Developing a web-based dashboard for adaptive radiotherapy workflows

Abstract

Purpose: Adaptive radiotherapy (ART) is limited by the absence of reliable thresholds to trigger necessary plan adaptation. This work develops and verifies a web-based, database-backed dashboard that unifies geometric, dosimetric, and radiobiological metrics to support day-to-day ART decisions. Methods: Ten retrospectively selected patients were analyzed (Head and Neck (H&N), n = 5, 30–35 fx; prostate MR-linac, n = 5, 5 fx). Daily dose was recalculated on CBCT-derived pseudo-CTs (pCT) (H&N) or on daily MR images (prostate). A Python/Dash dashboard with a PostgreSQL backend takes in RTStruct/RTDose, then performs structure name harmonization, computes per-fraction DVHs, and derives five radar-chart metrics: organ central point displacement, interfraction Dice (reported as 1–Dice), intrafraction Dice (OAR–PTV overlap vs reference), objective score (deviation from plan objectives), and a radiobiology score using TCP/NTCP to calculate PI, PB, and P+. Residuals, defined as the difference between the reference metric (MIM) and the corresponding dashboard calculated value (MIM—Dashboard), were calculated to assess agreement. Calculations were verified against MIM Maestro using residual plots, paired t-tests (α = 0.05), and effect sizes. Results: Central point residuals were negligible (mean < 0.05 mm on all axes; max 0.45 mm). Interfraction Dice mean absolute difference was 0.08 (max 0.83); intrafraction Dice differences were smaller (overall mean residual −0.01). For score card dose endpoints, the largest mean difference was at D99.9% (1.31 Gy; 2.08%—below TG-114′s 5% action level). Percent-volume endpoints showed small residuals (overall mean +0.32%); absolute-volume endpoints were near zero (overall mean +0.05 cm3). Radiobiology residuals were modest (mean: PI +1.78%, P+ −1.11%, PB +0.86%) with occasional outliers (max PI 8.69%). Three metrics reached statistical significance (central point, intra-/inter-fraction Dice), but effect sizes were negligible to small. Conclusions: The dashboard reproduces geometric, DVH, objective, and radiobiology metrics within acceptable limits relative to MIM, providing a credible foundation for ART decision support. Limitations include the projected DVH being a visualization (not dose accumulation) and radiobiology sensitivity to DVH sampling in steep gradients.