Characteristics and Mechanism of Upper Airway Collapse Revealed by Dynamic MRI During Natural Sleep in Patients with Severe Obstructive Sleep Apnea

Introduction

Obstructive sleep apnea (OSA) is characterized by repetitive collapse of the upper airway during sleep, resulting in hypopnea or apnea.^1,2 The primary etiologies of OSA are thought to be linked to abnormal upper airway anatomy and reduced neural control of pharyngeal muscles,^3–6 leading to dynamic collapses of multiple patterns.^7–9 Surgical intervention is one of the main interventions for OSA. However, the selection of surgical approaches is influenced significantly by various clinical factors, such as the pattern of upper airway collapse and the patient’s clinical status. Surgery’s success largely depends upon the personal judgement of the doctor and the utilization of appropriate techniques.¹⁰ Investigating the mechanism of upper airway collapse in OSA patients is of great clinical significance to understand the pathogenesis and select optimal treatments.

Magnetic resonance imaging (MRI) is commonly used to assess anatomical abnormalities in the upper airway or surrounding tissues in OSA patients.^11,12 MRI studies have shown that OSA patients are characterized by an oversized tongue and soft palate, narrowed cross-sectional areas of the airway, downshifted hyoid, and increased airway length compared to healthy individuals.^13–19 Although conventional MRI has significant advantage in identifying anatomical features in patients with OSA, its scan time is often much longer than the duration of the respiratory cycle or upper airway obstruction, and therefore it cannot capture the dynamic process of upper airway obstruction. The results obtained by MRI may vary widely when a particular tissue changes significantly during airway collapse. Therefore, conventional MRI has obvious limitations in studying the process of upper airway collapse during sleep.

Dynamic MRI, which capture a series of images in quick succession to create a movie-like sequence of tissue in motion,²⁰ can record the process of airway changes and tissue deformation during respiratory cycles or obstruction events.²¹ Dynamic MRI studies in awake state show changes in airway morphology during breathing in OSA patients compared with healthy subjects, including increased airway compliance and reduced airway ellipticity.^22,23 Dynamic MRI studies during drug-induced or natural sleep can capture airway obstruction in OSA patients, such as retropalatal and retroglossal collapse, airway space changes, and tongue or soft palate movement. Upper airway obstruction can be classified according to the collapse site.^7,8 Most studies focus on the changes of the airway itself during breathing or collapse, but ignore the movement of its surrounding tissues.^24–26 Because the airway is a cavity enclosed by the head and neck tissues, and its collapse is caused by the movement of surrounding tissues, it is crucial to uncover tissue movement patterns during the collapse process to understand the mechanism behind airway collapse. Previous studies using dynamic MRI have also observed the dynamic changes in the tongue, soft palate, and uvula, showing that the range of tongue and soft palate movements during sleep was greater in OSA patients than in healthy individuals.^27,28 However, these studies have not analyzed the movements and morphological changes of tissues during airway collapse, which is of great significance for understanding the mechanism of different types of airway obstruction.

The present study aims to investigate the process of upper airway collapse during natural sleep in patients with severe OSA using dynamic MRI. We will analyze the characteristic changes of the upper airway and surrounding tissues under different obstruction patterns, compare the anatomical features during normal breathing in OSA patients with different obstruction patterns, and establish correlations between anatomical structural features and dynamic motion characteristics, which can lead to a better understanding of the mechanisms of airway obstruction and help in the development of personalized treatments for OSA.

Materials and Methods

Subjects and Dynamic MRI Examinations

All subjects are Asian males. Exclusion criteria include lung disease, upper airway tissue lesions, psychiatric disorders, and severe cardiovascular disease. None of them had received any surgical treatment for OSA prior to joining the study. All patients were monitored overnight using a noninvasive polysomnography (SOMNO HD, SOMNOmedics, Germany) under standard procedures. Apnea-hypopnea index (AHI) was confirmed by a professional sleep physician based on polysomnography data. Finally, 31 patients with severe OSA diagnosed with AHI >30 were enrolled for dynamic MRI scans during natural sleep.

