Machine Learning-driven Elemental Material Decomposition of DECT scans for Enhanced Radiation Treatment Planning

Material decomposition from dual-energy computed tomography (DECT) is crucial for enhancing radiation treatment planning (RTP) by determining tissue compositions for accurate dose calculations. Traditional methods often struggle with limitations such as noise and anatomical variability, reducing their overall effectiveness. This study investigates machine learning-based elemental material decomposition using a dense neural network for voxel-level analysis and a fully convolutional neural network (CNN) for image-level processing, both trained on synthetic datasets. The inclusion of the XCAT phantom ensured realistic tissue variability, enhancing the relevance of the findings. The dense network achieved low prediction errors under noise-free conditions, while the CNN demonstrated robust performance across most elements, even with added noise. These results underscore the potential of machine learning to overcome challenges in material decomposition, enabling more accurate RTP, minimizing radiation exposure to healthy tissues, and ultimately improving patient outcomes in cancer therapy.