Automated determination of arterial input function in DCE-MR images of the kidney

This paper concerns the problem of estimating renal perfusion based on the Dynamic Contrast Enhanced MRI. Quantification of perfusion parameters is possible by the means of pharmacokinetic modeling. Several mathematical formulations of PK models have been proposed. In any case, it is important to determine the so-called arterial input function, i.e. the time-course of the contrast agent bolus in a main feeding artery. In case of the kidney it is the descending aorta. Usually, determination of AIF is performed manually. We propose the automatic procedure to determine AIF, thus reducing the involvement of a human observer in the image processing pipeline. Our proposed method uses a combination of image processing and machine learning algorithms firstly to identify all voxels potentially belonging to the descending aorta and secondly to select those voxels which are free from the inflow artifact. The tests of our method performed for 10 DCE-MRI datasets show its effectiveness in terms of the resulting perfusion parameters measurements.