【摘要】 目的：针对肝肿瘤热消融手术中呼吸运动引起的肿瘤位置变化导致肿瘤中心位置难以标记的情况，设计一种基于光学多点追踪的肝肿瘤热消融呼吸运动实时建模系统。方法：肝肿瘤热消融呼吸运动实时建模系统硬件部分主要由光学定位系统和呼吸模拟装置构成，并配合上位机平台完成系统控制与数据处理；软件部分采用Python3.8语言编写，使用scikit-learn中的多元线性回归算法进行模型训练与预测，通过Matplotlib实现预测与真实值的可视化对比。通过标准实验、噪声扰动实验和真实人体数据验证系统的应用效果。结果：在标准实验数据条件下，该系统对标准呼吸、快速呼吸、深呼吸3种呼吸模式的运动预测平均绝对误差（mean absolute error,MAE）均小于0.03 mm，决定系数R~2均大于0.87。在幅值不超过0.05 mm的噪声干扰条件下，R~2保持在0.82以上。在真实人体数据实验中，系统呼吸运动预测均方根误差控制在0.4 mm以内，R2大于0.94,MAE控制在0.3 mm以内。结论：该系统能够准确、实时地预测中心点的肿瘤位移，具有优异的抗干扰性能和临床适用性，可为肝肿瘤热消融实时呼吸运动补偿及术中实时位移预测提供可靠的技术支持。更多还原

【Abstract】 Objective To design an optical multi-point tracking-based respiratory motion modeling system for liver tumor thermal ablation, so as to address the challenge of accurately marking the center of a tumor during thermal ablation of liver tumors where respiratory motion causes the tumor’s position to shift.Methods The optical multi-point tracking-based respiratory motion modeling system had its hardware composed of an optical positioning system and a respiratory device,which worked in conjunction with a host computer platform to handle system control and data processing; the software of the system was programmed with Python 3.8 language, which used the multiple linear regression algorithm from scikit-learn for model training and prediction, and employed Matplotlib to visualize the comparison between predicted and actual values.The system’s performance was validated through standard experiments, noise perturbation experiments and real human data.Results Under standard experimental data, the mean absolute errors(MAEs) of the system for motion prediction in standard breathing, fast breathing and deep breathing modes were all lower than 0.03 mm, and the coefficients of determination R~2 were all higher than 0.87. R~2 stayed above 0.82 under the noise interference with an amplitude not exceeding 0.05 mm. In real human data experiments, the root mean square error of respiratory motion prediction was within 0.4 mm, R~2 was higher than 0.94 and MAE was lower than 0.3 mm.Conclusion The system accurately predicts the displacement of the tumor center at real time, which features high anti-interference capability and clinical applicability and provides reliable technical support for real-time respiratory motion compensation and intraoperative displacement prediction in liver tumor thermal ablation.更多还原