Investigation and Modelling of Radiation Dermatitis Related to Radiobiological Dose In Proton Therapy of Head and Neck Cancer
Abstract
Purpose
Radiation dermatitis (RD) is a primary acute toxicity in proton therapy for H&N cancer patients. Since skin typically resides in the distal fall-off region of the posterior or posterior oblique beams, the radiobiological dose may increase and lead to more severe RD. This study aims to investigate the skin biological dose and its effects on radiation dermatitis, and develop a prediction model.
Methods
We retrospectively analyzed 52 H&N proton patients, stratified into severe (CTCAE Grade ≥ 2, n=22) and mild (Grade ≤ 1, n=32) RD groups based on clinical follow-up. Skin dosimetric parameters were calculated and compared between the constant RBE model (RBE1.1) and the LET-based variable RBE model (RBELET). Biological dose increase was defined as DΔ=DRBELET–DRBE1.1. Robust features were selected using LASSO-based stability selection. The predictive performance of a combined model (incorporating DRBE1.1 and DΔ features) was compared against a traditional physical model (DRBE1.1 features only). Based on the combined model, a clinically applicable risk prediction formula was constructed, and an optimal stratification threshold was determined.
Results
DRBELET was significantly higher than DRBE1.1 (all indices P 70Gy versus only 35% with DRBE1.1_max. DRBE1.1_max and DΔ_20cc were identified as the most robust features. The combined model outperformed the traditional physical model. The final clinical prediction model achieved an AUC of 0.877 (Formula: Logit(P)=0.0045×DRBE1.1_max+0.0065×DΔ_20cc–31.46); a calculated risk probability of P=1/(1+e-Logit(P))>0.494 indicates the high-risk group.
Conclusion
The combined model, incorporating skin dose calculated with RBE1.1 and the LET-based dose increase, provides a clinical risk stratification tool to predict RD in H&N proton therapy, potentially leading to better dose optimization in treatment planning and reduced RD severity.