Impulsive pattern recognition of a myoelectric hand via Dynamic Time Warping

Ersin Topta\c{s}; Gazi Akg\"un; Mustafa Can Kadilar

REVIEW 4 major objections 4 minor 18 references

Impulsive pattern recognition of a myoelectric hand via Dynamic Time Warping

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read One EMG sensor plus Dynamic Time Warping can recognize four myoelectric hand commands at 92 percent accuracy.

desk verdict A plausible single-sensor DTW control scheme, but the 92% accuracy claim is not backed by a clean validation protocol; the idea deserves a careful referee, but only with the expectation of major revision. read the letter →

arxiv 2504.15256 v1 pith:M3WYSDJQ submitted 2025-04-21 physics.bio-ph

Mustafa Can Kadilar , Ersin Toptaş , Gazi Akgün This is my paper · ORCID

classification physics.bio-ph

keywords myoelectricprosthetichandsingle-sensorEMGDynamicTimeWarpingpatternrecognitionimpulsesignalsbinarycommandencodingconfusionmatrixcontrol

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

This paper aims to show that a myoelectric prosthetic hand can be controlled with a single EMG sensor instead of the many sensors used in conventional systems. The key move is to replace continuous multi-channel data with brief, deliberate muscle impulses, then classify those impulses with Dynamic Time Warping. In a four-command test, the system got 46 of 50 trials right, for 92 percent accuracy, which the authors say is comparable to multi-sensor systems. If the claim holds, it would mean simpler, cheaper, and easier-to-deploy prosthetic hands, especially valuable in disaster settings where amputations cluster and rapid fitting matters.

What carries the argument

The load-bearing object is Dynamic Time Warping, an algorithm that measures similarity between two time series by allowing the time axis to be stretched and compressed so that misaligned events still match. Here it compares short EMG impulse signals against pre-recorded template patterns; when the DTW similarity value falls below 1.12, the signal is judged close enough to the searched pattern, and a binary classification event is triggered. Those binary events, separated by a small delay, encode the four command patterns.

What would settle it

Fix the DTW templates and the threshold at 1.12, run the same four-pattern protocol on a new set of 50 trials from the same user on a later day, and compute the confusion matrix; if the fresh-trial accuracy falls substantially below 92 percent, the reported accuracy was an in-sample result rather than a generalizable property of the single-sensor method.

Watch

Extended reading notes

Core claim

The paper claims that a single surface EMG sensor placed on the palmaris longus muscle, paired with Dynamic Time Warping as the classifier, is sufficient for reliable myoelectric control. The user makes short muscle contractions that the system sorts into two basic signals, pattern 0 and pattern 1, and sequences of these bits form four commands: "00", "01", "10", and "11". A confusion matrix over fifty trials shows 46 correct classifications, giving 92 percent accuracy, which the authors state is similar to conventional multi-sensor systems. The paper also reports a response delay of 414 to 644 milliseconds between the user's signal and the motor moving, which it attributes to DTW waiting until the live signal's similarity value drops below a threshold of 1.12 relative to the recorded template.

Load-bearing premise

The central claim assumes that the similarity threshold and the recorded template patterns, which were set during the same session that produced the 92 percent accuracy, will classify new muscle contractions from fresh sessions and users just as reliably.

Editorial extensions

If this is right

A prosthetic hand can be controlled with one EMG sensor, cutting hardware cost, weight, and setup complexity relative to multi-sensor systems.
The four binary commands demonstrated can be extended into longer bit sequences, potentially generating many more control patterns from the same single sensor.
DTW's tolerance for time-axis misalignment could make the system more robust to natural variability in how a user times their muscle contractions.
The simple hardware and short training process are well matched to rapid deployment in disaster-affected regions where advanced multi-sensor prosthetics are impractical.

