Spectral arc length as a method to quantify pharyngeal high-resolution manometric curve smoothness

Abstract

Background

Pharyngeal high-resolution manometry (HRM) has emerged over the last decade as a valuable assessment tool for oropharyngeal dysphagia. Data analysis thus far has focused primarily on measures of pressure and duration within key anatomic regions. We apply spectral arc length (SPARC), a dimensionless metric for quantifying smoothness felt to indirectly reflect neuromuscular coordination, as a new method of describing manometric curves. We then use it to distinguish swallows from healthy subjects and those with dysphagia related to stroke.

Methods

Previously collected pharyngeal HRM data from eight subjects with history of stroke and eight age- and sex-matched controls were reviewed. Receiver operating characteristic (ROC) analysis was used to optimize SPARC inputs. SPARC was then computed for the velopharynx, tongue base, hypopharynx, and upper esophageal sphincter (UES), and the values were compared between the two subject groups.

Results

Optimized parameter settings yielded an ROC curve with area under the curve (AUC) of 0.953. Mean SPARC values differed between control and stroke subjects for the velopharynx (t = 3.25, p = 0.0058), tongue base (t = 4.77, p = 0.0003), and hypopharynx (t = 2.87, p = 0.0124). Values were similar for the UES (t = 0.43, p = 0.671).

Conclusions

In this preliminary study, SPARC analysis was applied to distinguish control from post-stroke subjects. Considering alternative methods of analyzing pharyngeal HRM data may provide additional insight into the pathophysiology of dysphagia beyond what can be gleaned from measures of pressure and duration alone.

Introduction

Swallowing is pressure-driven and can be evaluated through pharyngeal high-resolution manometry (HRM), which uses 36 sensors to record pressure along the pharynx.¹ Previous studies report pressure peaks, durations, and rates of change.^{2, 3} However, smoothness of pressure changes remains unexplored. Smoothness can reflect stability of muscle contractions generating those pressures. Less smooth movements may indicate weakness, incoordination, or tremor.⁴ No study has described smoothness of swallowing pressures.

Spectral Arc Length (SPARC) was developed in 2011 to estimate motion smoothness⁵ and modified in 2015 to make SPARC independent of temporal scaling.⁶ SPARC is computed by applying the arc length formula to a signal’s Fourier magnitude spectrum (representation of the signal in the frequency domain). More complex spectra have greater arc lengths and represent less smooth signals. Conventionally, a SPARC value closer to zero represents a smoother signal than one with a more negative value.⁵

SPARC analysis was originally used to evaluate arm movements after stroke.⁷ Compared to jerk-based smoothness metrics, SPARC is sensitive to differences in motor function and minimally affected by noise.⁵ SPARC was also used to assess arm movements in cerebral palsy,⁸ gait smoothness in Parkinson’s disease,⁹ and stair negotiation in elderly subjects.¹⁰ It has not been used to assess signals associated with cranial motor functioning; however, threshold and frequency range adjustability make new applications feasible.

Our objective was to introduce SPARC for assessing smoothness of pharyngeal pressure curves, interpreted to reflect neuromuscular coordination of swallowing control. We also determined if SPARC can distinguish between participants with normal swallowing versus those with dysphagia following stroke. Previous studies on limb movement and gait used speed profiles (first derivative of movement); we use the speed profile of pressure change. We hypothesize that control subjects will have less negative SPARC values compared to stroke subjects, reflective of smoother pressure changes.

Materials and Methods

Participants

Previously collected data from sixteen participants were included with Institutional Review Board approval. Eight had dysphagia related to prior stroke (2 females, 6 males; age: 61–95; Table 1), and eight were age- (±6 years) and sex-matched controls without swallowing, neurological, or gastrointestinal disorders (2 females, 6 males, 61–89 years). Participants avoided eating for four hours and drinking for two hours before testing.

Table 1.

