Khazaee, ”Detection of premature ventricular contractions using MLP neural networks: A comparative study,” Measurement, vol. 43, no. 1, pp. 103–112, 2010, doi: 10.1016/j.measurement.2009.07.002

    work page doi:10.1016/j.measurement.2009.07.002 2010

  12. [12]

    D. E. Moghaddam, A. Muguli, M. Razavi, and B. Aazhang, ”A graph-based cardiac arrhythmia classification methodology using one-lead ECG recordings,”Intelligent Systems with Applications, vol. 22, p. 200385, 2024. doi: 10.1016/j.iswa.2024.200385

    work page doi:10.1016/j.iswa.2024.200385 2024

  13. [13]

    Bertsimas, L

    D. Bertsimas, L. Mingardi, and B. Stellato, ”Machine learning for real-time heart disease prediction,” IEEE J. Biomed. Health Inform., vol. 25, no. 9, pp. 3627-3637, Sept. 2021. doi: 10.1109/JBHI.2021.3066347

    work page doi:10.1109/jbhi.2021.3066347 2021

  14. [14]

    J. Yu, X. Wang, X. Chen, and J. Guo, ”Automatic Premature Ventricular Contraction Detection Using Deep Metric Learning and KNN,” Biosensors, vol. 11, no. 3, p. 69, 2021, doi: 10.3390/bios11030069

    work page doi:10.3390/bios11030069 2021

  15. [15]

    D. Lai et al., ”Non-Standardized Patch-Based ECG Lead Together With Deep Learning Based Algorithm for Automatic Screen- ing of Atrial Fibrillation,” IEEE Journal of Biomedical and Health Informatics , vol. 24, no. 6, pp. 1569–1578, June 2020, doi: 10.1109/JBHI.2020.2980454

    work page doi:10.1109/jbhi.2020.2980454 2020

  16. [16]

    P. N. Singh and R. P. Mahapatra, ”A novel deep learning approach for arrhythmia prediction on ECG classification using recurrent CNN with GWO,”Int. J. Inf. Technol., vol. 16, pp. 577–585, 2024. doi: 10.1007/s41870-023-01611-1

    work page doi:10.1007/s41870-023-01611-1 2024

  17. [17]

    Anand, S

    R. Anand, S. V . Lakshmi, D. Pandey, et al., ”An enhanced ResNet-50 deep learning model for arrhythmia detection using electrocardiogram biomedical indicators,”Evolving Systems, vol. 15, pp. 83–97, 2024. doi: 10.1007/s12530-023-09559-0

    work page doi:10.1007/s12530-023-09559-0 2024

  18. [18]

    Rajkumar, M

    A. Rajkumar, M. Ganesan, and R. Lavanya, ”Arrhythmia classification on ECG using Deep Learning,” in Proc. 5th Int. Conf. Adv. Comput. Commun. Syst. (ICACCS), Coimbatore, India, 2019, pp. 365–369. doi: 10.1109/ICACCS.2019.8728362

    work page doi:10.1109/icaccs.2019.8728362 2019

  19. [19]

    Y . D. Daydulo, B. L. Thamineni, and A. A. Dawud, ”Cardiac arrhythmia detection using deep learning approach and time frequency repre- sentation of ECG signals,”BMC Med. Inform. Decis. Mak., vol. 23, no. 232, 2023. doi: 10.1186/s12911-023-02326-w

    work page doi:10.1186/s12911-023-02326-w 2023

  20. [20]

    J. Chen, X. Zhang, L. Xu, V . H. C. de Albuquerque, and W. Wu, ”Implementing the confidence constraint cloud-edge collaborative computing strategy for ultra-efficient arrhythmia monitoring,” Applied Soft Computing, vol. 154, 111402, 2024, doi: https://doi.org/10.1016/j. asoc.2024.111402

    work page doi:10.1016/j 2024

  21. [21]

    Deep learning-based multidimensional feature fusion for classification of ECG arrhythmia,

    J. Cui, L. Wang, X. He, et al., “Deep learning-based multidimensional feature fusion for classification of ECG arrhythmia,”Neural Computing and Applications, vol. 35, pp. 16073–16087, 2023. doi: 10.1007/s00521-021-06487-5

    work page doi:10.1007/s00521-021-06487-5 2023

  22. [22]

    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, May-June 2001. doi: 10.1109/51.932724

    work page doi:10.1109/51.932724 2001

  23. [23]

