Abstract. An efficient clinical image segmentation framework is proposed by combining a pattern classifier, hierarchical and coupled level sets. The framework has two stages: training and segmentation. During training, first, representative images are segmented using hierarchical level set. Then the results are used to train a pattern classifier. During segmentation, first the image is classified by the trained classifier, and then coupled level set functions are used to further segment to get correct boundaries. The classifier provides an initial contour which is close to correct boundary for coupled level sets. This speeds up the convergence of coupled level sets. A hybrid coupled level set method which combines minimal variance functional and Laplacian edge detector is proposed. Experimental results show that by the proposed framework, we achieve accurate boundaries, with much faster convergence. This robust autonomous framework works efficiently in a clinical setting where there are limited types of medical images.