INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The changes that occur in upper airway caliber during sleep are fundamental to understanding the pathogenesis of obstructive sleep apnea (OSA). Upper airway dimensional changes during sleep must be the result of changes in the position, motion, or configuration of the soft-tissue structures surrounding the airway. During sleep, there is a reduction in the activity of the upper airway dilator muscles (1), which results in a decrease in airway size and an increase in upper airway resistance (5). During sleep, airway narrowing may occur in an anteroposterior (AP) direction as a result of thickening or posterior motion of the soft palate and tongue (5). Alternatively, lateral collapse of the airway may occur as a result of thickening of the lateral pharyngeal walls (9, 10).

The tongue, soft palate, and lateral pharyngeal muscles have all been implicated in the pathogenesis of OSA. To date, however, the specific soft-tissue structures that are responsible for airway narrowing during sleep have not been identified. Some insight into this has been obtained from previous studies conducted during wakefulness (9). Studies using magnetic resonance imaging (MRI) of awake subjects have shown that the primary soft-tissue variable mediating changes in upper airway caliber in both normal and apneic subjects is the thickness of the lateral pharyngeal walls (9). Moreover, thickening of the lateral pharyngeal walls is the predominant anatomic structural change accounting for airway narrowing in apneic as compared with normal subjects during wakefulness (9). It has also been demonstrated that in normal subjects, upper airway dimensional changes are significantly greater in the lateral than in the anteroposterior direction in response to incremental levels of continuous positive airway pressure (CPAP) (11). These lateral changes in the airway result from progressive thinning of the lateral pharyngeal walls with CPAP, rather than from changes in the tongue and soft palate (11). Studies with electron-beam (cine) computed tomography (CT) have demonstrated that lateral upper airway changes are not only important under static conditions, but also under dynamic conditions (i.e., respiration) (12, 13). The respiratory-related changes in upper airway dimensions demonstrated in these studies were predominantly in the lateral dimension (12, 13). Thus, we hypothesize that the critical structures controlling the dimensions of the upper airway are the lateral pharyngeal walls. This leads us to postulate that airway narrowing during sleep will also be related to thickening of these lateral walls.

Studies during sleep are particularly important not only in demonstrating changes related to the state of sleep in upper airway caliber and in the surrounding soft-tissue structures, but also because it is known from other investigations that the site of upper airway narrowing during wakefulness does not necessarily correlate with the site of obstruction during sleep (14, 15). Previous studies have primarily focused on state- related airway changes in patients with OSA (15). Through rapid electron-beam CT, airway occlusion has been demonstrated in the retropalatal (RP), retroglossal (RG), and epiglottal regions during apneic episodes (16). Suto and colleagues (15) described sagittal changes in the airway during sleep, using ultrafast MRI in both normal and apneic subjects, and demonstrated multiple sites of airway narrowing or obstruction. These studies, however, evaluated only state-dependent airway changes, and did not address the fundamental issue of which soft-tissue structures mediate these changes during sleep.

Our study was designed to evaluate normal subjects during wakefulness and sleep with MRI in order to fully characterize state-dependent upper airway and soft-tissue changes in both axial and sagittal planes. Evaluation of the changes in upper airway caliber and in the surrounding soft-tissue structures during wakefulness and sleep in normal subjects will allow the beginning of better identification of the mechanisms underlying airway closure in patients with sleep-disordered breathing (SDB).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Study candidates were sought through advertisements within the University of Pennsylvania complex. The study was approved by the Institutional Review Board of the University of Pennsylvania. Exclusionary criteria for normal subjects included an age of less than 18 yr, pregnancy, a history of SDB, loud snoring, or a weight of more than 300 pounds (the table weight limit of the magnetic resonance scanner), in addition to the usual contraindications to magnetic resonance scanning, including claustrophobia or the presence of ferromagnetic clips or pacemakers and other implants. All subjects were evaluated first with a history and physical examination, which included measurement of weight, height, and neck collar size. Informed consent, outlining the objectives of the study and experimental tasks, was obtained from each subject. Subjects were admitted to the General Clinical Research Center of the University of Pennsylvania Medical Center for supervised sleep deprivation. Sleep deprivation helped ensure that the subjects would fall asleep during the magnetic resonance scan on the following day. The nursing staff made sure that subjects maintained wakefulness by constantly monitoring them throughout the night. MRI of the subjects during wakefulness and asleep was performed on the following day.

Polysomnography

Single-night polysomnography was performed in the sleep disorders center with a Nihon Kohden EEG 4418 A/K Polygraph (Nihon Kohden Corp., Tokyo, Japan) or a computerized polysomnography system (Sandman; Mellville Diagnostics, Ottawa, ON, Canada) according to a standard protocol that allowed 8 h of sleep. As described previously, standard parameters were monitored during polysomnography (9). The polysomnograms were scored by a registered technologist according to the criteria of Rechtschaffen and Kales (19), using 30-s epochs. Apneas were defined as lack of airflow for > 10 s, and hypopneas were defined as a 50% reduction in airflow for more than 10 s associated with a > 4% decrement in oxygen saturation and/or an arousal. The respiratory disturbance index (RDI) (defined as the mean number of apneas and hypopneas per hour of sleep) and oxyhemoglobin saturation nadirs were obtained. Subjects were included in the analysis if on polysomnography they had an RDI < 3 events/h and O₂ saturation nadirs > 90%.

Monitoring Sleep and Wakefulness During Magnetic Resonance Scanning

Sleep onset and maintenance when patients were in the magnetic resonance scanner were monitored behaviorally with a tactile stimulus (20). Subjects were outfitted with a latex balloon secured to the palm of the hand with a tight-fitting nylon glove. Lack of response to periodic inflation of the balloon signified sleep. The balloon was attached to a continuous positive airway pressure (CPAP) unit (Model 318; Puritan Bennett, Lenexa, KS) that delivered 20 cm of H₂O pressure to inflate the balloon when the investigator manually opened the valve. A strip-chart recorder (Ohmeda, Boulder, CO) sensed pressure changes from the inflated balloon if it was squeezed by the subject (Figure 1). The pressure signal was flat if the inflated balloon remained undisturbed. Subjects were instructed to squeeze the balloon three times after it was inflated. In order to minimize the startle from the noise of magnetic resonance image acquisition, subjects listened continuously to a prerecorded tape of magnetic resonance gradient noise delivered to the subject via pneumatic tubing attached to adapted foam earplugs. Wakefulness was confirmed with the behavioral stimulus during the sagittal localizing scan. The subject was then allowed at least 20 min to fall asleep. The balloon was inflated before, during, and after each scan to confirm sleep and wakefulness throughout the period of imaging.