Clinical characteristics of the post-stroke subjects

Number	Age	Stroke Location	Deficits
1	73	Lateral medulla	Unilateral vocal fold immobility, cricopharyngeal dysfunction; s/p medialization thyroplasty with arytenoid adduction
2	67	Multiple brainstem infarcts	Facial paresis, impaired base of tongue movement, reduced hyolaryngeal excursion, impaired pharyngeal constriction, cricopharyngeal dysfunction
3	95	Left middle cerebral artery, left caudate nucleus, right external capsule	Decreased tongue base retraction, reduced hyolaryngeal excursion, cricopharyngeal dysfunction
4	70	Lateral medulla	Palatal weakness, decreased tongue base retraction, reduced hyolaryngeal excursion, cricopharyngeal dysfunction
5	61	Lateral medulla, vertebral artery	Unilateral vocal fold immobility, hemipharyngeal weakness, cricopharyngeal dysfunction
6	71	Left precentral gyrus	Delayed initiation, reduced tongue base retraction, pharyngeal weakness
7	63	Left cerebellum, dorsolateral medulla	Unilateral vocal fold immobility, delayed initiation, palatal weakness, reduced hyolaryngeal excursion, cricopharyngeal dysfunction; s/p vocal fold injection augmentation
8	71	Lateral medulla	Unilateral vocal fold immobility, hemipharyngeal weakness, cricopharyngeal dysfunction

Equipment

The ManoScan ESO HRM system (Medtronic) was used. Circumferential sensors are 1 cm apart on a catheter with 2.7 mm outer diameter. Fidelity is 2 mmHg with range of −20-300 mmHg and sampling rate of 50 Hz.

Procedure

Topical 2% viscous lidocaine (<1 cc) was applied to the catheter, which was passed through the nasal cavity to the proximal esophagus. Participants rested prior to data collection. Control group swallowed five 10 cc boluses of barium sulfate (International Dysphagia Diet Standardization Initiative (IDDSI) Level 0), delivered via syringe and held until cued by the experimenter to swallow. Participants swallowed with normal effort and head in neutral position while sitting upright. Data on 2 cc water boluses were available for 6/8 healthy participants for bolus volume analysis.

For participants with stroke, bolus size varied from saliva to 10 cc IDDSI 0, and number of swallows varied from 1 to 5 depending on swallowing safety.

Data processing

Regions of interest included velopharynx, tongue base, hypopharynx, and UES.^{11, 12} Velopharynx is the superior-most pharyngeal swallow-related pressure change. Tongue base is caudal to the velopharynx and features a single broad pressure peak. Hypopharynx features multiple peaks secondary to contraction of middle and inferior pharyngeal constrictor muscles and laryngeal movement.¹² UES represents baseline high-pressure zone that decreases during swallow and then increases rapidly before returning to baseline.

Maximum pressure, duration of pressure above baseline, and pressure integral were calculated for the velopharynx, tongue base, and hypopharynx (Table 2). For the UES, maximum pre- and post-pressures, nadir pressure, and nadir duration were calculated (Table 3).

Table 2.

Summary of pressure and timing data for the velopharynx, tongue base, and hypopharynx regions

Region	Control			Stroke
	Max pressure (mmHg)	Duration (s)	Integral (mmHg*s)	Max Pressure (mmHg)	Duration (s)	Integral (mmHg*s)
Velopharynx	139.63 ± 30.84	0.79 ± 0.17	99.25 ± 49.3	168.25 ± 61.89	1.02 ± 0.31	171.75 ± 103.93
Tongue Base	129.63 ± 22.33	0.48 ± 0.12	89 ± 31.92	152.13 ± 61.89	0.86 ± 0.30	211.63 ± 109.39
Hypopharynx	195.38 ± 52.44	0.58 ± 0.16	97.63 ± 52.94	233.38 ± 112.89	0.74 ± 0.14	165.63 ± 76.89

Table 3.

Summary of pressure and timing data for the upper esophageal sphincter (UES).

