Extraction and application of deformation-based feature in medical images

Abstract One of the most challenging issues that hinder the development of accurate medical image segmentation system is the insufficiency of features which are relevant to actual anatomical meaning from the images. Although deformation of normal structures caused by compression from abnormal structures has usually been considered as undesired or even a problem to be tackled in medical image segmentation tasks, it is actually relevant to the correlation between normal and pathological structures. With the objective of investigating the feasibility of extracting and applying deformation-based features of such type, we propose an approach to estimate feature from the correlation between brain lateral ventricular (LaV) deformation and tumor and apply the extracted feature for computerized magnetic resonance (MR) image tumor segmentation. Experimental results on feature extraction show the relevancy between LaV deformation and location of tumor; comparative experiments on tumor segmentation suggest that, in most cases, tumor segmentation accuracy improves when the extracted feature is applied.