3119 - Workflow-Integrated Beam's-Eye-View Surface Mapping in Proton Therapy: Population-Level Accuracy Shift

Purpose/Objective(s): Daily CBCT-to-simulation CT surface mapping quantifies setup and anatomic variation during proton therapy; however, clinical implementation requires stable beam’s-eye-view (BEV) z_diff (depth difference along the beam axis) heatmaps suitable for multidisciplinary review. We evaluated whether formal integration of BEV surface maps into routine chart rounds was associated with measurable shifts in surface deviation metrics and adaptive plan modification activity across the treated population.

Materials/Methods: BEV surface mapping was incorporated into clinical practice beginning June 2025. An automated algorithm removed immobilization devices using binary image processing by isolating the largest centroid-containing inner contour contiguous with the patient exterior. Cleaned contours were reconstructed via Poisson surface reconstruction, and pointwise z_diff was computed on a fixed BEV grid comparing daily CBCT to planning CT surfaces. Values with |z_diff| > 30 mm were excluded; beams with >60% point loss were omitted. Per-beam deviation was summarized using percentile families (P98/P90/P75/P50) for absolute deviation (AbsDev), positive deviation (PosDev), negative deviation (NegDev), and maximum-directional deviation (MdxDev). Data spanned January 2025 to February 2026 across all proton patients at one institution (324 patients; 14,671 beam-date records). A changepoint sweep identified the optimal before/after split by maximizing the Welch t-statistic on AbsDev_P98. Plan turnover rate was defined as the proportion of treated patients with =1 non-alternating new plan name per period.

Results:

Changepoint analysis temporally aligned with workflow integration, identifying 2025-08-13 as the optimal split (t = 10.6). Following integration, AbsDev_P98 decreased by 13.6% (6.34 to 5.48 mm, p < 0.001). Similar reductions were observed across additional percentile tiers, indicating distribution-wide tightening rather than isolated tail reduction. The proportion of beam-dates exceeding 5 mm at P98 declined from 39.6% to 35.6%. Plan turnover rate was defined as the proportion of treated patients within each pre/post window with =1 non-alternating plan name change during that window.

Conclusion: Workflow integration of BEV surface maps coincided with significant distribution-wide reductions in surface deviation metrics and increased adaptive plan modification activity. These findings support quantitative surface mapping as an operational decision-support tool within adaptive proton therapy workflows. Table 1. Before/After Comparison (split: 2025-08-13)

†Proportion of patients with =1 plan modification; excludes alternating-beam swaps but includes scheduled reductions and other routine plan changes unrelated to surface deviation.

Metric	Before	After	p
Coverage <5 mm (%)	87.7	91.4	<0.001
Beam-dates with P98 >5 mm (%)	39.6	35.6	<0.001
Plan turnover rate† (% patients)	23.0	30.9	—