Deep-Learning Estimation of Second-Generation Pharmacokinetic-Model Parameters in DCE-MRI

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is a promising method for the evaluation of tissue perfusion. Current standard is fitting of a pharmacokinetic model to the acquired signals. Most commonly, first generation models are used (Tofts, extended Tofts model) providing stable results, however, only a limited set of parameters. Second generation models allow estimation of a larger parameter set, thus a more complete description of the perfusion state, however, they require high data quality and their application is more computationally demanding. Overall, the lack of standardization of DCE-MRI, its computational time and reliability hinders its routine clinical application. Deep learning methods allow fast parameter estimation and bring new possibilities into this field. In this study, we have explored the application of a convolutional neural network for the prediction of second-generation model parameters. The network was tested for different noise levels and sampling periods on a simulated dataset, and the results were validated on a real preclinical dataset. The proposed method provided more stable and robust results compared to the conventional model fitting.