    A. L. Goldberger, L. A. N. Amaral, L. Glass, et al., ”PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals,”Circulation, vol. 101, no. 23, pp. e215–e220, 2000. doi: 10.1161/01.CIR.101.23.e215

    work page doi:10.1161/01.cir.101.23.e215 2000

  24. [24]

    Sarshar and M

    N. Sarshar and M. Mirzaei, ”Premature ventricular contraction recognition based on a deep learning approach,” J. Healthc. Eng., vol. 2022, Article ID 1450723, Mar. 2022. doi: 10.1155/2022/1450723. PMID: 35378947; PMCID: PMC8976634

    work page doi:10.1155/2022/1450723 2022

  25. [25]

    De Marco, D

    F. De Marco, D. Finlay, and R. R. Bond, ”Classification of premature ventricular contraction using deep learning,” in 2020 Computing in Cardiology, Rimini, Italy, 2020, pp. 1-4. doi: 10.22489/CinC.2020.311

    work page doi:10.22489/cinc.2020.311 2020

  26. [26]

    Mastoi, M

    M. Mastoi, M. S. Memon, A. Lakhan, et al., ”Machine learning-data mining integrated approach for premature ventricular contraction prediction,”Neural Comput. Appl., vol. 33, pp. 11703–11719, 2021. doi: 10.1007/s00521-021-05820-2

    work page doi:10.1007/s00521-021-05820-2 2021

  27. [27]

    T. J. Jun et al., ”Premature Ventricular Contraction Beat Detection with Deep Neural Networks,” in2016 15th IEEE International Conference on Machine Learning and Applications (ICMLA), Anaheim, CA, USA, 2016, pp. 859-864. doi: 10.1109/ICMLA.2016.0154

    work page doi:10.1109/icmla.2016.0154 2016

  28. [28]

    Testing the manifold hypothesis,

    C. Fe fferman, S. Mitter, and H. Narayanan, “Testing the manifold hypothesis,” Journal of the American Mathematical Society , vol. 29, pp. 983–1049, 2016. https://doi.org/10.1090/jams/852

    work page doi:10.1090/jams/852 2016

  29. [29]

    van der Maaten and G

    L. van der Maaten and G. Hinton, ”Visualizing Data using t-SNE,” Journal of Machine Learning Research , vol. 9, no. 86, pp. 2579-2605,

    work page

  30. [30]

    Available: http://jmlr.org/papers/v9/vandermaaten08a.html

    [Online]. Available: http://jmlr.org/papers/v9/vandermaaten08a.html

    work page

  31. [31]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    L. McInnes, J. Healy, and J. Melville, ”UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction,” arXiv, 2020. [Online]. Available: https://arxiv.org/abs/1802.03426

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  32. [32]

    Time series data augmentation for neural networks by time warping with a discriminative teacher,

    L. Weijler, M. Diem, M. Reiter, and M. Maurer-Granofszky, ”Detecting Rare Cell Populations in Flow Cytometry Data Using UMAP,” in Proceedings of the 25th International Conference on Pattern Recognition (ICPR) , Milan, Italy, 2021, pp. 4903-4909. doi: 10.1109/ICPR48806.2021.9413180

    work page doi:10.1109/icpr48806.2021.9413180 2021

  33. [33]

    Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types,

    V . van Unen, T. H¨ollt, N. Pezzotti, et al., “Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types,”Nature Communications, vol. 8, p. 1740, 2017. doi: 10.1038/s41467-017-01689-9

    work page doi:10.1038/s41467-017-01689-9 2017

  34. [34]

    Becht, L

    E. Becht, L. McInnes, J. Healy, et al., ”Dimensionality reduction for visualizing single-cell data using UMAP,” Nature Biotechnology, vol. 37, pp. 38-44, 2019. doi: 10.1038/nbt.4314

    work page doi:10.1038/nbt.4314 2019

  35. [35]

    Data-driven identification of prognostic tumor subpopulations using spatially mapped t-SNE of mass spectrometry imaging data,

    W. M. Abdelmoula, B. Ballu ff, S. Englert, J. Dijkstra, M. J. T. Reinders, A. Walch, L. A. McDonnell, and B. P. F. Lelieveldt, “Data-driven identification of prognostic tumor subpopulations using spatially mapped t-SNE of mass spectrometry imaging data,” Proceedings of the National Academy of Sciences of the United States of America, vol. 113, no. 43, pp....

    work page doi:10.1073/pnas.1510227113 2016

  36. [36]

    On the use of t-distributed stochastic neighbor embedding for data visualization 17 and classification of individuals with Parkinson’s disease,

