Deep learning applications in motion management for radiotherapy

The aim of radiotherapy (RT) is to deliver prescribed doses to tumors while sparing neighboring organs at risk. As the demand for treatment precision increases in modern RT, intrafractional motion management becomes critical for achieving high precision, particularly for tumors in the thorax and abdomen that are affected by respiration and other physiological motions. Deep learning (DL) has demonstrated strong potential in addressing the limitations of conventional methods by enabling rapid and accurate motion detection, tumor location prediction, and management measures. This review provides a comprehensive overview of recent DL-based applications in intrafractional motion management, which are categorized into three areas: 1) motion detection for real-time tumor localization, 2) tumor location prediction to compensate for the latency of the motion management system, and 3) management measures, including gating and real-time adaptive radiotherapy. In addition, this review discusses key challenges and their potential solutions for DL-based motion management.