The participants were asked to sleep less the night before so that they could fall asleep naturally during the MRI scan without the use of any sedatives. Each patient was supine with his heads in an 8-channel head and neck combination coil secured with sponge pads and earplugs in both ears. The nasal airflow was inputted into the sleep monitor (EMBLA N7000, Reykjavik, Iceland) via a nasal cannula. If the nasal airflow signal indicated repetitive pauses lasting more than 10 seconds, the subject was deemed asleep, and the dynamic MRI scan was initiated. Dynamic MRI scans were carried out with a fast gradient-echo sequence using a 3.0T MRI scanner (GE Healthcare, Milwaukee, Wisconsin, USA) on the mid-sagittal plane of the head and neck. The scan parameters included TR/TE = 7.6/1.9 ms, a flip angle of 12°, a matrix of 256×256, and a layer thickness of 16 mm. The average scan time for each participant was 120 s. The total number of mid-sagittal images acquired was 128. Patients who exhibited excessive head movement during dynamic MRI or did not show upper airway obstruction were excluded. Ultimately, 23 male patients with severe OSA were included in the study. The protocol was approved by the Ethics Committee of Capital Medical University, Beijing, China (2013SY67). All subjects signed the informed consent prior to participation. This study was conducted in accordance with the Declaration of Helsinki.

Acquisition of Dynamic Parameters

We manually segmented structures frame-by-frame from the dynamic images. Figure 1A shows the tissues and airway extracted from the images, which include the tongue, soft palate, hyoid bone, retropalatal and retroglossal airway. Then, morphological parameters were measured, and the centers of the tongue, soft palate, and hyoid bone were calculated separately to characterize the position of each structure.

Figure 1 Schematic diagram of the tissues (A) and morphological parameters (B) to be extracted. RP: Retropalatal; RG: Retroglossal region; LLAX: The long axis of the tongue; LSAX: the short axis of the tongue; α: the angle between the hard palatal and the soft palate; β: the angle between the soft palate and the uvula; LSP: The length of soft palate. Coordinate system is built by the front upper point of C3 and the front lower point of C4.

To make the results from different subjects comparable, the vertebral coordinate system was established based on the C3 and C4 vertebrae, as shown in Figure 1B, and the coordinates of all tissue centers were in the vertebral coordinate system. The morphological parameters of the tissues, including the long axis of the tongue (L_LAX), the short axis of the tongue (L_SAX), the hard palate-soft palate angle (α), the soft palate-tip of the uvula angle (β), and the soft palate length (L_SP), as shown in Figure 1B, were measured. The aspect ratio of the tongue, AR_Tg = L_LAX/L_SAX, was calculated. The upper airway parameters included the minimum width of the retropalatal space (minW_RP) and the minimum and mean width of the retroglossal space (minW_RG and mW_RG). We also measured the distance between the tongue center and the tip of the uvula (L_TgU) to assess the relative position between the tongue and the uvula.

The stable values of the parameters during airway obstruction were measured at each obstruction event. The average values of the parameters in a time interval of approximately 4 seconds before obstruction were calculated as the baseline at the time of airway opening for this obstruction event. The change of a parameter due to the obstruction event was then obtained by subtracting the value of the obstruction state from the value of the opening state.

Classification of Upper Airway Obstruction Pattern

Based on dynamic MRI images of the mid-sagittal plane, we classified the types of airway obstruction according to the motion characteristics of the soft palate and tongue during collapse. We grouped the patients based on their obstruction type for our statistical analysis. If a patient had multiple types of airway obstruction, the patient was classified based on the longest cumulative time for the type of obstruction. As well, we characterized the airway anatomical parameters in normal breathing status of the patients using the mean value of the parameters during airway opening.

Statistical Analysis

The data of all subjects were analyzed using SPSS software (SPSS Inc, Chicago, USA). Paired-sample nonparametric tests were conducted to compare various parameters in both the airway opening and collapse states. The independent sample nonparametric test was used to evaluate the parameter changes of different types of obstruction events and the anatomical parameters in normal breathing status of subjects with different types of obstruction. When p-values < 0.05, the difference was considered statistically significant. Spearman correlation analysis was used to test the correlations of anatomical parameters and parameter changes.

Results

A total of 112 obstruction events were extracted from the images of 23 patients. The basic characteristics of the patients were given in Table 1.

