Sci-Sat AM(1): Imaging-06: Proximity-based modification to an automatic method for tumor delineation using MRSI

Quantifying the relative levels of Choline (Cho) to N-Acetylaspartate (NAA) has been the method of choice by many groups as a mean to biologically identify tumors in the brain. Mcknight et al. have introduced an automatic technique for delineating tumors biologically using MRSI. A statistical model is used to separate tumors from normal tissue based on the relative concentrations of Choline and NAA in both tissue types; that method is commonly referred to as the Choline-to-NAA Index (CNI). In their work, it is assumed that the variation in the relative levels of Cho to NAA in normal brain is unnoticeable to within 2 standard deviations of the mean. However, developments in MRSI sequences have enabled the detectablity of more variations within the relative levels of Cho to NAA in normal tissue. With the uncertainty in the Cho to NAA levels of normal tissue increasing, it is essential to modify the CNI method to improve its specificity. This work introduces a modification to the CNI method developed by McKnight et al. that would address such increase in uncertainty. Instead of relying on an arbitrary CNI value of 2 to define the tumor boundaries, our method defines a high certainty tumor volume, surrounded by a region of uncertainty. Then based on their proximity to high certainty tumor regions, the voxels in the uncertainty region are segmented to either tumor or normal tissue. Preliminary results suggest that the proposed modified method decreases the number of false positive resulting from the original CNI method.