A content based retrieval method for MR brain images

Content-based medical image retrieval allows exploring same images appearance with different diagnosis. It also allows searching through large collections of disease-related illustrations using their visual attributes. This provides a very convenient environment for practitioners to benefit from the retrieved images. Approaches that are based on one-to-one regions of image matching are sensitive to rotation, shifting and scaling. In an improvement effort in this direction, we propose a new approach based on Hungarian algorithm which compares one window from the query image to all windows in each image in the image database to return the closet matching. This comparison is based on features vector of gray-level intensity for each window. We used K-mean clustering algorithm to separate each window into K clusters. Based on the histogram of each cluster, the mean, variance and skewness are computed. We believe that the proposed content-based medical image retrieval approach will be helpful in assisting practitioners in the decision making process in clinical pathology.