Parameter	Control	Stroke
Pre-UES max pressure (mmHg)	78.88 ± 16.95	131.75 ± 68.74
Post-UES max pressure (mmHg)	169.88 ± 50.36	145.25 ± 79.84
Nadir pressure ¼ sec (mmHg)	8.25 ± 5.04	10.00 ± 5.52
Nadir duration (s)	0.67 ± 0.12	0.89 ± 0.23

Spectral arc length

SPARC values were calculated using a custom LabVIEW (NI) program. The most proximal and distal sensors are located to select time bounds. Swallow start was defined as the timepoint with near-zero slope before the pressure increase on the proximal-most velopharyngeal sensor. Swallow ended at the pressure peak on the distal-most UES sensor. On each sensor, time derivatives were taken to obtain speed profiles and the Fourier spectrum was computed. Each spectrum was normalized to its maximum amplitude, and SPARC values are calculated for each spectrum. SPARC values were averaged over each regional sensor group and over all sensors. SPARC was calculated using Equation 1 where ωc is the spectrum frequency, V(ω) is the Fourier magnitude spectrum of the rate of change in pressure (I.e., the first derivative of the pressure trace). Adaptive frequency cutoff is defined through Equation 2. This process is depicted in Figure 1.

Figure 1:

A: Sample of control subject; B: Sample of stroke subject; 1: The pharyngeal swallow is isolated in the spatiotemporal plot; 2: The swallow is trimmed down to the point of near-zero slope in the most proximal sensor and the peak of the most distal sensor (this area is outlined by black dashed lines on the spatiotemporal plot); 3: Time derivatives are taken to obtain speed profiles of each sensor; 4: The Fourier spectrum is computed for each sensor’s velocity profile, and each spectrum is normalized to its maximum amplitude; 5: The spectra are trimmed based on the adaptive threshold described in Equation 2; 6: SPARC values are calculated for each trimmed spectrum.

$$\mathrm{SPARC}=-{\int }_0^{{\omega }_c}{\left[{\left(\frac{1}{{\omega }_c}\right)}^2+{\left(\frac{d\widehat{V}\left(\omega \right)}{d\omega }\right)}^2\right]}^{\frac{1}{2}}d\omega ;\widehat{V}\left(\omega \right)=\frac{V\left(\omega \right)}{V\left(0\right)}$$ Equation 1:

$${\omega }_c=min\left\{{\omega }_c^{max},min\left\{\omega ,\widehat{V}\left(r\right)〈\stackrel{-}{V}\forall r〉\omega \right\}\right\}$$ Equation 2:

Parameter optimization

Two threshold values are determined for spectral arc length calculation: (1) upper frequency threshold that bounds range of movement, encompassing both normal and abnormal profiles; and (2) amplitude threshold that defines where the noise floor of the spectrum is, thus controlling the trade-off between sensitivity and noise contamination.^{5, 6} To ensure that we captured the full range of relevant frequencies, the testing range for the higher bound threshold (${\omega }_c^{max}$) was from 1 to 5 Hz, in 1 Hz increments. To capture a complete range of possible amplitude thresholds ($\stackrel{-}{V}$), the testing range used was from 0.01 to 0.1 in increments of 0.01.^{6, 9}

Statistical analysis

To determine the optimal combination of parameters for differentiating controls and participants with stroke, receiver operating characteristic (ROC) analysis was performed on various permutations of settings.¹³ The combination that yielded the highest area under the curve (AUC) was chosen for further analysis.

Independent samples t tests were performed to compare SPARC values for each region and the entire swallow between the two groups. A significance level of α = 0.05 was used. Cohen’s d was calculated for effect size.¹⁴

Smoothness of pharyngeal pressure is not expected to vary significantly across bolus volumes within a typically swallowed bolus size range; however, to ensure that any differences observed between the controls and participants were due to differing swallow function and not bolus size, a secondary analysis comparing 2 cc swallows to 10 cc swallows within the control group was performed using paired samples t tests with a significance level of α = 0.05. No formal statistical analysis was performed for the typical HRM parameters as that was not the study objective.

Results

Parameter settings yielding the highest AUC were ${\omega }_c^{max}=3\mathrm{Hz}$ and $\stackrel{-}{V}=0.01$ (unitless), with an AUC of 0.953. These parameters were used for subsequent analyses.

