Multi-scale feature based medical image classification

In order to describe the characteristics of medical image more fully in different scales and solve the problem of automatic image category annotation, multi-scale feature based medical image classification is discussed. A set of complementary image features in various scales, including gray-level, texture, shape features and features extracted in the frequency domain is used. An ensemble learning based classification framework is proposed and applied to the medical image classification task with the feature extracted. The features and their combination are used for classification and the most commonly used classifiers are chosen to compare the results of classifications. The experiment results show that, generally, the proposed classification approach with multiple complementary features has achieved higher accuracy than traditional medical image classification methods.