    F. H. M. Oliveira, A. R. P. Machado, and A. O. Andrade, “On the use of t-distributed stochastic neighbor embedding for data visualization 17 and classification of individuals with Parkinson’s disease,” Computational and Mathematical Methods in Medicine , vol. 2018, Article ID 8019232, 2018. doi: 10.1155/2018/8019232

    work page doi:10.1155/2018/8019232 2018

  37. [37]

    Colombo, R

    G. Colombo, R. J. A. Cubero, L. Kanari, et al., ”A tool for mapping microglial morphology, morphOMICs, reveals brain-region and sex- dependent phenotypes,”Nature Neuroscience, vol. 25, pp. 1379-1393, 2022. doi: 10.1038/s41593-022-01167-6

    work page doi:10.1038/s41593-022-01167-6 2022

  38. [38]

    Fleischer and M

    J. Fleischer and M. T. Islam, ”Identifying and phenotyping COVID-19 patients using machine learning on chest X-rays,” European Respira- tory Journal, vol. 56, suppl. 64, p. 4151, 2020. Available: https://doi.org/10.1183/13993003.congress-2020.4151

    work page doi:10.1183/13993003.congress-2020.4151 2020

  39. [39]

    M. T. Islam and J. W. Fleischer, ”Outlier Detection in Large Radiological Datasets Using UMAP,” inTopology- and Graph-Informed Imaging Informatics, C. Chen, Y . Singh, and X. Hu, Eds., TGI3 2024, Lecture Notes in Computer Science, vol. 15239. Cham: Springer, 2025. Available: https://doi.org/10.1007/978-3-031-73967-5_11

    work page doi:10.1007/978-3-031-73967-5_11 2024

  40. [40]

    Merdjanovska and A

    E. Merdjanovska and A. Rashkovska, ”A framework for comparative study of databases and computational methods for arrhythmia detection from single-lead ECG,”Scientific Reports, vol. 13, no. 11682, 2023, doi: 10.1038/s41598-023-38532-9

    work page doi:10.1038/s41598-023-38532-9 2023

  41. [41]

    I. I. Christov, ”Real-time electrocardiogram QRS detection using combined adaptive threshold,” BioMed Engineering Online, vol. 3, p. 28,

    work page

  42. [42]

    [Online]

    doi: 10.1186/1475-925X-3-28. [Online]. Available: https://doi.org/10.1186/1475-925X-3-28

    work page doi:10.1186/1475-925x-3-28

  43. [43]

    Tompkins

    J. Pan and W. J. Tompkins, ”A Real-Time QRS Detection Algorithm,” IEEE Transactions on Biomedical Engineering, vol. BME-32, no. 3, pp. 230-236, Mar. 1985. doi: 10.1109/TBME.1985.325532

    work page doi:10.1109/tbme.1985.325532 1985

  44. [44]

    Lau, Jan C

    D. Makowski, T. Pham, Z. J. Lau, et al. , ”NeuroKit2: A Python toolbox for neurophysiological signal processing,” Behavior Research Methods, vol. 53, pp. 1689-1696, 2021. doi: 10.3758/s13428-020-01516-y

    work page doi:10.3758/s13428-020-01516-y 2021

  45. [45]

    Neighborhood preservation in nonlinear projection methods: An experimental study,

    J. Venna and S. Kaski, “Neighborhood preservation in nonlinear projection methods: An experimental study,” inInternational Conference on Artificial Neural Networks, pp. 485–491, Springer, 2001. doi: 10.1007/3-540-44668-0 66

    work page doi:10.1007/3-540-44668-0 2001

  46. [46]

    Sex matters: A comprehensive comparison of female and male hearts,

    S. R. St Pierre, M. Peirlinck, and E. Kuhl, “Sex matters: A comprehensive comparison of female and male hearts,” Frontiers in Physiology, vol. 13, p. 831179, Mar. 2022. doi: 10.3389/fphys.2022.831179

    work page doi:10.3389/fphys.2022.831179 2022

  47. [47]

    Sex differences in heart: from basics to clinics,

    C. Prajapati, J. Koivum ¨aki, M. Pekkanen-Mattila, and K. Aalto-Set ¨al¨a, “Sex differences in heart: from basics to clinics,” European Journal of Medical Research, vol. 27, no. 1, p. 241, Nov. 2022. doi: 10.1186/s40001-022-00880-z

    work page doi:10.1186/s40001-022-00880-z 2022

  48. [48]

