Texture-based 3d image retrieval for medical applications

Although content-based image retrieval (CBIR) has been researched for more than two decades, retrieving 3D datasets has just begun recently and is mainly focusing on 3D shapes, i.e., iso-surfaces of objects. This is in part due to the fact that the major application area of CBIR is to those images available over the internet. In the medical domain however, more and more images are in three or more dimensions with important information spreading from both shapes to anatomic locations. This study aims to help surgical planning for image-guided neurosurgery, by which the anatomic location plays more important role than the shape of a tumour. Texture-based methods are therefore exploited with four approaches including 3D Grey Level Co-occurrence Matrices (3D GLCM), 3D Wavelet Transform (3D WT), 3D Gabor Transform (3D GT) and 3D Local Binary Pattern (3D LBP) working on a database consisted of around 100 image volumes with both normal and lesioned brains. Preliminary results have shown that 3D LBP performs best with precision recall of 65% and processing time of less than 1 second for both feature extraction and retrieval.