CoChat

Non-invasive identification of swallows via deep learning in high resolution cervical auscultation recordings

Open access
Khalifa, Yassin, Coyle, James L., Sejdić, Ervin · 2020 · Scientific Reports

The DOI matches its Crossref record; 55 of the 55 references it cites were found in a registry or an index; 1 of the 16 sources its claims cite were checked for support.

Abstract

High resolution cervical auscultation is a very promising noninvasive method for dysphagia screening and aspiration detection, as it does not involve the use of harmful ionizing radiation approaches. Automatic extraction of swallowing events in cervical auscultation is a key step for swallowing analysis to be clinically effective. Using time-varying spectral estimation of swallowing signals and deep feed forward neural networks, we propose an automatic segmentation algorithm for swallowing accelerometry and sounds that works directly on the raw swallowing signals in an online fashion. The algorithm was validated qualitatively and quantitatively using the swallowing data collected from 248 patients, yielding over 3000 swallows manually labeled by experienced speech language pathologists. With a detection accuracy that exceeded 95%, the algorithm has shown superior performance in comparison to the existing algorithms and demonstrated its generalizability when tested over 76 completely unseen swallows from a different population. The proposed method is not only of great importance to any subsequent swallowing signal analysis steps, but also provides an evidence that such signals can capture the physiological signature of the swallowing process.

Overview

CoChat Summary

TL;DR

The paper proposes an automatic swallowing-event segmentation algorithm for high-resolution cervical auscultation using spectrogram features and a deep feedforward neural network. It reports >95% window-level detection accuracy on 3,144 manually labeled swallows and promising generalization to 76 unseen swallows from a different population.

Key findings

  • The best-performing signal was the anterior-posterior (A-P) accelerometer axis, with best window sizes of 800–900 samples (200–225 msec); the algorithm achieved over 95% accuracy at the window level, with sensitivity and specificity also reported as similar.
  • On 3,144 swallows from 248 patients (average duration 862.6 ± 277 msec), detection exceeded 90% across single, multiple, and sequential swallows under several overlap criteria; for single swallows at 800 samples, detected swallows were 100% for 2 SD below average and 98.3% for 90% and 95% overlap.
  • On 76 completely unseen swallows from a different population (average duration 1011 ± 216 msec), detection was 97.4% using a 2 SD below-average overlap, 84.2% using 1 SD below-average overlap, and 65.8% using 90% or greater overlap.
  • Manual validation of more than 6,500 detected segments found the algorithm captured more than 94% of swallows and detected 353 swallows not captured/labeled in videos; false positives included hyoid bone movement (434), head and neck movement (266), coughing (134), reduced oral containment/premature spillage (38), and unexplained events (403).

Methods

  • Design: an online-style sliding-window segmentation framework that computes short-time Fourier transform spectrograms for each window and feeds the magnitude spectrogram into a deep feedforward neural network to classify swallow vs. non-swallow.
  • Data: 248 patients (148 males, 100 females, age 63.8 ± 13.7) referred for videofluoroscopic swallowing assessment; 3,144 swallows (603 stroke, 2,541 other) manually labeled by speech-language pathologists; signals included microphone and tri-axial accelerometer (A-P, S-I, M-L), sampled at 20 kHz and downsampled to 4 kHz.
  • Analysis: four signal streams were tested independently; window sizes 500–1500 samples with and without 50% overlap were evaluated; the best configuration was retrained with 10-fold cross-validation and tested on 76 unseen swallows from 20 healthy adults (9 males, 11 females, age 65.8 ± 11.4).

Limitations

  • Stated: performance dropped on the unseen dataset, possibly because average swallow duration differed between datasets and because multiple swallows were included; false positives such as coughing and head movement remained, and future work was proposed to fuse signal lines and use recurrent neural networks.
  • Stated: the dataset “might be available upon reasonable request” from the corresponding author and with permission from clinical collaborators, so it is not openly available.
  • Evident from the text: manual validation of false positives and visually uncovered events was performed for a specific configuration (window size 900 and 90% overlap), and the paper notes that unlabeled swallows during video pause periods were treated as false positives, which may lower reported sensitivity/specificity.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-020-65492-1