    ECG analysis: a new approach in human identification,

    L. Biel, O. Pettersson, L. Philipson, and P. Wide, “ECG analysis: a new approach in human identification,” IEEE Transactions on Instrumen- tation and Measurement, vol. 50, no. 3, pp. 808-812, June 2001, doi: 10.1109/19.930458

    work page doi:10.1109/19.930458 2001

  49. [49]

    Right Bundle Branch Block: Current Considerations,

    T. Ikeda, “Right Bundle Branch Block: Current Considerations,” Current Cardiology Reviews, vol. 17, no. 1, pp. 24–30, 2021, doi: 10. 2174/1573403X16666200708111553

    work page 2021

  50. [50]

    Electrocardiographic Changes in Primary Hyperparathyroidism,

    A. Alzate, M. M. Urbano, and K. Buitrago-Toro, “Electrocardiographic Changes in Primary Hyperparathyroidism,” International Journal of Clinical Cardiology, vol. 2, no. 6, Art. no. 068, Dec. 2015, doi: https://doi.org/10.23937/2378-2951/1410068

    work page doi:10.23937/2378-2951/1410068 2015

  51. [51]

    A Case of J Wave Syndrome Due to Severe Hypercalcemia with Ventricular Fibrillation Storm and Successful Treatment of Isoproterenol Infusion,

    M. Shiozaki, M. Sumiyoshi, H. Tabuchi, H. Hayashi, H. Tamura, K. Inoue, and T. Minamino, “A Case of J Wave Syndrome Due to Severe Hypercalcemia with Ventricular Fibrillation Storm and Successful Treatment of Isoproterenol Infusion,”International Heart Journal, vol. 62, no. 4, pp. 924–926, 2021, doi: https://doi.org/10.1536/ihj.20-798

    work page doi:10.1536/ihj.20-798 2021

  52. [52]

    Learning a parametric embedding by preserving local structure,

    L. van der Maaten, “Learning a parametric embedding by preserving local structure,” in Proceedings of the Twelfth International Conference on Artificial Intelligence and Statistics , Proceedings of Machine Learning Research, vol. 5, pp. 384–391, 2009. https://proceedings. mlr.press/v5/maaten09a.html

    work page 2009

  53. [53]

    Parametric UMAP embeddings for representation and semi-supervised learning,

    T. Sainburg, L. McInnes, and T. Q. Gentner, “Parametric UMAP embeddings for representation and semi-supervised learning,”arXiv preprint arXiv:2009.12981, 2021. [Online]. Available: https://arxiv.org/abs/2009.12981

    work page arXiv 2009

  54. [54]

    Manifold-aligned neighbor embedding,

    M. T. Islam and J. W. Fleischer, “Manifold-aligned neighbor embedding,” arXiv preprint arXiv:2205.11257 , 2022. [Online]. Available: https://arxiv.org/abs/2205.11257

    work page arXiv 2022

  55. [55]

    Vectorcardiographic diagnostic and prognostic information derived from the 12-lead electrocardiogram: Historical review and clinical perspective,

    S. Man, A. C. Maan, M. J. Schalij, and C. A. Swenne, “Vectorcardiographic diagnostic and prognostic information derived from the 12-lead electrocardiogram: Historical review and clinical perspective,” Journal of Electrocardiology, vol. 48, no. 4, pp. 463–475, 2015. https: //doi.org/10.1016/j.jelectrocard.2015.05.002

    work page doi:10.1016/j.jelectrocard.2015.05.002 2015

  56. [56]

    Clinical IoT in Practice: A Novel Design and Implementation of a Multi-functional Digital Stethoscope for Remote Health Monitoring,

    B. Baraeinejad, M. Shams, M. S. Hamedani, et al., “Clinical IoT in Practice: A Novel Design and Implementation of a Multi-functional Digital Stethoscope for Remote Health Monitoring,” TechRxiv, Nov. 7, 2023. doi: 10.36227/techrxiv.24459988.v1

    work page doi:10.36227/techrxiv.24459988.v1 2023

  57. [57]

    Descriptor: Heart and Lung Sounds Dataset Recorded from a Clinical Manikin using Digital Stetho- scope (HLS-CMDS),

    Y . Torabi, S. Shirani, and J. P. Reilly, “Descriptor: Heart and Lung Sounds Dataset Recorded from a Clinical Manikin using Digital Stetho- scope (HLS-CMDS),”IEEE Data Descriptions, 2025, doi: 10.1109/IEEEDATA.2025.3566012. Appendix A. Example Appendix Section To further demonstrate that many distinct clusters in Figure 6 correspond to di fferent individu...

    work page doi:10.1109/ieeedata.2025.3566012 2025