Radiation Dose Optimization. The Role of Artificial Intelligence

Artificial intelligence (AI) is an emerging technique impact- ing the world we live in, including medical imaging. Its current massive popularity hinges on three intertwined pillars: big data availability, constant increase of computational power, and algorithms. Among its various tasks, imaging denoising is the one holding the highest expectations for CT radiation dose optimization. AI is a generic term indicating a science that mimics cognitive human functions, aiming at performing tasks that are typical of human intelligence. Machine learning (ML) is a subset of AI in which algorithms build their own training data, requiring no explicit program- ming. Feature learning is a subset of machine learning in which the algorithm automatically selects the features useful to detect and classify data. Lastly, deep learning (DL) is a subset of representation whose architecture is formed by the composition of multiple-level mathematical equations, known as deep neural networks (DNNs), based on the demands of the specific task that must be accomplished in order to achieve what ML does. Despite continuous technical advancements, CT radiation dose and the potential risk of induced carcinogenesis are still a matter of concern in terms of public health, both in the scien- tific community and in the lay press. In addition to the implementation of decision guidelines and the education of radiology personnel, referring clinicians, and patients, CT scanning parameters such as tube voltage and tube cur- rent modulation, prospective ECG gating in cardiac CTA (CCTA), and high-pitch acquisition protocols are determinant factors in radiation dose optimization. Among the CT technical parameters, image reconstruction plays a funda- mental role in the final image quality and, consequently, in the management of radiation exposure.