pith. sign in

arxiv: 2601.10191 · v1 · pith:2DRVIQ3Rnew · submitted 2026-01-15 · 💻 cs.AI

How does downsampling affect needle electromyography signals? A generalisable workflow for understanding downsampling effects on high-frequency time series

Mathieu Cherpitel , Janne Luijten , Thomas B\"ack , Camiel Verhamme , Martijn Tannemaat , Anna Kononova This is my paper

Pith reviewed 2026-05-21 15:14 UTC · model grok-4.3

classification 💻 cs.AI
keywords downsamplingneedle electromyographyneuromuscular diseasesshape distortion metricsmachine learningtime seriessignal processingworkflow
0
0 comments X

The pith

Shape-aware downsampling preserves peak structure in needle electromyography signals better than standard decimation while keeping classification performance for neuromuscular disease detection.

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

The paper develops a workflow to measure information loss when downsampling high-frequency needle electromyography signals for automated neuromuscular disease analysis. The workflow pairs shape distortion metrics with results from feature-based machine learning models on a three-class classification task and examines changes in the feature space. It finds that downsampling methods built to respect signal shape retain waveform peaks and overall morphology more effectively than basic decimation. This approach allows substantial cuts in data size and computation time, opening the door to near real-time diagnostic support without erasing the signals' clinical value, and the same evaluation steps can apply to other high-rate time series.

Core claim

The paper establishes a generalisable workflow that combines shape-based distortion metrics, classification accuracy from feature-based machine learning models, and feature space analysis to quantify the effects of different downsampling algorithms and factors on high-frequency needle electromyography signals. Applied to a three-class neuromuscular disease task, the workflow shows that shape-aware downsampling algorithms outperform standard decimation by better preserving peak structure and overall signal morphology, thereby maintaining predictive performance while reducing computational load for near real-time analysis.

What carries the argument

A workflow that pairs shape-based distortion metrics with machine learning classification outcomes and feature space analysis to assess how downsampling affects waveform integrity and diagnostic content in high-frequency time series.

If this is right

  • Downsampling configurations selected via the workflow enable near real-time nEMG analysis by cutting data volume while preserving classification performance.
  • Shape-aware algorithms are preferable to standard decimation because they retain peak structure and signal morphology more faithfully.
  • The workflow supplies concrete guidance for choosing rates and methods that reduce computational demands without loss of predictive power.
  • The same evaluation steps can be reused on other high-frequency time series to balance data reduction against model accuracy.

Where Pith is reading between the lines

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

  • The workflow could be adapted to other biomedical signals such as EEG or ECG to achieve similar efficiency gains in real-time monitoring.
  • Direct testing on hardware-constrained devices would show whether the recommended downsampling supports portable diagnostic tools.
  • Extending the workflow to deep learning models might expose different sensitivities to downsampling than those seen with feature-based approaches.

Load-bearing premise

The three-class neuromuscular disease classification task and the chosen feature-based machine learning models capture the diagnostic information present in the original high-frequency signals.

What would settle it

A direct comparison in which signals downsampled with the workflow-recommended settings produce measurably lower diagnostic accuracy in actual clinical review by neurologists than the original high-frequency recordings.

Figures

Figures reproduced from arXiv: 2601.10191 by Anna Kononova, Camiel Verhamme, Janne Luijten, Martijn Tannemaat, Mathieu Cherpitel, Thomas B\"ack.

Figure 1
Figure 1. Figure 1: Five steps of the proposed workflow to investigate the effects of downsampling on time series. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effects of five downsampling methods on a syn [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Accuracy per downsampling factor on the EMGLAB [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Measured per-class performance for each down [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Reported SHAP values for the distance metrics used [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of the LTTB and Decimate downsam [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Feature importance clusters at increasing downsam [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Trade-off between measured feature extraction time [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
read the original abstract

Automated analysis of needle electromyography (nEMG) signals is emerging as a tool to support the detection of neuromuscular diseases (NMDs), yet the signals' high and heterogeneous sampling rates pose substantial computational challenges for feature-based machine-learning models, particularly for near real-time analysis. Downsampling offers a potential solution, but its impact on diagnostic signal content and classification performance remains insufficiently understood. This study presents a workflow for systematically evaluating information loss caused by downsampling in high-frequency time series. The workflow combines shape-based distortion metrics with classification outcomes from available feature-based machine learning models and feature space analysis to quantify how different downsampling algorithms and factors affect both waveform integrity and predictive performance. We use a three-class NMD classification task to experimentally evaluate the workflow. We demonstrate how the workflow identifies downsampling configurations that preserve diagnostic information while substantially reducing computational load. Analysis of shape-based distortion metrics showed that shape-aware downsampling algorithms outperform standard decimation, as they better preserve peak structure and overall signal morphology. The results provide practical guidance for selecting downsampling configurations that enable near real-time nEMG analysis and highlight a generalisable workflow that can be used to balance data reduction with model performance in other high-frequency time-series applications as well.

