Entropy-Based Automatic Segmentation and Extraction of Tumors from Brain MRI Images

We present a method for automatic segmentation and tumor extraction for brain MRI images. The method does not require preliminary training, and uses an extended concept of image entropy. The latter is computed over gray levels (which are in fixed number) instead of single pixels. The obtained measure can be assumed as a measure of homogeneity/similarity of regions, and therefore be the base for image segmentation. Being independent from the number of pixels in the image, its computation is highly scalable with respect to image resolution. As a matter of fact, it is always carried out over the fixed number of 256 gray levels. Moreover, being region-based rather than pixel-based, the measure is also more robust to slight differences in orientation.