Abstract
Deuterium oxide (D2O) had been used as contrast agent in early perfusion MRI due to its diffusibility and nontoxicity. However, the detection of D2O in MRI is limited to its sensitivity and the signal-to-noise ratio of perfusion image is low. Recently, a novel contrast-enhanced strategy has been proposed to address this issue. D2O is indirectly detected by monitoring the signal change of 1H signal after D2O administration. In order to quantify the cerebral blood flow precisely, a reliable arterial input function (AIF) is essential. Thus we aim to develop an automatic process in AIF selection to avoid the bias from manual selection. K-means cluster analysis was used to determine the most suitable AIF of D2O perfusion in mice experiments. The selected AIF time curve was then fitted using bi-exponential model to obtain a derived AIF [19-20]. However, we found that the wash out part of the derived AIF has a high concentration level. As a result, we extracted the fast rate constant from the bi-exponential model to generate a novel AIF with mono-exponential decay and compared their fitting performances. The AIF was then used to calculate the perfusion parameters by both model-dependent and model-free method. Our result shows that two-compartment model with fast-exponential AIF could describe the D2O perfusion properly. A similar result can be found in the analysis by SVD method. This could be reasonable since the shape of fast-exponential AIF is similar to unit impulse response (delta function). The deconvolution process with fast-exponential AIF can bring to a more accurate tissue residual function. In summary, our method has improved the selection process and also provides more accurate perfusion information.