1Diagnostic Radiology, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong, China; 2Radiology, Kwong Wah Hospital, Hong Kong, China; 3Philips Healthcare, Hong Kong, China
To more precisely segment high-cellularity tumor tissues in heterogeneous lesions and therefore more accurately measure volumes and ADCs, we proposed a semi-automatic method based on thresholding both the b0 images and the ADC maps. Using k-means clustering algorithm, B0 images and ADC maps in the contoured regions were separately classified into three clusters (with low, intermediate and high value). The pixels with low intensities on b0 images and those with high ADC values on ADC maps were excluded, leaving only the probable high-cellularity tumor tissues. The volumes measured using the proposed method had perfect concordance with those in PET/CT. Furthermore, stronger correlations between ADC values and SUV values were achieved using this method.