Automated dose-gradient curve and dose-volume histogram analysis platform: development, validation, and clinical decision support with TG-119 datasets

Abstract

Background Highly conformal radiotherapy techniques, such as stereotactic radiosurgery and stereotactic ablative radiotherapy, require steep dose fall-off to spare organs at risk (OARs). The dose-volume histogram (DVH) provides limited spatial information, whereas the dose gradient curve (DGC) offers quantitative assessment of dose fall-off but has seen limited clinical use due to time-consuming manual processing. Purpose This study aimed to develop and validate an automated DGC and DVH analysis platform to enable accurate and rapid dose gradient evaluation for radiotherapy plan assessment. Methods The platform automatically processes DICOM RT dose and RT structure files, with optional RT Plan input for direct prescription dose extraction, to compute differential and cumulative dose gradient index (dDGI and cDGI), DVH metrics, and clinically informed dose gradient quality grading. Computational validation was performed using AAPM TG-119 phantom datasets, including prostate, head and neck, C-shape, and multi-target cases, at 1.25 mm and 2.5 mm dose grid resolutions. Clinical validation was conducted using 142 Gamma Knife stereotactic radiosurgery plans for vestibular schwannoma from the publicly available Vestibular-Schwannoma-SEG dataset (The Cancer Imaging Archive). Results The platform reduced computation time from over 2 hours (original R-based workflow) to under 10 seconds, representing >99.8% efficiency improvement. All TG-119 cases met or exceeded recommended PTV coverage and OAR tolerances. Using prescription doses extracted from DICOM RT Plan file and marching-cubes-based surface area estimation, dDGI at the prescription dose ranged from 0.312 to 0.792 mm on the 1.25 mm grid and from 0.327 to 0.659 mm on the 2.5 mm grid; cDGI at 50% of the prescription dose ranged from 12.4 to 15.7 mm (1.25 mm) and from 11.7 to 15.5 mm (2.5 mm). Clinical validation with 142 vestibular schwannoma Gamma Knife SRS cases (median target volume 0.881 cc, range 0.036-7.183 cc; Rx = 12-13 Gy) yielded dDGI at Rx = 0.294 +/- 0.089 mm with Paddick conformity index = 0.584 +/- 0.075 and gradient index = 2.770 +/- 0.218. The dDGI was strongly correlated with target volume (Pearson r=0.885, p < 10(-48)) but independent of conformity index ( r=0.040, p = 0.634), confirming that gradient steepness and conformity assess distinct plan quality aspects. Paired T1-T2 MRI-based reproducibility analysis ( n=30) demonstrated excellent agreement (dDGI r=0.987, mean difference 3.3%). Conclusions The proposed platform enables rapid, automated DGC analysis practical for routine plan evaluation. The integrated quality grading bridges the gap between quantitative dose gradient metrics and actionable clinical decisions, complementing conventional DVH-based assessment.