Multi-class Surveillance for Acute Respiratory Distress Syndrome using Belief Functions

The high incidence of pathologies implies the necessity of developing and implementing health surveillance technologies. This paper proposes a multi-class surveillance approach for a particular pathology, which is the acute respiratory distress syndrome. The multi-class model uses parameters extraction and belief functions theory applied on four vital signs. Vital signs are heart rate, respiratory rate, blood oxygen saturation and blood pressure. Thus, different linear and nonlinear parameters are extracted from these vital signs. A modeling of each class according to each parameter is performed in the framework of the belief functions theory. Then, these models are affected by a measure of confidence according to each parameter and combined together to lead finally to one model that distinguishes between the multi classes. This multi-class surveillance approach has shown interesting performances for the prediction of ARDS.