Purpose: Deformable image registration (DIR) facilitates cumulative dose estimation in re-irradiation of liver tumours. We quantify the accuracy of DIR in liver CT scans taken at multiple time points through use of contemporaneous intravenous contrast CTs.
Methods: We selected patients who had undergone repeat courses of radiation therapy to liver tumours, who had contrast CTs (conCTs) acquired contemporaneously to non-contrast treatment planning CTs (non-ConCTs) for each course. Using VelocityAI (Varian Medical Systems, USA), rigid image registration (RIR) and DIR were performed between the two time-points based on either non-conCTs alone or conCTs alone. Deformable multipass (DMP), extended DMP (EDMP) and structure guided (SG) were evaluated. SG was based on liver contours or five random landmarks. Landmarks were identified on the conCTs and transferred to the contemporaneous nonConCTs via RIR and used to evaluate the accuracy of the RIR and DIR based on Target Registration Error (TRE). Landmarks used in SG registration were excluded from TRE analysis of SG registration.
Results: Mean RIR TRE was 5.8 mm for conCT and 6.0 mm for non-conCT registrations. The mean TRE was 3.9/6.0 mm for conCT/non-conCT, 4.3/5.9 mm for conCT/non-conCT for DMP and EDMP respectively. The mean TRE was 7.7/7.3mm for conCT/non-conCT and 3.4/3.9 mm for conCT/non-conCT for SG based on liver contours or five landmarks respectively.
Conclusion: SG DIR based on five landmarks achieved the lowest TRE, and SG DIR based on the liver contour the highest TRE. DIR based on contrast-CTs was more accurate than for non-contrast-CTs for DMP and EDMP DIR algorithms. For some landmarks, DIR increased the TRE compared to the RIR. Use of landmarks to guide the DIR improves accuracy of both contrast and non-contrast CT DIR.
Funding Support, Disclosures, and Conflict of Interest: This study was supported by the Peter MacCallum Cancer Centre Foundation. The authors acknowledge the support received for this research through the provision of an Australian Government Research Training Program Scholarship. N Hardcastle receives funding from Varian Medical Systems for an unrelated project.
Deformation, Contrast, Quality Assurance