4D CT image acquisition errors in SBRT of liver identified using correlation

In the AAPM Report 80, the imaging modality of 4D CT and respiration-correlated CT was declared a "promising solution for obtaining high-quality CT data in the presence of respiratory motion". To gather anatomically correct data over time, the existence of correlation between the internal organ movement and an external surrogate has to be assumed. For the in-house evaluation of such correlation, we retrospectively analyzed 21 four-dimensional computer tomography (4D CT) scans of five patients, out of which the artifacts experienced in three patients are shown here. To provide context and a baseline for the analysis of patient motion, a real-tissue liver phantom was used with a solid water block and liver tissue. The superior-inferior motion of fiducials in phantom and patients was correlated to the recorded anterior-posterior motion of an external surrogate marker on the chest. The use of a solid water block yielded a measurable correlation coefficient of 0.98 or better using a sinusoidal animation pattern. With sinusoidally-animated liver tissue, the minimum correlation observed was 0.96. Comparing this to retrospective patient data, we found three cases of a change in the correlation coefficient, or simply a low correlation. The source of this low correlation was investigated by careful examination of the breathing traces and the CT-phase assignments used to reconstruct the datasets. Consequences of nonregular breathing are elaborated on. We demonstrate the impact of wrong phase assignments and missing image information in the 4D CT phase sampling processes. We also show how daily patient-based correlation analysis can indicate changes in breathing traces, which can be significant enough to decrease, or completely eliminate, previously existing correlation.