View larger version (83K): [in this window] [in a new window]   Figure 1.   Overnight polysomnography with tactile balloon-pressure trace, showing comparison of different characteristic responses during wakefulness and Stage 2 sleep. In wakefulness, the subject squeezes the balloon in response to inflation. This response produces an oscillatory change in pressure in the balloon. During Stage 2 sleep, there is no response to the inflated balloon, and the pressure signal is flat.

MRI

Studies were done with a 1.5-Tesla MRI scanner (Signa Advantage; General Electric, Milwaukee, WI). Since flexion or extension of the neck influences upper airway size (21), subjects were placed in a neutral head position. This was defined by aligning the Frankfurt plane, a plane from the orbit of the eye to the superior portion of the tragus of the ear, perpendicularly to the table. Foam padding was used to insulate the subject's head from the surrounding neck coil. Subjects were instructed to breathe through their nose with the mouth closed, and to refrain from swallowing during scanning. Scans were repeated if swallowing occurred.

A receive-only signal volume neck coil (General Electric Medical Systems, Milwaukee, WI) was used to maximize signal-to-noise ratio. All studies were initiated with a sagittal localizing scan during wakefulness to confirm neutral airway position, using the parameters of spin echo repetition time (TR) = 400 ms; echo time (TE) = 16 ms; 256 × 128 matrix; 1 number of signal averages (NEX); flip angle = 90 degrees; field of view (FOV) = 24 cm; thickness = 5.0 mm; and skip = 1.5 mm. Contiguous axial T1-weighted spin echo images (TR = 400 ms, TE = 16 ms, ¹/₂ NEX, flip angle = 90 degrees, FOV = 24 cm, thickness = 5.0 mm, and skip = 0.0 mm) were obtained from above the hard palate to the larynx during wakefulness and sleep. In nine subjects, sagittal data were also obtained during wakefulness and sleep, using the magnetic resonance settings of TR = 500 ms; TE = 23 ms; 0.5 NEX; flip angle = 90 degrees; FOV = 24 cm; thickness = 3.0 mm, and and skip = 0.0 mm. Scan acquisition times for both the axial and sagittal images, utilizing this conventional T1-weighted spin echo magnetic resonance pulse sequence, were approximately 3.5 min.

Anatomic Definitions and Image Processing

Anatomic definitions based on axial magnetic resonance images are shown in Figure 2. The caudal tip of the soft palate divides the RP from the RG region as shown in the midsagittal image in Figure 3. Airway measurements obtained from axial images included: (1) cross-sectional area; (2) AP distance through the centroid (center of the airway); and (3) lateral distance through the centroid. Soft-tissue measurements from axial data included: (1) intermandibular width, defined as the distance between the right and left rami of the mandible; (2) distance between the RP lateral-wall fat pads; and (3) lateral pharyngeal wall thickness, defined as the soft-tissue width between each lateral-wall fat pad and the airway, measured along a horizontal line drawn through the centroid on each side of the airway. Airway volume was computed from three-dimensional views generated from contiguous axial images.

View larger version (107K): [in this window] [in a new window]   Figure 2.   A representative T1-weighted spin echo magnetic resonance axial image in the RP region of a subject during wakefulness. The important anatomic structures are indicated. Note that fat is white on magnetic resonance imaging (high signal).

View larger version (125K): [in this window] [in a new window]   Figure 3.   A representative T1-weighted spin echo magnetic resonance midsagittal image of a subject during wakefulness. Three anatomic regions are identified: RP, RG, and epiglottal (EPG). For definition of regions, see text.

Sagittal soft-tissue structures. Figure 4 displays the upper airway and soft-tissue measurements obtained from midsagittal images. The midsagittal measurements of the tongue included (see Figure 4 for all line definitions): (1) the cross-sectional area; (2) the oblique width measured from the mandible through the tongue centroid to the posterior edge of the tongue along line B; (3) the horizontal distance measured along line C between the tongue centroid (center) and the anterior surface of the second vertebral body (C2); and (4) the horizontal distance measured along line C between the posterior edge of the tongue and the anterior surface of C2. The midsagittal measurements of the soft palate included (Figure 4): (1) the cross-sectional area; (2) the curved length measured from the hard palate to the caudal tip of the soft palate; (3) the vertical height measured from the base of the hard palate to the caudal tip of the soft palate; (4) the horizontal width measured through the soft palate centroid along line E; (5) the oblique width of the soft palate measured along line A; (6) the horizontal distance measured along line E between the soft palate centroid and the anterior surface of C2; and (7) the horizontal distance measured along line E between the posterior edge of the soft palate and the anterior surface of C2.

View larger version (157K): [in this window] [in a new window]   Figure 4.   Representative T1-weighted magnetic resonance midsagittal image of a subject during wakefulness and sleep (top images). During sleep, airway caliber is reduced in the RP region and the soft palate moves posteriorly. The bottom image demonstrates measurements of the dimensions of the soft tissues along lines A to E. The large dot represents the tongue centroid, and the small dot represents the soft palate centroid. Line A connects the mandible to line D through the centroid of the soft palate. Line B connects the mandible to line D through the centroid of the tongue. Line C connects the centroid of the tongue and intersects line D perpendicularly. Line D is a posterior landmark, traversing the anterior portion of C2. Line E connects the centroid of the soft palate and intersects line D perpendicularly.

