Fast and unbiased 3D calibration method of arbitrary scintillator based PET detectors

In high resolution positron emission tomography (PET) more and more scintillator crystal geometries other than pixelated arrays are being investigated. For pixelated crystal arrays an intrinsic spatial calibration can be used as the crystal bins are separable in a center-of-gravity 2D flood histogram. Monolithic crystals, on the other hand, can not be calibrated without a collimated beam or using a model for the light distribution describing the scintillation position. In this paper we present a method which allows to gather calibration data in a PET crystal without the need for a model of the light distribution. The method uses reference calibration beams from different directions which are electronically collimated. To obtain reference calibration lines we combine a high resolution 22Na point source with a diameter of 0.25mm and electronically collimate it using a high resolution preclinical LYSO crystal array with a pitch of 1mm using a light guide for light sharing between crystals. Interactions of the gammas are known to lie on this calibration line. We use at least two intersecting calibration lines and identify the coincident singles at the point of geometrical intersection. These events are then used as calibration data at the intersection point in the crystal. The method operates on the channel space defined by the 64 readout channels of the digital silicon photomultiplier (dSiPM) PDPC DLS 3200-22 sensor. We use the eucledian distance in the channel space to compare two singles. A k-nearest neighbors (k-NN) algorithm is employed to identify the events which define to the point-of-intersection of two calibration lines. This corresponds to finding the point in channel space with the highest value for the multiplied single density. We show, as a proof of concept, that we can calibrate a traditional preclinical crystal array without the use of a clustering algorithm operating on a center-of-gravity position histogram.