Segmentation of Magnetic Resonance Images According to Contrast Agent Uptake Kinetics Using a Competitive Neural Network

The kinetics of the uptake of a magnetic resonance (MR) contrast agent (gadolinium—DTPA) can be monitored via the effect on signal intensity in a series of MR images. This can be used to estimate blood flow in normal tissues or tumours. We have used a competitive neural network approach to identify image pixels with similar patterns of contrast agent kinetics. This allows automatic (and un-supervised) visualisation of regions with similar blood flow. Averaging of contrast agent kinetics over self-similar pixels was useful for fitting to model functions to obtain quantitative measures of blood flow in untreated tumours and in tumours treated with the anti-vascular drug, combretastatin.