Table 1 Basic Characteristics of 23 Asian Male Participants

Upper Airway Obstruction Pattern

We first categorized airway obstruction into three major patterns based on dynamic MRI in the mid-sagittal plane, as shown in Figure 2. These patterns included retropalatal obstruction caused by a backward movement of the soft palate separated from the tongue (Type A), retropalatal obstruction due to a backward movement of the soft palate attached to the tongue (Type B), and retropalatal and retroglossal obstruction arising from a combined backward movement of the soft palate and the tongue (Type C).

Figure 2 Schematic diagram of upper airway opening and different types of obstruction.

However, there were two distinct tissue movement patterns within type B obstruction. This prompted us to further subdivide type B obstruction into type BI and type BII. Both subtypes were characterized by retropalatal obstructions resulted from tongue pushing against the soft palate, but they exhibited different tissue movements. Supplementary dynamic schematic diagrams of the four obstruction patterns are presented in the attached Video File, which can clearly distinguish between BI and BII obstruction.

Figure 3 shows the distribution of different airway obstruction patterns. Subfigures A and B are based on subjects and events, respectively. Of the 23 patients, 82.7% exhibited a single type of obstruction, and 17.3% showed multiple types of airway obstruction. More than half of the subjects presented with type B obstruction, much higher than the others. Notably, type BI obstruction event occurred most frequently, accounting for approximately 44.6% of all obstruction events.

Figure 3 Distribution features of airway obstruction in 23 subjects (A) and in 112 obstruction events (B). The data in brackets at the top of each rectangular diagram is the percentage of a given obstruction type.

During the MRI scan, 10 of the 23 people had oral breathing. Six of them were breathing orally at the time of the airway obstruction. Oral breathing had the greatest impact on type A, accounting for half of the obstruction events (16/32). In patients with BI and BII types, 16% (8/50) and 14% (2/14), respectively, had oral breathing at the time of obstruction, while no oral breathing occurred in type C patients.

Dynamic Changes of Upper Airway Obstruction

Figure 4A and B show the displacement of the hyoid center in the X and Y directions, ΔX_Ho and ΔY_Ho, respectively, during airway obstruction. It could be seen that the hyoid remained stable during type A obstruction without significant displacement. However, the hyoid moved significantly upward in type BI, and downward and toward the posterior wall of the airway in type BII. The hyoid moved upward in type C, and the displacement was significantly larger than that in type BI.

Figure 4 Dynamic characteristics of the displacement of each tissue centroid during airway obstruction of type A, BI, BII, and C. (A and B) ΔXHo and ΔYHo, displacement of the hyoid center in the X-direction and Y-direction. (C and D) ΔXSP and ΔYSP, displacement of the soft palate center in the X-direction and Y-direction. (E and F) ΔXTg and ΔYTg, displacement of the tongue center in the X-direction and Y-direction. The hash #Indicates that the change of the centroid location itself during airway obstruction is significant, p-value < 0.05. The asterisk indicates that the displacement with different types of obstruction is significantly different. **p-value <0.01, ***p-value <0.001.

Figure 4C and D show the displacement of the soft palate center in the X and Y directions, ΔX_SP and ΔY_SP, respectively, during airway obstruction. Although the soft palate moved toward the posterior airway wall in type A, its displacement was comparatively smaller than in the other three types of obstruction. The soft palate moved downward and toward the posterior airway wall in type BI, and the displacement trend in type BII was similar to that in type BI, with the exception of a greater downward displacement than that in type BI. The soft palate exhibited upward and posterior displacement in type C.

Figure 4E and F show the displacement of the tongue center in the X and Y directions, ΔX_Tg and ΔY_Tg, respectively, during airway obstruction. Type A obstruction resulted in a posterior displacement of the tongue, which was comparatively smaller than that in the other three types of obstruction. The tongue moved upward and toward the posterior airway wall in type BI. But the tongue moved downward and toward the posterior airway wall in type BII. The displacement trend of the tongue in type C was the same as that in type BI, except that the posterior displacement of the tongue was larger in type C obstruction.

