Partial Volume Estimation and the Fuzzy C-means Algorithm
Dzung L. Pham, Jerry L. Prince · 1998
Partial volume averaging (PVA) is present in nearly all practical imaging situations, medical imaging in particular. One method that has been used to account for the effects of PVA is the fuzzy c-means algorithm (FCM). We propose a new method for estimating the partial volume coefficient of each class at each voxel in a given image using a Bayesian statistical model. A prior probability on the partial volume coefficients is used to reflect how most voxels in the image are expected to be pure. We then show that the results obtained by this method are quite similar and in some cases equivalent to results obtained using FCM. Both algorithms are demonstrated on a magnetic resonance image of the brain. 1. Introduction An important application in medical imaging is the segmentation and volumetric quantification of anatomical structures in an acquired image. Nearly all images, however, are subject to the partial volume averaging (PVA) artifact, which occurs when multiple tissues are presen...