Image processing. Airway and soft-tissue geometric analysis was done with customized three-dimensional image-processing software (VIDA: Volumetric Image Display and Analysis). These techniques have been previously validated and described in detail (9, 11) and are summarized as follows:

1. Airway geometry. The three-dimensional upper airway is first identified from a dataset of contiguous magnetic resonance image slices, using a "three-dimensional fill" computer algorithm. T1-weighted magnetic resonance images have a large signal gradient at the boundaries of the airway surface and the surrounding soft-tissue structures. This allows the algorithm to grow out from a starting airway seed point and "fill" the entire airway in three dimensions. The algorithm has been previously described and validated (22). Next, an algorithm computes a three-dimensional centerline of the upper airway (Figure 5), which is a curvilinear line traversing the center of the long axis of the airway. From this curvilinear centerline, a series of oblique image slices are generated which are locally perpendicular to the centerline. This permits "straightening" of the airway and allows airway and soft-tissue measurements in image slices locally perpendicular to the long axis of the airway (22). Conventional axial MRI slices tend to overestimate these measurements, since they do not account for airway curvature relative to the magnetic resonance scanning direction.
2. Soft-tissue identification. Individual soft-tissue structures and boundaries (defined previously) are identified through a semiautomated "full width, half maximum" edge-detection algorithm previously described and validated (9). In the edge-detection algorithm, a polygon is drawn manually by the operator on the computer screen and approximates the edge of the airway or soft-tissue structure. The program then adjusts each point around the circumference of this original polygon (using a full-width, half-maximum criterion) to determine an objective edge of the structure at that location, and a new polygon is generated. It is then possible to quantify the area of the final, resultant polygon.
3. Airway three-dimensional volumetric analysis. The three-dimensional volumetric shape of the upper airway is identified through the three-dimensional fill algorithm previously described in the section on airway geometry. This algorithm defines the total extent of the upper airway within the magnetic resonance image slices. This volume is then subdivided into regional volumes, using anatomic landmarks. The RP volume is defined as the airway region from the hard palate to the caudal tip of the soft palate. The RG volume is defined as the airway region from the caudal tip of the soft palate to the tip of the epiglottis. These regional volumes were compared during wakefulness and sleep.

View larger version (156K): [in this window] [in a new window]   Figure 5.   A three-dimensional reconstruction of the head, neck, and airway of a subject from sagittal MRI data. The airway, extracted from a seed point in three dimensions, is shown in white. The black dotted line represents the calculated centerline of the airway from which axial images perpendicular to this centerline are obtained to allow measurement of dimensions of the airway and soft-tissue structures.

Data Analysis

Contiguous axial images obtained during sleep were analyzed with the image processing techniques described previously. The airway geometry programs produce graphs of upper airway area and lateral and AP dimensions throughout the airway. From these data, the level of the smallest airway area during sleep was noted in three distinct anatomic regions; RP, RG, and epiglottal. These images were then matched with the corresponding images during wakefulness, using anatomic landmarks. The airway area and AP and lateral distances were then measured in these waking images.

The soft-tissue analysis was done with the edge-detection algorithm; the airway centroid was computed and a horizontal line was drawn through it for each axial oblique image during sleep and the corresponding image during wakefulness. In the RP region, distances from the right to the left ramus of the mandible, the mandible to the airway wall, fat pad to fat pad, and fat pad to airway wall (lateral-wall thickness) were measured along this horizontal line (Figure 2). In the RG region, mandible-to-mandible and mandible-to-airway wall distances were recorded.

Midsagittal images during wakefulness and sleep were compared, and changes in soft palate and tongue dimensions were computed with the edge-detection algorithm. Soft palate area, width, height, and length were measured during wakefulness and sleep. In order to assess the changes in AP airway dimension in the RP region, the distances from the soft palate centroid and from the posterior edge of the soft palate to the anterior surface of the C2 vertebral body were measured (Figure 4). The C2 vertebral body was considered a fixed landmark. In the RG region, tongue area and oblique distance were measured during wakefulness and sleep. In addition, AP airway dimensional changes were computed by measuring the distances from the tongue centroid and posterior edge of the tongue to a line drawn perpendicularly along the anterior surface of C2 (Figure 4).

Upper airway regional volumes were calculated as described in the section on three-dimensional volumetric analysis of the airway. For images made during both sleep and wakefulness, the position of the tip of the soft palate was used to define the transition from the RP to the RG region. The hard palate defined the most superior border of the RP region, whereas the tip of the epiglottis defined the most inferior border of the RG region. The distance used to calculate airway volume along the centerline from the reference position (tip of the soft palate) was the same for the images made during sleep and wakefulness. Therefore, upper airway volumetric changes from sleep to wakefulness could not be attributed to a difference in airway length.

Statistical Analysis

Two approaches were taken to analysis of the study data. The first approach focused on intrasubject differences between wakefulness and sleep in upper airway structures. For axial measures, mixed-model analyses of variance (ANOVA) were first used to test the hypothesis that state-induced mean changes (minimum area during sleep versus anatomically matched images during wakefulness) differed by anatomic level. When the interaction of anatomic level and sleep state was significant, paired t tests were used to assess the significance of sleep state separately for each anatomic level. Differences in state- dependent sagittal measures were assessed with paired t tests. Additionally, plots of the distribution of differences were examined. When skewed distributions or outliers were present, significance tests were repeated, using the nonparametric Wilcoxon's signed-rank test, and "typical" values were characterized by median rather than by mean values.

The second approach to data analysis utilized multiple linear regression to identify state-dependent changes in axial and sagittal measurements independently associated with changes in RP airway area. First, separate forward stepwise regression analyses were performed for candidate axial and sagittal variables, using an alpha level of 0.10. Then, significant variables from these two domain-specific models were combined and variables not significant at the p  0.05 level were removed one at a time.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subject Demographics and Polysomnographic Measurements

We performed upper airway imaging during wakefulness and sleep in 16 subjects. One subject was excluded from analysis because of motion artifact. The age of the 15 normal subjects (14 men and one woman) was 25.1 ± 4.1 yr (mean ± SD). The mean body mass index (BMI) was within the normal range, at 24.2 ± 2.5 kg/m². SDB was excluded by overnight polysomnography. The mean RDI was 0.3 ± 0.4/h and the mean oxyhemoglobin desaturation nadir was 93.2 ± 1.9%. Quiet snoring was noted in seven of the subjects. Snoring did not result in arousals in these subjects. The mean neck collar size was 15.0 ± 0.8 in.

Monitoring Sleep During MRI

As described in the METHODS section, sleep onset and maintenance in the magnetic resonance scanner were monitored with a pneumatically inflated latex balloon placed in a glove covering the subject's hand. The accuracy of this behavioral stimulus in identifying sleep was validated through a direct comparison with standard polysomnography in 10 other subjects (nine with sleep apnea and one with a seizure disorder) (20). The tactile stimulus and polysomnography data were collected simultaneously (Figure 1), allowing a direct comparison of the results of electroencephalographic sleep staging (19) and responses to balloon inflation. Table 1 demonstrates the percentage of sleep stages correctly identified by the tactile stimulus as sleep. Overall, this technique was highly effective (> 95%) in documenting that subjects were sleeping during stage 2, delta, and rapid eye movement (REM) sleep, but was only 70% effective in documenting stage 1 sleep.

