Project Summary/Abstract
A major advantage of coincidence detection of annihilation photons from positron-emitting radiotracers is the
availability of time-of-flight (TOF) information, and the ability to measure TOF differences to better localize the
positron emitter. Normally for positron emission tomography (PET), TOF information is used as a weighting
kernel during image reconstruction and results in an effective sensitivity gain that can be used to reduce radiation
dose, improve signal-to-noise ratio, or reduce scan duration. The magnitude of these benefits depend on the
TOF resolution, which is governed by the timing performance of the detectors. Current state of the art for PET
scanners is ~220 ps which corresponds to a localization of ~3.3 cm. A transformational change would occur,
however, if a TOF resolution of <30 ps could be achieved. This would localize events within 4.5 mm, allowing
images to be directly generated without a reconstruction algorithm at a spatial resolution that matches what is
achieved in clinical PET scanners today. We refer to this as direct positron emission imaging (PEI). With this
superb TOF resolution and reconstruction-free imaging, we enter a new regime where we expect major increases
in image signal-to-noise, both due to the additional TOF information, and the removal of noise amplification
inherent in reconstructing noisy data with noisy corrections from projection data. We propose to develop a first
proof-of-concept imaging system that uses ultra-fast detectors to directly produces cross-sectional images
without reconstruction and to quantify the performance of PEI both through simulations and experimentally.
Since direct PEI does not have the same sampling constraints for data collection as PET, it creates opportunities
for portable, and flexible imaging devices, with implications for patient-tailored or task-specific imaging
applications (i.e. cardiac or breast imaging), as well as open designs for general purpose applications.
To achieve the unprecedented TOF capabilities needed for direct PEI, we will exploit promptly emitted Cerenkov
radiation that is generated with <10 ps in certain materials, including scintillators, in response to a 511 keV
photon interaction. Our proposed novel detector design integrates a Cerenkov radiator directly into the entrance
window of an ultra-fast microchannel plate photomultiplier tube, which is the fastest photon detector currently
available with a response time of 25 ps. This approach eliminates all optical reflections between the point of light
generation and the photocathode, preserving the prompt timing nature of Cerenkov photons. We then combine
the integrated Cerenkov radiator detector with auxiliary photodetector read-out for robust coincidence detection,
and complement this with advanced signal processing algorithms we have pioneered using convolutional neural
networks to extract all possible timing information from the digitized detector waveforms and ultimately to perform
reconstruction-free imaging using only the digitized waveforms as input. In summary, we aim to prove that direct
PEI is possible, to characterize its properties and to provide the technological and algorithmic foundations for
eventual translation for human imaging.