Figure 5 shows the alterations in the morphological parameters of the tongue and soft palate during airway obstruction. As shown in Figure 5A, there was no significant change in the aspect ratio of the tongue, ΔAR_Tg, during type A obstruction. However, ΔAR_Tg in type BI exhibited an obvious decline, which was mainly due to the shortening of ΔL_LAX and the lengthening of ΔL_SAX, as shown in Figure 5B and C. By contrast, ΔAR_Tg exhibited a significant increase in type BII, which was primarily driven by a large increase in ΔL_LAX. The trend of ΔAR_Tg in type C was similar to that in type BI. The difference was that the ΔAR_Tg change in type C was greater. Figure 5D shows the change in the length of the soft palate, ΔL_SP, during airway obstruction. Notably, all four types of obstruction exhibited an elongation in ΔL_SP, with the least significant increase in type A and the most pronounced increase in type BII.

Figure 5 Dynamic characteristics of tongue and soft palate morphology changes during airway obstruction. (A) ΔARTg, variation of the aspect ratio of the tongue. (B) ΔLLAX, variation of the long axis of the tongue. (C) ΔLSAX, variation of the short axis of the tongue. (D) ΔLSP, variation of the length of soft palate. (E) Δα, variation of the angle between the hard and soft palate. (F) Δβ, variation of the angle between the soft palate and uvula. The hash #Indicates that the change of the centroid location itself during airway obstruction is significant, p-value <0.05. The asterisk *Indicates that the displacement with different types of obstruction is significantly different. *p-value < 0.05, **p-value <0.01, ***p-value <0.001.

Figure 5E and F show the change in the angles of the hard palate-soft palate, Δα, and soft palate-tip of the uvula, Δβ, during airway obstruction. The angle Δα increased significantly during type A, BI and C obstruction, while it decreased significantly in type BII obstruction. For the angle Δβ, there was no significant change in type A and BII obstruction, reflecting the constant curvature of the soft palate. However, Δβ decreased significantly in type BI and C obstruction, indicating increased curvature of the soft palate in response to airway obstruction.

Figure 6A and B show the percentage changes in the mean and minimum width of the retroglossal space, ΔmW_RG and ΔminW_RG, during airway obstruction. Both ΔmW_RG and ΔminW_RG experienced significant reductions during all types of obstruction. Type A obstruction resulted in the lowest reduction in both width parameters of the retroglossal space. Types BI and BII resulted in greater reduction, whereas type C resulted in complete closure of the retroglossal space.

Figure 6 Dynamic characteristics of the mean width ΔmWRG (A) and the minimum width ΔminWRG (B) of the retroglossal space changes during airway obstruction. The hash #Indicates that the change of the centroid location itself during airway obstruction is significant, p-value <0.05. The asterisk *Indicates that the displacement with different types of obstruction is significantly different. *p-value <0.05, **p-value <0.01, ***p-value <0.001.

Anatomical Features

Table 2 shows the characteristics of the study participants, including age, BMI, AHI, and upper airway anatomical parameters measured during normal breathing. No significant differences were observed in age, BMI, and AHI in these patients. Among the 23 patients, 7 were dominated by type A obstruction, 9 by type BI obstruction, 2 by type BII obstruction, and 5 by type C obstruction. The differences of X_Ho, L_TgU, minW_RP, and mW_RG among the patients were statistically significant.

Table 2 Anatomical Measurement Results in Patients with Different Obstruction Types A, BI, BII, and C

During normal breathing, patients with type A dominated obstruction exhibited larger L_TgU and |X_Ho|, which can be used to estimate roughly the distance from the hyoid to the posterior wall of the airway, and smaller minW_RP compared with other patients, indicating that in patients with type A dominated obstruction the hyoid was more anteriorly positioned, the distance between the tongue and uvula was longer, and the retropalatal space was smaller. The value of mW_RG was larger in patients with type BI dominated obstruction, suggesting a larger retropalatal space compared to other patients, especially patients with type C dominated obstruction. No significant anatomical features were identified in patients with type BII dominated obstruction, possibly due to the relatively smaller number of patients in this group. Patients with type C dominated obstruction displayed a significantly larger retropalatal space, smaller retroglossal space, and smaller distance between the hyoid and the posterior airway wall compared with patients with type A dominated obstruction.

