| 多智能体深度强化学习驱动的医学影像病灶检测方法 |
 点此下载全文 |
| 引用本文：赵仁楷1,赵晗宇2.多智能体深度强化学习驱动的医学影像病灶检测方法[J].计算技术与自动化,2026,(1):103-107 |
| 摘要点击次数: 243 |
| 全文下载次数: 198 |
|
|
| 中文摘要:医学影像病变检测因训练时间长、收敛慢及早期检测中难以捕捉细微特征而具有挑战性。为此，本文提出一种多智能体强化学习系统用于三维医学影像标志点检测。该系统融合优先经验回放（PER）与Nash-A3C算法，以提升检测精度与效率。通过PER优化经验回放，并利用Nash-A3C引入智能体间竞争以稳定状态。在832例成人MRI与72例胎儿超声脑成像数据集上实验表明，该框架使检测精度较以往模型提升约1 mm，各智能体数量下误差均缩小0.2 mm以上。 |
| 中文关键词:优先经验回放 Nash-A3C 多智能体竞争 病变检测 |
| |
| Medical Imaging Lesion Detection Method Driven by Multi-agent Deep Reinforcement Learning |
|
|
| Abstract:Medical imaging lesion detection is challenging due to long training times, slow convergence, and the difficulty in capturing subtle features in early detection. To address this, this paper proposes a multi-agent reinforcement learning system for 3D medical image landmark detection. This system integrates priority experience replay (PER) and the Nash-A3C algorithm to improve detection accuracy and efficiency. PER optimizes experience replay, while Nash-A3C introduces inter-agent competition to stabilize the state. Experiments on a dataset of 832 adult MRI cases and 72 fetal ultrasound brain imaging cases show that this framework improves detection accuracy by approximately 1 mm compared to previous models, and reduces the error by more than 0.2 mm for each number of agents. |
| keywords:prioritized experience replay Nash-A3C multi-agent competition lesion detection |
| 查看全文 查看/发表评论 下载pdf阅读器 |
