CoChat

Tremor Detection Using Parametric and Non-Parametric Spectral Estimation Methods: A Comparison with Clinical Assessment

Open access
Martinez Manzanera, Octavio, Elting, Jan Willem, van der Hoeven, Johannes H., Maurits, Natasha M. · 2016 · PLOS ONE

The DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index.

Abstract

In the clinic, tremor is diagnosed during a time-limited process in which patients are observed and the characteristics of tremor are visually assessed. For some tremor disorders, a more detailed analysis of these characteristics is needed. Accelerometry and electromyography can be used to obtain a better insight into tremor. Typically, routine clinical assessment of accelerometry and electromyography data involves visual inspection by clinicians and occasionally computational analysis to obtain objective characteristics of tremor. However, for some tremor disorders these characteristics may be different during daily activity. This variability in presentation between the clinic and daily life makes a differential diagnosis more difficult. A long-term recording of tremor by accelerometry and/or electromyography in the home environment could help to give a better insight into the tremor disorder. However, an evaluation of such recordings using routine clinical standards would take too much time. We evaluated a range of techniques that automatically detect tremor segments in accelerometer data, as accelerometer data is more easily obtained in the home environment than electromyography data. Time can be saved if clinicians only have to evaluate the tremor characteristics of segments that have been automatically detected in longer daily activity recordings. We tested four non-parametric methods and five parametric methods on clinical accelerometer data from 14 patients with different tremor disorders. The consensus between two clinicians regarding the presence or absence of tremor on 3943 segments of accelerometer data was employed as reference. The nine methods were tested against this reference to identify their optimal parameters. Non-parametric methods generally performed better than parametric methods on our dataset when optimal parameters were used. However, one parametric method, employing the high frequency content of the tremor bandwidth under consideration (High Freq) performed similarly to non-parametric methods, but had the highest recall values, suggesting that this method could be employed for automatic tremor detection.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pone.0156822