Reading between the lines

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

The paper tests one user and one session, so the 92 percent figure is best read as a feasibility proof; the claim that it matches multi-sensor systems will only be fully established when tested across multiple users, sessions, and electrode re-placements.
The binary-impulse vocabulary suggests a Morse-code-like control scheme; longer command sequences could be classified without changing the sensor hardware, at the cost of slower command throughput.
Because the decision rests on a fixed similarity threshold, an adaptive calibration that re-estimates the threshold per session could trade acceptance delay against accuracy, a testable extension the paper does not explore.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The idea is genuinely nice: one EMG sensor, two impulse patterns, DTW to glue them into four binary commands. That is a clever way to cut hardware complexity, and the motivation for low-cost, fast-deployable prosthetic control in disaster settings is reasonable. The prototype itself seems to work—the hand moves, the delays are measured, and the DTW approach is well matched to time-varying EMG signals. Credit where it is due: this is a sensible engineering application of a known method, not a new theory, but it does not need to be new theory to be useful.

The problem is the evidence. The paper reports 46/50 correct trials (92%), but gives no subject count, no session count, no error bars, and no cross-validation. The methods describe recording filtered signals to an SD card and then searching live EMG with recorded templates, but never state that a separate calibration set was used to fix the DTW similarity threshold of 1.12. Given that same threshold is reported in the results section immediately after the confusion matrix, it is hard to avoid the conclusion that the 92% is an in-sample fit. With four balanced classes, chance is 25%, so 92% on 50 trials is promising, but without confidence intervals or a held-out test you cannot even bound it. The comparison to Castro et al. (10 sensors, different gesture set) is a cross-study comparison and cannot support the abstract's claim of being "similar to conventional multi-sensor systems." That claim needs a matched baseline or a much more careful discussion.

The writing is rough in places—typos, incomplete sentences, a reference list with a placeholder DOI—but that is minor relative to the validation gap. The discussion does acknowledge some limitations (long-term use, muscle fatigue), which is honest, but it does not address the threshold selection or the lack of a proper test protocol.

Who gets value from this paper? A researcher working on low-cost myoelectric prosthetics might find the single-sensor DTW encoding worth replicating. But as it stands, it reads as a preliminary proof of concept, not a rigorous comparative study. It deserves a serious referee because the idea has merit and the community would benefit from a properly executed version. My recommendation: send it to peer review, but the referee should insist on multi-subject data, separate calibration and test sets, and an honest discussion of what the 92% actually represents.

Referee Report

4 major / 4 minor

Summary. This paper presents a myoelectric prosthetic hand control strategy based on a single EMG sensor and short impulse-like muscle activation patterns, classified with Dynamic Time Warping (DTW). Four binary command patterns ("00", "01", "10", "11") are formed from two detected muscle-pulse types, and the classifier triggers the hand motor when the DTW similarity to a stored template falls below a threshold. A confusion matrix over 50 trials reports 46 correct classifications (92% accuracy), and measured motor-activation delays range from 414 to 644 ms. The authors argue that this single-sensor, impulse-based approach achieves accuracy comparable to conventional multi-sensor myoelectric systems while reducing hardware complexity and cost, and they position it as particularly relevant for prosthetic deployment in disaster-affected regions.

Significance. If the reported accuracy is confirmed in a properly validated protocol, the single-sensor DTW approach would be a meaningful simplification over multi-sensor myoelectric pattern recognition, with practical implications for low-cost prosthetics. The paper includes a working prototype and an explicit control latency measurement, which are useful engineering contributions. However, the central quantitative claim (92% accuracy comparable to multi-sensor systems) rests on a small, statistically underdescribed dataset and an in-sample-selected threshold, so the current evidence is preliminary. The concept is promising, but the strength of the claim as written exceeds what the experimental protocol supports.

major comments (4)

