Abnormality detection of specific brain structure in MR images based on multi-atlas and texture descriptor

An abnormal brain structure might contain different types of lesions with different shapes, areas or textures, and the same normal brain structure in different persons might exhibit individual differences in shape or tissue texture. This makes it a challenging problem to judge whether or not a brain structure in a MR image is normal. In this paper, we present a framework for abnormality detection of the brain structure of interest (IBS) in MR images based on the multi-atlas and the texture descriptor. In the framework, a set of normal brain MR images are first collected, where different images are aligned and the atlases of different brain structures are marked in each image. Then, multi-atlas information is used to locate and segment the IBS in the test MR image. Subsequently, textural features of the IBS are computed in the test image and in each one of the collected normal images. Finally, by analyzing whether or not there is a significant difference between textural features of the test image and those of the collected images, a conclusion about the abnormality detection is drawn. This framework has been applied to MRI brain images with different abnormal subjects, and detection results are acceptable.