Use of Tracer Kinetic Models for Selection of Semi-Quantitative Features for DCE-MRI Data Classification

The aim of this study was to identify, on the basis of simulated tracer kinetic data, the best subset of semi-quantitative features suitable for classification of dynamic contrast-enhanced magnetic resonance imaging data. 1926 time concentration curves (TCCs) of Type III, IV and V [according to the classification of Daniel et al. (Radiology 209(2): 499-509 (1998))] were simulated using the gamma capillary transit time model and the Parker's arterial input function. TCCs were converted in time intensity curves (TICs) corresponding to a gradient echo sequence. Seventeen semi-quantitative shape descriptors were extracted from each TIC. Feature selection in combination with classification and regression tree was adopted. Several acquisition parameters (total duration, time resolution, noise level) were used to simulate TICs to evaluate the influence on the features selected and on the overall accuracy. The highest accuracy (99.8 %) was obtained using 5 features, total duration 9 min and time resolution 60 s. However, an accuracy of 93.5 % was achieved using only 3 features, total duration 6 min and time resolution 60 s. This latter configuration has the advantage of requiring the smallest number of features (easily understandable by the radiologist) and not a very long duration (reduced patient discomfort).