Aerodynamic-Induced Effects of Artificial Subglottic Stenosis on Vocal Fold Model Phonatory Response

Abstract

Objective:

Subglottic stenosis (SGS) is characterized by a narrowing of the trachea near the cricotracheal junction and impairs breathing. SGS may also adversely affect voice quality, but for reasons that are not fully understood. The purpose of this study is to provide experiment-based data concerning the effects on phonation of airway obstruction due to SGS.

Study Design:

Basic science

Methods:

A device simulating a SGS of adjustable severity ranging from 36% to 99.8% obstruction was created. Self-oscillating synthetic VF models were mounted downstream of the device and data were acquired to evaluate the effects of the obstruction on phonatory response.

Results:

Onset pressures were relatively insensitive to obstructions of up to approximately 80% to 90% reductions in subglottic airway area and sharply increased thereafter. Flow rate (under conditions of constant pressure), flow resistance, and fundamental frequency all exhibited similar degrees of sensitivity to SGS obstruction as onset pressure. High-frequency noise became significant by 80% obstruction. Glottal area appeared to be less sensitive, not being affected until approximately 90% obstruction.

Conclusion:

Consistent with previous computational studies, this study found that aerodynamic, acoustic, and vibratory responses of self-oscillating VF models were largely unaffected by SGS until approximately 80% to 90% obstruction, and significantly affected at higher obstructions. This suggests that Grades I and II stenoses are unlikely to introduce subglottic airway aerodynamic disturbances that are sufficient in and of themselves to significantly alter phonatory output. The SGS model introduces a framework for future benchtop studies involving subglottic and supraglottic airway constrictions.

INTRODUCTION

Subglottic stenosis (SGS) is a narrowing of the airway just below the vocal folds (VFs) near the cricotracheal junction [1]. SGS is commonly a result of prolonged medical intubation or laryngeal trauma, but can also be caused by granulomatosis with polyangiitis (an auto-immune disorder) or may be idiopathic [1]. The severity of a stenosis is typically described using the percent obstruction of the airway by area or classified using the Myer-Cotton scale, the latter of which categorizes severities into Grades I, II, III, or IV in ascending order of severity; these grades denote 0% to 50% obstruction, 51% to 70% obstruction, 71% to 99% obstruction, and 100% obstruction, respectively [2].

SGS causes dyspnea, and surgery is frequently necessary to open the airway. Further, SGS and associated treatments may adversely affect voice quality. For example, Hillel et al. [3] analyzed the change in voice quality of 27 subjects with laryngotracheal stenosis (including both subglottic and tracheal stenosis) before and after a balloon dilation operation to restore airway patency, and found that 19 of the 27 had improved Voice-Related Quality of Life (V-RQOL) scores, five had worse V-RQOL scores, and three remained unchanged. Tanner et al. [4] performed a similar study of female patients, although only on subglottic (not tracheal) stenosis, and following a modified cricotracheal resection operation instead of a balloon dilation. They found that, particularly for sustained vowels, voice quality improved. Ettema et al. [5] conducted a study on 31 subjects with subglottic stenosis and found that the subjects had a mean rating of 1.4 on the Grade-Roughness-Breathiness-Asthenia-Strain (GRBAS) scale, which corresponds to mild to moderate hoarseness. Furthermore, their study found that many factors other than simply the degree of stenosis can affect the mean GRBAS rating, including the presence of laryngeal inflammation, the presence of additional stenoses, history of previous airway surgery, and bilateral or unilateral VF motion impairment.

Computational simulations have been used to study the degree to which SGS-induced alterations in subglottic aerodynamic flow patterns may affect phonatory output. Smith and Thomson [6] simulated the effects of SGS on VF vibration using a two-dimensional computational fluid dynamics (CFD) model. They found that glottal width and flow rate waveforms were minimally affected by SGS at 60% obstruction but became noticeable at 90% and greater obstructions. Bodaghi et al. [7] similarly studied SGS and phonation using a three-dimensional high-fidelity CFD model, and as with Smith and Thomson [6], found that most measures, including flow rate and glottal area waveforms, fundamental frequency, signal-to-noise ratio, and flow resistance, were minimally changed at 75% obstruction but became significant at 90% and greater obstruction. Brouns et al. [8] performed a CFD simulation on the flow dynamics of a tracheal stenosis, reporting that the simulated pressure drop across the stenosis became most significant beyond 70% obstruction. A numerical flow, acoustic, and structural simulation was performed by van der Velden et al. [9] in which the acoustics generated by a subglottic stenosis were predicted using a realistic airway model derived from a CT scan. This study was focused on stridor, an abnormal noise that results from constricted breathing, and was not concerned with speech or VF vibration. They found that the stenosis caused an increase of approximately 15 to 20 dB for the broadband noise spectrum during inhalation. Using static cast models of excised human larynx airways, Grisel et al. [10] studied the effect of subglottic medialization on glottal airflow, finding turbulence intensity to be significantly affected only by large degrees of medialization.

