ベイズ推定を用いたラジオクロミックフィルムのADC値–線量変換法

Purpose: Radiochromic film is widely used for patient-specific intensity-modulated radiation therapy quality assurance (IMRT QA) because of its high spatial resolution. However, calibration for converting pixel values to dose requires irradiation at multiple known dose levels, resulting in a substantial workload. Consequently, in some institutions, calibration is performed only under limited circumstances, such as when the film lot changes. This study aimed to develop and validate a Bayesian inference model that estimates the calibration curve using previously acquired calibration datasets together with the unirradiated pixel value of the target film, thereby improving the efficiency of the calibration process. Methods: A total of 93 calibration datasets acquired using TomoHD were analyzed. A quadratic polynomial regression model was constructed with dose and elapsed time from irradiation to scanning as explanatory variables. Separate models were developed for each scanner orientation (portrait and landscape), and the intercept was estimated using the unirradiated pixel value of the target film. Bayesian inference was performed using Stan, and model convergence was evaluated by trace plots and the Rhat statistic. Model performance was validated using 10 EBT4 films from a different lot by comparing predicted and measured pixel values and evaluating dose errors. Results: All models showed good convergence, with Rhat values below 1.01. For the validation films, the pixel value error was less than 3.6%, and the coefficient of determination (R2) exceeded 0.99. The mean dose error was 2.7±1.3 cGy at 24.0 cGy and 37.5±12.8 cGy at 868.2 cGy. The maximum dose error was 66.6 cGy at 710.6 cGy. Conclusion: The proposed method demonstrated the feasibility of estimating the calibration curve using previously acquired calibration datasets and the unirradiated pixel value of the target film. This approach has the potential to improve the efficiency of radiochromic film calibration. Future studies should apply this method to patient-specific IMRT QA and investigate its impact on dose distribution evaluation.