Multi-scale Medical Image Segmentation Based on Salient Region Detection

In this paper, we introduce an adaptive medical image segmentation algorithm based on salient region detection. By implementing Gaussian Mixture Models and classifying the output in correlation comparison, our method is capable to localize the Region of Interest in multi-scales without adjusting any parameters. Two different types of medical image datasets, containing different scales of medical images are used for evaluate the method, comparing the other six general segmentation algorithms. Experiments prove that our method can self-adapt for different images, and outperform the other algorithms in all of the global, regional and vessel scales.