基于深度学习的手术场景运动感知

基于术中影像的运动感知是计算机辅助手术系统开发的重要研究内容，能够为运动补偿、软组织形变分析等应用提供有价值的信息，从而提高手术效率并增强手术安全。然而，手术影像中运动器械的遮挡降低了对局部区域估计的准确性。为解决这一难题，提出一种基于光流网络和解耦表示的运动感知方法，并结合自监督学习范式优化模型。制作了神经外科手术数据集，在PyTorch深度学习框架下对模型进行训练和验证。实验结果表明：该运动估计方法在复杂手术场景中具有稳定性强、准确度高的优点，在计算机辅助手术中具有较高的应用价值。

Motion Perception in Surgical Scenes Based on Deep Learning

Intraoperative-image-based motion perception is a critical research topic in the development of computer-assisted surgical systems. It can provide valuable information for applications such as motion compensation and soft tissue deformation analysis, thereby improving surgical efficiency and enhancing surgical safety. However, the occlusion of motion instruments in surgical images reduces the estimation accuracy in local regions. In order to address this issue, a motion perception method based on an optical flow network and decoupled representation was proposed, and the model was optimized using a self-supervised learning paradigm. A neurosurgical dataset was created, and the model was trained and validated in the PyTorch deep learning framework. Results demonstrate that the proposed motion estimation method achieves strong stability and high accuracy in complex surgical scenarios, which shows a high application value in computer-assisted surgery.