Landmark-guided deformable image registration for accurate cumulative dose evaluation in lung reirradiation

Dose summation in thoracic reirradiation is challenging due to anatomical changes and limitations of rigid image registration. Deformable image registration offers improved accuracy but requires careful evaluation before clinical implementation. This work aimed to develop and validate a clinically applicable workflow for deformable image registration-assisted dose accumulation in thoracic reirradiation.

Registration accuracy was evaluated using a digital phantom and 21 thoracic reirradiation clinical cases. Two deformable registration methods were assessed within progressively focused regions of interest. Registration accuracy was quantified using the American Association of Physicists in Medicine (AAPM) Task Group TG 132-recommended metrics, including target registration error, Dice similarity coefficient, mean distance-to-agreement, and Jacobian determinant. Cumulative dose volume metrics expressed as equivalent dose in 2 Gy fractions were calculated and compared between deformable and rigid image registration-based accumulation for relevant organs of interest.

Landmark-guided deformable registration using ten anatomical guiding landmarks within the irradiated lung volume (defined by the 10% isodose) with an additional 12.5 mm cranio-caudal margin, achieved the highest registration accuracy, meeting TG-132 quality tolerances in 18 of 21 cases. Compared with rigid registration, deformable image registration-based dose accumulation resulted in statistically significant differences in lung dose-volume metrics, including mean dose, V20 Gy and critical volume CV12.5 Gy (all p < 0.01).

Patient-specific deformable image registration using landmark guidance within targeted regions improved the accuracy of cumulative dose-volume metrics in thoracic reirradiation compared to rigid image registration. This validated workflow is feasible for clinical implementation and supports individualized reirradiation planning.