Cardiac Radiation Dose Reconstruction in the Study of Late Effects: A Comparison of Different Methods

Abstract

Radiation-related cardiovascular disease can occur several decades after radiotherapy. Cardiac radiation dose information is not available for most historic patient cohorts treated before individual CT-based treatment planning became widely available. Therefore, in order to investigate radiation-related cardiac late effects, it is necessary to reconstruct the doses delivered to the heart retrospectively. Several reconstruction methods have been published and here we evaluate their performance in Hodgkin lymphoma (HL) patients. Fifteen patients (8 females) treated with radiation for mediastinal HL, were selected for this study. Patients were scanned with arms akimbo under free breathing conditions. Two-dimensional digitally reconstructed radiographs (DRRs) were reconstructed to mimic simulation films, which are available for patients treated in the past, during the 2D planning era. These DRRs were used to reconstruct cardiac doses using three reconstruction methods: i) a simple patient-specific approach, where the mean dose to the heart is estimated from the percentage cardiac area exposed within the 2D simulation fields (%Heart)ii) a representative CT technique based on male and female anatomical data sets (RepCT)iii) a “2D to 3D” method using deformable image registration (Navigator)The estimated cardiac doses were compared to those calculated on each patient's individual CT (“Ground truth” method) using a paired t-test. The standard error of prediction for each method was also estimated using linear regression. The average mean heart dose (MHD) (±standard deviation) for the whole cohort estimated by the ”Ground truth“ method was 14.28 (±7.18) Gy and using the %Heart, RepCT and Navigator methods it was 13.08 (±6.69) Gy, 14.54 (±6.90) Gy and 14.27 (±7.25) Gy respectively. The %Heart method was the quickest method to use and for the MHD it had a standard error of prediction of 0.93 Gy. The RepCT and Navigator methods were more accurate than the %Heart method, with standard errors of prediction for MHD of 0.15 and 0.09 Gy respectively. The %Heart method can only estimate the dose to the whole heart but the RepCT and Navigator methods also have the potential to provide volumetric data or estimated dose to substructures of the heart as well. For MLVD they had standard errors of prediction of 0.96 and 0.12 Gy respectively. All three methods led to acceptable MHD estimates for the whole cohort. The %Heart method is the quickest way of estimating the MHD when 2D simulation films are available. However, the RepCT and the Navigator methods are more accurate and have the potential to provide volumetric data and estimates for cardiac substructures.