Low-Burden High Fidelity Obstructive Sleep Apnea Screening Using Oral Cavity Images

Soliish, Inc., Fairfax VA

Investigators

Abstract

PROJECT SUMMARY Obstructive Sleep Apnea (OSA) is a pervasive sleep disorder that affects about 70 million adults in the US and almost 1 Billion people globally, with an even larger estimated population of OSA patients who are undiagnosed. Obtaining treatment for OSA and preventing the onset of severe comorbidities is heavily dependent upon the early identification of symptoms through patient screening. However, current OSA screening solutions are limited in utility and scope and generally pose affordability, availability, and scalability challenges. One approach to improving OSA screening is through examination of the oral cavity. Deviations in oral cavity anatomy, including enlarged tonsils, elongated uvula, low-hanging or thickened soft palate, enlarged tongue (macroglossia), low visibility of the oropharynx (Mallampati), as well as bruxism and narrow dental arches (e.g., malocclusions) can directly relate to the occurrence of OSA. As such, dentists have been a critical component of OSA screening and care for several decades and oral cavity screenings are part of many clinicians’ physical evaluations for OSA. However, reliance on dentists and clinical evaluations of the oral cavity maintains the bottleneck of physician availability in limiting patient access to OSA screening. To address this need, Soliish is developing a novel digital tool for democratizing OSA screening at the population level. Our mission is to improve early detection of OSA and support clinical decisions during follow-up care. Soliish’s OSA Screening solution is an innovative digital platform that uses a deep learning model, OSA Net, to deliver a cost-effective, non-invasive, user-friendly, and privacy-preserving experience that outperforms outdated screening questionnaires. Unlike other artificial intelligence (AI) or deep learning models which rely upon medical imaging of the oral cavity to predict OSA risk, our models utilize native images captured on a user’s smartphone at multiple angles -- all through a single peripheral-free platform. Here, Soliish aims to build upon the successes of their prior models for craniofacial OSA screening to include automated 2D image analysis of the oral cavity – this approach will improve the accuracy of our current OSA risk assessment models and enable us to capture features predictive of OSA likelihood that are not visible from the exterior. We will separately train and evaluate three models classifying Mallampati scores, malocclusion types, and macroglossia. Completion of these objectives will result in a comprehensive suite of models capable of analyzing anatomical structures within the oral cavity and independently predicting OSA likelihood. Success at this stage will support subsequent Phase II activities focused on amalgamating individual model predictions and evaluating cumulative model performance using data collected directly from patients in a clinical study with our partners.