Probabilistic modeling of the oxygen saturation pattern for the detection of anomalies during clinical interventions

In this paper, we propose a Markov model-based methodology aimed at detecting in real time the anomalies that the oxygen saturation pattern suffers during clinical interventions or procedures. To this end, we first extract a reference pattern from the patient in nominal conditions before the procedure takes place. Then, in a second stage, a measurement of the similarity between the reference pattern and the pattern of the epoch to be tested is obtained through the Williams' Index. This measurement is compared with a threshold to determine the normal/abnormal character of the pattern under test. Experiments on real data show that the proposed methodology is sensitive to the anomalies induced when the respiratory function is impaired; this is accomplished through the simulation of several situations (shortness of breath, interrupted breathing, hyperventilation and CO2 increasing in blood) in which the respiratory impairment is manually emulated.