Considering the abovementioned human and computational studies, it is clear that phonatory output can be adversely affected by high degrees of SGS; however, the relationship is not yet fully understood. Human studies are subject to person-to-person variability and uncontrolled variables. Computational studies can be precisely controlled, but they rely on mathematical models (which are only approximations) to predict highly complex physical phenomena such as flow separation, turbulence, radiated acoustics, large deformation, and medial surface collision. In benchtop experiments, these phenomena are directly present, are not approximated, and are measurable. Consequently, physical experiments using self-oscillating VFs or VF models are essential for validating the SGS simulation results summarized above, but such experiments have not yet appeared in the SGS literature. Additionally, while in vivo and excised larynges have the advantage over synthetic models of the actual vocal folds being the centerpiece of the experiments, synthetic model experiments are generally more controllable and parameterizable than in vivo and excised larynx experiments. Furthermore, studies can be repeated with duplicate synthetic models that likely exhibit less model-to-model variability than the subject-to-subject variability of human larynges. Synthetic models also generally exhibit superior vibratory longevity. Considered together, each of these three approaches—computer simulation, synthetic model experimentation, and in vivo/excised larynx experimentation—comes with particular limitations, but each also has unique advantages that complement the others.

The purpose of the present study is to introduce an experimental SGS setup and present subsequent experimental data in order to contribute to an improved overall understanding of the influence of SGS on phonation. A device that simulated SGS with varying degrees of constriction was developed. Synthetic, self-oscillating VF models were mounted downstream of the stenosis. Experiments were conducted in which flow rate and pressure data were measured, radiated sound data were recorded, and high-speed videos were acquired. These data were analyzed to assess the influence of the subglottic airway obstruction on phonatory output.

In the following sections, the SGS device and experimental setup and procedure are described. Results are then presented and discussed, and conclusions are summarized. Limitations of the work are discussed, and opportunities for future research are suggested.

METHODS

An overview of the experimental setup and procedures are provided in this section. For additional data and details, the reader is referred to Hilton [11].

Subglottic stenosis device

A computer-controlled mechanical device for simulating varying degrees of SGS was developed. The device consisted of a silicone sleeve constricted by four sets of cords which were tightened by stepper motors as shown in Fig. 1. The sleeve represented the SGS. Each motor constricted the sleeve in an elliptical shape; if all motors were actuated, the constriction was nearly circular. The device was designed in this way to enable future research of non-circular constrictions. The device was fitted with a mounting plate for attaching VF models (described in the “Vocal fold models” and “Experimental setup” sections below). For videos of the SGS constricting, see supplemental videos 1, 2, and 3.

FIGURE 1.

Top: SGS device. I: Partially constricted SGS sleeve circumscribed by cords. II: Cords. III: Stepper motors. IV: Mounting plate for VF models (models not shown). V: Pressure sensor port. Bottom left: Illustration of mode of SGS sleeve constriction as viewed from above, with flow out of the page (not to scale). Shown is one cord (orange) wrapped once around sleeve (blue). Sleeve constriction is achieved by pulling on the cord ends. The SGS device included four such cords oriented orthogonally every 90° around the sleeve. Bottom center: Sketch of unconstricted silicone sleeve and protective outer layer (not to scale; dimensions in mm). Bottom right: Image of silicone cast of constricted SGS interior airway.

The silicone sleeve was cast from Ecoflex 00-30 (Smooth-On, Inc.), which has a 100% modulus of 69 kPa as per manufacturer specifications. The sleeve had an inner diameter of 20 mm and walls that were 5.5 mm thick (see Fig. 1). However, due to plastic deformation after several initial uses, the sleeve could only return to an inner diameter of 16 mm (i.e., an obstruction of 36% of the initial area) when allowed to relax. Consequently, 36% obstruction was the smallest testable obstruction. An additional 1 mm-thick protective outer layer of the stiffer Dragon Skin 10 (Smooth-On, Inc.) was cast where the cords would circumscribe the sleeve to prevent them from cutting into the softer Ecoflex.

