Efficient content-based CT brain image retrieval by using region shape features

Presents an efficient content-based computed tomography (CT) brain image retrieval system for medical training or diagnosis. We focus on the analysis of lateral ventricle features to help physicians to distinguish hydrocephalus and atrophy from normal cases. Our system is composed of four modules: (1) ventricle region segmentation, (2) feature computation, (3) feature indexing, and (4) database search. In the first module, we apply the Otsu's (1979) thresholding plus gradient vector flow snake to segment out ventricles. Four ventricle features are computed automatically based on the segmented result and then used for database indexing and search. The Linde-Buzo-Gray (1980) algorithm is employed to cluster the sample feature vectors for constructing an efficiently indexed database. In database retrieval, we propose a fast matching algorithm based on feature ranking or ordering to speed up the search significantly. Experiments show that our system is capable of achieving 30% of computational load with respect to the traditional database system.