Correlation of Anatomical Features and Dynamic Changes

Figure 7 shows the heat map of correlation analysis between the anatomical features in normal breathing and dynamic changes during airway obstruction. The results show that, among the anatomical features, the X-direction position of the hyoid, X_Ho, the distance between the tongue center and the tip of the uvula, L_TgU, and the minimum width of the retropalatal space, minW_RP, were significantly correlated with many dynamic changes during airway obstruction. These correlations highlight the important influence of these anatomical features on the process of upper airway obstruction.

Figure 7 Correlation of the anatomical parameters during normal breathing and the parameter changes during airway obstruction. *P < 0.05, **P < 0.01. R value is represented by color. The horizontal axis shows the anatomical parameters of the upper airway during normal breathing. The vertical axis shows the changes of parameter during airway obstruction. Abbreviations: XHo and YHo, the position of the hyoid center in the X-direction and Y-direction; XSP and YSP, the position of the soft palate center in the X-direction and Y-direction; LSP, the length of soft palate; α, the angle between the hard and soft palate; β, the angle between the soft palate and uvula; XTg and YTg, the position of the tongue center in the X-direction and Y-direction; LLAX, the long axis of the tongue; LSAX, the short axis of the tongue; ARTg, the aspect ratio of the tongue; LTgU, the distance between the tongue center and the tip of uvula; minWRP, the minimum width of the retropalatal space; mWRG, the mean width of the retroglossal space; minWRG, the minimum width of the retroglossal space; Δ, the change in the parameter.

As shown in Figure 8A–F, X_Ho was positively correlated with Δα, ΔX_Tg, and ΔL_SAX, as well as negatively correlated with ΔAR_Tg, ΔmW_RG, and ΔminW_RG. These findings imply that in patients with a more posterior hyoid position, there is greater posterior displacement and deformation of the tongue, increased elevation of the soft palate, and greater reduction of the retroglossal space during airway collapse. Figure 9A–F show that L_TgU was positively correlated with ΔL_LAX, ΔAR_Tg, ΔminW_RP, ΔmW_RG, and ΔminW_RG, as well as negatively correlated with ΔX_SP, ΔX_Tg, and Δα. This suggests that in patients with a longer distance between the tongue and the uvula, there is less posterior displacement and deformation of the tongue and soft palate, and less reduction of the retropalatal and retroglossal space during airway obstruction. As shown in Figure 10A–E, the correlation between minW_RP and dynamic changes was similar to that of X_Ho. This observation indicates that in patients with larger retropalatal space, there is greater posterior displacement and deformation of the tongue, as well as increased reduction of the retroglossal space.

Figure 8 Correlation of the X-direction position of the hyoid, XHo with (A) the difference of the hard palate-soft palate angle, Δα, (B) the displacement of the X-direction position of the tongue, ΔXTg, (C) the difference of the length of the long axis of the tongue, ΔLLAX, (D) the difference of the aspect ratio of the tongue, ΔARTg, (E) the difference of the mean width of the retroglossal space, ΔmWRG, (F) the difference of the minimum width of the retroglossal space, ΔminWRG in all subjects. Solid lines are the fitting results for data from all subjects.

Figure 9 Correlation of the distance between the tongue and uvula, LTgU with (A) the displacement of the X-direction position of the soft palate, ΔXSP, (B) the displacement of the X-direction position of the tongue, ΔXTg, (C) the difference of the hard palate-soft palate angle, Δα, (D) the difference of the length of the long axis of the tongue, ΔLSAX, (E) the difference of the aspect ratio of the tongue, ΔARTg, (F) the difference of the minimum width of the retropalatal space, ΔminWRP, (G) the difference of the mean width of the retroglossal space, ΔmWRG, (H) the difference of the minimum width of the retroglossal space, ΔminWRG in all subjects. Solid lines are the fitting results for data from all subjects.

Figure 10 Correlation of the minimum width of the retropalatal space, minWRP with (A) the displacement of the X-direction position of the tongue, ΔXTg, (B) the difference of the length of the short axis of the tongue, ΔLSAX, (C) the difference of the aspect ratio of the tongue, ΔARTg, (D) the difference of the mean width of the retroglossal space, ΔmWRG, (E) the difference of the minimum width of the retroglossal space, ΔminWRG in all subjects. Solid lines are the fitting results for data from all subjects.