The interior stenosis profile was qualitatively evaluated by casting silicone into the cavity at 91% obstruction. An image of this cast is shown in Fig. 1. Further quantitative analysis of the obstruction cross-sectional area was performed by imaging (D5100 SFR camera, Nikon) the constriction from above at a set of programmed constrictions and then calculating the areas using a custom MATLAB script and calibration image for pixel-to-length conversion. Desired (programmed) and actual areas were converted into percent obstruction using the relationship

$$PercentObstruction=\frac{A_0-A}{A_0}\times 100(\%)$$ (1)

where A₀ is the cross-sectional area when the sleeve was undeformed (314 mm² based on 20 mm diameter) and A is the constricted area. Five trials from 36% to 100% obstruction were measured; the results are shown in Fig. 2. The obstruction error (i.e., the difference between measured and desired percent obstruction) is seen to have ranged from approximately −1 to +4% across the five trials, with lower errors at greater obstructions.

FIGURE 2.

Top: Obstruction error (measured percent obstruction minus desired percent obstruction) vs. desired percent obstruction for five trials. Bottom: Images of SGS as viewed from above for Trial #5 (inset numbers denote desired percent obstruction).

Vocal fold models

Synthetic, self-oscillating VF models have been used to study various acoustic, aerodynamic, and vibratory aspects of voice production (e.g., [12–15]). One of their advantages is that they are typically able to vibrate for longer durations than excised larynges. In this study, synthetic VF model geometry and material properties were based on a previous study in which a two-dimensional computational self-oscillating VF model was optimized for closed quotient and frequency [16,17]. The models included four layers of silicone representing the human VF epithelium, superficial lamina propria (SLP), ligament, and body tissue layers. The layers were fabricated using the general sequential casting process described by Murray and Thomson [18] and consisted of silicone (either Dragon Skin 10 or Ecoflex 00-30) combined with different ratios of Silicone Thinner (Smooth-On, Inc.) to yield layers of differing stiffness. The body, ligament, and SLP layers were cast using molds, but the thin epithelial layer was poured on top and allowed to flow over the model prior to curing. The silicone compositions are listed in Table 1 and the geometry is shown in Fig. 3.

TABLE 1.

Silicone mixing ratios for four-layer VF models used in this study.

Layer	Mixing Ratio (A:B:Thinner)
Body	1:1:1 Dragon Skin 10
Ligament	1:1:5.5 Ecoflex 00-30
SLP	1:1:6.5 Ecoflex 00-30
Epithelium	1:1:1 Dragon Skin 10

FIGURE 3.

Vocal fold geometry design. The model was 17 mm in the anterior-posterior direction with a uniform cross section. I: inferior, S: superior, M: medial, L: lateral.

Material samples for each VF model layer were created. The body and the epithelial layer specimens were tested using an Instron 3342 tensile testing apparatus with a 50 N force transducer (model 2519-102, Instron, Inc.) and the much softer ligament and SLP layer specimens were tested using an AR2000ex rheometer (TA Instruments) with a 40 mm plate (Part number 511400.901, TA Instruments). The body and epithelial layer tensile elastic moduli were determined by a linear fit of stress-strain data between 0 and 3.6% strain. The ligament and SLP layer storage moduli were approximated by averaging measured shear moduli at eleven frequencies from 1 to 10 Hz and converting to tensile moduli using the relationship E′ = 2G′(1 + v), where E′ is the tensile storage modulus estimate, G′ is the real component of the measured complex shear modulus, and v is the Poisson’s ratio (assumed to be v = 0.49).

Three pairs of models were fabricated, here denoted pairs A, B, and C. Each pair consisted of two opposing VF models with layers that were poured at the same times using the same batches of silicone. The modulus estimates for each layer of each pair are listed in Table 2. Dragon Skin 10 for the epithelial layer was poured on twice; thus there were samples for both epithelial layers. Some rheometer samples were tested multiple times under slightly different conditions, in which case ranges are provided. Variation in material properties and material property measurements among synthetic VF model materials is common owing to these silicones being sensitive to environmental, temporal (i.e., time between curing and modulus testing), and other processing parameters, measurement conditions, and analysis assumptions; this is particularly the case for the Ecoflex 00-30 used in the SLP and ligament layers. Consequently, the variations seen in Table 2 between the modulus values of the same layers of different pairs are not unexpected, and the values here should primarily be interpreted as being representative of the layer stiffnesses.

TABLE 2.

Modulus values (kPa) of body and epithelium layers (elastic moduli from tensile tests) and ligament and SLP layers (storage moduli from rheometer tests).

