Data Driven Surrogate Signal Extraction for Dynamic PET Using Selective PCA

Respiratory motion correction is beneficial in PET. Methods of motion correction include gated reconstruction, where the acquisition is binned, based on a respiratory trace. To acquire these respiratory traces, an external device, like the Real Time Position Management System, or a data driven method, such as PCA, can be used. Data driven methods have the advantage that they are non-invasive, and can be performed post-acquisition. However, data driven methods have the disadvantage that they are adversely affected by the tracer kinetics of a dynamic acquisition. This work seeks to evaluate several adaptions of the PCA method, through which it can be used with dynamic data. The methods explored in this work include, using a moving window (similar to the KRG method of Schleyer et al. (PMB 2014)), extrapolation of the principal component from later time points to earlier time points, as well as a method to select and combine multiple respiratory components. The respiratory traces acquired, were evaluated on 21 patients, by calculating their correlation with a Real Time Position Management System surrogate signal. The results indicate that all methods produce better surrogate signals than when applying static PCA to dynamic data. Extrapolating a late principal component, produced more promising results than using a moving window, and selecting and combining components held benefits for all methods.