Airway Volume During Wakefulness and Sleep

The effect of sleep on airway volume was evaluated in the RP and RG regions. Figure 6 displays a three-dimensional volumetric reconstruction of a subject's upper airway in both the lateral and coronal planes during wakefulness and sleep. Compared with the images produced during wakefulness, the images during sleep show substantial narrowing in the RP region. The data, when analyzed across all 15 subjects, confirm airway narrowing during sleep in the RP region, with a mean airway volume of 2,256.9 ± 1,233.1 mm³ during wakefulness as compared with a mean volume of 1,615.3 ± 719.5 mm³ during sleep (Figure 7). Using a mixed model analysis of variance, we observed significant regional differences in state-dependent changes in airway volume. The mean difference (asleep  awake) in airway volume in the RP region was 641.6 ± 760.2 mm³ (t = 3.27, df = 14, p = 0.01). The mean percentage change ([asleep  awake]/awake) in airway volume was also significant in the RP region, with a mean of 19 ± 30% (t = 2.44, df = 14, p = 0.03). However, the changes in airway volume in the RG region were not significant in terms of either absolute or percentage differences. The mean arithmetic difference in RG airway volume was 16.4 ± 729.4 mm³ (p = 0.93), whereas the mean percentage difference was 4 ± 34%, p = 0.63). These data indicate that in these normal subjects, there are regional state-dependent volumetric differences in airway narrowing. Airway volume is significantly reduced in the RP region but not in the RG region.

View larger version (121K): [in this window] [in a new window]   Figure 6.   A three-dimensional volumetric reconstruction of a subject's airway in both lateral and coronal planes during wakefulness and sleep. Narrowing of the airway during sleep is evident in the RP region in both planes. Conversely, there is little change in the size of the airway in the RG region.

View larger version (14K): [in this window] [in a new window]   Figure 7.   A bar-graph representation of changes in RP and RG airway volume during sleep (mean ± SD) in 15 subjects. State- dependent changes in airway volume in the RP region were significant (p = 0.01). State-dependent changes in airway volume in the RG region were not significant (p = 0.93).

State-dependent Changes in Sagittal Soft Palate and Tongue Dimensions

Using sagittal MRI data, images obtained during sleep were matched anatomically (midline position of C2) with images during wakefulness. A representative example of the state- dependent sagittal changes in soft-tissue measurements is shown in Figure 4. In this subject the retropalatal airway narrows during sleep and the soft palate moves posteriorly. Figure 8 confirms these findings in nine normal subjects. We were able to obtain sagittal images during sleep in only nine of the 15 subjects. The mean state-dependent change in the distance from the posterior edge of the soft palate to the anterior edge of C2 was 1.99 ± 2.15 mm, which was significant (p = 0.02). Both the mean difference from wakefulness to sleep in the soft palate oblique distance (1.24 ± 1.64 mm) and in the distance from the soft palate centroid to C2 (1.76 ± 2.31 mm) approached statistical significance (p = 0.053 and p = 0.052, respectively). These data indicate that there is posterior movement and thickening of the soft palate with sleep. Changes in soft palate area, length, height, and width were not significantly different from wakefulness to sleep.

View larger version (21K): [in this window] [in a new window]   Figure 8.   A bar-graph representation of state-dependent sagittal soft-palate dimensions (mean ± SD) in nine subjects. Significant differences are shown in the soft palate posterior edge-to-C2 distance (*) (p = 0.02), centroid of the soft palate to C2 (+) (p = 0.052), and soft palate oblique distance (+) (p = 0.053). The other soft palate variables were not significantly different in wakefulness versus sleep.

Figure 9 displays sagittal dimensions of the tongue during both wakefulness and sleep in the same nine subjects. Univariate analysis showed no significant difference in tongue area and oblique distance from wakefulness to sleep. In addition, the distances from the tongue centroid or posterior tongue edge to a perpendicular line (Figure 4, line D) drawn from the anterior C2 vertebral body were also not significantly different when comparing sagittal images in wakefulness and sleep. These data indicate that changes in the soft palate may be more important than change in the tongue in mediating upper airway narrowing during sleep in normal individuals.

View larger version (18K): [in this window] [in a new window]   Figure 9.   A bar-graph representation of state-dependent sagittal tongue dimensions (mean ± SD) in nine subjects. There were no significant differences in any of these tongue variables in wakefulness versus sleep.

Axial Airway and Soft Tissue Measurements during Sleep as Compared with Wakefulness

Differences between anatomic levels. State-dependent changes in airway dimensions were evaluated separately, with investigation of the minimal airway area for each anatomic level. This was necessary after observing significant interaction in the state-dependent change in airway dimension by anatomic level with a mixed-model ANOVA, combining data from all anatomic levels. Specifically, differences in airway area at the level of the minimum airway area during sleep as compared with the corresponding axial slice during wakefulness were significantly dependent on anatomic level (F = 10.42, df = 2.26, p = 0.001). This was also true for differences in airway AP (F = 9.02, df = 2.26, p = 0.001) and lateral (F = 5.02, df = 2.26, p = 0.01) dimensions. Therefore, it was necessary to compare changes in airway area and AP and lateral dimensions during sleep with those during wakefulness separately in the RP, RG, and epiglottic regions.

Axial airway measurements. Figure 10 shows axial magnetic resonance images in the RP region from a single subject, demonstrating a substantial difference in airway cross-sectional area from sleep to wakefulness. These changes were due to both AP and lateral narrowing of the airway and thickening of the lateral pharyngeal walls. When all subjects were analyzed, the mean difference (awake  asleep) in minimum cross-sectional area was significant in the RP (mean = 28.8 ± 28.6 mm², p = 0.002) and RG (mean = 17.2 ± 28.5 mm², p = 0.04) regions (Figure 11). The mean percentage difference (arithmetic difference/asleep baseline) in airway area was also significant in the RP (mean = 228% ± 260%, p = 0.004) and RG (mean = 22% ± 33%, p = 0.02) regions. The mean arithmetic difference in airway AP (mean = 2.8 ± 3.2 mm, p = 0.01) and lateral (mean = 3.1 ± 4.3 mm, p = 0.02) dimensions was significant in the RP region (Figure 12). Neither mean absolute nor percentage differences in airway diameter in AP or lateral projections were significant in the RG and epiglottic regions (Figure 12). These data show that minimal airway cross-sectional area is significantly reduced in sleep at both the RP and RG regions, although the change in the RP region is much larger than in the RG region. In the RP region, both AP and lateral changes in airway dimensions contribute to state- dependent alteration in airway area.