Pair	Body	Ligament	SLP	First Epithelium	Second Epithelium
A	64.2	0.50	0.22-0.46	72.8	70.8
B	64.2	0.40-0.51	0.30-0.48	68.8	66.8
C	64.2	0.81-0.97	0.15-0.23	72.8	70.8

Experimental setup

The experiment was conducted in a custom sound booth using the setup illustrated in Fig. 4. The SGS device was mounted to an optics table with air flowing upwards. The VF models were adhered to mounting plates using Sil-poxy (Smooth-On, Inc.), placed on top of the SGS device at the superior end of the SGS silicone sleeve, tightened by clamps, and sealed with vacuum grease to prevent air leakage. A 1/2” free-field microphone (378B02, PCB) was mounted off-axis 28 cm from the VF models and 16 cm from the vocal tract model exit when a vocal tract model (described below) was used. The acoustic signal was digitized at 51.2 kHz using a cDAQ-9178 chassis (National Instruments) and BYU Acoustic Field Recorder software. The microphone was calibrated before all experiments (42AG Multifunction Sound Calibrator, GRAS Sound & Vibration). Acoustic data analysis was based on five-second segments using Hanning windowing with 50% overlap.

FIGURE 4.

Left: Schematic of setup (not to scale), including sub-stenosis and supra-stenosis pressure sensors. The distance between the center of the supra-stenosis sensor and the bottom of the VF models was 4 mm. Right: Vocal tract model cross section, the full length of which was 167 mm.

As illustrated in Fig. 4, pressure sensors were inserted below and above the stenosis (respectively, models 6CF6G and 6AF2G, Honeywell). The sensors were connected to the cDAQ chassis and the values were read in LabVIEW. The sub-stenosis and supra-stenosis pressure sensors were respectively located approximately 62 mm below and 23 mm above the location of maximum constriction. The supra-stenosis sensor was located in the pressure sensor adapter plate (also see Fig. 1). The sub-stenosis sensor was inserted into a 3D-printed tube adapter just upstream of the SGS sleeve. Compressed air was fed to the tube adapter via a system of tubes and an expansion chamber, similar to that which has been described elsewhere [19]. The flow was controlled by a valve and measured using a flow meter (FMA-A2323, Omega Engineering, Inc.)

Model vibration was imaged using a high-speed camera (Edgertronic SC2+, monochrome, Nikon 50 mm lens with 12 mm tube extension, F/4 f-stop, 1/50000 s shutter speed) mounted approximately 20 cm above the device. Two LED lights (900420H, Visual Instrumentation Corporation) provided illumination. Video was recorded at 4000 fps with 1024×592 pixel resolution. The high-speed camera was removed when not in use.

In some experiments a vocal tract model was used which was a modification of the simplified vocal tract models of Arai [20]. As illustrated in Fig. 4, this vocal tract consisted of a straight tube with two lengths of differing diameters simulating the /a/ vowel. The vocal tract base geometry was adjusted to fit over the VF models. Specifically, the base had a rectangular 22 mm × 28 mm cross section to avoid interference with the vibration of the VF models. The vocal tract transitioned to a larger diameter 82 mm above the base and was a total of 167 mm long. The vocal tract was printed using polylactic acid (PLA) filament on a 3D printer (Maker Select Plus, Monoprice) with a solid infill.

Experimental procedure

Generally, for a given VF model pair using this setup, as the SGS was constricted the sub-stenosis pressure increased while the flow rate stayed approximately the same. This led to two possible experimental approaches, one in which the flow rate was constant and the sub-stenosis pressure varied as the constriction increased, and the other in which the flow rate was adjusted so as to maintain constant sub-stenosis pressure. In this study, both methods were used, hereafter referred to as “constant pressure” and “constant flow.” In the constant pressure experiments, the pressure below the stenosis was maintained at the same value over all degrees of stenosis by adjusting the flow rate. In the constant flow experiments the flow rate was maintained at the same value over all degrees of stenosis while the pressure below the stenosis changed. The constant pressure experiments were likely more physiologically relevant, but constant flow results are here included for a more comprehensive study of the effects of SGS on vibration.

As noted above, three model pairs were tested. Before tests were performed with each model pair, onset pressure (using the sub-stenosis sensor) and onset flow rate, both with 36% SGS obstruction, were recorded. Once these baseline values were recorded, the following tests were conducted for obstructions between 36% and 99.8%:

1. Flow and pressure:

For each percent obstruction and without the vocal tract model, (i) onset pressure was recorded (average of five trials), (ii) sub- and supra-stenosis pressures were recorded with the flow rate maintained at 120% of onset flow rate, and (iii) flow rate and supra-stenosis pressure were recorded with the sub-stenosis pressure maintained at 120% of onset pressure.

