Meeting Banner
Abstract #1087

Combining Nonlinear Least Squares & Random Forest Regression to Increase the Accuracy & Precision of DCE-MRI Tracer Kinetic Model Parameter Estimates

Jakub Palowski1,2, Chris James Rose1,2

1The University of Manchester Biomedical Imaging Institute, the University of Manchester, Manchester, Greater Manchester, United Kingdom; 2Manchester Academic Health Science Centre, Tthe University of Manchester, Manchester, Greater Manchester, United Kingdom


Nonlinear regression is commonly used in MRI to estimate physiological quantities that cannot be measured directly, but can be modelled as a function of measureable phenomena: e.g., modelling the tracer kinetics of contrast-agents as in DCE-MRI. We present a novel nonlinear regression algorithm that estimates the parameters of the extended Tofts model, based on a machine learning method called random forests. Using simulated gadopentate dimeglumine (Gd-DTPA) concentration time series we show that, compared to conventional least squares, the proposed method can estimate all model parameters with greater accuracy and comparable precision, and in less time.