Repeated Least Squares Analysis of Simulated Xenon Computed Tomographic Measurements of Regional Cerebral Blood Flow

Computer simulations were done as a feasibility study of xenon computed tomographic measurements of regional cerebral blood flow. Accuracy of initial least squares estimates of gray matter and white matter rate constants from a two-compartment model depended very little on the number and timing of scans, but did depend significantly on the enhancement-to-noise ratio as well as on the true values of the rate constants and gray/white ratio. Nonlinear least squares gives an optimal fit of the predicted wash-in-wash-out curve to the data rather than optimal estimates of the rate constants. A polynomial correction factor was obtained by regressing initial estimates on the true values used to generate the simulations. The correction factor substantially reduced error in the estimates and, in particular, eliminated large outliers.