Computer-Aided Prognosis Based on Temporal Dependencies

Data analysis and mining, through computer-based systems, achieved particular interest in the area of healthcare in the last years, where it has been shown to produce high levels of accuracy. The variety of available techniques is nowadays applied interchangeably with success on almost every healthcare domain, but mostly for diagnosis. Indeed, the results achieved on prognosis through the same techniques are much more modest. In this paper, we argue that the difference of success on diagnosis and prognosis, by mining techniques, is mainly due to the inadequacy of those techniques for dealing with the inherent temporal information attached to clinical data. Moreover, we discuss a new approach to address this issue, independent of the domain, and present some preliminary results on two different datasets.