Indicator dilution theory provides a framework for the measurement of several cardiovascular parameters. Recently, dynamic imaging and contrast agents have been proposed to apply the method in a minimally invasive way. However, the use of contrast-enhanced sequences requires the definition of regions of interest (ROIs) in the dynamic image series; a time-consuming and operator dependent task, commonly performed manually. In this work, we propose a method for the automatic extraction of indicator dilution curves, exploiting the time domain correlation between pixels belonging to the same region. Individual time intensity curves were projected into a low dimensional subspace using principal component analysis; subsequently, clustering was performed to identify the different ROIs. The method was assessed on clinically available DCE-MRI and DCE-US recordings, comparing the derived IDCs with those obtained manually. The robustness to noise of the proposed approach was shown on simulated data. The tracer kinetic parameters derived on real images were in agreement with those obtained from manual annotation. The presented method is a clinically useful preprocessing step prior to further ROI-based cardiac quantifications.