Abstract: The manual method for the detection of arterial input function(AIF)in cerebral perfusion based on DSC-MRI technique was not only time-consuming but also user-dependent， meanwhile， the accuracy and reproducibility were not very satisfactory. To solve this problem， a semi-automatic AIF detection method based on k-means cluster analysis is suggested. The pixels in the region of interest(ROI)were divided into several clusters and the mean curve of each cluster was calculated. A measure，［peak value /(time to peak× full width at half maximum)］， was calculated for each mean curve， and the one with the maximum measured value was used to determine the AIF. Twenty subjects were taken part in the research. By comparing with the result derived from the traditional manual method， the clinical feasibility was validated. The result demonstrated that the AIF obtained from the semi-automatic method based on k-means cluster analysis was superior to that based on traditional manual method. In conclusion， the semi-automatic selection of AIF based on the k-means cluster analysis can not only reduce the analysis time and observer dependence， but also improve the calculation accuracy and reliability.

Key words: cerebral perfusion, dynamic susceptibility contrast, arterial input function, hemodynamic parameters, cluster analysis