Discussion

In this study, we investigated the collapse process of the upper airway during natural sleep in patients with severe OSA using dynamic MRI. This study identified four distinct types of airway obstruction, including type A, type BI, type BII, and type C based on the obstruction location and motion characteristics. By analyzing the anatomical features of the upper airway and its surrounding tissues, we clarified the tissue movements and morphological changes in each obstruction pattern. Moreover, we evaluated the differences in the anatomical features of patients with different obstruction patterns during normal breathing and explored the correlation between anatomical features and dynamic changes, which shed light on the key anatomical structures that influence the process of airway obstruction.

Patients with type A obstruction have a significantly smaller retropalatal space and a greater distance between the tongue and uvula during normal breathing compared to patients with other obstruction types. In such a case, even a small posterior displacement of the soft palate can result in retropalatal obstruction. The flow rate through the very narrow retropalatal space can be very high, resulting in a large negative pressure there during inspiration, which pulls the tip of the uvula towards the posterior wall of the airway quickly to block the airway without the additional assistance of tongue displacement. When the retropalatal airway is blocked, there is no further change in downstream airway pressure, and thus, the position of the tongue can remain basically unchanged. Therefore, the primary dynamic characteristic of type A obstruction is the upward rotation of the soft palate around the junction of the hard and soft palates.

Type B obstruction refers to a retropalatal obstruction caused by the movement of the soft palate attached to the tongue. During airway obstruction in type BI, the tongue pushes the soft palate backwards and upwards. As the tongue moves back and up, the hyoid moves up, the soft palate rotates upward around the junction between the hard and soft palate, and the soft palate itself becomes more curved. We found two patients with type B obstruction, where the tongue moved backward and downward, more of a retropalatal obstruction due to uvula elongation. In these cases, the hyoid also moves backward and downward and the soft palate rotates downward around the junction between the hard and soft palate while its curvature remains essentially constant. Therefore, we speculate that the underlying mechanism is different from that of type BI obstruction and classify it as type BII obstruction. However, no significant difference is found in the anatomical characteristics of patients with type BI and type BII during normal breathing, and their specific mechanisms need further study.

In patients with type C obstruction, the hyoid is located closer to the posterior wall of the airway during normal breathing than in patients with other types of obstruction, with a larger retropalatal space and a smaller retroglossal space. The larger retropalatal space requires a larger posterior displacement of the tongue to cause retropalatal obstruction, which may result in simultaneous obstruction of the smaller retroglossal space. The movement characteristics of type C airway obstruction are more like an enhanced pattern of type BI. The tongue, soft palate, and hyoid move in same manner as in type BI, but to a greater extent. Importantly, the soft palate rotates more upward, and the soft palate center exhibits a posterior upward displacement due to excessive compression of the tongue.

Our study reveals significant differences in both the anatomical and motor characteristics of the hyoid in different types of airway obstruction. Specifically, from type A to type BI to type C, the hyoid is successively located posterior and inferior during normal breathing, and is successively moved upward with increasing magnitude during airway obstruction. Type BII obstruction differs in that the hyoid is in a similar position to type BI during normal breathing, but moves posteriorly and inferiorly during airway obstruction. These findings suggest that hyoid position may help determine the type of obstruction in OSA patients. Previous imaging studies related OSA have confirmed that the hyoid position in OSA patients differs from that of healthy individuals, and the severity of OSA is related to the size of the downward position of the hyoid.^29,30 This is consistent with our opinion that hyoid plays an important role in the development of OSA. Our study extends upon this and shows that the hyoid location is also a key factor in distinguishing among different types of airway obstruction in OSA patients. Hyoid suspension is often used in clinical practice to treat OSA, and our findings suggest that altering the hyoid position can positively improve obstruction in OSA patients. Previous studies have shown that hyoid suspension significantly reduces the severity of OSA,³¹ and is more effective when used in combination with other surgical options.^32,33 Stuck et al reported that hyoid suspension did not improve the upper airway anatomy in OSA patients while awake,³⁴ but this does not mean that hyoid suspension is ineffective. Our study indicates that the hyoid position is strongly correlated with multiple dynamic characteristics during airway obstruction, including the changes in tongue movement and morphology, the reduction in the retroglossal space, and the changes in the hard-soft palate angle. This suggests that improving hyoid position may not significantly alter the anatomy of the upper airway when awake, but may influence tissues movement during airway obstruction. Therefore, it is recommended that more attention be paid to hyoid-related procedures in future OSA treatment options.

