Automated Determination of Arterial Input Function Areas in Perfusion Analysis

Perfusion in biological system refers to capillary-level blood flow in tissues, and is a critical parameter used for detecting physiological changes. Medical imaging provides an effective way to measure tissue perfusion. Quantitative analysis of perfusion studies requires the accurate determination of the arterial input function (AIF), which describes the delivery of intravascular tracers to tissues. Automating the process of finding the AIF can save operating time, remove the inter-operator variability, and correct the errors in the presence of the dispersion of the arterial system. Even though several methods are currently developed for automatically extracting an AIF, they are specific to a single modality and particular to a certain tissue.

In this thesis, we developed an algorithm to automatically determine an AIF by classifying the characteristic parameters of image pixels' dynamic evaluation curves between blood feeding areas and tissues. This automated AIF determination can be used to facilitate the generation of parametric maps for perfusion studies based on various imaging modalities and covering a variety of tissues. Automatic AIF determination was accomplished by extracting characteristic parameters such as maximum slope, maximum enhancement, time to peak, time to wash-out, and wash-out slope. Multi-dimensional data containing the characteristic parameters were converted and reduced into two-dimensional (2-D) representations, which were presented as a plurality of 2-D plots. Then physiological phases were localized within the simplified representations. Automated segmentation of non-AIF tissues and determination of AIF areas were accomplished by automatically finding peaks and valleys of each physiological phase on the plurality of 2-D plots. The algorithm was tested in CT myocardial perfusion studies, in which a pig was used as a model of myocardial ischemia and perfusion. PET gastrointestinal (GI) perfusion studies were performed using this algorithm, in which GI perfusion was evaluated when cardiac outputs were controlled with four modes. This automated AIF determination study was compared with manual selection of AIF in PET imaging and microsphere studies to assess the effectiveness of this algorithm. In the CT myocardial perfusion study, the perfusion of infarcted myocardium was significantly lower than that of non-infarcted areas and lower than that when it was normal. In the PET abdominal perfusion study, PET imaging data gives lower value of standard deviation relative to the mean than that in microsphere results. In the manual AIF selection study, a slight change in selecting the AIF region caused a big influence on the result. On the contrary, the automated AIF selection remains consistent in the entire study and reduces inter-operator variation.

A conclusion was made that this technique is applicable to several imaging modalities, such as PET, CT and MRI, and is effective on many tissues. In addition, this algorithm is straightforward and provides consistent results. More importantly, this automated AIF determination technique replaces the conventional spatial classification method with the functional classification method, taking more physiological considerations and explanations involved.