View larger version (72K): [in this window] [in a new window]   Figure 10.   A comparison of T1-weighted RP axial magnetic resonance images of a representative subject during wakefulness and sleep. The airway during sleep is reduced in the AP and lateral dimensions. Thickening of the lateral walls is demonstrated during sleep.

View larger version (16K): [in this window] [in a new window]   Figure 11.   A bar-graph representation of the minimal cross-sectional airway areas (mean ± SD) in the RP, RG, and epiglottal (EPG) regions during sleep in 15 subjects. These values are compared with airway areas during wakefulness obtained from anatomically matched axial images in each region. There are significant changes in airway area in both the RP (p = 0.002) and RG (p = 0.04) regions, but not in the EPG region (p = 0.89).

View larger version (24K): [in this window] [in a new window]   Figure 12.   A bar-graph representation of airway AP and lateral dimensions (mean ± SD) measured through the airway center (centroid) in the RP, RG, and epiglottal (EPG) regions in 15 subjects. These values are compared with airway dimensions during wakefulness obtained from anatomically matched axial images in each region. The data show significant AP (p = 0.01) and lateral (p = 0.02) airway dimensional differences in the RP region during sleep. There are no significant changes in airway dimensions in the RG and EPG regions.

Axial soft-tissue measurements. Lateral pharyngeal wall thickness, measured from the lateral edge of the airway to the fat pads in the RP region, exhibited a significant increase during sleep as compared with wakefulness (mean: 2.6 ± 3.7 mm, p = 0.02). The mean percentage increase of 7 ± 13% in wall thickness was also significant during sleep as compared with wakefulness (p = 0.04). These differences in wall thickness were not due to state-dependent changes in the distances between the fat pads (p = 0.67) or mandibles (p = 0.31) in the RP region (Figure 13). In the RG region there was no significant difference in lateral pharyngeal wall thickness (mean 0.4 ± 2.2 mm, p = 0.52) during sleep as compared with wakefulness. These data indicate that there are regional state-dependent differences in the thickness of the lateral pharyngeal walls.

View larger version (16K): [in this window] [in a new window]   Figure 13.   A bar-graph representation of measurements of axial soft-tissue dimensions (mean ± SD) along a horizontal line through the airway centroid in the retropalatal region in 15 subjects. The lateral pharyngeal wall thickness (bilateral, cumulative distance from the lateral edges of the airway to the adjacent fat pads) was significantly larger (p = 0.02) in the axial image of the minimal cross-sectional area during sleep than in the anatomically matched axial image during wakefulness. The distances between the fat pads and the mandibles, measured along the same horizontal axis, were not significantly different in sleep versus wakefulness.

Regression Analysis of State-dependent Axial and Sagittal Soft-Tissue Dimensions Versus Airway Caliber

We performed a regression analysis to determine the soft tissue and craniofacial changes associated with reduction in airway caliber in the RP region during sleep. Table 1 shows Pearson's correlation coefficients for all of the candidate axial and sagittal measurement associated with change in RP area with sleep among the nine subjects for whom both axial and sagittal data were collected. State-dependent changes in RP airway area were significantly correlated with changes in lateral-wall thickness (p = 0.001), changes in the distance from the soft palate centroid to the C2 vertebra (p = 0.013), and tongue oblique distance (p = 0.003) (i.e., as tongue oblique distance and lateral wall thickness increased during sleep and as distance from the soft palate centroid to C2 decreased during sleep, RP area decreased). Next, separate forward, stepwise regressions were performed, using all domain-specific (axial and sagittal) measurements, in order to obtain a parsimonious set of variables affecting state-dependent changes in RP airway area. For the axial model, lateral pharyngeal wall thickness entered the model at Step 1, explaining 79% of the change in airway area in the RP region (p = 0.001). Fat pad and intermandibular distances measured on axial images did not contribute significantly in this model.

For the sagittal model, tongue oblique distance entered the model at Step 1, explaining 73% of the change in the RP airway area (p = 0.003). At Step 2, the soft palate centroid-to-C2 distance explained an additional 12% (p = 0.08) of the change in RP airway area, and at Step 3, soft palate width entered the model, explaining an additional 10% of the variance (p = 0.02).

The combined axial and sagittal model explained 97% of the state-dependent change in airway area in the RP region (R² = 0.97, p < 0.0001). In this axial and sagittal model, state-dependent changes in the soft palate centroid-to-C2 distance (p = 0.02) and in soft palate width (p = 0.02) retained their significance. In contrast, the partial correlation coefficients for oblique tongue distance and lateral pharyngeal wall thickness were no longer significant (both p = 0.17), indicating that one or the other of these variables, but not necessarily both, contributed to explaining the variance in RP airway area during sleep. Further investigation revealed that state-dependent changes in tongue oblique distance and lateral pharyngeal wall thickness were highly correlated (r = 0.87, p = 0.003), indicating a colinearity between these structures. When either the lateral walls or tongue oblique distance were removed from the model, the other structure was significantly associated with state-dependent airway change in the RP region (p = 0.016 for tongue oblique distance and p = 0.019 for the lateral pharyngeal walls). The squared partial coefficients were 0.81 for the soft palate centroid-to-C2 distance, 0.68 for the soft palate width, and 0.71 for both the tongue oblique distance or lateral pharyngeal wall thickness (depending on which variable was included in the model), indicating that all of these variables made a unique contribution to airway narrowing in the RP region during sleep. The regression analysis indicated that changes in lateral wall thickness, tongue oblique distance, and soft palate position/width contribute significantly to state-dependent RP airway narrowing.

