List-mode maximum-likelihood reconstruction for the ClearPEM system

A dedicated implementation of list-mode maximum-likelihood expectation-maximization (MLEM) reconstruction for the ClearPEM system is presented. The system is composed of two face-to-face detectors, which can be rotated to acquire data from different angular positions. Due to the specific design with irregular sampling and depth of interaction capability, the possible number of lines of response (LOR) is significantly greater than the number of detected events in a standard clinical study. Because reconstruction methods based on data histogramming to sinogram lead to a high computational cost and/or a loss of the intrinsical system resolution, it is necessary to consider the processing of events in list-mode during the reconstruction. The presented method adopted EM algorithm to maximize the logarithmic likelihood function that is expressed in list-mode. The voxel efficiency is corrected by pre-calculated efficiency maps based on flood phantom acquisitions. The method is also implemented with parallelization by distributing the calculation of the acquired events into different threads for significantly increasing computational speed. The results of a Derenzo phantom study show that the presented algorithm can achieve a similar result as 3D-OSEM reconstruction based on data histogramming with significantly lower reconstruction time (6 times faster with one thread, 20 times faster with 8 threads distributed in 8 CPU cores). In clinical studies with lower acquired events, the acceleration ratio can be even higher. The result from a breast phantom study shows that lesions with 15 mm in diameter, each, as well as a small lesion with 5 mm in diameter are clearly visible and can be characterized. The mouse imaging studies show also great potential of the system in small animal applications.