Image Segmentation

The mean photon flux from the BLI data was calculated over an automated region of interest (ROI) using M3Vision software (Biospace Lab, France). All data sets acquired with US, low-field MRI and high-field MRI were manually segmented into tumour and normal liver tissue by two researchers with at least four years of preclinical imaging experience, using Amira 5.4 (FEI, Oregon, USA). Segmentation was performed blind, following randomisation of the day and mouse number.

Whole liver volumes from low- and high-field MRI data were analysed. However, whole liver analysis wasn’t practically feasible for US images: the three-dimensional acquisition provided a very fine step thickness, and sagittal coverage of the whole liver typically yielded in excess of 250 slices. To enable segmentation within a reasonable timeframe, the data were compressed such that every three slices were averaged into one to yield a final slice thickness of 0.228 mm. In order to match the number of slices segmented by MRI, three representative samples of ten consecutive slices were chosen by each user, to account for heterogeneity of tumour engraftment. The relative tumour burden was resultantly used to compare the MRI and US for analysis of model development, calculated as the ratio of observed tumour to observed liver volume: