Evaluation of the Role of Low Level and High Level Features in Content Based Medical Image Retrieval

Abstract

The Content based medical image retrieval, which aims at searching the image database using invariant features, is an important research area for manipulating large amount of medical images. So designing and modeling methods for medical image search is a challenging task. This paper proposes an approach by combining DICOM header information (high level features) and content features (low level features) to perform the retrieval task. A novel approach of rotation invariant contourlet transform (CT) is proposed for texture feature extraction and fixed resolution format is used to derive the shape features. Initially the DICOM header information is extracted which is used to perform a pre-filtering on the original image database. Content based search is performed only on these pre-filtered images which speed up the retrieval process. The retrieval performance of this method is tested using a large medical image database and measured using commonly used performance measurement.