Automated PAP mask sizing: a pathway for future mask designs

Abstract

For suffers of sleep apnoea, Positive Airway Pressure (PAP) therapy is the gold standard in treatment and with regular use is extremely effective in preventing the symptoms as well as reducing the occurrence of other comorbidities. Regular usage however can be challenging, particularly during the first few weeks when many patients either sporadically use adhere to therapy or discontinue treatment altogether and often cite issues related to PAP mask fit such as air pressure leaks or skin irritations as the reasons for quitting.The work presented in this thesis document provides a pathway for improving clinical outcomes through a readily available and completely automated sizing process. This project has developed several methods for automatically determining a patient’s most appropriate mask size using a single facial image as an input. Each method demonstrates the progression from a semi-automated system requiring some degree of human input, to where no manual intervention is required. These systems, built using machine learning models and a generic dataset produced a single nasal PAP mask size accuracy of 62% and a within one size mask sizing accuracy of 95% when compared to the manually allocated sizes. These studies demonstrated the potential of an automated mask sizing system and provides a pathway for developing a clinically applicable methodology.