TH‐EF‐BRA‐02: A Novel Markerless 4D CBCT Projection Phase Sorting Technique Using Prior Knowledge and Patient Motion Modeling: A Feasibility Study

Purpose: To investigate the feasibility of a novel marker‐less motion‐modeling based method for automatic 4D‐CBCT projection phase sorting. Methods: Patient on‐board image volume at any instant is considered as a deformation of one phase of the prior planning 4D‐CT. The deformation field map(DFM) is represented as a linear combination of three major deformation patterns extracted from the planning 4D‐CT using principle‐component‐analysis(PCA). The PCA coefficients are solved for each single projection based on data fidelity constraint, and are used as motion information for phase sorting. Projections at the valleys of the Z direction coefficient are sorted as phase 0/100% and projection phases in between are linearly interpolated. 4D‐digital‐extended‐cardiac‐torso(XCAT) phantoms and 3 patient cases were used for evaluation. XCAT phantoms simulated different patient respiratory and anatomical changes from prior 4D‐CT to on‐board image volume, including changes of tumor size, locations, motion amplitudes and motion directions. Three patient cases include 2 full‐fan slow‐rotation and one half‐fan normal‐rotation case. Manual phase sorting based on visual inspection was used as the gold standard. The average absolute phase difference, and the pass rate (percentage of projections sorted within 10% phase error) were used to evaluate sorting accuracy. Results: The amplitude of PCA coefficient motion curve correlated with the actual motion amplitude. The algorithm was robust against respiratory and anatomical changes from prior to on‐board imaging. For all XCAT cases, the average phase errors were lower than 1.43%, and the pass rate was 100%. The patient data set showed average phase error of 2.47%, 1.90% for full fan slow rotation case and 2.78% for half fan normal rotation case, respectively. The corresponding pass rates were 99.4%, 98.5% and 99.5%, respectively. Conclusion: Preliminary results demonstrated the robustness and high accuracy of the marker‐less PCA based phase sorting algorithm for different patient scenarios and 4D‐CBCT scanning protocols.