An medical image feature extraction algorithm based on texture shape and texture information is proposed. The Contourlet transform combines non-separable and directional filters banks, and has multiscale and directional properties. It is used for texture feature extraction. As a region-based shape descriptor, Zernike moments have good rotation invariance. It is used for shape feature extraction. Then, combining the two features . In order to verify the effectiveness of the algorithm, a content-based medical image retrieval prototype is built. Experimental results show that it improves the retrieval accuracy of the system.