Estimation of Reference Tissue Based Arterial Input Function Using Neural Network

Jin Zhang¹, Alana Amarosa¹, Andrew B. Rosenkrantz¹, Sungheon Kim¹

Arterial Input Function (AIF) plays an important role in pharmacokinetic model analysis of dynamic contrast enhanced (DCE)-MRI data. Reference tissue approaches have been proposed as a means to estimate AIF using a two compartment model and literature values of transfer constant and interstitial space volume. In this study, neural network was introduced to delineate reference tissue concentration curve and consequently to reconstruct AIF. Clinical and preclinical DCE-MRI studies were performed to compare different reference tissue approaches. The results demonstrated the proposed method has improved reproducibility as well as flexibility in estimating different shapes of AIF.