Results: 40 patients were identified. Fifteen hippocampi (19%) exceeded constraints in 12 patients (30%); 74 metastases were treated in these 12 patients. Hippocampal sparing was achieved in 10 of 12 replanned cases (83%). Decreasing dmin was associated with exceeding dose constraints (8.0 vs 24.0 mm, PZ0.002). Each additional millimeter of distance between the closest metastasis and the hippocampus decreased the odds of exceeding constraints by 13% (OR 0.87, 95% CI 0.79-0.96). There was a statistical trend between exceeding constraints and total target volume (PZ0.057, odds ratio 1.14, 95% CI 1.00-1.32), but no association with number of metastases (PZ0.409) or prescribed dose (PZ0.584). For reoptimized plans there were no significant differences in PTV coverage or V12, though the CI and median brain dose were lower (Table 1). Conclusion: A substantial minority of hippocampi receive high radiation dose from SRS for 4-10 brain metastases. Decreasing distance of the closest metastasis and potentially total target volume are associated with exceeding hippocampal constraints. Reoptimizing these plans yielded hippocampal-sparing SRS plans with acceptable CI and PTV coverage.

No abstract

No abstract

Materials/Methods: Pre-treatment CBCTs for 327 fractions from 39 gated patients (17 liver, 12 pancreas, 5 abdomen, 5 lung) were collected. Markers ranged from 2-12 mm long. Aperture contours expanded from the markers from the planning CT were imported into the program and used as allowed marker positions for appropriate "beam on." The marker locations in each CBCT projection, as tracked by the DP algorithm, were used to estimate 3D trajectories using a probability-based method and to optimize the patient couch position and gating window parameters by balancing beam accuracy and efficiency. For each patient, the markers were manually contoured in all images of one fraction and used as the ground truth for comparison. The algorithm-optimized patient couch positions were compared to those used clinically during treatment. Finally, a physician analyzed the outputs of the algorithm to determine whether the parameters were acceptable for treatment. Results: The mean error of the DP tracking algorithm was 1.3AE1.0 mm over all markers for all projections, with larger markers having larger errors. The 3D trajectories showed motions of 9.6AE5.0 mm (SI), 4.3AE2.5 mm (AP), and 2.9AE2.0 mm (LR), with 95-percentile motions of 29, 20, and 15 mm. The optimized couch locations differed from those used clinically by 2.8AE2.2 mm (SI), 2.7AE2.3 mm (AP), and 2.2AE1.8 mm (LR). A physician determined that 323 (99%) of the fractions were acceptable for treatment. The entire algorithm takes less than 1 second to run. Conclusion: We showed that our DP algorithm can accurately track fiducial markers in CBCT projections. These tracked locations can be used to estimate 3D motion magnitudes, providing invaluable information for the physician to determine whether motion management is needed, as well as be used to automate the patient setup. Due to an aperture contour expansion of 3-5 mm in our data sets, a wide range of couch positions was acceptable. Therefore, the physician confirmation of the acceptability of the optimized shift in 99% of the fractions is the best measure of performance. We are in the process of testing the algorithm for real time use in the clinic.