[Section III-B, Table II] The reported 92% accuracy (46/50) is presented without any information about the number of subjects, number of sessions, or trials per class. With four balanced classes, chance accuracy is 25%, and no confidence interval or variance measure is given, so the result cannot be distinguished from a chance-level performance with statistical significance. The authors should report subject/session counts, per-class trial counts, and a proper error analysis (e.g., Wilson interval or repeated-session variability).
[Section III-B, Table III] The threshold value 1.12 is described as the similarity at which "the signal is determined to be similar enough to the searched pattern," but the paper never states whether this threshold was fixed independently of the 50 trials in Table II or whether the same trials that produced the confusion matrix were used to select the threshold. If the threshold and the recorded templates were chosen from the same data, the 92% figure is an in-sample fit and is not a valid estimate of generalizable accuracy. A train/test split or cross-validation protocol is required to support the headline claim.
[Section III-B, Table II and text] The comparison with Castro et al. [17] is not a matched comparison: it uses a different number of sensors (10 vs. 1), a different gesture vocabulary (6 or 10 vs. 4 binary patterns), and presumably different subjects and recording protocols. The statement that the single-sensor approach is "similar to that of conventional multi-sensor systems" is therefore not supported by the data presented. The authors should either provide a same-protocol multi-sensor baseline or explicitly soften the claim to a feasibility demonstration with an appropriate qualifying statement.
[Section II-A and II-B] Several experimental details necessary for reproducibility are missing: the number of template recordings per pattern, how templates were selected or averaged, the EMG sampling rate and filter characteristics, the duration of each impulse signal, and the exact procedure for setting the threshold 1.12. Without these details, other groups cannot replicate the experiment, and the influence of these choices on the reported accuracy cannot be assessed. This is a load-bearing omission because the DTW-based classifier's performance depends critically on the template and threshold definitions.

minor comments (4)

[Throughout] The manuscript contains numerous typographical and grammatical errors (e.g., "algrrorithm," "sequances," "Disscussion," "degress of freedom," "instantces"), which should be corrected in a thorough editorial pass.
[Section II-B, Equation (1)] The DTW recurrence in Equation (1) is not fully explained: the variables $D_{i,j}$, $x_i$, and $y_j$ are not defined clearly, and the example in Table I does not show how the initial conditions (e.g., $D_{0,1} = \infty$) are set. The caption and surrounding text should be clarified.
[Section II-A, Figures 3 and 4] The text says "The three EMG probes are connected to the forearm muscle palmaris longus ... measures the biological signals..." with an incomplete sentence. The sentence should be rewritten, and Figure 5 is cited as "unfiltered data" but is said to be retrieved from [13], which is inconsistent with the claim that the figure shows data from this study.
[References] Reference [13] is given in Turkish and appears to be a conference abstract rather than a full citation; the venue and page numbers are missing. Reference [12] is a textbook and should be cited with a specific chapter or page if used to justify electrode placement.

Circularity Check

1 steps flagged · score 6.0 of 10

92% accuracy is an in-sample fit: the DTW threshold (1.12) is set on the same runs that produce Table II.

fitted input called prediction [Section III (Results), B. Accuracy and Statistics, Table II/Table III discussion]
"The default state of the muscles provides a 1.85~ similarity signal to the searched pattern, when the similarity starts changing, it indicates the contraction is beginning. When the similarity value reaches below a certain number, in this case the value 1.12. The signal is determined to be similar enough to the searched pattern and motor starts to move the fingers with the intended outcome."

The 92% accuracy in Table II ('After 46 successes and 4 failures, the algorithm provides a 92% accuracy') is obtained with a decision threshold (1.12) and template patterns that are described only by this same observed similarity behavior. The manuscript never describes a calibration session or a held-out set that fixes the threshold before scoring; the same signal runs that define the threshold and templates appear to be the runs counted in the confusion matrix. Consequently the headline accuracy is an in-sample fit of the decision boundary rather than an independent prediction of new muscle activations.

full rationale

The core DTW similarity computation is an external, standard method (Berndt and Clifford, [14]) and is not circular. The self-citations ([1], [13]) are not load-bearing: [13] is a feature-extraction/classification study and [1] is a prior presentation of this work. However, the central quantitative claim — 92% single-sensor accuracy, asserted to match conventional multi-sensor systems — depends on the confusion matrix in Table II and the threshold 1.12 in Table III. The paper never states that the threshold or templates were fixed on a separate calibration set or that the 50 scored trials were held out. The threshold is presented as the observed similarity value at which the signal 'is determined to be similar enough,' and the accuracy is then counted from the same experimental runs. This makes the reported accuracy an in-sample result rather than an independent prediction, so the generalizability claim in the abstract is not yet established. No additional circularity was found in the mechanical or algorithmic derivation.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The system rests on standard DTW math, an unvalidated assumption about single-muscle EMG informativeness, and a threshold that is effectively fitted. No new physical entities are introduced.

