Artificial intelligence, signal and data processing towards greater end-user flexibility through increased sensitivity and detection efficiency in PET imaging

点击翻译按钮获取中文摘要

This proposal is part of a novel vision where digital signal and data processing, artificial intelligence and distributed modular analysis can substantially improve both the intrinsic performance and end-user flexibility of Positron Emission Tomography (PET), a vision symbiotic and interdependent on the usual advances in technology, comprehension of system-level issues and socio-economic efficiency constraints. In the short term, the research program will mitigate the pressure on the flexibility paradigm emerging as side-effects of other resolution, TOF and detector breakthroughs. In the mid to long term, the vision will uniquely benefit all PET end-users through innovative complements to existing technology, helping this molecular and genomic imaging modality continue to grow in the next decades through the emergence of new radiotracers or new routine care and research protocols. Depending on scanner performance like resolution, contrast and sensitivity, the end-user faces a significant trade-off between image quality, injected dose and scan length, directly impacting the flexibility, scientific return and cost/benefit ratio of any given imaging protocol, all of primordial importance in the continued availability of the modality in healthcare worldwide. Some new protocols will require high sensitivity, others good contrast and resolution, or both: the cost effective evolution of PET requires flexibility, which an all-digital architecture and the aforementioned vision inherently provide. The program focuses on sensitivity and detection efficiency because, when considering the above end-user trade-off, these are currently the most under-studied and wide-reaching factors through which end-user flexibility can be gained quickly. Varied interwoven aspects of PET technology will immediately benefit from the proposed vision: 1) optimization of a novel artificial intelligence method I recently demonstrated to improve detection efficiency by 50 to 200%; 2) refinement of image reconstruction methods to account for the improved efficiency, minimizing its impact on image quality; 3a) algorithmic maximization of new SPAD detectors’ dynamic range to improve sensitivity; 3b) implementation of the new algorithms in modular embedded power as a 3D ASIC affixed to the detector itself instead of a central computer; and 4) design of an advanced stereoscopic DOI technique to decrease localization uncertainties and gain both contrast and efficiency. Overall this vision will lead to new scientific if not commercial endeavors, allowing adaptations of digital techniques and artificial intelligence mostly unheard of in PET until now to circumvent many known problems and limitations and to open new possibilities for the modality.