Denoising approaches by SubtlePET™ artificial intelligence in positron emission tomography (PET) for clinical routine application

Abstract

Positron emission tomography (PET) plays an important role in the diagnosis and surveillance of neoplastic diseases. PET images may show higher noise levels than other imaging modalities, especially in a dose- or time-saving approach. Artificial Intelligence techniques can improve the signal-to-noise ratio in PET image reconstruction. Deep learning approaches have made significant advances in comprehensive data retrieval and de-noising. Artificial Intelligence de-noising in PET is a very promising approach that could allow shorter scan times or lower radiopharmaceutical dose administration. We reviewed studies about the de-noising AI-driven PET images, i.e., by SubtlePET™ AI tool, according to the following items: (1) retrieval of complete PET data acquired with reduced scan time; (2) reconstruction of PET images with low-count statistics by reducing radiopharmaceutical doses; (3) impact of artificial intelligence-based de-noising on PET radiomics. We evaluated their implementability in PET image reconstruction to increase the signal-to-noise ratio and image definition. This approach seems promising to positively impact patient healthcare—especially in pediatric patients—and overall diagnostic procedures reducing the cost of radiopharmaceuticals and increasing productivity and efficiency.