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

1 major / 2 minor

Summary. The manuscript introduces a workflow for assessing downsampling effects on high-frequency needle electromyography (nEMG) signals. It integrates shape-based distortion metrics, feature-based machine learning classification performance on a three-class neuromuscular disease (NMD) task, and feature space analysis to quantify information loss and identify downsampling configurations that maintain waveform integrity and predictive utility while lowering computational demands. Experiments demonstrate that shape-aware downsampling algorithms better preserve peak structure and signal morphology compared to standard decimation.

Significance. If the results hold, the work offers practical guidance for selecting downsampling parameters in nEMG analysis to support near real-time applications without substantial loss of diagnostic content. The generalisable workflow combining multiple evaluation axes (shape metrics plus classification) is a strength and could apply to other high-frequency biomedical time series. Explicit credit is due for grounding the evaluation in both morphological fidelity and downstream task performance rather than relying on a single metric.

major comments (1)
  1. [Abstract and experimental evaluation section] The central claim that downsampling configurations preserve diagnostic information rests on the three-class NMD classification task and chosen feature-based ML models serving as a sufficient proxy for clinical utility. This assumption is load-bearing because a coarse three-class setup may rely on lower-frequency or non-specific features rather than the high-frequency elements (MUAP morphology, fibrillation potentials, recruitment patterns) that neurologists use. A concrete test against expert-labeled finer-grained data or direct comparison to clinical diagnostic criteria would be required to substantiate that preserved classification accuracy equates to preserved diagnostic content. (Abstract and the experimental evaluation section describing the three-class task.)
minor comments (2)
  1. [Methods] Clarify the exact sampling rates of the original nEMG signals and the specific downsampling factors tested, as these details are essential for reproducibility and for readers to map the findings to their own hardware constraints.
  2. [Shape-based distortion metrics subsection] The description of the shape-based distortion metrics would benefit from explicit formulas or pseudocode in the main text rather than relying solely on references, to make the workflow more self-contained.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for their constructive feedback. We address the major comment below, indicating planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract and experimental evaluation section] The central claim that downsampling configurations preserve diagnostic information rests on the three-class NMD classification task and chosen feature-based ML models serving as a sufficient proxy for clinical utility. This assumption is load-bearing because a coarse three-class setup may rely on lower-frequency or non-specific features rather than the high-frequency elements (MUAP morphology, fibrillation potentials, recruitment patterns) that neurologists use. A concrete test against expert-labeled finer-grained data or direct comparison to clinical diagnostic criteria would be required to substantiate that preserved classification accuracy equates to preserved diagnostic content.

    Authors: We acknowledge that the three-class classification task functions as a proxy and does not fully replicate the detailed clinical diagnostic process, which relies on expert interpretation of specific high-frequency elements such as MUAP morphology, fibrillation potentials, and recruitment patterns. The feature set in our models draws from established nEMG analysis practices that target these characteristics, and the shape distortion metrics offer an orthogonal, direct measure of morphological fidelity. Nevertheless, the available dataset is restricted to three-class labels, precluding a direct test on finer-grained expert annotations or explicit mapping to clinical criteria. We will revise the abstract and experimental evaluation section to explicitly state this scope limitation and frame the results as evidence of preserved utility for the evaluated task rather than equivalence to full clinical diagnosis. revision: partial

standing simulated objections not resolved
  • We do not have access to finer-grained expert-labeled nEMG datasets that would enable the suggested direct comparison to clinical diagnostic criteria.

