Electronic Portal Imaging Devices (EPIDs) are important in real-time radiotherapy (RT) monitoring, offering advantages such as speed, resolution, and stability. However, effective use of EPIDs for in vivo dosimetry (IVD) requires modeling their response to estimate Portal Dose (PD). This study aims to validate and develop a robust dataset for training a Deep Learning (DL) based methodology to convert EPID responses into PD images. Monte Carlo Monaco software, widely used in Treatment Planning Systems (TPS) within clinical settings, was used to generate the output data. To simulate EPID in TPS, the detection area was extended beyond the computer tomography (CT), with a 4.22 cm water-equivalent thickness. The scoring layer of the virtual EPID is placed in the middle. Monaco software validation was performed by comparing the simulated data with the experimental measurements obtained using 7 Gafchromic EBT3 films (GAF) placed at the EPID level. A dataset of 200 EPID images and their corresponding simulated dose distributions was collected, showcasing how the innovative application of the Monaco TPS facilitates the development of accurate deep learning-based dose prediction models, employing a trained U-net architecture. This enables the reliable conversion of grayscale EPID images into PD images for clinical applications.
Mozzi et al. (Wed,) studied this question.