Cites

55 of 55 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 System’s parameter selection process.
Figure 1 System’s parameter selection process.CC BY · PubMed Central
Figure 2 The labeling process of a sample swallow sound signal. Red windows represent the swallow segments identified by human expert SLP’s. Green windows represent different positions of the sliding window. The 1st and 3rd positions are labeled as swallows due to large overlap.
Figure 2 The labeling process of a sample swallow sound signal. Red windows represent the swallow segments identified by human expert SLP’s. Green windows represent different positions of the sliding window. The 1st and 3rd positions are labeled as swallows due to large overlap.CC BY · PubMed Central
Figure 3 Sample raw sound and acceleration signals as recorded from the sensors attached on the anterior neck for each patient. The onset and offset of the swallow segment are marked in red dotted lines and the rest are non-swallow segments with the segment marked with the blue dotted lines as an example. (a) Microphone signal (b) A-P acceleration signal.
Figure 3 Sample raw sound and acceleration signals as recorded from the sensors attached on the anterior neck for each patient. The onset and offset of the swallow segment are marked in red dotted lines and the rest are non-swallow segments with the segment marked with the blue dotted lines as an example. (a) Microphone signal (b) A-P acceleration signal.CC BY · PubMed Central
Figure 4 Spectrogram of non-swallow and swallow segments for both acceleration and sound.
Figure 4 Spectrogram of non-swallow and swallow segments for both acceleration and sound.CC BY · PubMed Central
Figure 5 Quality measurements of the full run for the system. (a) Accuracy. (b) Specificity. (c) Sensitivity.
Figure 5 Quality measurements of the full run for the system. (a) Accuracy. (b) Specificity. (c) Sensitivity.CC BY · PubMed Central
Figure 6 Flow of the training (a) and testing (b) paths of the proposed system.
Figure 6 Flow of the training (a) and testing (b) paths of the proposed system.CC BY · PubMed Central
Figure 7 Temporal enhancement process: (a) shows the flowchart of the process. (b) A sample area under the spectrogram curve of a swallowing mic signal.
Figure 7 Temporal enhancement process: (a) shows the flowchart of the process. (b) A sample area under the spectrogram curve of a swallowing mic signal.CC BY · PubMed Central
Figure 8 Possible swallow segmentation results. (a) Sample swallowing sound signal and definition of the swallowing segment (in blue). (b–e) Examples of correctly identified swallow segments (in red). (f,g) Examples of incorrectly identified swallow segments.
Figure 8 Possible swallow segmentation results. (a) Sample swallowing sound signal and definition of the swallowing segment (in blue). (b–e) Examples of correctly identified swallow segments (in red). (f,g) Examples of incorrectly identified swallow segments.CC BY · PubMed Central
Figure 9 A clean A-P acceleration record. The red segments represent swallows as labeled by SLP. Black boxes are segments detected by the algorithm. Images on each corner are simultaneous VFSS snapshots of the signal events.
Figure 9 A clean A-P acceleration record. The red segments represent swallows as labeled by SLP. Black boxes are segments detected by the algorithm. Images on each corner are simultaneous VFSS snapshots of the signal events.CC BY · PubMed Central
Figure 10 Number of swallows in the dataset for each PA score.
Figure 10 Number of swallows in the dataset for each PA score.CC BY · PubMed Central

Verification

Claims the paper makes

1 Valid · 1 Not valid · 10 Not checked · 72 Own results and interpretations

35 more claims

Create a free account to see all 84 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 26
  • It may state two separate claims as one. 27

Create a free account to see these claims.

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.

1 of 1

Citations

Cites

Show all 55 references

Cited by

See all 82

Metadata

Authors
Khalifa, Y., Coyle, J. L., & Sejdić, E.
Affiliations
University of Pittsburgh (US)
Corresponding author
Sejdić, Ervin
Venue
Scientific Reports
Published
2020-05-26
Volume
10
Issue
1
Pages
8704-8704
ISSN
2045-2322 (ISSN-L)
OpenAlex
W3030237993
Cited by
82
References
53
Citations by year
2026: 10 · 2025: 11 · 2024: 13 · 2023: 13 · 2022: 14 · 2021: 14 · 2020: 7
FWCI
17.05
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
U.S. Department of Health & Human Services | NIH | Eunice Kennedy Shriver National Institute of Child Health and Human Development (R01HD074819, R01HD092239)
Topics
  • Health Sciences › Health Professions › Speech and Hearing › Dysphagia Assessment and Management
  • Health Sciences › Medicine › Pulmonary and Respiratory Medicine › Phonocardiography and Auscultation Techniques
  • Health Sciences › Health Professions › Speech and Hearing › Voice and Speech Disorders
Keywords
swallowing detection, cervical auscultation, dysphagia screening, aspiration detection, swallowing sounds, deep feedforward neural network, online detection
MeSH terms
Deep Learning, Aged, Auscultation, Deglutition, Deglutition Disorders, Female, Humans, Male, Middle Aged, Signal Processing, Computer-Assisted

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-30, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-30 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

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: Non-invasive identification of swallows via deep learning in high resolution cervical auscultation recordings. https://app.cochat.ai/papers/10.1038/s41598-020-65492-1
The paper
Khalifa, Y., Coyle, J. L., & Sejdić, E. (2020). Non-invasive identification of swallows via deep learning in high resolution cervical auscultation recordings. Scientific Reports. https://doi.org/10.1038/s41598-020-65492-1

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.