Accurate prediction of tumor motion for motion adaptive radiotherapy has been a challenge as respiration-induced motion is non-stationary in nature and often subjected to irregularities. Despite having a plethora of works for predicting this motion, their tracking capabilities are usually prone to large prediction errors due to the time-varying irregularities and intra-trace variabilities. To overcome this, prediction models are re-trained at regular intervals. This solution however demands a trade-off between the re-training interval and prediction accuracy in estimating the future tumor location. This is because re-training with small interval increases the computational requirements whereas a larger interval hampers the prediction performance. To address these issues, a prediction model that relies on random convolution nodes (RCN) governed by local receptive fields (LRFs) is proposed for respiratory motion prediction. The innate nature of LRFs extracts the features that contribute to the local-patterns as well as the non-stationary patterns in recent samples and subsequently learn them using extreme learning machine (ELM) theories. To address the re-training issue, we propose an online sequential learning framework (OS-fRCN) that can update the model parameters at regular intervals. Suitability of the proposed OS-fRCN for respiratory motion prediction is evaluated on 304 respiratory motion traces. Performance analysis conducted at four prediction horizons (in-line with the commercially available radiotherapy systems) demonstrated that the proposed OS-fRCN method requires less computational complexity and yields robust, accurate prediction performance when compared with existing prediction methods.

对于运动适应性放射疗法，准确预测肿瘤运动一直是一个挑战，因为呼吸诱导的运动本质上是非平稳的，经常会出现不规则现象。尽管有大量的工作可以预测这种运动，但是由于随时间变化的不规则性和迹线内的变化性，它们的跟踪能力通常容易产生较大的预测误差。为了克服这个问题，预测模型会定期重新训练。然而，该解决方案需要在重新训练间隔和预测精度之间进行权衡以估计未来的肿瘤位置。这是因为以较小的间隔进行重新训练会增加计算要求，而较大的间隔则会影响预测性能。为了解决这些问题，提出了一种依赖于由局部感受野（LRF）控制的随机卷积节点（RCN）的预测模型来进行呼吸运动预测。LRF的天生本质会提取有助于局部模式以及最近样本中的非平稳模式的特征，然后使用极限学习机（ELM）理论学习它们。为了解决再培训问题，我们提出了一个在线顺序学习框架（OS-fRCN），该框架可以定期更新模型参数。在304条呼吸运动轨迹上评估了拟议的OS-fRCN对呼吸运动预测的适用性。