Prof. Joanna Polanska will give a talk on October the 2nd
by Susanne Bornelöv
My opponent Prof. Joanna Polanska from the Silesian University of Technology will give a talk tomorrow (October 2) at 14:15 in the meeting room at the B7:2 corridor.
Title:DELINEATION OF BRAIN TUMOUR MARGINS USING GAUSSIAN MIXTURE MODEL BASED SEGMENTATION OF THE APPARENT DIFFUSION COEFFICIENT MAPS
Abstract: The automatic detection of tumour tissue on apparent diffusion coefficient (ADC) map is of big importance for both cancer diagnosis and therapy planning. The main goal of the study is to develop a method for data driven ADC threshold estimation allowing for delineation of tumour margins. The proposed approach relies on mathematical modelling of ADC signal distribution in the form of Gaussian mixture (GM), combined with the standard unsupervised clustering technique. The developed method was applied to the clinical dataset. Training dataset consists of 26 diffusion weighted images (DWIs) of patients with different types of brain tumour (dataset A). The estimated ADC threshold was validated on two additional datasets – 26 another DWIs from the same patients (dataset B) and 24 DWIs from another patients (dataset C), and it allowed for tumour identification with average Dice index equal to 95.24%, 95.46%, and 78.70% for datasets A, B, and C respectively. The comparison to other segmentation methods used for the ADC map analysis ranks the proposed GMM based method among the most accurate techniques.
Keywords: ADC maps, brain tumour, DWI MR, Gaussian mixture model, k-mean