Graph neural networks and deep reinforcement learning for simultaneous beam orientation and trajectory optimization of Cyberknife

Abstract

Objective. Despite the high-quality treatment, the long treatment time of the Cyberknife system is believed to be a drawback. The high flexibility of its robotic arm requires meticulous path-finding algorithms to deliver the prescribed dose in the shortest time. Approach. We proposed a Deep Q-learning based on Graph Neural Networks to find the subset of the beams and the order to traverse them. A complex reward function is defined to minimize the distance covered by the robotic arm while avoiding the selection of close beams. Individual beam scores are also generated based on their effect on the beam intensity and are incorporated in the reward function. Main results. The performance of the presented method is evaluated on three clinical cases suffering from lung cancer. Applying this approach leads to an average of 35% reduction in the treatment time while delivering the prescribed dose provided by the physicians. Significance. Shorter treatment times result in a better treatment experience for individual patients, reduces discomfort and the sides effects of inadvertent movements for them. Additionally, it creates the opportunity to treat a higher number of patients in a given time period at the radiation therapy centers.

Graph Neural Networks in Cancer and Oncology Research: Emerging and Future Trends

“…Several recent studies applied GNN representation and learning to radiotherapy optimization and planning. Kafaei et al [115] developed a GNN / reinforcement learning model for simultaneous beam orientation and trajectory optimization of Cyberknife, achieving shorter treatment times without compromising the efficacy of radiotherapy. Shao et al [116] used a GNN representation (from a single onboard x-ray projection) of a liver surface model that accurately translated, via real-time biomechanical modeling, to liver tumor localization, thus optimizing image-guided radiotherapy.…”

Districting and Vehicle Routing: Learning the Delivery Costs

“…A recent and popular architecture for solving approximately routing problems is the so-called Graph Neural Network (GNN) [11,39]. Similar to convolutional neural networks that are dedicated to learn from spatial data such as images, graph neural networks are specialized to learn from data having a graph structure, such as in many combinatorial problems [40,41,42,43]. Joshi [44] has implemented a Graph Neural Network (GNN) with beam search to solve the TSP, achieving an average gap of 1.39% for 100 nodes, improving the previous learning mechanisms to solve the problem but still remaining far from standard optimization approaches.…”