Upper esophageal sphincter opening segmentation with convolutional recurrent neural networks in high resolution cervical auscultation

May combine two claims.

The method reached over 90% accuracy, with sensitivity and specificity close to human ratings, including on swallows from a separate clinical experiment.

“The proposed method achieved more than 90% accuracy and similar values of sensitivity and specificity when compared to human ratings even when tested over swallows from an independent clinical experiment.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Of the 116 patients, 15 had stroke and 101 had other medical conditions unrelated to stroke.

“Of the sample, 15 patients were diagnosed with stroke while the remaining 101 patients were diagnosed with different medical conditions unrelated to stroke.”

May combine two claims.

The chosen network architecture was the most stable among tested configurations and fastest to converge, with the smallest number of parameters while achieving detection accuracy that did not sharply change when adding layers or increasing layer sizes.

“The chosen architecture was found to be the most stable among the tested configurations and the fastest to converge. In other words, it included the smallest number of parameters to be optimized while achieving a detection accuracy that doesn’t sharply change when adding more layers or increasing the layer sizes.”

May combine two claims.

The study yielded 710 swallows (132 from stroke patients, 578 from other diagnoses) with average pharyngeal bolus transit duration 869.5±221 msec and average DUESO 604.9±150 msec.

“This study yielded 710 swallows (132 from patients diagnosed with stroke and 578 from patients with other diagnoses) with an average duration of pharyngeal bolus transit of 869.5±221 msec and an average DUESO of 604.9±150 msec.”

May combine two claims.

In the independent test set, about 84% of swallows had UES opening detected within 100 ms, and about 66% had closure detected within that tolerance.

“The results showed that UES opening and closure were detected within a 100 msec tolerance in around 84% and 66% of the swallows in the independent test set respectively.”

May combine two claims.

In 82.6% of swallows, the opening of the UES fell within 100 ms of the human ratings, and in 90% it fell within 133 ms.

“The comparison shows that for 82.6% of the swallows, the opening of UES was detected within a 100 msec (≈ 3 frames at 30 FPS) of the human ratings, and within a 133 msec (≈ 4 frames at 30 FPS) for 90% of the swallows (Fig. 5 (a)).”

May combine two claims.

The proposed system achieved an average accuracy of 0.9039, sensitivity of 0.9145, and specificity of 0.9119 across all folds on the whole dataset.

“The average accuracy of all folds across the whole dataset was 0.9039 with 0.9145 sensitivity and 0.9119 specificity.”

May combine two claims.

The network's detections of UES opening and closure occurred, on average, 33 ms and 16 ms before the true events, respectively, as determined by swallow kinematic analysis.

“On average, the network detected UES opening 33 msec earlier and closure 16 msec earlier than true opening and closure as measured by swallow kinematic analysis.”

May combine two claims.

The system detected as much as 93.6% of the true DUESO, with few false positives or negatives, which appeared only at the edges of the DUESO.

“Based on the results, we can clearly see that the proposed system accurately detected up to 93.6% of the actual DUESO with low rates of false positives and negatives occurring only at the borders of DUESO as shown in Fig. 4 (b).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Validation participants reported no neurological disorder, no head or neck surgery, and no possibility of pregnancy.

“Participants in this validation sample also had no history of neurological disorder, surgery to the head or neck region, or chance of being pregnant based on participant report.”

May combine two claims.

For the 50 swallows in the independent clinical experiment, the system achieved average per swallow accuracy 0.8880, sensitivity 0.8559, and specificity 0.9356.

“for the 50 swallows, the system achieved an average per swallow accuracy of 0.8880, an average per swallow sensitivity of 0.8559, and an average per swallow specificity of 0.9356.”

May combine two claims.

The clinical validation included 15 healthy adults (8 males, 7 females, mean age 63.7 ± 6.2) with no reported current or prior swallowing difficulties.

“In order to evaluate the proposed system in a clinical environment, it was tested during the workflow of an ongoing clinical experiment performed on 15 (8 males, 7 females, age: 63.7 ± 6.2), community dwelling healthy adults who provided informed consent, and who had no reported current or prior swallowing difficulties.”

May combine two claims.

The system achieved average accuracy above 90% of the swallow width and sensitivity above 91% of the DUESO width, with few false positives.

“The proposed system yielded an average accuracy of more than 90% of the swallow width and more than 91% of the DUESO width (sensitivity) with a low false positive rate.”

May combine two claims.

The system's testing accuracy on the independent set (0.888) was close to the validation accuracy on the original dataset (0.9035), and sensitivity and specificity were similar.

