BOLD-contrast functional MRI (fMRI) has been used to assess the evolution of tumor oxygenation and blood flow after treatment. The aim of this study was to evaluate K-means-based cluster analysis as a exploratory, data-driven method. The advantage of this approach is that it can be used to extract information without the need for prior knowledge concerning the hemodynamic response function. Two data sets were acquired to illustrate different types of BOLD fMRI response inside tumors: the first set following a respiratory challenge with carbogen, and the second after pharmacological modulation of tumor blood flow using flunarizine. To improve the efficiency of the clustering, a power density spectrum analysis was first used to isolate voxels for which signal changes did not originate from noise or linear drift. The technique presented here can be used to assess hemodynamic response to treatment, and especially to display areas of the tumor with heterogeneous responses.
Copyright 2003 Wiley-Liss, Inc.