Automatic volumetric segmentation of the upper airway and parapharyngeal fat in head and neck CT scans

Abstract

Obstructive sleep apnea (OSA) is a debilitating condition that affects a very large percentage of the adult population. It is a condition that is both difficult to detect and highly impactful on the lives of patients. Through the years different neck structures have been associated with the pathogenesis and development of the condition. In this research we attempt to look into the relationship between the volume and composition of the upper airway and the parapharyngeal fat pads with OSA. Furthermore, an innovative way of measuring it using image processing methods is proposed. To this end, we computed the volumetric segmentation of the upper airway and the pha- ryngeal fat pads using computational image processing techniques. This approach was chosen due to the difficulty in obtaining volumetric training examples as they require a high amount of expertise and time. For this propose we had access to 3D computed tomography (CT) scans and polysomnographic clinical data for 176 patients that participated a Colciencias ob- structive sleep apnea study. Of the 176 patients in the study, 61 had to be removed because they had teeth amalgams that generated a large amount of interference in the CT scans and a further 31 had to be removed due to lack of clinical data, leaving us with a sample of 84 patients with viable data. After obtaining the segmentations a correlations study was performed with the intention of researching the relationship of the parapharyngeal fat pads and the Apnea-hypoapnea index (AHI). Despite performing a multitude of correlation tests, no significant correlation was found between the measured volumes and the AHI.