School of
Information Technology and Electrical Engineering

Modelling Instantaneous Characteristics and Temporal Dynamics of Upper Airway using Hidden Markov Models

Dulip HerathMon, 21/10/2013 - 10:00
A/Prof Udantha Abeyratne

Obstructive Sleep Apnea (OSA) is a highly prevalent breathing disorder with serious consequences. It is caused by the partial or full obstruction of the upper airway during sleep. The Polysomnography (PSG) is the standard reference diagnosis study for OSA. The main outcome of PSG is Apnea/Hypopnea Index (AHI) which is the average number of apnea (full) or hypopnea (partial) obstructions that occur per hour.  Alternatively, snoring sound based OSA diagnostic methods have been proposed by the biomedical engineering community. This performance of these methods has nearly approached the gold standard, i.e. AHI from PSG. However, these methods can provide only a summarized single-point description about the disease severity of OSA.

With the intention of going beyond AHI-based diagnosis, we are planning to develop modes to describe the instantaneous characteristics of the upper airway to capture its temporal dynamics. During a breath cycle, a number of upper airway characteristics are dynamically changing for example intrapulmonary pressure, airflow rate, upper airway cross-sectional area and volume. These changes affect the snoring and breathing sounds as well. Therefore, we expect to characterize the upper airway by using the intra-episodic snoring or breathing sounds. The intra-episodic behaviour of snoring or breathing sounds are modelled by using Hidden Markov Model (HMM) with Mel Frequency Cepstral Coefficients (MFCC). 


Dulip Herath holds a BSc in Computer Science (2004) from the University of Colombo, Sri Lanka. He also holds an MA in Linguistics (2006) from the University of Kelaniya, Sri Lanka and an MPhil in Speech & Language Processing (2008) from The Computer Laboratory, University of Cambridge, UK. For the past 10 years he has been working in the area of speech and language processing, statistical machine translation, linguistic resource development for local languages in Sri Lanka.

Seminar Type: 

PhD Confirmation Seminar