Segmentation Of Hippocampus Based On ROI Atlas Registration

Hippocampus belongs to limbic system of brain and the change of hippocampus’volume has a close relationship with many nervous and mental diseases, such asAlzheimer’s disease (AD), epliepsy, major depression, and schizophrenia. Therefore,the study of hippocampal volume can provide useful outcome clinical trail of mentaldiseases.Magnetic resonance (MR) images of brain can provide more information aboutsoft tissue and have high contrast of soft tissue. MR image can be used to study therelationship between the disease and hippocampus volume. Many methods can beused to segment MR images.Atlas-based segmentation performs registration between an atlas image and targetimage that is to be segmented. It only requires a registration framwork and a numberof pre-segmentation data without landmarkings and complex training procedures.Individual differences among variability of brain images and mislabeling can bothaffect the segmentation accuracy. The volume of hippocampus is about2~3ml, whichis hard to segment. A novel multi-atlas segmentation method based on ROI image ispresented according to the texture and structure characters of hippocampus.By making use of multiple atlas image registrated to template image, multi-atlasbased segmentation can generate several initial segmentation results. Combination ofthe initial segmentation can achieve finally segmentation. Registration framework andatlas combination strategy are key steps of multi-atlas based segmentation. This paperfocuses on the atlas combination strategy and compares the majority voting, STAPLE(Simultaneous Truth Performance Level Estimation) and COLLATE (Consensus Level,Labeler Accuracy and Truth Estimation). Evaluation of segmentation accuracy isperformed by comparing the segmentation result to manual segmentation with DICEsimilariy coefficient. The experimental results prove that our method is morepracticable and effective.