“In terms of global measurements, the system achieved a close testing accuracy compared to the validation done through the folds of the original dataset (0.888 vs. 0.9035) and the same for sensitivity and specificity.”

May combine two claims.

For 72.3% of swallows, UES closure was detected by the network within 100 ms, and for over 80% it was detected within 133 ms.

“Likewise, the network accurately detected UES closure within a 100 msec (≈ 3 frames at 30 FPS) for 72.3% of the swallows and within a 133 msec (≈ 4 frames at 30 FPS) for more than 80% of the swallows (Fig. 5 (b)).”

Classified as probably not a substantial claim.

UES opening permits ingested material to move from the pharynx into the esophagus.

“UES opening enables food and liquid to enter the esophagus [19]–[22].”

• cites [19]–[22]Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims.

Signals and video were acquired simultaneously with complete end-to-end synchronization.

“Both signals and video were acquired simultaneously using LabView’s Signal Express (National Instruments, Austin, Texas) with a complete end-to-end synchronization.”

May combine two claims.

Sensor placement over the cricoid cartilage is reasonable because most pharyngeal swallowing activity arises from structures there, and this site has been reported to give the best signal-to-noise ratio for swallowing signals.

“Attaching the sensors around the area of cricoid cartilage is logical given that most of the pharyngeal swallowing activity is produced by the anatomical structures present at this level and it has been reported to yield the best signal-to-noise ratio for the acquisition of swallowing signals [35], [36], [56], [57].”

• cites [35], [36], [56], [57]Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

UES opening is produced by hyolaryngeal excursion, bolus propulsion, and neural inhibitory relaxation of the UES.

“UES opening is the product of hyolaryngeal excursion, bolus propulsion, and neural inhibitory relaxation of the UES itself [22], [23].”

• cites [22], [23]Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

HRCA has been used to monitor pharyngeal bolus propulsion from when the bolus passes the mandible until the UES closes.

“HRCA has been used in multiple studies to monitor the pharyngeal bolus propulsion during swallowing from the moment the bolus passes the mandible till the UES closes [33]–[36].”

• cites [33]–[36]Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims.

RNNs have been extensively used for time series modeling because they can carry information from long contexts, selectively transfer information across time steps, and scale affordably compared with stochastic models and feed-forward networks.

“Recurrent neural networks (RNNs) have been extensively employed for the time series modeling in the recent years, due to their capability of carrying information from arbitrarily long contexts, selective information transfer across time steps, and affordable scalability when compared to stochastic models and feed-forward networks [41]–[44].”

• cites [41]–[44]Not checked yet

  Not checked in the free run.

May combine two claims.

Because pharyngeal bolus propulsion, hyolaryngeal excursion, and UES closure are parts of the UES opening mechanism, HRCA signals have potential to detect UES opening.

“The formerly mentioned events are all parts of the UES opening mechanism which proves the potential of HRCA signals in detection of UES opening.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Hyolaryngeal excursion has been investigated as the origin of HRCA signals and later used to track hyoid bone location during swallowing.

“Furthermore, hyolaryngeal excursion has been investigated to be the origin of HRCA signals in many occasions [37]–[39], and later they were successfully used to actually track the location of the hyoid bone during swallowing [32].”

• cites [37]–[39]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [32] Mao S, Zhang Z, Khalifa Y, Donohue C, Coyle JL, and Sejdić E,´ “ Neck sensor-supported hyoid bone movement tracking during swallowing,” Royal Society Open Science, vol. 6, no. 7, p. 181982, 7 2019. 31417694 PMC6689594 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Reduced UES opening diameter, delayed opening onset, or premature closure shorten UES opening duration and can cause pharyngeal residue, which may enter the upper or lower airway and is a known risk factor for pneumonia and airway obstruction.

“Reduced UES opening diameter, delayed onset of opening, or premature closure attenuate UES opening duration and can result in pharyngeal residue that in turn can enter the upper (laryngeal penetration) or lower (aspiration) airway, which are known risk factors for pneumonia and airway obstruction [23].”

• cites [23] Kim Y, Park T, Oommen E, and McCullough G, “ Upper esophageal sphincter opening during swallow in stroke survivors,” American Journal of Physical Medicine and Rehabilitation, vol. 94, no. 9, pp. 734– 739, 9 2015. 25768066 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The University of Pittsburgh IRB approved the study, and every patient gave informed consent, including consent to publish, before enrollment.