Cites

20 of 20 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig 1 Building the consensus signal. Top: The signal to be analyzed is constructed from segments where two clinicians agreed on the presence or absence of tremor. These segments were concatenated. This signal is composed of the tremor segments (in purple) and the “no tremor” segments (in green) of the 14 patients. It is important to note that the segments were first rearranged for each patient, in the following order: all “no tremor” segments followed by all “tremor” segments. Bottom: The bottom signal is the consensus between both clinicians. Above this signal, the four iterations used to evaluate each method are described. For each iteration, dark green segments represent the segments employed to obtain the optimal parameters and light green segments represent the segments employed to evaluate the optimal parameters.
Fig 1 Building the consensus signal. Top: The signal to be analyzed is constructed from segments where two clinicians agreed on the presence or absence of tremor. These segments were concatenated. This signal is composed of the tremor segments (in purple) and the “no tremor” segments (in green) of the 14 patients. It is important to note that the segments were first rearranged for each patient, in the following order: all “no tremor” segments followed by all “tremor” segments. Bottom: The bottom signal is the consensus between both clinicians. Above this signal, the four iterations used to evaluate each method are described. For each iteration, dark green segments represent the segments employed to obtain the optimal parameters and light green segments represent the segments employed to evaluate the optimal parameters.CC BY · PubMed Central
Fig 2 Frequency responses of autoregressive models. Frequency responses of 20 AR models (1st to 20th order) of one segment of 4 s (green lines). The DFs were determined according to five criteria: 1) High Amp: the DF corresponds to the frequency of the highest peak among all frequency responses (yellow dot). 2) High Freq: the DF corresponds to the frequency of the peak with the highest frequency among all frequency responses (pink dot). 3) Akaike: the DF corresponds to the frequency of the highest peak of the model selected by Akaike criterion (black dot). 4) Mode High Amp: the DF corresponds to the value of the mode of the frequencies of the highest peak of each model (yellow diamond). 5) Mode High Freq: the DF corresponds to the value of the mode of the frequencies of the peak with the highest frequency of each model (yellow diamond). Note that for this segment Mode High Amp and Mode High Freq have the same value. The PSD employing the periodogram, Welch (2), Welch (3) and Welch (8) methods are shown in dark red, red, blue and dark blue respectively. Also, the DF of each method is shown with a dot of its corresponding color.
Fig 2 Frequency responses of autoregressive models. Frequency responses of 20 AR models (1st to 20th order) of one segment of 4 s (green lines). The DFs were determined according to five criteria: 1) High Amp: the DF corresponds to the frequency of the highest peak among all frequency responses (yellow dot). 2) High Freq: the DF corresponds to the frequency of the peak with the highest frequency among all frequency responses (pink dot). 3) Akaike: the DF corresponds to the frequency of the highest peak of the model selected by Akaike criterion (black dot). 4) Mode High Amp: the DF corresponds to the value of the mode of the frequencies of the highest peak of each model (yellow diamond). 5) Mode High Freq: the DF corresponds to the value of the mode of the frequencies of the peak with the highest frequency of each model (yellow diamond). Note that for this segment Mode High Amp and Mode High Freq have the same value. The PSD employing the periodogram, Welch (2), Welch (3) and Welch (8) methods are shown in dark red, red, blue and dark blue respectively. Also, the DF of each method is shown with a dot of its corresponding color.CC BY · PubMed Central
Fig 3 Example of dominant frequencies and tremor classification of each segment for patient 1. Top: identified DFs during the analysis of approximately 23 minutes of tremor data using Periodogram (pink), High Freq (purple) and Welch (2) (green). Each circle corresponds to the dominant frequency of one segment. This figure illustrates that the High Freq method was not constrained to a specific frequency resolution like the Periodogram or Welch methods. A segment was classified as “tremor” if the DF was located in the 2.5–12 Hz frequency band (pale green) for this example. Bottom: The tremor classification according to the consensus between both clinicians (black: agreement on tremor presence, white: agreement on tremor absence, red: lack of agreement), and for every automatic detection method are shown for the same dataset (tremor bandwidth = 2.5–12 Hz, power threshold = 0 m2/s4). The tremor classification corresponds to individual segments, before rearranging the data (as illustrated in Fig 1).
Fig 3 Example of dominant frequencies and tremor classification of each segment for patient 1. Top: identified DFs during the analysis of approximately 23 minutes of tremor data using Periodogram (pink), High Freq (purple) and Welch (2) (green). Each circle corresponds to the dominant frequency of one segment. This figure illustrates that the High Freq method was not constrained to a specific frequency resolution like the Periodogram or Welch methods. A segment was classified as “tremor” if the DF was located in the 2.5–12 Hz frequency band (pale green) for this example. Bottom: The tremor classification according to the consensus between both clinicians (black: agreement on tremor presence, white: agreement on tremor absence, red: lack of agreement), and for every automatic detection method are shown for the same dataset (tremor bandwidth = 2.5–12 Hz, power threshold = 0 m2/s4). The tremor classification corresponds to individual segments, before rearranging the data (as illustrated in Fig 1).CC BY · PubMed Central
Fig 4 ROC and PR curves for each method for one of the training folds. Top: Precision and Recall are employed to build the PR curves of each method (gray lines). The optimal point of each method according to these metrics is plotted with a purple dot. Periodogram (red lines) was the method with the largest F1 score (its optimal point is the one closest to the (1, 1) coordinate). High Freq is plotted in blue. Bottom: Sensitivity and 1-Specificity are employed to build the ROC curves for each method (gray lines). The optimal point of each method according to these metrics (best combination of sensitivity and specificity) is marked with a red dot. Periodogram (red lines) is the method with the highest sensitivity and specificity combination (its optimal point is the one closest to the (0, 1) coordinate). High Freq is plotted in purple. To illustrate that both methods are not equivalent the optimal points for each method employing the PR curves are plotted among the ROC curves (purple) and the optimal points for each method employing the ROC curves are plotted among the PR curves (red).
Fig 4 ROC and PR curves for each method for one of the training folds. Top: Precision and Recall are employed to build the PR curves of each method (gray lines). The optimal point of each method according to these metrics is plotted with a purple dot. Periodogram (red lines) was the method with the largest F1 score (its optimal point is the one closest to the (1, 1) coordinate). High Freq is plotted in blue. Bottom: Sensitivity and 1-Specificity are employed to build the ROC curves for each method (gray lines). The optimal point of each method according to these metrics (best combination of sensitivity and specificity) is marked with a red dot. Periodogram (red lines) is the method with the highest sensitivity and specificity combination (its optimal point is the one closest to the (0, 1) coordinate). High Freq is plotted in purple. To illustrate that both methods are not equivalent the optimal points for each method employing the PR curves are plotted among the ROC curves (purple) and the optimal points for each method employing the ROC curves are plotted among the PR curves (red).CC BY · PubMed Central
Fig 5 Average F1 scores. F1 scores for each combination of the lower limit for the tremor bandwidth and power thresholds for the High Freq method averaged across the four iterations. The highest F1 score is obtained for lower limits between 2.45–2.90 Hz and a power amplitude threshold of 0 m2/s4.
Fig 5 Average F1 scores. F1 scores for each combination of the lower limit for the tremor bandwidth and power thresholds for the High Freq method averaged across the four iterations. The highest F1 score is obtained for lower limits between 2.45–2.90 Hz and a power amplitude threshold of 0 m2/s4.CC BY · PubMed Central
Fig 6 Estimated F1 scores. The optimal parameters for each method were derived for each of the four test iterations. The corresponding F1 scores (purple), precision (green) and recall (pink) of each fold are plotted. For all methods except High Freq there is considerably higher precision than recall. High Freq is the only method where precision (dots covered by recall dots) and recall have a similar performance.
Fig 6 Estimated F1 scores. The optimal parameters for each method were derived for each of the four test iterations. The corresponding F1 scores (purple), precision (green) and recall (pink) of each fold are plotted. For all methods except High Freq there is considerably higher precision than recall. High Freq is the only method where precision (dots covered by recall dots) and recall have a similar performance.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 20 references

