Subin Erattakulangara1, Wahidul Alam1, Douglas Van Daele2, Junjie Liu3, and Sajan Goud Lingala1,4
1Roy J Carver Department of Biomedical Engineering, University of Iowa, iowa city, IA, United States, 2Department of Otolaryngology, University of Iowa, iowa city, IA, United States, 3Department of Neurology, University of Iowa, iowa city, IA, United States, 4Department of Radiology, University of Iowa, iowa city, IA, United States
Synopsis
Keywords: Segmentation, Segmentation
Motivation: The motivation for this research study is to better understand and characterize upper-airway collapse during sleep in patients with obstructive sleep apnea (OSA). The study aims to provide valuable insights into the dynamics of airway collapse.
Goal(s): The main goal is to quantitatively assess upper-airway collapse dynamics in both OSA and normal individuals, developing imaging phenotypes.
Approach: The research methodology involves data collection using MRI, manual analysis for quantitative imaging phenotypes, and presenting qualitative and quantitative findings.
Results: The study effectively visualizes airway collapse patterns in normal and OSA patients, develops quantitative imaging phenotypes, and distinguishes various collapse patterns.
Impact: The study's results provide researchers with new, quantitative insights into upper-airway collapse in sleep apnea. This may enable more precise diagnosis and treatment, stimulate further research into non-CPAP therapies, and improve the quality of care for patients with sleep apnea.
Introduction
Obstructive sleep apnea (OSA) is characterized by
dynamic, breathing-related obstructions of the upper airway during sleep.
Continuous positive airway pressure (CPAP) therapy is the first-line treatment of OSA, but
about 30-70% of OSA patients cannot use CPAP consistently. Various non-CPAP
therapies have emerged as alternatives (eg. surgical resection of intraluminal
obstructive tissues, and hypoglossal nerve stimulation therapy1). A first clinical step in such non-CPAP therapies is
to screen patient candidacy by characterizing the dynamics of upper-airway
collapse, particularly at the levels of velopharynx, oropharynx, tongue base,
epi-glottis, and distinguishing between anterior-posterior; lateral; concentric
collapse spatio-temporal patterns. The clinical standard of using drug-induced sleep endoscopy has notable
challenges (eg. use of anesthesia to induce sleep, lack of quantitative
assessment of collapse)2,3. Dynamic 3-dimensional MRI of the upper-airway is a
promising new alternative to visualize the collapsing airway during natural
sleep in OSA4–6. In this work, we perform preliminary feasibility of
extracting quantitative imaging phenotypes to characterize dynamics of airway
collapse from 3-D dynamic upper-airway MRI datasets.Methods
Experiments
were performed on a 3T GE Premier scanner equipped with high-performance
gradients (80 mT/m amplitude and 150 mT/m/ms slew rate) using either a 21
channel head-neck coil or a 16 channel custom airway coil. One normal volunteer
performed the Muller’s manuever mimicking airway collapse in the awake state. One OSA patient was recruited and imaged during natural sleep.
Condition of sleep was confirmed by monitoring changes in simultaneously
recorded physiological signals (eg. breathing effort levels, 02 saturation
levels). A 3D Cartesian GRE sequence was implemented with sparse variable
density view-ordering in the ky-kz plane. The sequence was prescribed with
parameters; FOV: 20cm x 20 cm x 8 cm; spatial resolution: 2mm x 2mm x 2mm; flip
angle: 5 degrees; minimum TR and TE. The raw k-space vs time data was resorted
by using 209 points in the ky-kz plane per time frame, which corresponds to a
time resolution of 543.4 ms. Reconstruction was performed using sparse SENSE
employing finite difference constraints along space and time.
