6533b7dbfe1ef96bd12700f7

RESEARCH PRODUCT

Multiple Mean Models of Statistical Shape and Probability Priors for Automatic Prostate Segmentation

subject

description

International audience; Low contrast of the prostate gland, heterogeneous intensity distribution inside the prostate region, imaging artifacts like shadow regions, speckle and significant variations in prostate shape, size and in- ter dataset contrast in Trans Rectal Ultrasound (TRUS) images challenge computer aided automatic or semi-automatic segmentation of the prostate. In this paper, we propose a probabilistic framework for automatic initialization and propagation of multiple mean parametric models derived from principal component analysis of shape and posterior probability information of the prostate region to segment the prostate. Unlike traditional statistical models of shape and intensity priors we use posterior probability of the prostate region to build our texture model of the prostate and use the information in initialization and propagation of the mean model. Furthermore, multiple mean models are used compared to a single mean model to improve segmentation accuracies. The proposed method achieves mean Dice Similarity Coefficient (DSC) value of 0.97±0.01, and mean Mean Absolute Distance (MAD) value of 0.49±0.20 mm when validated with 23 datasets with considerable shape, size, and intensity variations, in a leave-one-patient-out validation framework. The model achieves statistically significant t-test p-value<0.0001 in mean DSC and mean MAD values compared to traditional statistical models of shape and texture. Introduction of the probabilistic information of the prostate region and multiple mean models into the traditional statistical shape and texture model framework, significantly improve the segmentation accuracies.