Automatic selection of arterial input function in DSC-MRI measurements for calculation of brain perfusion parameters using parametric modelling

¹ Department of Electric and Power Engineering, Electronics and Automation, Faculty of Technical Sciences, University of Warmia and Mazury, Olsztyn, Poland
² Department of Biomedical Engineering, Faculty of Electronics, Telecommunications and Informatics, Gdańsk University of Technology, Gdańsk, Poland

^* Corresponding author: seweryn.lipinski@uwm.edu.pl

Accepted: 4 December 2018

Abstract

A novel method and algorithm of automatic selection of arterial input function (AIF) is presented and its efficiency is proved using exemplary DSC-MRI measurements. The method chooses AIF devoted to a particular purpose, which is calculation of perfusion parameters with the use of parametric modelling of DSC-MRI data. The quality of medical diagnosis made on the basis of perfusion parameters depends on the quality of these parameters, which in turn is determined by the quality of the AIF signal. The proposed algorithm combines physiological requirements for AIF with mathematical criteria. The choice of parametric approach, instead of black-box modelling, allows better understanding of the investigated system functioning, as model parameters may be credited with physical interpretation. Furthermore, using multi-compartmental model of the DSC-MRI data with AIF regression function in an exponential form, gives direct, analytic results concerning the basic descriptors of AIF. The method chooses candidates for AIF on the basis of the descriptors quality. This step allows rejecting measurements which do not fulfil fundamental requirements concerning AIF from the physiological point of view. As these requirements are met, the next criterion can be adopted, that is the quality of fitting the regression function to measurements. The final step is choosing the AIF for calculating perfusion parameters with the best accuracy, which is attainable thanks to implementing the AIF devoted particularly to parametric modelling.