Detection of Measurement Outliers in Tracer Kinetics

In tracer kinetic studies for extracting biological information, measurement noise and error in the kinetics could affect the reliability of the estimated biological parameters. While the effect of statistical noise has been studied extensively before, the effect of outliers has not been addressed as much. In this study, we described an algorithm for detecting and removing outliers. Computer simulation was used to generate kinetic data sets corresponding to those of the FDG tracer used in PET studies to evaluate the effect of outliers and the performance of the outlier detection algorithm. Results show that outlier could increase drastically the variability of the estimated rate constants of FDG transport and phosphorylation. The outlier detection algorithm imbedded in a regular model fitting procedure was found to have a low probability of missing outliers in the kinetics. The probability of falsely identifying non-outliers as outliers was high, but these false positive detections did not affect the reliability of the biological estimates. With the outlier detection, the variability of the parameter estimates in the simulated FDG kinetics with outliers could be reduced by a factor of more than 5. The present study demonstrated the importance of outlier detection in interpreting tracer kinetics. Some areas for future studies are also discussed.