Improving patient treatment accuracy using transit dosimetry with Electronic Portal Imaging Device images and deep learning.

Background and Purpose: Electronic portal imaging devices (EPID) can be utilized for transit in vivo dosimetry (IVD). A major limitation of standard amorphous Silicon EPID detectors for usage IVD is their non-linear detector response to water equivalent dose. Labor-intensive calibration and correction procedures are needed, limiting their practicality for routine clinical use. The purpose of this work was to develop a deep learning (DL) framework, based on data acquisitions and treatment planning system (TPS) dose simulations, which reconstructed water equivalent dose from transit EPID images. Materials and methods: A database of over 200 EPID measurements and corresponding portal dose simulations from the Monaco TPS was created, employing different phantoms and irradiation fields. A 2D U-Net model was developed that transforms EPID transit images into water-equivalent portal dose images. The performance was evaluated through the mean absolute error (MAE) and 2D Results: The mean MAE value over all test cases was Conclusions: Our DL framework enabled fast and accurate reconstruction of portal dose from transit EPID images. Once trained on clinical data, it may reduce complex corrections, supporting safe and effective treatments that meet modern regulatory standards.