How to choose biomarkers in view of parameter estimation.

In numerous applications in biophysics, physiology and medicine, the system of interest is studied by monitoring quantities, called biomarkers, extracted from measurements. These biomarkers convey some information about relevant hidden quantities, which can be seen as parameters of an underlying model. In this paper we propose a strategy to automatically design biomarkers to estimate a given parameter. Such biomarkers are chosen as the solution of a sparse optimization problem given a user-supplied dictionary of candidate features. The method is in particular illustrated with two realistic applications, one in electrophysiology and the other in hemodynamics. In both cases, our algorithm provides composite biomarkers which improve the parameter estimation problem.