Data-Driven Timing Calibration of TOF PET

Improved performance has been shown in time-of-flight (TOF) PET compared with conventional PET. However, constant timing calibrations are required to keep optimum system performance. Data-driven timing calibrations have the potential to reduce the number of required phantom-based calibrations. Non-TOF image reconstruction is first performed, followed by a TOF projection to acquire the estimated TOF projection. For each line-of-response (LOR), the shift between the estimated and the measured TOF projection is calculated. A maximum-likelihood approach is applied to the estimated LOR-based timing shift to acquire the crystal-block-based timing shift. 40 patients scanned using the same scanner were included. Artificial timing drift was introduced in the data for validation. Motion-induced attenuation-activity mismatch effects were studied. The variance of the estimation is calculated both theoretically and validated with the patient data. Reconstructed images with and without the self-calibration method were compared quantitatively with the non-TOF image reconstruction. The estimated block-based timing shift was comparable with the added timing drift. The reconstructed TOF image with the self-calibration was quantitatively more consistent with the non-TOF image. The study using the added attenuation and activity mismatch suggested that the average error was ~2 ps per millimeter mismatch. The timing drifts acquired using lung region were different compared with other regions, while the results from other regions were consistent with each other, indicating the effects of respiratory motion. The variance of the timing drifts calculated based on the theoretical method is consistent with patient-based results, indicating it is possible to use the theoretical estimation to predict the precision of the method. We have developed a data-driven timing calibration method. The robustness of this method is demonstrated by the study of noise and attenuation activity mismatch impact.