
Research Article
Reinforcement Learning for Multifocal Tumour Targeting
@INPROCEEDINGS{10.1007/978-3-031-43135-7_3, author={Yi Hao and Zhijing Wang and Minghao Liu and Yifan Chen and Yue Sun}, title={Reinforcement Learning for Multifocal Tumour Targeting}, proceedings={Bio-inspired Information and Communications Technologies. 14th EAI International Conference, BICT 2023, Okinawa, Japan, April 11-12, 2023, Proceedings}, proceedings_a={BICT}, year={2023}, month={9}, keywords={Reinforcement Learning Biological Gradient Field Markov Rewards Multifocal Tumours}, doi={10.1007/978-3-031-43135-7_3} }
- Yi Hao
Zhijing Wang
Minghao Liu
Yifan Chen
Yue Sun
Year: 2023
Reinforcement Learning for Multifocal Tumour Targeting
BICT
Springer
DOI: 10.1007/978-3-031-43135-7_3
Abstract
This paper implements a reinforcement learning (RL) targeting strategy for multifocal tumour lesions in the framework of computational nanobiosensing (CONA). Multi-tumours are promoted by the metastatic interaction between the surrounding tissues and the tumour suppressor. Nanorobots, regarded as computing agents, aim to search the multi-tumour lesions within the complicated vessel network. The Biological information gradient fields (BGFs) indicate the formation of the tumour microenvironment regulated by the nearby vessel network. By using reinforced learning and applying the knowledge of BGFs, this work achieves a higher tumour targeting efficiency than the previous work. The Markov and BGFs rewards are included in the total RL reward, in which the Markov reward is utilized for training nanorobots to find the path and avoid colliding with vessel walls, allowing them to learn the vascular network’s topology, whereas the knowledge of BGFs incentive benefits faster convergence of the searching process. Therefore, this method enables the discovery of the path planning for the multi-tumour in a heterogeneous vessel network by combining viable vessel path planning with BGFs information.