Abstract:

Data mining methods are widely used to analyze medical images in current research. Commonly used mining methods first need to extract features from medical images and then do classification analysis. At present, the statistical features extracted from images are mostly applied, however, it has a strong dependence on the extracted features. We propose a new classification method based on convolution restricted Boltzmann machine (CRBM), which can train the CRBM model by the fast continuous contrastive divergence algorithm. The method can directly and automatically learn features from the mammography image and use these features to do classificature. Experimental results show that the proposed method can improve the classification accuracy of medical images.

Key words: medical image classification, convolution restricted Boltzmann machine, fast continuous contrastive divergence, accuracy of classification