Content-based image retrieval (CBIR) technique for digital image searching has been applied in medical images which are called content-based medical image retrieval (CBMIR). In this paper, we combine texture, density and shape features for CBMIR based on 3D multi-phase contrast enhanced CT images according to radiologists' clinical experience. And a database of 132 focal liver lesions (FLLs) with confirmed pathology type is constructed in this work. We implement bag of visual words (BoVW) model to extract texture features from FLLs based on 3D local binary pattern (LBP) and combine it with conventional intensity-based BoVW. Density and temporal density are designed based on clinical observation. Principle Component Analysis (PCA) is used to extract the sphericity of the lesions as shape features. Experiments performed on the FLLs database show the efficiency of the proposed system and features.
Financed by the National Centre for Research and Development under grant No. SP/I/1/77065/10 by the strategic scientific research and experimental development program:
SYNAT - “Interdisciplinary System for Interactive Scientific and Scientific-Technical Information”.