2. Acoustics:

For each percent obstruction, (i) radiated sound was recorded as the flow rate was maintained at 120% of onset flow rate (no vocal tract model), (ii) radiated sound was recorded as the sub-stenosis pressure was maintained at 120% of onset pressure (no vocal tract model), and (iii) with the vocal tract model in place, sound was recorded as the flow rate was maintained at 120% of onset flow.

3. Vibration (high-speed imaging):

For each percent obstruction and without the vocal tract model, VF model vibration was imaged (i) with the flow rate maintained at 120% of onset flow and (ii) with the sub-stenosis pressure maintained at 120% onset pressure.

The test order was varied between the three VF model pairs. For pair A, flow and pressure tests (#1 above) were performed first, followed by acoustic tests (#2), and then vibration tests (#3). For pair B, vibration tests (#3) were performed first, followed by flow and pressure tests (#1), and then acoustic tests (#2). For pair C, acoustic tests (#2) were performed first, followed by vibration tests (#3), and then flow and pressure tests (#1). All sub-experiments were performed in the order listed above within each of the three test types.

RESULTS AND DISCUSSION

Baseline onset pressures

Baseline onset pressure and flow rate results with 36% obstruction, measured at the beginning of tests for each pair, are listed in Table 3, along with corresponding 120% values used in subsequent experiments as noted above. For reference, Baken and Orlikoff [21] reported onset pressures for human vocal fold vibration in the range of 0.29 to 0.49 kPa, and onset pressures for various synthetic VF models in the range of 0.27 to 4.6 kPa have been reported (e.g., [19, 22–24]).

TABLE 3.

Baseline onset pressures (sub-stenosis; kPa) and flow rates (liters per minute; LPM) for all VF model pairs measured at the beginning of tests for each pair.

Pair	Onset Pressure	120% Onset Pressure	Onset Flow Rate	120% Onset Flow Rate
A	1.58	1.89	53.4	64.1
B	0.93	1.12	18.3	21.9
C	1.60	1.92	41.5	49.8

Flow and pressure measurements

Onset pressure

Sub- and supra-stenosis pressures at onset vs. percent obstruction for the three VF model pairs are shown in Fig. 5a. Onset sub-stenosis pressures were fairly consistent from 36% obstruction to approximately 80% to 90% obstruction, beyond which the onset pressures sharply increased. The onset pressures at the highest obstruction (99.8%) were between 34% and 56% higher for the three pairs than their corresponding onset pressures at 36% obstruction. For a patient with SGS, increases in onset pressure would translate to greater effort being required for voicing.

FIGURE 5.

(Top) Sub-stenosis (solid lines) and supra-stenosis (dashed lines) onset pressures for VF model pairs A, B, and C vs. percent obstruction. (Bottom) Pressure drop across the stenosis at onset vs. percent obstruction.

In contrast to the sub-stenosis pressures, supra-stenosis pressures at onset significantly decreased as obstruction increased from 90% to 99.8% obstruction. This occurred as sub-stenosis pressures were significantly increasing. This can be seen by inspecting the pressure drop data across the stenosis (i.e., sub-stenosis pressure minus supra-stenosis pressure) shown in Fig. 5b. For all pairs there was a pressure drop across the stenosis, even at lower degrees of stenosis, which is largely attributed to so-called “minor” flow losses [25] associated with such flow obstructions. This pressure drop was fairly minimal (between 0.038 and 0.07 kPa for all models) until 80% obstruction, but then grew to as much as 1.4 kPa (pair A) at the greatest obstruction. It might have been expected that as the obstruction increased, the sub-stenosis pressure would have adjusted in order to maintain the same supra-stenosis pressure for onset, assuming that onset may have been governed by supra-stenosis pressure. This, however, was not the case. For example, vibration was initiated for VF model pair A at low levels of obstruction with a supra-stenosis pressure of 1.55 kPa and at 96% obstruction with a supra-stenosis pressure of 1.06 kPa. The supra-stenosis pressure for pair B decreased only slightly at 90% and greater obstructions, but the supra-stenosis pressure for pair C decreased from 1.32 kPa at 75% obstruction to 1.02 kPa at 96% obstruction. Such decreased supra-stenosis pressures may be attributable to the Bernoulli effect associated with increased velocities through the narrower airways, but the precise mechanisms would need to be explored further.

