An improved kernel for analytical time-of-flight PET reconstruction

We demonstrate a new weighting kernel for linear time-of-flight (TOF) positron emission tomography (PET) reconstruction that produces lower image noise variance in finite-sized objects than "confidence weighting" (CW) or any other kernel previously reported in the literature. The proposed kernel is adaptive to object size, but is always broader than the CW kernel. For 1.25 nsec TOF resolution, the new kernel improves image noise variance in a 10 cm diameter cylinder by 8%. Small but significant gains in image noise variance are also seen in clinical reconstructions.