Initial results of applying automatic channel fault detection and diagnosis on small animal APD-based digital PET scanners

Optimal image quality in small animal positron emission tomography (PET) is critical to ensure accuracy and reliability of results obtained in biological studies. Indeed, unstable image quality over time can jeopardize longitudinal studies. This is why quality control (QC) procedures are of the utmost importance in order to keep PET scanners at an optimal performance level. Unfortunately, as the scanner technology evolves increasing the number of acquisition channels, so does the scanner operator's effort to keep up with adequate QC procedures. With scanners using one-to-one crystal to photodetector coupling to achieve enhanced spatial resolution and contrast to noise ratio (CNR), the QC workload rapidly increases to unmanageable levels due to the number of independent channels involved. An intelligent system (IS) was proposed to help reduce the QC workload by performing automatic channel fault detection and diagnosis. The IS consists of four high-level modules that employ machine learning methods to perform their tasks: Parameter extraction, Fault detection, Fault prioritization and Fault diagnosis. Ultimately, the IS presents a prioritized list to the operator containing the faulty channels and proposes actions that should be taken to correct them. To validate that the IS can perform QC procedures with minimal operator intervention, it was deployed on a LabPET™ scanner in Sherbrooke and image quality metrics were extracted before and after the channel corrections proposed by the IS where applied. After a single iteration of corrections on sub-optimal scanner settings, a 6.3 % increase in the CNR was observed as well as a 7.0 % decrease of the uniformity percentage standard deviation. These results indicate that the IS can improve scanner performance and further iterations are expected to make the scanner converge towards optimal settings.