3589 - Technical Validation of STRIDE (Systemic Toxicity Risk from Immune Dose Exposure) V1.0: A Cross-Platform Framework for Immune-Relevant Cardiac Dosimetry in Locally Advanced NSCLC
Presenter(s)
J. Z. Pierre-Charles1, J. Castle2, A. Holtman3, D. A. Cheek4, M. E. Bernard5, D. Fabian5, J. Wang6, Y. Wu6, K. Shang7, and W. Yan8; 1University of Kentucky, Department of Radiation Medicine,, Lexington, KY, 2Aurora St. Lukes's Medical Center, Milwaukee, WI, 3Department of Radiation Oncology, The Ohio State University Wexner Medical Center, Columbus, OH, 4University of Kentucky Department of Radiation Medicine, Lexington, KY, 5University of Kentucky, Department of Radiation Medicine, Lexington, KY, 6Department of Radiation Oncology, the Fourth Hospital of Hebei Medical University, Shijiazhuang, Hebei, China, 7Department of Radiotherapy, the Fourth Hospital of Hebei Medical University, Shijiazhuang, China, 8Department of Radiation Oncology, University of Kentucky, Lexington, KY
Purpose/Objective(s):
Radiation-induced lymphopenia (RIL) is associated with inferior survival in patients with locally advanced non–small cell lung cancer (LA-NSCLC) treated with concurrent chemoradiation. Circulating lymphocytes pass through cardiac structures during treatment and may receive unintended radiation dose, supporting the concept of the heart as an immune organ-at-risk (iOAR). However, consistent and reproducible cardiac substructure dose extraction remain technically challenging and varies across institutions and treatment planning systems. STRIDE V1.0 was developed as a research software platform to standardize immune-relevant cardiac dose analysis using automated segmentation and DICOM-based extraction of RTSTRUCT, RTDOSE, and RTPLAN data.Materials/Methods:
Fifteen anonymized LA-NSCLC cases from two independent sites were retrospectively analyzed. STRIDE V1.0 integrates automated cardiac segmentation with DICOM-based extraction of RTSTRUCT, RTDOSE, and RTPLAN data to compute cardiac substructure dose metrics. Geometric segmentation agreement was assessed against expert manual contours using Dice similarity coefficient (DSC) and Hausdorff distance (HD). DVH extraction reproducibility was evaluated by comparing STRIDE-derived dose metrics against outputs from two independent commercial treatment planning systems. Primary validation endpoint: mean dose (Dmean) agreement within ±2%. Secondary metrics: V5, V10, and V20. Statistical analyses were performed using Python/R statistical software. A large language model (Claude Code) assisted with code drafting and debugging. All analyses and outputs were reviewed and validated by the authors and are fully reproducible from the underlying data.Results:
STRIDE V1.0 demonstrated high dosimetric reproducibility across all evaluated structures and cases. Dmean error was <1% and all cases met the predefined ±2% acceptance threshold. Minor differences were attributed to interpolation and voxel sampling variation between systems. Whole-heart segmentation demonstrated strong agreement with manual contours (median DSC >0.90). Smaller vascular and conduction-related substructures showed greater variability, reflecting inherent anatomic complexity and contouring sensitivity.Conclusion: