Quantifying daily residual motion of bilateral diaphragmatic domes using an iterative detection algorithm for breath-hold radiation therapy.

Abstract

Breath-hold (BH) techniques reduce respiratory motion and planning margins in thoracic and abdominal radiotherapy, but residual motion during BH can still compromise targeting accuracy. Direct tumor tracking remains technically challenging for many abdominal sites; tracking the diaphragm, which may correlate with implanted fiducial markers, could provide a practical surrogate. Because cone-beam computed tomography (CBCT) is routinely acquired for patient setup, its projection images offer a way to quantify diaphragm motion without additional imaging dose. However, automated diaphragm dome detection on CBCT projections remains limited. We developed and evaluated a fully automatic, iterative local-global algorithm to identify bilateral diaphragmatic domes on CBCT. In a retrospective cohort of 207 CBCT scans from 17 pancreatic cancer patients, initial dome locations were estimated from planning contours. Local robust parabola fitting was used to detect dome positions, followed by a global fitting step to enforce consistency. A second local fitting stage refined the detections, and a final global fitting step with outlier exclusion provided robust dome positions. Superior-inferior dome coordinates were computed for each projection and analyzed within BHs and across paired BHs within each CBCT. Digitally reconstructed radiographs were generated and tested as static patient scans. A secondary dome detection algorithm based on random sample consensus was implemented as part of the verification. Across 41 019 projection frames, identification succeeded in 93.8% of frames for the right dome and 89.7% for the left dome. Residual motion was observed in all patients, with maximum within- BH standard deviations of 6.5 mm (right) and 6.8 mm (left), and mean standard deviations across patients of 2.4 mm (right) and 2.6 mm (left). Between the two BHs within a single CBCT scan, the mean difference in average dome position was 2.5 mm. This method enables routine, CBCT-based monitoring of diaphragm residual motion and supports large-cohort analyses of diaphragm-target correlation.