In the normal
Muller’s maneuver dataset, we have manually segmented a time-varying airway
area function from the airway cross-sectional area in every fourth axial cut
from the nasopharynx to the oropharynx, one time instance of which is shown in
Fig. 1. From these area functions, we derived the following phenotypes to
quantitate the pattern of airway shaping in each axial cut: (a) Distensibility,
D= (Areamax – Areamin)/Areamax);
(b) Collapse in the Anterior-Posterior (AP) direction, CAP= (APmax –
APmin)/APmax; (c) Collapse in the Lateral (Lat)
direction, CLateral= (Latmax – Latmin)/Latmax; (d) Concentric collapse CConc= (CAP + CLat)/2; where Areamax, Areamin are the maximum and minimum areas of the
airway over the course of a collapse event in sleep state. (APmax,APmin) and (Latmax,Latmin) are the maximum, and minimum dimensions of
the airway respectively in the anterior-posterior and lateral directions. All
the indices are established as a percent measure. D gives a measure
of the overall airway compliance where a high D signifies the
airway being highly distensible (or highly collapsible), and vice-versa.
Similarly CAP, CLat, Cconc respectively characterize the percent
collapse in the anterior-posterior, lateral, and concentric directions. We have
done a similar analysis on the OSA patient dataset but on representative 4 axial
cuts in each of the nasopharynx, velopharynx, oropharynx, hypopharynx regions.
All segments were performed manually in the 3D SLICER software.
Results
Fig. 2
shows an example of the sparse view ordering in the ky-kz plane, and the
corresponding 3D reconstructions of simulated airway collapse in the normal
volunteer during Muller’s manuever. For example, nasopharynx, velopharynx,
oropharynx, and hypopharynx cuts are shown faithfully depicting the airway collapse
patterns in the awake state.
Fig. 3
shows the 3D dynamics of the airway collapse in the naturally sleeping OSA
patient. Qualitatively, we observe concentric collapse in several of the axial
cuts (particularly at the velopharynx level).
In Table 1(Fig. 4), we observe a high degree of concentric collapse (>87%)
at all axial cuts, except the nasopharynx level (where the lateral collapse was
dominant at 77%). In Table 2 (Fig. 5) (on the OSA patient dataset), We observe
an increasing amount of concentric collapse in all the axial cuts (>87), except at the oropharynx level, where the anterior-posterior collapse is dominant (85%)Conclusion
This
work presented the preliminary applicability of manual segmentation on 3D+time
deforming upper-airway datasets to extract imaging phenotypes quantitatively
characterizing airway collapse in the anterior-posterior; lateral; and
concentric directions. Future work includes automating the pipeline by
efficient deep learning-based segmentations, and further analysis of multiple
OSA patient-datasets.Acknowledgements
This work was conducted on an MRI instrument
funded by 1S10OD025025-01References
1. Costantino A,
Rinaldi V, Moffa A, et al. Hypoglossal nerve stimulation long-term clinical
outcomes: a systematic review and meta-analysis. Sleep Breath. Published
online 2020. doi:10.1007/s11325-019-01923-2
2. Chao TN, Thaler ER. Predictors of
success in hypoglossal nerve stimulator implantation for obstructive sleep
apnea. World J Otorhinolaryngol - head neck Surg. 2020;7(1):40-44.
doi:10.1016/J.WJORL.2020.02.007
3. Kellner P, Herzog B, Plößl S, et al.
Depth-dependent changes of obstruction patterns under increasing sedation
during drug-induced sedation endoscopy: results of a German monocentric
clinical trial. Sleep Breath. Published online 2016.
doi:10.1007/s11325-016-1348-6
4. Kim Y-C, Lebel RM, Wu Z, Ward SLD, Khoo
MCK, Nayak KS. Real-time 3D magnetic resonance imaging of the pharyngeal airway
in sleep apnea. Magn Reson Med. 2014;71(4):1501-1510.
doi:10.1002/mrm.24808
5. Alam W, Rusho RZ, Liu J, Van Daele D,
Jacob M, Lingala SG. Accelerated imaging of airway collapse in obstructive
sleep apnea with variable density spirals and variational manifold
regularization. In: ISMRM Workshop on Data Sampling and Image Reconstruction.
; 2023.
6. Javed A, Kim YC, Khoo MCK, Ward SLD,
Nayak KS. Dynamic 3-D MR visualization and detection of upper airway
obstruction during sleep using region-growing segmentation. IEEE Trans
Biomed Eng. Published online 2016. doi:10.1109/TBME.2015.2462750