Absolute cerebral blood flow with 15O-water and PET: determination without a measured input function

PET cerebral blood flow (CBF) methods require tissue and arterial blood radioactivity measurements to yield absolute values. The authors have developed a method to estimate CBF without a measured input function. For N pixels and M scan frames, the authors estimate N+M parameters (N flow values and M input function integrals) from N×M measurements with weighted least squares using the iterative Gauss-Newton (GN) algorithm. Tracer distribution volume is assumed to be known. This method was tested with simulated and human image data. Simulation GN errors in whole brain CBF were -3±2%, with uniform percent errors for all flow values. GN image quality was comparable to that obtained from algorithms which require the measured input function. Results with actual scan data (8 subjects, 4 studies each) had errors in global flow of -77±3% due to violations of the model assumptions, particularly tissue heterogeneity. Use of a modified algorithm which included inter-pixel variations in the distribution volume to account for heterogeneity reduced the bias but the results are overly sensitive to the assumed value of distribution volume variability. Although this method can theoretically provide absolute CBF, it will be useful in practice only if its large sensitivity to model inaccuracies can be controlled