free parameters (1)

DTW similarity threshold = 1.12
Used in Section III.B / Table III to decide when a live EMG signal matches the searched pattern; no derivation or independent validation is given.

assumptions (3)

standard math DTW dynamic programming recurrence D(i,j)=d(x_i,y_j)+min(D(i-1,j),D(i,j-1),D(i-1,j-1)) yields a valid similarity measure.
Invoked in Equation (1), Section II.B; standard result from Berndt and Clifford [14], not proved in this paper.
domain assumption Surface EMG recorded from palmaris longus contains enough information to distinguish the four binary impulse patterns.
Assumed in Sections II.A and III; no cross-user or cross-session validation is provided.
ad hoc to paper A fixed threshold of 1.12 remains valid for the same user over time.
Introduced in Table III as the operating point; the paper acknowledges long-term stability is unresolved (Section IV).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Impulsive pattern recognition of a myoelectric hand via Dynamic Time Warping." pith.science (2026). https://pith.science/paper/M3WYSDJQ

@misc{pith2026250415256,
  author       = {Pith},
  title        = {Pith review of: Impulsive pattern recognition of a myoelectric hand via Dynamic Time Warping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M3WYSDJQ}},
  note         = {Machine review of arXiv:2504.15256}
}

read the original abstract

Although myoelectric prosthetic hands provide amputees with intuitive control, their reliance on many EMG sensors limits accessibility and makes them complex and expensive. To address this problem, this work presents a different perspective that makes use of a single EMG sensor and brief impulse signals in conjunction with Dynamic Time Warping (DTW) for accurate pattern detection. Conventional techniques rely on real-time data from multiple sensors, which can be costly and bulky. The method presents high accuracy while lowering hardware complexity and expense. A DTW-based system that reliably identifies muscle activation patterns from short EMG signals was created and tested. Results show that this single-sensor approach obtained an accuracy rate of 92%, which is similar to that of conventional multi-sensor systems. This research provides a more straightforward and economical approach that can be used to obtain enhanced myoelectric control. These findings provide a different perspective on more easily accessible and user-friendly prosthetic devices, which will be especially helpful in disaster-affected areas where quick deployment is essential. Future improvements would investigate this system's dependability over time and wider implementations in real situations, to take prosthetic technology one step further.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 15 canonical work pages

[17]

Widodo, A., & Yang, B.-S. (2007). Support vector machine in machine condition monitoring and fault diagnosis. Mechanical Systems and Signal Processing, 21(6), 2560–2574. https://doi.org/10.1016/j.ymssp.2006.12.007

doi:10.1016/j.ymssp.2006.12.007 2007
[1]

C., Toptaş, E., & Akgün, G

Kadılar, M. C., Toptaş, E., & Akgün, G. (2024). An EMG-based prosthetic hand design and control through dynamic time warping. International Journal of Advanced Natural Sciences and Engineering Researches, 8(2), 339– 349.https://doi.org/10.1234/ijnser.2024.339

work page doi:10.1234/ijnser.2024.339 2024
[2]

International Orthopaedics, 37(6)

Hernigou, P.(2013).Ambroise ParéIV: The early history of artificial limbs (from robotics to prostheses). International Orthopaedics, 37(6). https://doi.org/10.1007/s00264-013-1884-7

work page doi:10.1007/s00264-013-1884-7 2013
[3]

Abdhul,A.A.,Subramani,D.,Ganesan,J.,Subramaniam, S., & Dharani, K. G. (2020). Design and development of EMG-based prosthetic arm. In 2020 6th International ConferenceonAdvancedComputing and Communication Systems (ICACCS) (pp. 1–5). IEEE. https://doi.org/10.1109/ICACCS48705.2020.9074206

arXiv 2020
[4]