Cited by

See all 22

Metadata

Authors
Martinez Manzanera, O., Elting, J. W., van der Hoeven, J. H., & Maurits, N. M.
Affiliations
University Medical Center Groningen (NL)
Venue
PLOS ONE
Published
2016-06-03
Volume
11
Issue
6
Pages
e0156822
ISSN
1932-6203 (ISSN-L)
OpenAlex
W2425917746
Cited by
22
References
20
Citations by year
2026: 1 · 2025: 1 · 2024: 1 · 2023: 2 · 2021: 4 · 2020: 5 · 2019: 4 · 2018: 2 · 2017: 2
FWCI
2.00
Type
journal article
Language
en
Open access
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • University of Groningen research database (University of Groningen / Centre for Information Technology) (submittedVersion, cc-by)
Funders
Consejo Nacional de Ciencia y Tecnología
Topics
  • Health Sciences › Medicine › Clinical Neurology › Deep Brain Stimulation and Movement Disorders
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Motor Control and Adaptation
Keywords
tremor detection, tremor disorders, accelerometer data, spectral estimation, clinical evaluation, differential diagnosis, clinician consensus
MeSH terms
Aged, Aged, 80 and over, Electromyography, Female, Humans, Male, Middle Aged, Signal Processing, Computer-Assisted, Tremor, Essential Tremor, Accelerometry

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Tremor Detection Using Parametric and Non-Parametric Spectral Estimation Methods: A Comparison with Clinical Assessment. https://app.cochat.ai/papers/10.1371/journal.pone.0156822
The paper
Martinez Manzanera, O., Elting, J. W., van der Hoeven, J. H., & Maurits, N. M. (2016). Tremor Detection Using Parametric and Non-Parametric Spectral Estimation Methods: A Comparison with Clinical Assessment. PLOS ONE. https://doi.org/10.1371/journal.pone.0156822

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.