This study also reveals that the minimum width of retropalatal space is associated with multiple dynamic characteristics during airway obstruction. Specifically, the OSA patients with a larger retropalatal space show greater changes in the tongue movement and morphology, as well as greater reductions in airway size during airway obstruction. These findings may help explain the currently low success rate of retropalatal surgery for OSA. Previous studies have shown that patients who do not respond to retropalatal surgery experience more severe airway obstruction.³⁵ Our study suggests that merely expanding the retropalatal space may not alleviate OSA in some patients and may actually lead to more severe airway obstruction, as previously predicted using the finite element simulation.³⁶ Therefore, multilevel surgery may be needed to increase the success rate of the surgery in patients with B type obstruction.

Drug-induced sleep endoscopy (DISE) is a widely used clinical technique for the observation of upper airway obstruction in OSA patients. The advantage of DISE is the ability to visually observe the location and degree of airway obstruction in OSA patients. Previous DISE studies have shown that retropalatal obstruction is the dominant site of obstruction.^37,38 However, DISE examines upper airway obstruction under drug-induced sleep, which may differ from natural sleep in terms of upper airway dynamics.³⁹ DISE also does not provide information on the movement of surrounding tissues such as the tongue and soft palate, which are directly responsible for airway obstruction. Some dynamic studies of the upper airway collapse process have been conducted using CT or MRI. Li et al dynamically observed the process of upper airway collapse under drug-induced sleep through CT, and found that OSA patients had multilevel obstruction, in which the retropalatal airway was the predominant site of obstruction. Complete retropalatal obstruction occurred in 86% of moderate to severe OSA patients.⁴⁰ Studies with dynamic MRI for upper airway obstruction indicated that retropalatal obstruction occurred in 98% of OSA patients and retroglossal obstruction in 41% of OSA patients. Most patients with retroglossal obstruction showed concurrent retropalatal obstruction, while only 4% of patients was hypopharynx obstruction.²¹ The segmental pattern of upper airway obstruction observed in our study is generally consistent with previous findings in DISE, CT, and MRI studies, highlighting the prevalence of retropalatal obstruction in OSA patients. Our study showed that all patients with severe OSA presented with retropalatal obstruction, and 21.74% of them also had retroglossal obstruction. We should indicate that the four categories we made were purely based on the events of airway obstruction captured in this study. We do not rule out other types of obstruction, such as the obstruction in retroglossal area only or in epiglottis region. However, due to the limited data and the differences in the characteristics of patients participated in different studies, we did not capture these types of obstruction, and therefore, they were not included in the classification of the current study. An important difference from previous studies on airway obstruction is that our study not only classifies the patterns of upper airway obstruction, but also provides insights into the specific movement process and mechanism of each type of obstruction.

There are still some limitations in this study. In order to clearly distinguish the characteristic differences, only Asian male patients with severe OSA are included in this study. Whether the study results are applicable to OSA patients of other races, female OSA patients, or patients with mild to moderate OSA needs further investigation. In addition, this study only examine the process of airway collapse in the mid-sagittal plane of the head and neck, and therefore cannot assess the lateral collapse of the upper airway. The method of using nasal airflow signals to determine whether a OSA patient is asleep can effectively serve the purpose of the current study, but may have limitations for studies that require determination of sleep degree. Due to the technical difficulties in acquiring dynamic MRI during natural sleep, the sample size in this study is relatively small. The 120-second scan is relatively short in duration, likely capturing mainly the light sleep stage, and the four types of obstruction consist of just 112 obstruction events. Larger sample sizes can help identify more distinctive features of different types of obstruction, especially in patients with newly identified type BII obstruction.

Conclusions

This study proposed a novel categorization of upper airway obstruction on the sagittal plane and distinguished four different patterns with varying behavioral characteristics. The motions of the tongue, soft palate and hyoid, as well as associated airway dynamic changes were analyzed in the different patterns of collapse. Moreover, this investigation scrutinized the anatomical features of the upper airway during normal breathing among in patients with different types of obstruction and identified the hyoid position and retropalatal space as the key factors influencing upper airway obstruction.