Reconstruction of underlying physiological structures from a sequence of images is a long-standing problem which has been solved by factor analysis with a success. This paper tries to return to roots of the problem, to exploit the available findings and to propose an improved paradigm.