Improving apparent diffusion coefficient estimates and elucidating tumor heterogeneity using Bayesian adaptive smoothing.

The spatial distribution of apparent diffusion coefficient (ADC) estimates in tumors is typically heterogeneous, although this observed variability is composed of both true regional differences and random measurement uncertainty. In this study, an adaptive Bayesian adaptive smoothing (BAS) model for estimating ADC values is developed and applied to data acquired in two murine tumor models in vivo. BAS models have previously been shown to reduce parameter uncertainty through the use of a Markov random field. Here, diffusion data acquired with four averages was used as an empirical gold standard for evaluating the BAS model. ADC estimates using BAS displayed a significantly closer accordance with the gold standard data and, following analysis of uncertainty estimates, appeared to even outperform the gold standard. These observations were also reflected in simulations. These results have strong implications for clinical studies, as it suggests that the BAS postprocessing technique can be used to improve ADC estimates without the need to compromise on spatial resolution or signal-to-noise or for the adaptation of acquisition hardware. A novel measure of tumor ADC heterogeneity was also defined, which identified differences between tumors derived from different cell lines, which were reflected in histological variations within the tissue microenvironment.