Circularity Check

0 steps flagged

No circularity: empirical workflow uses independent external metrics

full rationale

The paper describes an empirical workflow that evaluates downsampling effects by combining shape-based distortion metrics with classification performance from feature-based ML models on a three-class NMD task. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains are present that would reduce any central claim to its own inputs by construction. The analysis relies on external benchmarks (shape metrics and classification accuracy) that are independent of the downsampling configurations being tested, rendering the evaluation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract does not introduce or rely on any explicit free parameters, axioms, or invented entities beyond standard assumptions of signal processing and supervised classification.

pith-pipeline@v0.9.0 · 5780 in / 1083 out tokens · 39994 ms · 2026-05-21T15:14:16.486822+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 1 internal anchor

  1. [1]

    Preston and Barbara E

    David C. Preston and Barbara E. Shapiro.Elec- tromyography and Neuromuscular Disorders. 4 edi- tion, April 2020

    work page 2020

  2. [2]

    A glance into the future of myositis therapy.Therapeutic Advances in Muscu- loskeletal Disease, 14:1759720X221100299, January

    Ilaria Chiapparoli, Claudio Galluzzo, Carlo Sal- varani, and Nicol` o Pipitone. A glance into the future of myositis therapy.Therapeutic Advances in Muscu- loskeletal Disease, 14:1759720X221100299, January

    work page

  3. [3]

    Publisher: SAGE Publications

    work page

  4. [4]

    Helmar Christoph Lehmann, David Burke, and Satoshi Kuwabara. Chronic inflammatory demyeli- nating polyneuropathy: update on diagnosis, im- munopathogenesis and treatment.Journal of Neu- rology, Neurosurgery & Psychiatry, 90(9):981–987, September 2019. Publisher: BMJ Publishing Group Ltd Section: Neuromuscular

    work page 2019

  5. [5]

    Miller, Merit E

    Timothy M. Miller, Merit E. Cudkowicz, Angela Genge, Pamela J. Shaw, Gen Sobue, Robert C. Bucelli, Adriano Chi` o, Philip Van Damme, Al- bert C. Ludolph, Jonathan D. Glass, Jinsy A. Andrews, Suma Babu, Michael Benatar, Christo- pher J. McDermott, Thos Cochrane, Sowmya Chary, Sheena Chew, Han Zhu, Fan Wu, Ivan Nestorov, Danielle Graham, Peng Sun, Manjit Mc...

    work page doi:10.1056/nejmoa2204705 2022

  6. [6]

    Amato, and Machiel Zwarts.Electrodiagnostic Medicine

    Daniel Dumitru, Anthony A. Amato, and Machiel Zwarts.Electrodiagnostic Medicine. 2 edition, September 2001

    work page 2001

  7. [7]

    Pushpa Narayanaswami, Thomas Geisbush, Lyell Jones, Michael Weiss, Tahseen Mozaffar, Gary Gronseth, and Seward B. Rutkove. Critically re-evaluating a common technique.Neurology, 86(3):218–223, January 2016. Publisher: Wolters Kluwer

    work page 2016

  8. [8]

    Potters, and Camiel Ver- hamme

    Sterre de Jonge, Wouter V. Potters, and Camiel Ver- hamme. Artificial intelligence for automatic classi- fication of needle EMG signals: A scoping review. Clinical Neurophysiology, 159:41–55, March 2024

    work page 2024

  9. [9]

    Tannemaat, Mario Kefalas, Victor J

    Martijn .R. Tannemaat, Mario Kefalas, Victor J. Geraedts, Linda Remijn-Nelissen, A.J.M. Verschu- uren, Milan Koch, Anna V. Kononova, Hao Wang, and Thomas H.W. B¨ ack. Distinguishing normal, neuropathic and myopathic EMG with an automated machine learning approach.Clinical Neurophysiol- ogy, 146:49–54, February 2023

    work page 2023

  10. [10]

    A Review on Machine Learning for EEG Signal Processing in Bioengineering.IEEE Re- views in Biomedical Engineering, 14:204–218, 2021

    Mohammad-Parsa Hosseini, Amin Hosseini, and Kiarash Ahi. A Review on Machine Learning for EEG Signal Processing in Bioengineering.IEEE Re- views in Biomedical Engineering, 14:204–218, 2021

    work page 2021

  11. [11]

    Geometric feature performance under downsampling for EEG classifi- cation tasks, February 2021

    Bryan Bischof and Eric Bunch. Geometric feature performance under downsampling for EEG classifi- cation tasks, February 2021. arXiv:2102.07669 [cs]

    work page arXiv 2021

  12. [12]

    Exploring Best Practices for ECG Pre-Processing in Machine Learn- ing, May 2025

    Amir Salimi, Sunil Vasu Kalmady, Abram Hindle, Osmar Zaiane, and Padma Kaul. Exploring Best Practices for ECG Pre-Processing in Machine Learn- ing, May 2025. arXiv:2311.04229 [eess]. 14

    work page arXiv 2025

  13. [13]

