Evaluation of Intrafractional Anatomical Variations During Liver Magnetic Resonance-guided Adaptive Radiation Therapy Under Abdominal Compression Using Optical Flow Calculation.

Abstract

To investigate the intrafractional anatomical variations during liver magnetic resonance-guided adaptive radiation therapy (MRgART) under abdominal compression (AC) using optical flow calculations. This study included 27 consecutive patients who underwent liver MRgART under AC. Overall, 387,566 slices from 145 single-slice cine magnetic resonance series obtained from 29 different treatment plans were analyzed in the coronal and sagittal planes through tumor centers. After defining the square regions as 12 pixels centered on the isocenter for the tumor and 8 pixels between the inspiratory and expiratory phases, excluding the lung/liver boundary for the diaphragm, the vectors were calculated using Farnebäck optical flow. The intrafractional superior-inferior (SI) motion range and the root-mean-square error (RMSE) between the position of the tumor and the diaphragm in the coronal (SI_coronal and RMSE_coronal) and sagittal planes (SI_sagittal and RMSE_sagittal) were classified according to the Couinaud-based tumor regions (segments I + IV, II + III, V + VIII, and VI + VII). Statistical significance was determined using the Wilcoxon signed-rank test with Holm-Bonferroni corrections (P < .05). The median SI_coronal and SI_sagittal motion ranges of the tumor were 6.1 mm (range, 1.5-18.0 mm) and 8.1 mm (range, 1.0-21.0 mm), respectively (P < .05). When classified according to tumor location, segments VI + VII showed the largest difference, with the median SI_coronal and SI_sagittal motion ranges of 6.5 mm (range, 2.3-17.7 mm) and 10.6 mm (range, 4.8-21.0 mm), respectively (P < .05). The median RMSE_coronal and RMSE_sagittal values were the largest in segments VI + VII, showing significant differences of 2.6 and 2.2 mm, respectively (P < .05). These differences were caused by the sliding motion of dorsally located tumors. Optical flow analysis underestimated the SI motion range in the coronal plane compared with that in the sagittal plane during liver MRgART under AC. Tumor motion should be monitored in the sagittal plane, considering the sliding motion of the liver, with individualized margins according to tumor location.