This paper presents a new approach to noise reduction and contrast enhancement for different types of medical images. An anisotropic scheme is used to iteratively reduce noise as well as to define image regions and enhance region contrast. This allows an easier semiautomatic segmentation of the regions of interest. The process is performed in three stages, which consists in reducing noise, enhancing contrast and segmenting regions. The final values of the regions are automatically extracted from the image histogram, thus providing a fast method to obtain the most significant information in the image and a good approximation to region boundaries. This technique can be applied, not only to multiple region image segmentation, but also to certain processes of computer aided diagnosis which include several types of feature extraction and shape analysis.