Uluöz, M., & Gökmen, M. Y. (2023). Management of pediatric musculoskeletal injuries in the first month after the2023Turkeyearthquake.*Children*.Advanceonline. https://doi.org/10.3390/children10111733

work page doi:10.3390/children10111733 2023
[5]

Ergani,H.M.,Özmut,Ö.,Yıldırım, F.,Çit, R.,Yaşar,B., & Ünlü, R. E. (2023). Single‑center plastic surgery response to the 2023 Kahramanmaraş earthquake. *Journal of Disaster and Reconstructive Surgery*. Advanceonline.https://doi.org/10.52312/jdrs.2023.1107

work page arXiv 2023
[6]

J., Ephraim, P

Ziegler‑Graham, K., MacKenzie, E. J., Ephraim, P. L., Travison, T. G., & Brookmeyer, R. (2008). Projected prevalenceoflimb lossintheUnitedStatesbetween 2005 and 2050. *Archives of Physical Medicine and Rehabilitation*, 89(3), 422–429. https://doi.org/10.1016/j.apmr.2007.11.005

work page doi:10.1016/j.apmr.2007.11.005 2008
[7]

W., Kristoffersen, M

Franzke, A. W., Kristoffersen, M. B., Bongers, R. M., Murgia, A., Pobatschnig, B., Unglaube, F., & van der Sluis, C. K. (2019). Users’ vs. therapists’ views on multi‑function myoelectric prostheses: traditional control compared with pattern‑recognition. *PLoS ONE*, 14(8), Article e0220899. https://doi.org/10.1371/journal.pone.0220899

work page doi:10.1371/journal.pone.0220899 2019

Show all 18 references

[8]

(2016, October)

Atasoy, A., Kaya, E., Toptas, E., Kuchimov, S., Kaplanoglu, E., & Ozkan, M. (2016, October). 24 DOF EMG controlled hybrid actuated prosthetic hand. In 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 2452–2455). IEEE...

2016
[9]

Design of multi-grip patterns prosthetic hand with single actuator.https://doi.org/10.1109/TNSRE.2018.2829152

Wattanasiri,P.,Tangpornprasert,P.,&Virulsri,C.(2018). Design of multi-grip patterns prosthetic hand with single actuator.https://doi.org/10.1109/TNSRE.2018.2829152

2018
[10]

F., Tanveer, M

Ahmed, S. F., Tanveer, M. H., Kiwarkis, I. J., & Basy, H. B. (2020). Design and controlling of low-cost prosthetic hand using force sensor. IEEE. https://doi.org/10.1109/ICICT50521.2020.00061

2020
[11]

D., Ariyanto, M., Suriyanto, A., & Caesarendra, W

Ismail, R., Wijaya, G. D., Ariyanto, M., Suriyanto, A., & Caesarendra, W. (2018, December). Development of myoelectric prosthetic hand based on Arduino IDE and Visual C# for trans-radial amputee in Indonesia. IEEE. https://doi.org/10.1109/INCAE.2018.8579393

2018
[12]

S.(2011).Clinical anatomy by regions (9th ed.)

Snell,R. S.(2011).Clinical anatomy by regions (9th ed.). LippincottWilliams&Wilkins. IEEETransactionsonMedicalRoboticsandBionicsVOL.7,2025

2011
[13]

Pattern 0

addressed how myoelectric prosthetic hand control is constrained even with machine learning algorithms. On the other hand, this work used a user interface (UI) based calibration technique with a more stable algorithm as another control strategy. Compared to DTW, the other patt...

