To avoid biopsies, doctors use non invasive medical techniques such as the computed tomography. Even that, the detection of the liver remains a big challenge because of the gray level and shape variations which depend on patients and acquisition modalites. In this work, we propose to create a 3D liver model in the training phase of 3D active shape model algorithm. This training model will be deformed according to any given 3D data for liver segmentation. The contribution of our work is the use

more »

... f the Non-rigid registration with a B-spline registration on the training phase. We tested our method on an open access database ("3D-IRCADb") and on our database obtained from the radiology department of the National Oncology Institute of Tunis. Both data-sets showed the reliability of the method with an accuracy equal to 69.98% and 71.18% respectively for our database and "3D-IRCADb".