The onset pressures measured at the outset of tests for each pair in Table 3 can be compared with the 36% obstruction onset pressures in Fig. 5 as one indication of possible model fatigue throughout the course of the experiments. Onset pressures for pairs A and B at 36% obstruction in Fig. 5 very closely correspond to those in Table 3, a positive indication of model consistency. A change in onset pressure is more evident in pair C (1.30 kPa in Fig. 5 vs. 1.60 kPa in Table 3, a 19% decrease), possibly because the flow and pressure measurements were recorded last for this pair, after the acoustic and high-speed experiments (as discussed in the “Experimental procedures” section). For pair A the flow and pressure measurements were recorded first, and for pair B they were recorded second.

Constant flow rate experiments

Sub- and supra-stenosis pressures at varying degrees of constriction as the flow was maintained at 120% onset flow rate are shown in Fig. 6. Similar to Fig. 5, the divergence of the sub-stenosis and supra-stenosis pressures can be seen as obstruction increased. For all VF model pairs, the pressure drop across the stenosis was less than 0.1 kPa until 80% obstruction. At severe obstructions, the sub-stenosis pressures and the pressure drops across the stenosis were much greater than in the onset pressure experiments. During the course of the experiments, it was observed that pair B had a lower onset pressure and flow rate than pairs A and C. Consequently, to further explore the behavior of pair B, in addition to being tested at 21.9 LPM (120% onset flow), it was tested at 45.1 LPM, which was closer to 120% of onset flows for pairs A (64.1 LPM) and C (49.8 LPM). This data set, denoted B*, is included in Fig. 6. The results for B* coincide very well with those for pairs A and C. It is thus evidently possible that the responses of the three pairs would have exhibited less variation had they been tested at the same pressures and flow rates regardless of their individual onset pressures, but this would need to be confirmed through further investigation.

FIGURE 6.

For constant flow rate cases, (a) sub-stenosis (solid lines) and supra-stenosis (dashed lines) pressures, and (b) pressure drop across the stenosis vs. percent obstruction.

Constant pressure experiments

Supra-stenosis pressure and flow rate data at varying degrees of obstruction as the sub-stenosis pressure was maintained at each model’s respective 120% onset pressure (based on the onset at 36% obstruction) are shown in Fig. 7. Decreases in flow rate and supra-stenosis pressure began to be evident in pairs A and C between approximately 70% and 80% obstruction, but not until beyond approximately 90% obstruction in pair B. Further, at high degrees of obstruction for some cases, 120% of onset pressure became barely enough to sustain vibration (or, in a few cases, was insufficient to sustain vibration at the highest degrees of obstruction). This is in contrast to constant flow cases, in which all pairs vibrated for all obstructions tested.

FIGURE 7.

Supra-stenosis pressure (top) and flow rate (bottom) vs. percent obstruction for constant pressure cases.

Flow resistance

Flow resistance along a section of confined flow (e.g., across an orifice or along a length of tubing) is defined as the pressure drop along the section divided by the flow rate [6]:

$$R=\frac{P_1-P_2}{Q}$$ (2)

where R is the flow resistance, P₁ and P₂ are the upstream (sub-stenosis in this case) and downstream (supra-stenosis) pressures, respectively, and Q is the flow rate. Flow resistance calculations for both constant flow and constant pressure cases are shown in Fig. 8. The flow resistance began to significantly increase after approximately 90% for pair B and approximately 70% to 80% for pairs A and C. On average, the flow resistance increased by a factor of 67 from 36% to 99.8% obstruction; the increases ranged from a factor of 11 (pair B, constant pressure case) to a factor of 122 (pair A, constant flow case). This average increase is comparable to the approximately two-order of magnitude increase reported by Smith and Thomson [6] for two-dimensional computational simulations of SGS from 30% to 99% obstruction. Bodaghi et al. [7] found significant increases in flow resistance with SGS constrictions at 90% obstruction, and attributed this to the stenosis area at high constrictions being comparable to, or less than, the glottal area, thus resulting in the SGS resistance becoming comparable to the glottal resistance.

FIGURE 8.

Flow resistance, R, across the stenosis for constant pressure (top) and constant flow (bottom) cases.

Acoustic measurements

The acoustic spectra generated by all three VF model pairs, including cases with and without a vocal tract model, are shown in Fig. 9. Nonharmonic noise generally increased with increasing obstruction. In most instances both constant flow and constant pressure cases resulted in significantly more noise above 1000 Hz at 98.8% obstruction when compared to 36%. This was presumably caused by increased flow fluctuations through the stenosis and between the stenosis and the VF models. In addition, at levels of severe obstruction this jet may have impinged on the VFs, thereby altering aerodynamic loading and creating additional noise. Smith and Thomson [6] showed increases in jet velocity with increasing SGS obstruction, along with significant flow recirculation and vortex shedding between the stenosis and the glottis, beyond 95% constriction, but the simulation did not include turbulence modeling.