2025
[14]

Akgün, G., Demetgül, M., & Kaplanoglu, E. (2013). EMG sinyallerinin öznitelik çıkarımı ve geri yayılımlı yapay sinir ağı algoritması ile sınıflandırılması [Feature extraction of EMG signals and classification with a back-propagation neural network algorithm]. Journal/ConferenceName

2013
[15]

J., & Clifford, J

Berndt, D. J., & Clifford, J. (1994). Using dynamic time warpingtofindpatternsintimeseries.InKDDWorkshop (pp.359–370)

1994
[16]

Hoshigawa, S., Jiang, Y., Kato, R., Morishita, S., Nakamura, T., Yabuki, Y., & Yokoi, H. (2015). Structure design for a two-DoF myoelectric prosthetic hand to realize basic hand functions in ADLs. IEEE. https://doi.org/10.1109/EMBC.2015.7319463

2015
[18]

Castro, M. C. F., Arjunan, S. P., & Kumar, D. K. (2015). Selection of suitable hand gestures for reliable myoelectric human–computer interface. BioMedical Engineering OnLine, 14(1), 30. https://doi.org/10.1186/s12938-015-0025-5

2015 doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [17]

Widodo, A., & Yang, B.-S. (2007). Support vector machine in machine condition monitoring and fault diagnosis. Mechanical Systems and Signal Processing, 21(6), 2560–2574. https://doi.org/10.1016/j.ymssp.2006.12.007

doi:10.1016/j.ymssp.2006.12.007 2007

[2] [1]

C., Toptaş, E., & Akgün, G

Kadılar, M. C., Toptaş, E., & Akgün, G. (2024). An EMG-based prosthetic hand design and control through dynamic time warping. International Journal of Advanced Natural Sciences and Engineering Researches, 8(2), 339– 349.https://doi.org/10.1234/ijnser.2024.339

work page doi:10.1234/ijnser.2024.339 2024

[3] [2]

International Orthopaedics, 37(6)

Hernigou, P.(2013).Ambroise ParéIV: The early history of artificial limbs (from robotics to prostheses). International Orthopaedics, 37(6). https://doi.org/10.1007/s00264-013-1884-7

work page doi:10.1007/s00264-013-1884-7 2013

[4] [3]

Abdhul,A.A.,Subramani,D.,Ganesan,J.,Subramaniam, S., & Dharani, K. G. (2020). Design and development of EMG-based prosthetic arm. In 2020 6th International ConferenceonAdvancedComputing and Communication Systems (ICACCS) (pp. 1–5). IEEE. https://doi.org/10.1109/ICACCS48705.2020.9074206

arXiv 2020

[5] [4]

Uluöz, M., & Gökmen, M. Y. (2023). Management of pediatric musculoskeletal injuries in the first month after the2023Turkeyearthquake.*Children*.Advanceonline. https://doi.org/10.3390/children10111733

work page doi:10.3390/children10111733 2023

[6] [5]

Ergani,H.M.,Özmut,Ö.,Yıldırım, F.,Çit, R.,Yaşar,B., & Ünlü, R. E. (2023). Single‑center plastic surgery response to the 2023 Kahramanmaraş earthquake. *Journal of Disaster and Reconstructive Surgery*. Advanceonline.https://doi.org/10.52312/jdrs.2023.1107

work page arXiv 2023

[7] [6]

J., Ephraim, P

Ziegler‑Graham, K., MacKenzie, E. J., Ephraim, P. L., Travison, T. G., & Brookmeyer, R. (2008). Projected prevalenceoflimb lossintheUnitedStatesbetween 2005 and 2050. *Archives of Physical Medicine and Rehabilitation*, 89(3), 422–429. https://doi.org/10.1016/j.apmr.2007.11.005

work page doi:10.1016/j.apmr.2007.11.005 2008

[8] [7]

W., Kristoffersen, M

Franzke, A. W., Kristoffersen, M. B., Bongers, R. M., Murgia, A., Pobatschnig, B., Unglaube, F., & van der Sluis, C. K. (2019). Users’ vs. therapists’ views on multi‑function myoelectric prostheses: traditional control compared with pattern‑recognition. *PLoS ONE*, 14(8), Article e0220899. https://doi.org/10.1371/journal.pone.0220899

work page doi:10.1371/journal.pone.0220899 2019