Overall SPARC values were significantly lower, or less smooth, in participants with stroke than controls (t(14) = 3.25, p = 0.0058). SPARC values in the velopharynx (t(14) = 2.25, p = 0.0415), tongue base (t(14) = 4.77, p = 0.0003), and hypopharynx (t(14) = 2.87, p = 0.0124) were significantly lower for participants with stroke, while there was no significant difference in the UES (t(14) = 0.43, p = 0.671). Effect sizes for significant comparisons were all greater than 1, indicating a large effect. Summary data are provided in Table 4. No significant differences were found between 2 cc and 10 cc swallows in the six control subjects (Table 5).

Table 4.

Summary spectral arc length (SPARC) results. Values are presented as mean ± SD. p-values represent results of independent samples t tests.

Region	Control	Stroke	p-value	Cohen’s d
Overall	−2.19 ± 0.53	−3.09 ± 0.50	0.0058	1.75
Velopharynx	−2.51 ± 0.37	−3.23 ± 0.91	0.0415	1.04
Tongue base	−2.09 ± 0.23	−3.16 ± 0.57	<0.001	2.46
Hypopharynx	−2.48 ± 0.19	−3.03 ± 0.44	0.0124	1.62
Upper esophageal sphincter	−2.94 ± 0.31	−3.02 ± 0.61	0.671	–

Table 5.

Comparison of 2 cc and 10 cc swallows. Values are presented as mean ± SD. p-values represent results of paired samples t tests.

Region	2 cc	10 cc	p-value
Overall	−2.04 ± 0.59	−2.38 ± 0.22	0.322
Velopharynx	−2.47 ± 0.32	−2.35 ± 0.44	0.488
Tongue base	−2.09 ± 0.26	−1.98 ± 0.19	0.412
Hypopharynx	−2.48 ± 0.19	−2.57 ± 0.19	0.549
Upper esophageal sphincter	−2.95 ± 0.33	−2.86 ± 0.35	0.538

Discussion

This pilot study suggests that SPARC, a measure of smoothness, can differentiate healthy and abnormal swallow pressures. Smoothness reflects neuromuscular coordination and may be valuable in neurologic causes of dysphagia. Subtle changes in oropharyngeal swallow and/or improvements related to therapy may not be captured by regional pressure values. Since SPARC measurement utilizes speed profiles of pressure changes in addition to an adaptive frequency cutoff, global neuromuscular coordination can be analyzed without concern for timing of pressure events.

ROC analysis was used to optimize analysis. Resulting 3 Hz cutoff represents the frequency range in which important spectral information is found for healthy and pathological participants. Above 3 Hz, the spectrum contains more noise that can artificially lower the SPARC value. With lower cutoffs, important spectral information would not be included, artificially increasing the SPARC value.

SPARC values were lower for participants with stroke at the velopharynx, tongue base, and hypopharynx. Lack of a difference at the UES was unexpected given documented cricopharyngeal dysfunction in seven of eight stroke subjects. Age-related changes within the UES in control participants may influence analysis; recent studies demonstrated increased pressure variability¹⁷ and lower basal UES pressure with aging.¹⁵

One advantage of metrics describing overall characteristics of the manometric trace rather than discrete points is that they may be less affected by non-pathologic variables such as bolus volume.^{12, 16–18} No changes in SPARC were observed for volume differences in our control sample. Further evaluation in a larger group of subjects and across a wider range of bolus volumes will be performed.

Several limitations are noted. This was an initial exploratory study with a modest sample size. There were differences in bolus volume between groups, as volume was limited in stroke subjects for safety.

Standard pharyngeal HRM analysis includes pressure amplitudes and durations within different regions,^{1, 2} which has disadvantages. Peak pressures are prone to sensor errors and are highly variable across participants. Additionally, patients with weakness in one region may compensate in others, leading to an unremarkable pressure trace.³ Analyzing the unexplored frequency domain allows us to evaluate muscle coordination and contraction smoothness without time being a factor.

Key points.

• Pressure analysis may benefit from the inclusion of a metric which address signal complexity which SPARC analysis provide.

• Complexity increase is identified in dysfunctional swallowing as demonstrated by SPARC analysis.

• Patients who have stroke related dysphagia will demonstrate prolonged spectral arc lengths in the pressure profiles.

Data Availability Statement

Data related to this study are available upon request.