Improved Nearest Neighbor Methods for Gamma Photon Interaction Position Determination in Monolithic Scintillator PET Detectors

Abstract

Monolithic scintillator detectors have been shown to provide good performance and to have various practical advantages for use in PET systems. Excellent results for the gamma photon interaction position determination in these detectors have been obtained by means of the k -nearest neighbor (k -NN) method. However, the practical use of monolithic scintillator detectors and the k -NN method is hampered by the extensive calibration measurements and the long computation times. Therefore, several modified k-NN methods are investigated that facilitate as well as accelerate the calibration procedure, make the estimation algorithm more efficient, and reduce the number of reference events needed to obtain a given lateral (x,y) -resolution. These improved methods utilize the information contained in the calibration data more effectively. The alternative approaches were tested on a dataset measured with a SiPM-array-based monolithic LYSO detector. It appears that, depending on the number of reference events, ~ 10% to ~ 25% better spatial resolution can be obtained compared to the standard approach. Moreover, the methods amongst these that are equivalent to calibrating with a line source may allow for much faster and easier collection of the reference data. Finally, some of the improved methods yield essentially the same spatial resolution as the standard method using ~ 200 times less reference data, greatly reducing the time needed for both calibration and interaction position computation. Thus, using the improvements proposed in this work, the high spatial resolution obtainable with the k-NN method may come within practical reach and, furthermore, the calibration may no longer be a limiting factor for the application of monolithic scintillator detectors in PET scanners.