FIGURE 9.

Acoustic power spectral density (PSD) for pairs A (top), B (middle), and C (bottom) for four obstructions between 36% and 98.8%. Shown are constant pressure cases without the vocal tract model (left), and constant flow cases without (middle) and with (right) the vocal tract model.

Fundamental frequencies and amplitudes are shown in Fig. 10 (recalculated for improved frequency resolution using an increased block size beyond that which was used for Fig. 9). The frequencies of all three pairs increased slightly until approximately 80% obstruction and then decreased. Most experienced a decrease in frequency from the smallest to the greatest obstruction of over 10 Hz (exceptions being pairs A and B at constant pressure). The amplitudes of the fundamental frequencies similarly were fairly uniform until approximately 80% to 90% obstruction, beyond which significant decreases were observed.

FIGURE 10.

Fundamental frequency (top) and corresponding power spectral density amplitude (bottom) for constant pressure (solid lines) and constant flow (dashed lines) cases.

Vibration measurements

Still images from the high-speed videos of model motion are shown in Fig. 11 for all pairs for the constant flow experiments at 36% obstruction. Analysis of these and subsequent videos acquired at greater obstructions resulted in the plot of maximum glottal area vs. percent obstruction shown in Fig. 12. For constant flow cases, the amplitude increased as obstruction increased. For example, pair A reached a glottal area of 32.5 mm² at 99.2% obstruction, approximately double the glottal area at 36% obstruction (16.3 mm²). The glottal area vs. percent obstruction increased for pairs B and C as well, although to a lesser degree. This is in contrast to constant pressure cases, in which maximum glottal area decreased for all three pairs at severe obstructions (above 90% or 95%). For constant pressure cases, the flow rate decreased substantially with obstruction. For pair A, the flow rate decreased from 68.1 LPM at 36% to 19.5 LPM at 98.8% obstruction, while the supra-stenosis pressure decreased from 1.84 kPa at 36% to 0.78 kPa at 98.8% obstruction. Consequently, the amplitude of vibration decreased with percent obstruction for these cases.

FIGURE 11.

Images from high-speed video for pairs A (top), B (middle), and C (bottom) for constant flow at 36% obstruction.

FIGURE 12.

Maximum glottal area for all model pairs vs. percent obstruction for constant pressure (solid lines) and constant flow (dashed lines) cases, respectively.

CONCLUSIONS

The SGS-associated mechanisms that govern voice quality are poorly understood but include multiple possible factors. One of the main potential considerations in need of focused exploration is the degree to which airway obstructions may cause aerodynamic changes that affect voice quality. Results from recent computer simulations [6,7] have suggested that the aerodynamic effects alone may not account for major voice changes, but since these simulations have relied on approximations of important physical phenomena (turbulence, radiated acoustics, etc.), experiments are needed to validate and complement the simulations. The purpose of this study was to perform such experiments in which the effects of aerodynamic obstructions could be isolated from other possible SGS-related factors. To accomplish this, a benchtop device for simulating SGS with an adjustable obstruction was developed. Synthetic VF models were mounted at the downstream end of the device to enable experimental study of the aerodynamic, acoustic, and vibratory consequences of the obstruction on VF model phonatory output. The device allowed for the percent obstruction to be changed without removing or adjusting the attached synthetic VF models.

Experiments were conducted in which flow and pressure were measured, radiated sound data were recorded, and visual data from a high-speed camera were captured as the percent obstruction varied from 36% to 99.8%. The effects of subglottic stenosis were quantified using measures of onset pressure, flow rate, pressures upstream and downstream of the stenosis, flow resistance, acoustic spectra, and glottal area. The obstruction caused flow changes across the stenosis that affected these measures, but generally not until high degrees of obstruction. Onset sub-stenosis pressures were fairly consistent up to approximately 80% to 90% obstruction, beyond which the onset pressures sharply increased. The onset pressures at the highest obstruction (99.8%) were 34% to 56% higher than corresponding onset pressures at 36% obstruction for the three models. Flow rate (under the condition of constant pressure), flow resistance, and fundamental frequency all exhibited similar degrees of sensitivity to percent SGS obstruction as onset pressure (i.e., only becoming significant beyond 80% to 90% obstruction). High-frequency noise components became significant by 80% obstruction. Glottal area appeared to be even less sensitive, not being affected until approximately 90% obstruction. Altogether, these results suggest that Grade I and Grade II stenoses (0% to 50% obstruction and 51% to 70% obstruction, respectively) are not likely to introduce aerodynamic disturbances sufficient to alter these measures of phonatory output. Consequently, it is recommended that future studies include the study of other SGS-associated factors, such as changes in underlying tissue properties or in functional voice use patterns, that may influence phonation at these grades by more directly affecting VF vibration.