    Bjørn-Jostein Singstad and Eraraya Morenzo Muten. Assessing the Impact of Downsampled ECGs and Al- ternative Loss Functions in Multi-Label Classifica- tion of 12-Lead ECGs.Cardiovascular Engineering and Technology, 16(5):599–610, October 2025

    work page 2025

  14. [14]

    Effect of decimation on the classification rate of non-linear analysis methods ap- plied to uterine EMG signals.IRBM, 34(4):326–329, November 2013

    Ahmad Diab, Mahmoud Hassan, Brynjar Karlsson, and Catherine Marque. Effect of decimation on the classification rate of non-linear analysis methods ap- plied to uterine EMG signals.IRBM, 34(4):326–329, November 2013

    work page 2013

  15. [15]

    The Refined Composite Down- sampling Permutation Entropy Is a Relevant Tool in the Muscle Fatigue Study Using sEMG Signals.En- tropy, 23(12):1655, December 2021

    Philippe Ravier, Antonio D´ avalos, Meryem Jabloun, and Olivier Buttelli. The Refined Composite Down- sampling Permutation Entropy Is a Relevant Tool in the Muscle Fatigue Study Using sEMG Signals.En- tropy, 23(12):1655, December 2021. Publisher: Mul- tidisciplinary Digital Publishing Institute

    work page 2021

  16. [16]

    Enhancing Classification Results of Slope Entropy Using Downsampling Schemes.Ax- ioms, 14(11):797, November 2025

    Vicent Molt´ o-Gallego, David Cuesta-Frau, and Mahdy Kouka. Enhancing Classification Results of Slope Entropy Using Downsampling Schemes.Ax- ioms, 14(11):797, November 2025. Publisher: Multi- disciplinary Digital Publishing Institute

    work page 2025

  17. [17]

    Ma- chine Learning for Predicting the Damaged Parts of a Low Speed Vehicle Crash

    Milan Koch, Hao Wang, and Thomas B¨ ack. Ma- chine Learning for Predicting the Damaged Parts of a Low Speed Vehicle Crash. In2018 Thirteenth In- ternational Conference on Digital Information Man- agement (ICDIM), pages 179–184, September 2018

    work page 2018

  18. [18]

    Machine Learning for Predicting the Impact Point of a Low Speed Vehicle Crash

    Milan Koch and Thomas B¨ ack. Machine Learning for Predicting the Impact Point of a Low Speed Vehicle Crash. In2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), pages 1432–1437, December 2018

    work page 2018

  19. [19]

    Automated Machine Learning for EEG-Based Classification of Parkin- son’s Disease Patients

    Milan Koch, Victor Geraedts, Hao Wang, Martijn Tannemaat, and Thomas B¨ ack. Automated Machine Learning for EEG-Based Classification of Parkin- son’s Disease Patients. In2019 IEEE International Conference on Big Data (Big Data), pages 4845– 4852, December 2019

    work page 2019

  20. [20]

    Au- tomated Machine Learning for the Classification of Normal and Abnormal Electromyography Data

    Marios Kefalas, Milan Koch, Victor Geraedts, Hao Wang, Martijn Tannemaat, and Thomas B¨ ack. Au- tomated Machine Learning for the Classification of Normal and Abnormal Electromyography Data. In 2020 IEEE International Conference on Big Data (Big Data), pages 1176–1185, December 2020

    work page 2020

  21. [21]

    M4: a visualization-oriented time series data aggregation.Proceedings of the VLDB Endowment, 7(10):797–808, June 2014

    Uwe Jugel, Zbigniew Jerzak, Gregor Hackenbroich, and Volker Markl. M4: a visualization-oriented time series data aggregation.Proceedings of the VLDB Endowment, 7(10):797–808, June 2014

    work page 2014

  22. [22]

    Thesis, June 2013

    Sveinn Steinarsson 1979.Downsampling Time Se- ries for Visual Representation. Thesis, June 2013. Accepted: 2013-05-31T12:15:35Z

    work page 1979

  23. [23]

    MinMaxLTTB: Leveraging MinMax-Preselection to Scale LTTB, April 2023

    Jeroen Van Der Donckt, Jonas Van Der Don- ckt, Michael Rademaker, and Sofie Van Hoecke. MinMaxLTTB: Leveraging MinMax-Preselection to Scale LTTB, April 2023. arXiv:2305.00332 [cs]

    work page arXiv 2023

  24. [24]

