A medical image retrieval framework

Many investigators are incorporating medical image feature analysis into computer-aided diagnosis (CAD) systems to increase the precision and accuracy of characterization by radiologists. Searching medical databases for images similar to a given query image that corresponds to a case under current study and enabling access to those other clinical data and known diagnoses from those similar cases is expected to have great impact in CAD systems. However, efficiently and accurately searching for similar medical images in database systems is a very challenging task. In this paper, we propose a two-step content-based medical image retrieval framework. A candidate subset is first created utilizing the wavelet decomposition. The actual retrieval process is then constrained within this candidate subset. Besides the improved efficiency due to the reduced searching space, this framework also leads to improved retrieval accuracy, as demonstrated with our experimental results.