“Permission for this study was granted by the institutional review board of the University of Pittsburgh and all participating patients provided informed consents including consent to publish before enrollment.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Dropout between layers and early stopping were used to control overfitting to training data.

“The used architecture employed also dropout between layers as well as early stopping techniques to control the network from over-fitting to the training data [72].”

• cites [72] Srivastava N, Hinton G, Krizhevsky A, Sutskever I, and Salakhut-dinov R, “ Dropout: A simple way to prevent neural networks from over-fitting,” Journal of Machine Learning Research, vol. 15, no. 56, pp. 1929– 1958, 6 2014. sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

Three expert judges with training in swallow kinematics marked the first frame of UES opening and the first frame of UES closure in the segmented videos.

“Three expert judges trained in swallow kinematic judgments, identified the video frame of first UES opening and the video frame of first UES closure in the segmented videos.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The band-limiting filter cut-off frequencies were chosen to suppress most body sway components below 0.2 Hz and pass signal components containing the vast majority of energy.

“The cut-off frequencies for the band-limiting filter were chosen so that most of body sway components below 0.2 Hz are suppressed and the signal components with the vast majority of energy are passed [35], [59]–[61].”

• cites [35], [59]–[61]Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Earlier work found that swallowing accelerometry signals have their peak energy below 100 Hz and a central frequency near or below 300 Hz.

“A more intense down-sampling could have been adopted as previous studies reported that the frequency with the maximum energy for swallowing accelerometry signals occurs below 100 Hz and the central frequency almost below 300 Hz [35], [63]–[65].”

• cites [35], [63]–[65]Not checked yet

  Not checked in the free run.

May combine two claims.

VFSS is the most frequent and best modality for clinical assessment of swallow kinematic events such as UES opening because it dynamically visualizes the UES during all swallow phases and gives exact opening and closing moments.

“VFSSs are most frequently and actually the best modality to clinically assess swallow kinematic events such as UES opening, because of the ability to dynamically visualize the UES during all phases of the swallow and give exact estimates of the moments when UES opens and closes [19], [20].”

• cites [19], [20]Not checked yet

  Not checked in the free run.

May combine two claims.

The contact microphone was chosen to have a flat frequency response over the audible range, which was shown to pass most frequencies encountered during swallowing.

“The contact microphone was chosen as well so that it produces a flat frequency response over the entire range of audible sounds which was proved to pass most of the frequencies encountered during swallowing [57]–[59].”

• cites [57]–[59]Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

UES dysfunction can result from neurological diseases that change UES relaxation timing and muscular traction delivery, or from impaired propulsive forces from the oropharyngeal pump.

“UES dysfunction may occur due to neurological diseases that alter the timing of UES relaxation and the delivery of muscular traction forces that act to distend the relaxed UES during swallowing, or due to impaired propulsive forces applied by the oropharyngeal pump [20], [23].”

• cites [20], [23]Not checked yet

  Not checked in the free run.

May combine two claims.

Raters kept intra- and inter-rater reliability above 0.xx during checks on 10% of swallows and were blinded to demographics, diagnosis, and bolus condition.

“All raters maintained intra- and inter-rater reliability throughout measurements on 10% of swallows with ICC’s over 0.xx and were blinded to participant demographics and diagnosis and any bolus condition information.”

May combine two claims.

RNNs are efficient at modeling temporal contexts in time series and have been used for biomedical prediction tasks, but training RNNs on raw signals is very difficult because error signals propagate through many time steps.

“RNNs are seemingly efficient in modeling temporal contexts in time series data and have been recently used to perform many biomedical prediction tasks of various complexities [45]–[48], but nevertheless using RNNs on raw signals is extremely hard to optimize because of the propagating of error signals through huge number of time steps [49], [50].”

• cites [45]–[48]Not checked yet

  Not checked in the free run.

• cites [49], [50]Not checked yet

  Not checked in the free run.

May combine two claims.

VFSS uses ionizing radiation, is unavailable or undesirable for many patients, is relatively expensive, and requires specialized instrumentation and trained clinicians, leaving many patients undiagnosed or inaccurately diagnosed and at ongoing risk of dysphagia-related complications.

“However, VFSSs, which use ionizing radiation to produce radiographic images with full temporal resolution, are unavailable or undesirable to many patients, are relatively expensive, and require specialized instrumentation and trained clinicians to perform and interpret, leaving many patients undiagnosed or inaccurately diagnosed, and exposed to ongoing risk of dysphagia-related complications [19].”

