The Application of a Statistical Shape Model to Diaphragm Tracking in Respiratory-Gated Cardiac PET Images

Abstract

Respiratory-induced diaphragm mismatch between positron emission tomography (PET) and computed tomography (CT) has been identified as a source of attenuation-correction artifact in cardiac PET. Diaphragm tracking in gated PET could therefore form part of a mismatch correction technique, where a single CT is transformed to match each PET frame. To investigate the feasibility of such a technique, a statistical shape model of the diaphragm was constructed from gated CT and applied to two gated 18F-FDG PET-CT datasets. A poor level of accuracy was obtained when the model was fitted to landmarks obtained from PET, with errors of 3.6 and 5.0 mm per landmark for the two patients, despite inclusion of the data within the model construction. However, errors were reduced to 2.4 and 1.9 mm with the incorporation of a single frame of CT landmarks. These values are closer to the baseline measure of fitting solely to CT landmarks, found to be 2.2 and 1.2 mm in this case. Excluding the datasets from the model yielded similar trends but with higher overall residual errors, indicating the need for a larger training set. Therefore, a highly trained diaphragm model could negate the need for a gated CT for diaphragm tracking, provided that information from a static CT is incorporated.