Computer-Vision and Computational Fluid Dynamics Analysis Pipeline to Improve Diagnosis in Pediatric Subglottic Stenosis

Pediatric subglottic stenosis (SGS) is a pathological narrowing of the airway, affecting up to 2% of infants. The
restriction of the airway requires a dramatic increase in breathing effort, and can result in reliance on
tracheostomy tubes, respiratory failure, and death if left untreated. Multiple treatment options presently exist for
SGS, and they range from conservative approaches (eg: observation) to invasive open airway reconstructive
surgeries. A patient's long-term trajectory depends on proper classification of disease severity and selection of
the optimal treatment modality. Diagnosis and treatment planning is currently performed via endoscopy and a
low-fidelity, four-tiered grading system, the Cotton-Meyer scale, which is derived from endotracheal tube sizing.
Cross-sectional imaging is not standard of care for this population and thus no accurate quantitative measures
of the obstruction size and location are available for diagnosis. Additionally, the current diagnostic workup does
not quantify respiratory effort, and there is limited information describing the relationship between obstruction
size and respiratory physiology (eg: breathing resistance). Clinicians are forced to risk-stratify patients and
select a treatment course based on an incomplete and imprecise descriptions of SGS severity. The
management of SGS patients, thus, remains a challenging and drawn-out process with patients often requiring
multiple treatments. We aim to improve the clinical care of this sensitive population by developing a novel
diagnostic pipeline that accurately quantifies stenosis morphology and a patient's respiratory effort. This
pipeline leverages a computer-vision algorithm, structure from motion (SfM), and computational fluid dynamic
(CFD) simulations to generate 3D reconstructions of the diseased airway and then compute the airflow
environment. Crucially, the SfM algorithm reconstructs the 3D airway anatomy directly from standard clinical
endoscopy and requires almost no modification to the current diagnostic protocol. In Aim 1 we will demonstrate
and validate our SfM+CFD analysis pipeline on a cohort of patients receiving both endoscopy and computed
tomography (CT) imaging of the airway; the CT images provide a validation dataset for surfaces derived from
endoscopy using SfM. In Aim 2 we will develop the relationship between parameters measured from our
pipeline (eg: stenosis size, breathing resistance) and clinical severity. Upon completion of this study, we will
have refined and validated our analysis pipeline and identified the metrics that best predict clinical severity in a
small cohort and will be well positioned for a prospective clinical validation study. As our method relies on
standard endoscopy and not cross-sectional imaging, it can be applied across the full spectrum of SGS
patients and be used to study patients longitudinally throughout their treatment course. The rich patient-specific
data-set will enable improved risk-stratification, the identification of predictive factors for surgery, investigation
into treatment failure mechanisms, and improve our understanding of SGS disease progression and the role of
various treatments within the therapeutic ladder.

Grant Number: 5R21HL172011-03
NIH Institute/Center: NIH
Principal Investigator: Michael Barbour