Using hidden Markov models for sleep disordered breathing identification

In this work, an automatic diagnosis system based on hidden Markov models (HMMs) is proposed to help clinicians in the diagnosis of sleep apnea syndrome. Our system offers the advantage of being based on solid probabilistic principles rather than a predefined set of rules. Conventional and new simulated annealing based methods for the training of HMMs are incorporated. The inference method of this system translates parameter values into interpretations of physiological and pathophysiological states. The interpretation is extended to sequences of states in time to obtain a state-space trajectory. Some of the measurements of the respiratory activity issued by the technique of polysomnography (brain activity, respiratory activity, oxygen levels, and cardiac activity) are considered for off-line and on-line detection of the different sleep apnea syndromes: obstructive, central and hypopnea. Experimental results using respiratory clinical data and some future perspectives of our work are presented.