    Ming Li, Xin Chen, Xin Li, Bin Ma, and P.M.B. Vitanyi. The similarity metric.IEEE Transactions on Information Theory, 50(12):3250–3264, Decem- ber 2004

    work page 2004

  25. [25]

    Kempa-Liehr

    Maximilian Christ, Nils Braun, Julius Neuffer, and Andreas W. Kempa-Liehr. Time Series FeatuRe Ex- traction on basis of Scalable Hypothesis tests (tsfresh – A Python package).Neurocomputing, 307:72–77, September 2018

    work page 2018

  26. [26]

    Kursa, Aleksander Jankowski, and Witold R

    Miron B. Kursa, Aleksander Jankowski, and Witold R. Rudnicki. Boruta – A System for Feature Selection.Fundamenta Informaticae, 101(4):271– 285, July 2010

    work page 2010

  27. [27]

    Archive Location: world

    XGBoost|Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Archive Location: world

    work page

  28. [28]

    McGill, Zoia C

    Kevin C. McGill, Zoia C. Lateva, and Hamid R. Marateb. EMGLAB: An interactive EMG decom- position program.Journal of Neuroscience Methods, 149(2):121–133, December 2005

    work page 2005

  29. [29]

    tsdownsample: High-performance time series downsampling for scalable visualization

    Jeroen Van Der Donckt, Jonas Van Der Donckt, and Sofie Van Hoecke. tsdownsample: High-performance time series downsampling for scalable visualization. SoftwareX, 29:102045, February 2025

    work page 2025

  30. [30]

    Kernel Principal Component Analysis

    Bernhard Scholkopf, Alexander Smola, and Robert Muller. Kernel Principal Component Analysis

    work page

  31. [31]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    Leland McInnes, John Healy, and James Melville. UMAP: Uniform Manifold Approximation and Pro- jection for Dimension Reduction, September 2020. arXiv:1802.03426 [stat]

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  32. [32]

    Tenenbaum, Vin De Silva, and John C

    Joshua B. Tenenbaum, Vin De Silva, and John C. Langford. A Global Geometric Framework for Nonlinear Dimensionality Reduction.Science, 290(5500):2319–2323, December 2000

    work page 2000

  33. [33]

    J. B. Kruskal. Nonmetric multidimensional scaling: A numerical method.Psychometrika, 29(2):115–129, June 1964

    work page 1964

  34. [34]

    Self-Tuning Spectral Clustering

    Lihi Zelnik-manor and Pietro Perona. Self-Tuning Spectral Clustering. InAdvances in Neural Infor- mation Processing Systems, volume 17. MIT Press, 2004

    work page 2004

  35. [35]

    Viualizing data using t-SNE.Journal of Machine Learning Research, 9:2579–2605, Novem- ber 2008

    Laurens van der Maaten, Geoffrey Hinton, and Yoe- soep Rachmad. Viualizing data using t-SNE.Journal of Machine Learning Research, 9:2579–2605, Novem- ber 2008. 15

    work page 2008

  36. [36]

    Information-driven bars for finan- cial machine learning: imbalance bars, May 2019

    Gerard Mart´ ınez. Information-driven bars for finan- cial machine learning: imbalance bars, May 2019

    work page 2019

  37. [37]

    Halford, David A

    Jonathan J. Halford, David A. Clunie, Benjamin H. Brinkmann, Dagmar Krefting, Jan R´ emi, Felix Rosenow, Aatif Husain, Franz F¨ urbass, J. An- drew Ehrenberg, and Silvia Winkler. Standard- ization of neurophysiology signal data into the DICOM®standard.Clinical Neurophysiology, 132(4):993–997, April 2021

    work page 2021

  38. [38]

    Dean, Ekrem Kutluay, Zeke Campbell, Sarah Schmitt, Nicola Donato, Jonathan J

    Filippo Battaglia, Mattia Galanti, Giovanni Gugliandolo, Stefan Rampp, Jan Remi, Alexandra Parashos, Sonali Sharma, Sonal Bhatia, Brian C. Dean, Ekrem Kutluay, Zeke Campbell, Sarah Schmitt, Nicola Donato, Jonathan J. Halford, and Giuseppe Campobello. Neurophysiology Signal Codecs for the DICOM®Standard: Preliminary Results. In2024 IEEE International Sympo...

    work page 2024