Pet Arterial Input Function Estimation Using Machine Learning

Dynamic positron emission tomography (PET) scans are often converted to parametric images measuring tracer exchange between tissue compartments. The kinetic parameters aid in clinical diagnosis, therapy monitoring and treatment planning across a multitude of diseases and disorders. An estimate of the arterial input function (AIF) is necessary to be able to predict kinetic parameters. Conventionally, the AIF is estimated using invasive blood sampling for the plasma tracer concentration, and existing non-invasive methods, including image derived input functions (using different imaging modalities) and population based AIFs, have shortcomings. This gap provides an opportunity of exploring the estimation of the subject specific AIF from PET scans alone using machine learning.