Gaussian Mixture Model for Texture-based Medical Image Analysis

L. Tesar, H. Kobatake, A. Shimizu (Japan), and D. Smutek (Czech Republic)


Gaussian mixture model, medical imaging, texture analy sis.


In this paper, a method for medical diagnosing using Gaus sian mixture model is presented. Gaussian mixture model is used for clustering of texture feature vectors. Medical diagnosing described here consists of learning part and in ference part. In learning part knowledge is accumulated from cases with known diagnosis, which involves mixture model parameters estimation for every diagnosis consid ered. Inference is done by calculating the mixture value and maximizing it over every diagnosis considered. EM al gorithm is used for Gaussian mixture model parameters es timation. For every possible diagnosis, parameters of one mixture model have to be calculated. To demonstrate the method, we present the experiment with real data, diagnos ing the common liver lesions using computed tomography (CT) images.

Important Links:

Go Back