Non Negative Matrix Factorization for Time Series of Medical Images Analysis

The application of non negative matrix factorization to time series of medical images analyze is investigated in this paper. Time series images of the urinary system are acquired by intravenous pyelography (IVP). Factorial analysis and principal component analysis are often used to extract time signatures or factors and associated compartments of factor images. Blind source separation methods such as independent component analysis (ICA) may be an alternative to these methods for which the orthogonality condition is replaced with a more general constraint: the independence. Since the positivity constraint must not be ignored, we focused on the non negative matrix factorisation approach. More than that, there are situations where only few units of either factor loads and/or factor images are effectively used to represent observed data vectors. In this case, sparseness constraint must be taken into account.