State-dependent Difference in Longitudinal Position of Minimal Airway Area in the RP and RG Regions along the Airway Centerline

During wakefulness, the smallest cross-sectional area of the airway from the hard palate to below the epiglottis was located in the RP region in 12 of 15 normal subjects (three subjects had the smallest area in the RG region). During sleep, the minimum airway area was located in the RP region in 13 of 15 subjects (two subjects had the smallest area in the RG region). We analyzed the absolute difference in position of the smallest RP and RG airway areas from wakefulness to sleep along the airway centerline. In the RP region, the mean change in position of the minimum airway area was 5.4 ± 3.7 mm, with a range of 0.4 to 17.3 mm. In the RG region, this mean difference was 5.0 ± 3.5 mm, with a range of 0.6 to 13.3 mm. In 11 of the 15 subjects, the minimum airway area in the RP region moved caudally during sleep with respect to the minimum airway area during wakefulness. The minimum airway area in the RG region moved caudally during sleep in seven of the 15 subjects. These data indicate that in most subjects, the smallest airway area occurs in the RP region in both wakefulness and during sleep. The longitudinal position of the minimum airway area moved caudally in the majority of subjects during sleep.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

MRI of 15 normal subjects following a period of sleep deprivation demonstrated significant changes in both the upper airway and the surrounding soft tissues during sleep. The volume of the RP airway was reduced by 19% during sleep. In contrast, airway volume in the RG region was not significantly reduced during sleep. This suggests that factors affecting airway caliber during sleep in the RP and RG regions may act independently. Cross-sectional airway area in the RP region was also significantly smaller during sleep than during wakefulness. The changes in RP airway area were due to a reduction in both AP and lateral airway dimensions. In the RG region, the only significant finding was a reduction in cross-sectional airway area during sleep; there were no significant changes in lateral or AP dimensions. The airway in the region of the epiglottis did not demonstrate sleep state-dependent changes in cross-sectional area or airway dimension.

Our study also demonstrated important changes in the upper airway soft-tissue structures with sleep. The reduction in the lateral airway diameter during sleep was mediated through thickening of the lateral pharyngeal walls. Our data show a significant increase in the thickness of the lateral pharyngeal walls in the RP region during sleep. Thickening of the lateral pharyngeal walls was not due to a change in position of the fat pads. Sagittal analysis showed that changes in soft palate position and width were also associated with airway narrowing during sleep. Measurements of tongue area and position did not significantly change during sleep. These data suggest that the soft palate and lateral pharyngeal walls are important in reducing airway caliber during sleep in normal subjects. Moreover, in normal subjects, state-dependent changes in airway and surrounding soft-tissue structures are much larger in the RP than in the RG region during sleep. The data collected in our study suggest that this regional difference is primarily related to changes in soft palate position and/or to lateral pharyngeal wall thickening during sleep.

Background

Several studies using MRI and CT have investigated upper airway changes during both wakefulness and sleep in normal and apneic subjects (15, 17, 18, 23). Stein and colleagues (16), using cine CT, demonstrated that the airway was narrowest in the RP region in awake apneic subjects. Airway obstruction occurred at multiple anatomic levels in the RP and RG regions during apneic events within sleep. Horner and colleagues (18) used CT to demonstrate that the minimal cross-sectional airway area was significantly reduced in apneic as compared with normal subjects during wakefulness. Airway obstruction during sleep in the apneic patients occurred in the RP region (18). These findings are analogous to our findings.

Suto and colleagues (15), using ultrafast MRI, evaluated sagittal images of the airway in both normal subjects and patients with OSA during wakefulness and "sleep" induced by intravenous hydroxyzine hydrochloride. They were able to identify nine sites of airway narrowing in seven awake apneic subjects, with all apneic subjects showing narrowing in the RP region. During sleep, 21 sites of obstruction were identified in 13 sleeping apneic subjects. All 13 of these subjects had obstruction in the RP region; six sites of obstruction were identified in the oropharynx (RG region) and two sites in the hypopharynx. The sites of airway narrowing during wakefulness correlated with airway obstruction during sleep in only 31% of the apneic subjects.

However, these CT and MRI studies of the upper airway have a number of limitations. The studies done with MRI described only sites of airway obstruction, and did not measure, as we did, cross-sectional airway area, airway volume, or soft-tissue dimensions (15). Quantitative analysis of soft-tissue changes during sleep was not done in the cine CT studies (16, 18). Sleep onset was not monitored in the CT environment. In the MRI environment, sleep was inferred from a recorded apnea or a lack of response to verbal calls (15). We therefore believe that our study represents a significant advance over previous studies, since we objectively monitored sleep and used computer software analysis of reconstructed magnetic resonance images to make precise measurements of airway and soft-tissue dimensions. Such computer software analysis has allowed us to objectively quantify state-dependent changes in upper airway and soft-tissue dimensions in order to accurately examine the mechanisms of airway narrowing during sleep.

Study Protocol and Limitations

Achieving and monitoring sleep during MRI created several problems in our experimental design. A night of supervised sleep deprivation was used to facilitate sleep during imaging. Sleep deprivation is known to alter sleep-stage architecture during recovery (24), which could potentially influence airway dimensions by favoring delta and/or REM sleep periods. Other investigators have relied on sedative medications to induce sleep (15, 23) during magnetic resonance scanning. We believe that such an experimental approach has limitations, since the use of sedatives prior to sleep has been shown to worsen SDB (25). Sedatives could also exaggerate airway dimensional changes during sleep. Instead, we utilized sleep deprivation and attenuation of magnetic resonance scanner noise to allow our subjects to fall asleep in the scanner and enable us to avoid the adverse effects of pharmacologic agents on upper airway structure and function. It should be noted, however, that sleep deprivation itself may affect upper airway caliber.

The second problem we faced was how to objectively document sleep during imaging in the magnetic resonance environment. This was accomplished through a previously validated behavioral tactile stimulus technique, although this approach did not enable us to identify specific sleep stages or episodes of brief arousals during imaging. In addition, this behavioral technique was validated in apneic and not in normal subjects. Nonetheless, the technique allowed us to objectively determine whether a subject was sleeping before, during, and after imaging. Other studies have correlated electroencephalographic sleep stage with responses to auditory stimuli, and have also shown that such behavioral responses accurately predict sleep, primarily Stages 3 and 4 and REM sleep (26, 27). Recently, investigators (28, 29) have developed a system to monitor electroencephalographic activity and accurately stage sleep in magnetic resonance scanners. Such a system, however, cannot monitor electroencephalographic activity during actual image acquisition, because of radio-frequency interference. Thus, future studies of sleep-related changes in airway function may utilize a combination of electroencephalographic signals (to determine sleep stages before and after scanning) and a behavioral technique to monitor sleep during image acquisition.

