基于放疗计划常规CT深度特征提取的高精度低剂量CBCT成像方法研究

Cone beam CT radiation causes injury to the patient and increases the incidence of many diseases, such as secondary cancer during imaging guided radiation therapy. In addition, due to the usage of large area detector, the reconstruction image artifact is serious, which affects the precise treatment of tumor. The key problem to be solved in this field is the proper method to improve the quality of the reconstructed image while reducing the dose of the patient's exposure. This project is designed to use deep convolution neural network, and innovatively put forward to achieve low dose reconstruction and image scatter artifact correction. Based on the partial projection data reconstruction and scatter correction method which have been solved in our clinical study previously, Encoding-Decode depth network is applied to extract the depth features between the prior planning CT data and cone beam CT data. The beam blocker is applied to reduce the patient’s radiation dose and correct the scatter artifacts. The new quantitative and low-dose new image guided radiotherapy methodology and technology developed in this project are expected to provide key technical support for achieving high precision and low dose of adaptive image guided radiotherapy.

在影像引导放射治疗中，CBCT辐射对病人造成伤害，提高了二次癌症等疾病的发病率。此外，由于大面积探测器的使用，重建图像伪影严重，影响肿瘤的精确治疗。如何在降低病人受照剂量的同时，提高重建图像质量是本领域亟待解决的关键问题。本项目创新性地采用深度卷积神经网络实现低剂量重建与图像散射伪影校正。基于我们在临床上已经实现的不完全数据精确重建与散射伪影消除方法，采用Encoding-Decoding深度网络对先验计划CT数据和CBCT数据进行深度特征提取；采用射束阻挡器降低病人受照剂量，同时对图像伪影进行修正。本项目提出的高精度低剂量成像方法，有望为实现高精准、附加剂量低的自适应影像引导放疗提供关键技术支撑。

在影像引导放射治疗中，锥束CT（CBCT）辐射对病人造成伤害，提高了二次癌症等疾病的发病率。此外，由于大面积探测器的使用，重建图像伪影严重，影响肿瘤的精确治疗。剂量高和图像质量差一直是限制CBCT在临床广泛应用的瓶颈问题。深度学习理论为突破这一瓶颈提供了崭新的解决方案，其中如何高效地提取计划CT中的深度特征是这一理论能成功应用的根本保障。本研究提出采用射束阻挡器结合深度卷积神经网络，实现低剂量重建与图像散射伪影校正。具体包括：1）研究基于深度卷积神经网络和自适应滤波的高精度低剂量CBCT成像；2）提出基于射束阻挡器的高精度低剂量CBCT成像；3）提出基于无监督学习的CBCT伪影消除技术；4）提出结合投影域与重建域的CBCT条状伪影消除技术。在病人影像数据上，CT值的误差降至20 HU以下。本项目提出的高精度低剂量成像方法，有望为实现高精准、附加剂量低的自适应影像引导放疗提供关键技术支撑。本项目已经发表了SCI期刊论文6篇，会议论文4篇，授权发明专利4项，已培养或正在培养的博士/硕士研究生4名。