These findings are consistent with the results from the two-dimensional computational simulations of Smith and Thomson [6] and the three-dimensional simulations of Bodaghi et al.[7]. Computational simulations enable precise control over input parameters, whereas in experiments, important physical phenomena (e.g., turbulence, radiated sound, large amplitude vibration, large strains) are directly present and are not simulated. However, experiments pose challenges in that models are imperfect and contain inherent variabilities. The advantages and limitations of these respective approaches justify the pursuit of both methodologies. The agreement between the experimental results presented here and the computational simulation results of [6,7] support the overall conclusion that aerodynamic-induced changes associated with SGS affect phonatory aerodynamics, acoustics, and vibration, but not until very high degrees of obstruction are reached.

The results shown above consistently demonstrate relatively minor influence of SGS until high degrees of obstruction. A few potential sources of error and variability are here noted. Over the course of the experiment, it is estimated that each VF model pair experienced vibration for at least one hour. Fatigue-related factors may have thus been present, notwithstanding the tests having been performed in different orders (see “Experimental procedures” section) to reduce such potential effects; this was evident in the difference between the initial and subsequent onset pressures for pair B. Regarding model integrity, at the end of the experiments all models had a slight bowing in the middle which was not present when initially fabricated, although there did not appear to be any damage or separation of the layers. It is noted that buckling or other medial surface defects occurred in one VF model each from pairs B and C, such as can be seen in Fig. 11, and it is possible that compression of the models in the anterior-posterior direction may have occurred when mounting. Finally, some hysteresis of the SGS constriction is observed in supplemental videos 1, 2, and 3. These videos were acquired after the experiments, so the degree to which hysteresis may have been present in the experiments is unknown, and further investigation is needed. Importantly, however, notwithstanding these factors, the overall trends in effects of SGS on the measured output variables were consistent.

Along with these potential sources of error and variability, some additional limitations of the present study are worth noting. First, due to fabrication process limitations, synthetic VF models such as those used in this study exhibit inter-model material and geometric variability. One example of this is the layer material property data in Table 2. These contributed to variations in experimental results presented herein, including variations in onset pressures seen in Table 3. Efforts to improve VF model fabrication processes and improve model repeatability are underway. Second, even when the models are manufactured consistently, further improvements are needed to create models that are more representative of human VFs. Third, this study does not take into consideration potential changes to tissue properties (e.g., stiffness) associated with SGS. Fourth, vibration of the SGS itself was not quantified. However, given the higher stiffness of the silicone used for the SGS region compared to that of the VF model cover, in addition to the wall thickness of the SGS sleeve, it is expected that any SGS vibrations would have been minor. Finally, unsteady glottal jet flow separation, which is largely governed by subglottic pressure magnitude and VF medial surface shape, plays an important role in sound production during phonation. Consequently, follow-on studies spanning a wider range of subglottic pressures are recommended. Further, in the present study glottal jet flow separation and medial surface profiles were not measured or characterized. It is thus possible that synthetic models with greater degrees of intraglottal divergence angles and flow separation may yield different results. This possibility would need to be evaluated in future studies.

The adjustable SGS device constitutes a new experimental approach that could be incorporated in future studies. For example, the design of the device may enable future work to study asymmetric SGS obstructions, and variations of the design may enable exploration of additional, noncircumferential SGS shapes. Notably, the SGS device could be incorporated in excised larynx benchtop studies, thereby leveraging the advantages of excised larynges over synthetic models (such as has been preliminarily demonstrated by Murphey [26] and Smith [27]). Additionally, perceptual studies on radiated acoustics using the SGS device and synthetic models or excised larynges may enable exploration of output variables that may be more sensitive to SGS presence than the variables explored here. Complementary voice assessments of patients with SGS, including pre- and postoperative voice evaluations, laryngoscopic examinations, aerodynamic assessments, and auditory-perceptual studies, are recommended. Finally, the concept for the SGS device may serve as pattern for future studies involving other types of subglottic or supraglottic airway constrictions, such as a vocal tract model with segments that have adjustable cross sections.