Hyperspectral mathematical morphology of DCE-MRI series for angiognesis imaging
G. Noyel (Centre de Morphologie Mathématique-Ecoles des Mines de Paris, France), J. Angulo (Centre de Morphologie Mathématique-Ecoles des Mines de Paris, France), D. Jeulin (Centre de Morphologie Mathématique-Ecoles des Mines de Paris, France), D. Balvay (C-A. Cuenod LRI-EA4062 Paris V Descartes, APHP/HEGP, Service de Radiologie Paris, France)
In the sequel we will present a complete treatment chain for DCE-MRI (dynamic contrast enhanced-MRI series) using multivariate mathematical morphology. The series are seen as hyperspectral images, i.e. to each pixel of a 2D image is associated a vector, the spectrum, with values in time. The treatment is performed on 25 series of small animals in angiogenesis imaging.
First of all denoising and data reduction methods based on data analysis and model approach are performed on the series. Then stochastic watershed segmentation is applied to reinforce anatomical contours. Finally computer aided detection is computed to detect the tumours and confidence maps of tumours detection are produced.