Blind source separation for the computational analysis of dynamic oncological PET studies

Abstract

The analysis of dynamic positron emission tomography (PET) studies provides clinically useful parametric information, but often requires complex and timeconsuming compartmental or non-compartmental techniques. Independent component analysis (ICA), a statistical method used for feature extraction and signal separation, is applied to dynamic PET studies to facilitate the initial interpretation and visual analysis of these large image sequences. ICA produces parametric images, where structures with different kinetic characteristics are assigned opposite values and readily discriminated, improving the identification of lesions and facilitating the posterior detailed kinetic analysis.