Improving feature extraction for automatic medical image categorization

Medical image annotation remains a challenging task. Many feature schemes have been experimented with limited success. In this paper, we propose to improve the image categorization prediction through the employment of better feature schemes assessed with feature analysis. A new edge descriptor based on the Canny detector is proposed along with modified MPEG-7 features. Some preliminary results are presented, clearly indicating the improved effectiveness of these feature schemes. Finally, we argue that perhaps only with more involvement of semantic analysis the research on medical image categorization can make clinical significance.