Clinical evaluation of subglottic stenosis in neonates using automatic segmentation of optical coherence tomography via dynamic programming (Conference Presentation)

Abstract

Subglottic stenosis is a severe and challenging disease to manage in neonates. Previous reports describe the usage of long-range optical coherence tomography (LR-OCT) to image the subglottis through an endotracheal tube and potentially identify subepithelial changes in subglottic mucosa which are correlated with edema or scar tissue. A major challenge associated with OCT imaging is that large volumes of data (1-2 GB) are acquired with each airway scan, with no existing automated method for image analysis and tissue measurement. We have developed an innovative MATLAB based auto-segmentation program which identifies and measures tissue layers within the mucosa. LR-OCT data sets of 21 neonates were analyzed for mucosal thickness of the proximal trachea, subglottis and larynx. The auto-segmentation measurements were compared with measurements from manual tracings by a single operator. We found statistically significant associations between the thickness of the mucosa (p<0.001) and submucosa (p<0.001) layers in the upper airway when comparing these two segmentation processes. The auto-segmentation program segmented the OCT images on average over 8 times faster than the manual segmentation software. Following auto-segmentation, OCT images were also analyzed for texture analysis properties using ANOVA. Automated segmentation and measurement of OCT data sets is an efficient and precise method to analyze large volume LR-OCT data stacks. This may ultimately help provide vital objective information about the airway in real-time, which would aid clinicians in making management decisions for intubated neonates.