• cites [19] Singh S and Hamdy S, “ The upper oesophageal sphincter,” Neurogastroenterology and Motility, vol. 17 Suppl 1, no. Suppl. 1, pp. 3– 12, 6 2005. 15836450 sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims.

Swallows were split at random into ten equal groups; ten-fold cross-validation trained on nine groups and tested on the held-out group, repeated ten times.

“The dataset was randomly divided into 10 equal subsets in terms of the number of swallows. A holdout method was repeated 10 times by training with 9 subsets and testing with the remaining one (10-fold cross validation).”

Probably not a substantial claim: may not pass S2 (checkable).

When anatomical diagnoses are suspected but imaging is not immediately available, HRCA may offer provisional information to guide care.

“This point is crucial in that some causes of dysphagia are indeed anatomically based, however in situations in which such diagnoses are suspected and imaging is not available immediately, HRCA certainly shows promise toward providing interim information that can guide management.”

• AssumptionRests on an assumption: Some dysphagia causes are anatomical.

Probably not a substantial claim: may not pass S3 (load-bearing).

A single earlier study noted that the acoustic part of HRCA signals changes as the bolus moves through the UES.

“One study also observed changes in the acoustic component of HRCA signals while the bolus passed through the UES [75].”

• cites [75] Moriniere S, Boiron M, Alison D, Makris P, and Beutter P, “ Origin of the sound components during pharyngeal swallowing in normal subjects,” Dysphagia, vol. 23, no. 3, pp. 267– 273, 9 2008. 18071792 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The largest anterior-posterior UES diameter occurs 70 to 170 ms after the UES opens, with bolus size and other factors influencing the timing.

“Maximal values of A-P UES diameter were found also to be reached after 70–170 msec of UES opening, depending on the bolus size and other factors [22].”

• cites [22] Cook IJ, Dodds WJ, Dantas RO, Massey B, Kern MK, Lang IM, Brasseur JG, and Hogan WJ, “ Opening mechanisms of the human upper esophageal sphincter,” American Journal of Physiology, vol. 257, no. 5 Pt 1, pp. G748– G759, 11 1989. 2596608 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The validation experiment used the same equipment and hardware as the main dataset.

“The experimental setup of this clinical experiment relied on the same equipment and hardware used for the collection of the main dataset as shown in Fig. 1.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The accepted tolerance for human frame selection is approximately ±2.48 frames at 30 FPS.

“The accepted tolerance for human frame selection ≈ ± 2.48 frames at 30 FPS [62].”

• cites [62] Lof GL and Robbins J, “ Test-retest variability in normal swallowing,” Dysphagia, vol. 4, no. 4, pp. 236– 242, 12 1990. 2209099 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The clinical use of VF is limited by radiation exposure and its lack of availability beyond clinical care settings.

“We have established the fact that UES opening can be best visualized using VF which is clinically impractical due to the delivered radiation doses and unavailability outside clinical care settings.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Certain earlier studies found that some HRCA signal features reach local peaks around the moments when the UES opens and closes.

“Some of these studies reported the presence of localized maxima of some HRCA signal features at UES opening and closure times [40], [74].”

• cites [40] Kurosu A, Coyle JL, Dudik JM, and Sejdić E,´ “ Detection of swallow kinematic events from acoustic high-resolution cervical auscultation signals in patients with stroke,” Archives of Physical Medicine and Rehabilitation, vol. 100, no. 3, pp. 501– 508, 3 2019. 30071198 PMC6402604 sourceNot checked yet

  Not checked in the free run.

• cites [74] Perlman AL, He X, Barkmeier J, and Van Leer E, “ Bolus location associated with videofluoroscopic and respirodeglutometric events,” Journal of Speech, Language, and Hearing Research, vol. 48, no. 1, pp. 21– 33, 2 2005. 15938058 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The system's accurate DUESO detection extends to data it has never seen, indicating strong generalization.

“This confirms that the high quality of DUESO detection can be carried over to completely unseen data and assures a high degree of generalization in the proposed system.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The Adam optimizer was used to train the network because it converges well without hyperparameter fine-tuning.

“We used the Adam optimizer to train the network due to its superiority in convergence without fine tuning for hyper-parameters [73].”

• cites [73] Bengio Y, “ Practical recommendations for gradient-based training of deep architectures,” in Neural Networks: Tricks of the Trade, 2nd ed., ser. Lecture Notes in Computer Science, Montavon G, Orr GB, and Müller K-R, Eds. Springer, 2012, vol. 7700, pp. 437– 478.Not checked yet

  Not checked in the free run.