3423 - Artificial Intelligence-Augmented Cardiac Transit Modeling Pilot Improves Risk Stratification for Radiation-Induced Lymphopenia: A Novel Translational Dosimetric Approach

Purpose/Objective(s): Radiation-induced lymphopenia (RIL) is independently associated with inferior outcomes in head and neck squamous cell carcinoma (HNSCC). This is a frustrating yet common side effect during radiation that can lead to breaks, decreased efficacy, and outcomes. Currently, actionable accurate predictors are lacking. Monte Carlo–based circulating blood dosimetry (ICE3) models lymphocyte exposure within neck vessels but requires accurate vascular segmentation. We tested whether whole-body dose–volume histogram (DVH) metrics or the Simplified Circulating Blood Exposure Index (SCBEI) can offer a non-invasive way and can outperform ICE3 modelling in predicting nadir absolute lymphocyte count (ALC). Simultanesouly, we tested whether clinician-drawn vessel contours improve ICE3 performance.

Materials/Methods: Ten consecutive locally advanced HNSCC patients receiving definitive or adjuvant radiotherapy (60–69.96 Gy in 30–33 fractions) were prospectively enrolled. Nadir ALC was recorded. Spearman correlations were calculated between ALC nadir and: (1) whole-body DVH parameters (V5–V50Gy), (2) SCBEI variants, and (3) ICE3 at three segmentation levels (original autosegmented internal carotid, fixed common carotid geometry, and clinician-drawn carotid/internal jugular vein contours; 100,000 particles; relative standard error ˜1.1%).

Results: Grade =3 lymphopenia occurred in 8/10 patients (80%), including Grade 4 in 4/10 (40%). In the primary analysis (n=9; excluding one patient with direct skull-base marrow irradiation), Body V30Gy and SCBEI_ID were the strongest predictors (? = -0.900, p = 0.0009). SCBEI_shape showed the highest overall correlation (? = -0.917, p = 0.0005). ICE3 was non-significant at all segmentation levels: autosegmented (? = -0.333, p = 0.381), fixed (? = -0.317, p = 0.406), and clinician-drawn (? = -0.524, p = 0.183). ICE 3 did not have statistical significance at any segmentation level. Full-cohort analysis (n=10) confirmed no ICE3 metric reached statistical significance.

Conclusion: Global whole-body DVH metrics with respect to low and intermediate dose and SCBEI strongly predict lymphopenia in HNSCC, whereas ICE3 does not—even with high-fidelity contouring by an experienced head and neck fellowship trained radiation oncologist. Cervical vascular dose may not be the dominant driver of RIL. SCBEI offers a segmentation-free, clinically implementable metric for lymphopenia-sparing planning without additive workflow in a busy teaching clinic. Prospective multi-institutional validation at academic centers is warranted.