Another important limitation of our study was that the axial and sagittal MRI scans were each performed over periods of several minutes during both wakefulness and sleep. The images were averaged over many respiratory cycles. Thus, our state-dependent upper airway measurements represent "mean" values for the size of the upper airway and surrounding soft-tissue structures. We have previously shown with cine CT that normal respiration results in dimensional changes in the upper airway and surrounding soft tissues (12, 13). Because of the slow acquisition of our magnetic resonance images, upper airway area may be more weighted to airway size in expiration than in inspiration (since expiration is longer than inspiration) and thus may not truly represent a "mean" area. In addition, our edge-detection algorithm, which was validated with a static phantom, has not been validated with a moving structure. Nonetheless, the analysis strategy used in our study was identical during wakefulness and sleep (state-dependent comparisons), which would help to reduce any systematic error resulting from the edge-detection algorithm. It is also known that respiration is altered during sleep onset, with a reduction in tidal volume (VT) and an irregular respiratory rate, particularly in Stages 1 and 2 sleep (30). Because our MR images were acquired from averaged data obtained over several minutes, the effects of state-dependent respiratory variations on airway and soft-tissue dimensions should be small. Moreover, the magnitude of the respiratory changes in normal subjects during sleep is small (primarily a reduction in VT), and should not significantly affect upper airway dimensions as compared with those in the corresponding baseline images (30). However, our data do not allow us to conclude whether the upper airway changes observed during sleep in our subjects occurred during inspiration, expiration, or both. Future studies are necessary to determine state-dependent dynamic changes in the upper airway and surrounding soft-tissue structures.

A quantitative comparison of upper airway dimensions during wakefulness and sleep would be inaccurate if subjects moved their head positions or opened their mouths during the protocol. Movement artifact was mitigated in our study through the use of foam padding to secure the subject's head in the coil with the neck in a neutral, midline position. Changes in airway caliber were not due to opening of the mouth during sleep, since this was not objectively identified in the images (if the subject's mouth had opened, the soft palate would have risen off the tongue). Moreover, to reduce error, we used fixed landmarks (teeth, parotid glands, mandible, spinal cord, vertebral bodies, and fat pads to compare images made during sleep with anatomically matched images made during wakefulness.

Another potential limitation of our study was our strategy for data analysis. By comparing the narrowest axial image during sleep with a matched image during wakefulness, we were potentially biasing our analysis so that the narrowest airway had to occur during sleep. Nonetheless, we believe that this strategy was appropriate, since the minimal airway during sleep is physiologically the most important, especially in patients with SDB. In addition, to ensure that we were not biasing the results, we analyzed the data in our comparison of the minimal airway during wakefulness with a matched image made during sleep. In this analysis, airway area in the RP region was still significantly smaller during sleep (33.3 ± 25.0 mm²) than during wakefulness (47.3 ± 39.6 mm²) (p = 0.03).

Anatomic Structures Mediating Airway Caliber during Wakefulness

The muscles of the upper airway mediate several physiologic functions, including respiration, speech, and coordination of swallowing. The oropharynx and hypopharynx are surrounded by a complex arrangement of muscles that work in concert to maintain airway caliber (31). These muscles make up the tongue, soft palate, and lateral pharyngeal walls, and the position and size of these structures mediates upper airway caliber. Previous studies, using CT and cephalometrics, have demonstrated enlargement of the soft palate and increased tongue volume in patients with sleep apnea (32). Our prior studies have confirmed these findings and have demonstrated that the lateral pharyngeal walls are also important in mediating airway caliber in both normal and apneic subjects (9). In addition, we have demonstrated in normal subjects that the effect of CPAP is largely mediated through the lateral pharyngeal walls (11). However, changes in the tongue, soft palate, or lateral pharyngeal walls during sleep have not been fully elucidated in either normal subjects or those with SDB.

Review of State-dependent Airway and Soft-Tissue Changes

Changes in airway caliber. A number of studies have evaluated the effect of sleep on upper airway resistance in normal subjects and have found increases in upper airway resistance during non-REM sleep (6, 35). The state-dependent increases in airway resistance demonstrated in these studies suggest that airway narrowing occurs during sleep. These studies, however, did not provide definitive data on the location of the airway narrowing. Studies done with fiberoptic upper airway endoscopy have provided such information (38). Direct visualization of the upper airway during sleep, through fiberoptic nasopharyngoscopy, has been used to demonstrate state-dependent changes in airway caliber (23, 38). In general, these studies suggest that in apneic individuals, airway narrowing occurs at multiple anatomic locations in the RP and RG regions. Pharyngeal narrowing in normal subjects during sleep has been less well studied than in apneic subjects. Badr and coworkers (41) used fiberoptic nasopharyngoscopy to study six normal subjects and six subjects with central sleep apnea during non-REM sleep. In the normal subjects, airway narrowing occurred reproducibly in the RP region during induced central apneas. Although our subjects did not have induced central apneas, Badr and coworkers' results (41) support the finding in our study of airway narrowing primarily in the RP region during sleep in normal subjects.

The studies cited previously (23, 38) in which fiberoptic nasopharyngoscopy revealed airway closure during sleep support our findings that the sleep state in normal subjects affects the RP and RG regions independently and to different degrees. We found a reduction in airway volume during sleep in the RP rather than the RG region in normal subjects. Our data indicate that the upper airway does not act as a homogenous entity. The RP region appears to be more likely to collapse during sleep than does the RG region, which may be important in the pathogenesis of sleep apnea. However, we believe that regional soft tissue structures surrounding the airway must be studied to further understand the physiology of airway narrowing during sleep.

