Developing hidden markov models for aiding the assessment of preterm babies- health

Physiological measurements are routinely taken for premature babies in intensive care. Because of their underdeveloped biological systems, these babies often manifest cardiorespiratory instabilities, such as lowering of oxygen level in the blood or variations in heart rate or respiratory patterns. At the same time, various paediatric illnesses may also affect cardiorespiratory functions. Therefore, the measurements that are regarded to be most useful for indicating the health of the babies are blood oxygen concentration, pulse rate and respiration rate. Health care professionals frequently monitor these physiological signals to assess the babies' health, but it can be difficult to determine whether a baby is simply premature or premature and ill. This impacts on decisions whether to treat for illnesses and, if treatment is given, also makes it hard to assess the efficacy of treatment. Furthermore, the physiological measurements are taken automatically by instruments at a rate of thirty a minute or higher, producing vast amounts of data that can be overwhelming. One way to try and improve the use of the physiological measurements and, consequently, the assessment of the babies' health, is to use mathematical models that link the measurements to state of health. We consider hidden Markov models (HMMs) because of their ability to model hidden or unobservable states of a system -in this case, a baby's state of health. We develop variants of HMMs for this purpose, and describe how inference can be performed in each case. To be useful, the resulting models must be computationally tractable and interpretable by health care professionals.