Evaluation of Data Driven Respiratory Signal Extraction Methods from Cone-Beam CT using MR-based Digital Phantoms

A comparative study to assess the performance of data-driven methods to estimate respiratory signals from Cone-Beam CT (CBCT) projection data was earlier implemented using a recently developed 4D MRI-based CBCT digital phantom dataset. However, a more extensive study is deemed necessary to assess the accuracy and robustness of each method rather than just rely on a single dataset. Thus, the purpose of this work is to further investigate the methods using the developed 4D MRI-based CBCT digital phantom datasets from five volunteers with actual different anatomical and breathing features. The effect of a wider range of justified projection region-of-interest (ROI) selection is also considered to further assess these methods. The results indicate that the Local Principal Component Analysis (LPCA) estimates respiration most accurately with an overall mean local correlation coefficient value of 0.875 compared to the Amsterdam Shroud (AS), Intensity Analysis (IA) and Fourier Transform (FT) methods. The LPCA method also demonstrated robustness with the highest mean local correlation across the different ROIs as long as the diaphragm is present. This is dissimilar to the high sensitivity of the FT method on the visibility of the diaphragm regardless of the presence of other organs.