Changes in soft-tissue structures. We had hypothesized that the tongue, soft palate, and lateral pharyngeal walls were all important in mediating changes in airway caliber during sleep. In our study, the state-dependent changes in airway caliber were related to both lateral and AP airway narrowing in the RP region, which is consistent with our original hypothesis. However, further analysis of the axial and sagittal data in the RP region demonstrated that the lateral pharyngeal walls and the position and width of the soft palate were most significantly associated with airway narrowing during sleep. The size and position of the tongue, except for tongue oblique distance, were not significantly altered during sleep, and did not significantly influence the caliber of the minimum RP airway area during sleep. These findings were made by comparing state-dependent differences in soft-tissue structures, and by multiple linear regression analysis. These data suggest that the tongue may be less important than previously thought in contributing to airway closure during sleep in normal subjects. However, in the multiple linear regression analysis, tongue oblique distance and lateral pharyngeal wall thickness were highly correlated. This suggests that there may be biomechanical interactions between the tongue and lateral pharyngeal walls. In addition, interrelationships may exist between the soft palate and lateral pharyngeal walls. Volumetric studies may be necessary for careful evaluation of the state-dependent biomechanical interactions between the tongue, soft palate, and lateral pharyngeal walls. Developing such knowledge is essential for fully understanding the mechanism of airway narrowing and the relative role of these different soft-tissue structures.

Although few data exist on anatomic changes in the upper airway soft-tissue structures during sleep, a number of studies have evaluated state-dependent changes in electromyographic activity of the upper airway muscles (1, 8, 42). Most of these studies have focused on activity of the genioglossus (GG) muscle and muscles of the soft palate during sleep. Several of the studies (1, 4, 42) have indicated that GG activity is reduced during sleep. Such a reduction in tone might result in a thickening of the muscle or posterior movement of the tongue, a finding not made our study. However, mechanical action cannot be inferred from electromyographic data, and our data may therefore not be incompatible with these electromyographic studies of the tongue (1, 4, 42).

Our data indicate that the soft palate plays an important role in mediating airway narrowing during sleep in normal subjects. Wheatley and coworkers (8) have studied state- dependent changes in electromyographic activity of the soft palate. In their investigation (8), electromyographic responses of the tensor palatini (TP) muscle (which functions to stiffen the soft palate) to negative pressure generations of rapid onset were evaluated in six normal subjects. As compared with the responses in wakefulness, peak electromyographic activity of the TP muscle was significantly reduced during sleep, and the latency of electromyographic activity was significantly prolonged. These state-dependent electromyographic changes in the soft palate suggest that the muscles of the soft palate relax during sleep. Such changes may result in thickening or posterior movement of the soft palate, in analogy to the findings in our investigation.

To our knowledge, only one study has focused on the electromyographic activity of muscles of the posterior and lateral walls of the pharyngeal airway during sleep (2). Since we have shown that the lateral pharyngeal walls are important mediators of airway caliber during sleep, we might expect a reduction in the electromyographic activity of these muscles during sleep. Kuna and colleagues (2) evaluated the electromyographic activity of the superior, middle, and inferior pharyngeal constrictor muscles, which comprise a portion of the lateral pharyngeal walls, in 10 normal adults during wakefulness and sleep. These muscles exhibited an increase in electromyographic activity during swallowing maneuvers, an absence of activity during non-REM sleep, and sporadic bursts of activity during REM sleep. Kuna and colleagues (2) postulated that these muscles can either constrict or dilate the airway, depending on airway size. However, their finding of decreased electromyographic activity of these muscles during sleep indicates a reduction in their tone. We have shown that the lateral walls thicken during sleep. Although the muscles that make up the lateral pharyngeal walls are complex, and their interaction is not completely understood, we would hypothesize that reduced electromyographic activity during sleep may cause relaxation of these structures, contributing to wall thickening and airway narrowing. Studies correlating electromyographic activity with anatomic changes in airway caliber and soft-tissue dimensions would further our understanding of the pathophysiology of airway narrowing during sleep. It should be noted, however, that thickening of the lateral walls during sleep may be associated with state-related changes in the conformation of the soft palate and tongue. Complex three- dimensional biomechanical interactions may exist between the tongue, soft palate, and lateral pharyngeal walls. Dynamic and volumetric imaging studies will be necessary to develop a biomechanical model of the state-dependent changes in these soft-tissue structures of the upper airway.

State-dependent Changes in Minimum Airway Area

We found that the minimum cross-sectional airway area during wakefulness and sleep was located in the RP airway in most of our normal subjects. If MRI is to be used as a diagnostic modality for SDB, more information must be obtained with regard to state-dependent airway changes in normal subjects and in patients with sleep apnea syndrome. In our study, the position of the smallest cross-sectional airway area along a longitudinal center line in normal subjects changed from wakefulness to sleep in both the RP and RG regions. The magnitude and the direction (rostral or caudal) of the positional change was variable in both regions, although in the majority of subjects, the smallest airway area moved caudally with sleep. In the RP region, the absolute difference in the position of the minimum airway area ranged from less than 1 mm to over 17 mm from wakefulness to sleep. In the RG region, this range was also variable (from less than 1 mm to over 13 mm). Therefore, we believe that simply determining the location of the minimal cross-sectional area during wakefulness cannot be used to precisely predict the location of airway narrowing during sleep, at least in normal subjects. If these findings can be extrapolated to subjects with SDB, upper airway imaging during sleep may be necessary to identify the site of airway closure in sleep apnea.

Conclusions

In summary, we have demonstrated several significant state-dependent changes in airway caliber and surrounding soft-tissue structures in normal subjects through MRI and computer software-based three-dimensional analytic techniques. Airway volume is significantly reduced during sleep in the RP region but not in the RG region. A comparison of the smallest cross-sectional airway area during sleep to the corresponding area in images made during wakefulness in three distinct regions showed significant differences in both the RP and RG regions, but not in the epiglottic region. However, the changes in airway area observed in the RP region were substantially greater than the changes observed in the RG region. The change in area in the RP region correlated with significant changes in both AP and lateral airway diameter. The lateral pharyngeal walls in the RP region were significantly thicker during sleep, mediating the reduction in lateral airway diameter. Sagittal data demonstrated posterior movement and thickening of the soft palate with sleep. Multiple linear regression analysis revealed that state-dependent changes in lateral pharyngeal wall thickness, tongue oblique distance, and soft palate position were associated with narrowing of the RP airway with sleep. We also found that the narrowest portion of the airway was in the RP region in most of our subjects. However, the longitudinal position of the smallest cross-sectional area in both the RP and RG regions varied from wakefulness to sleep. We plan to replicate our study in subjects with sleep apnea, and we postulate that the changes in airway caliber observed in these subjects will mirror the changes seen in normal subjects, but to a greater degree.