Analysis of Inter-Fraction Respiratory Variability Effect on Data-Driven Respiratory Signal Estimation Methods from CBCT Imaging

Irregular breathing patterns appearing as apnea-like features in respiratory signals can occur within lung cancer patients undergoing radiotherapy treatments. Various data-driven methods used to extract respiratory signals from raw Cone-Beam CT (CBCT) projection data usually assumes that the signals are sinusoidal during free-breathing. The performance of these methods was previously studied comparatively on a set of clinical patient data. However, it did not encompass irregular breathing patterns. Thus, the purpose of this work is to further analyze the performance of the four data driven methods on five more sets of projection data on the same patient but displaying the involuntary breath-holding represented as apnea-like features during respiration. Other than the quantitative overall correlation coefficient values for each method and fraction against a reference signal, the local correlation is also evaluated and visually assessed for qualitative analysis. The results indicate that the Amsterdam Shroud (AS) method produced a higher inter-fraction mean overall local correlation coefficient value at 0.778 compared to the Local Principal Component Analysis (LPCA), Intensity Analysis (IA) and Fourier Transform-Phase (FTP) methods. Additionally, the qualitative plots showed that the FTP method also performed well in preserving the apnea-like features within its extracted signal. Thus, suggesting that the performance of current data driven methods can be distinctive with respect to the inter-fraction respiratory motion variations that occur during patient treatment.