Development of a non-invasive real-time tumour motion tracking method using surface-guided radiation therapy

在放射治疗过程中管理肿瘤诱导的肿瘤运动仍然是一个重大挑战，在确保预期的辐射剂量被输送到肿瘤，同时确保附近的关键器官辐射剂量限制不被超过。放射治疗过程中的实时肿瘤跟踪代表了一种近乎理想的运动管理策略，因为它可以通过消除治疗边缘来实现对靶点的适形剂量分布，同时患者自由呼吸，最大限度地减少射束输送中断。实时肿瘤跟踪的成功实施需要准确的肿瘤位置识别、肿瘤运动预测（也考虑射束定位响应中的时间延迟）、射束重新定位技术和准确的4D剂量计算模型。目前可以进行动态肿瘤跟踪的方法涉及在移动的肿瘤中或附近植入多个标记物。不幸的是，标记物和植入手术的成本、与手术相关的潜在毒性（包括过度出血或气胸）、潜在治疗延迟和标记物迁移可能超过这些技术的受益。已经提出了使用外部替代物作为预测肿瘤位置的手段，但有两个主要的局限性。首先，大多数外部替代物是放置在任意位置的单个块，这些位置可能与内部运动不太相关。第二，呼吸运动模型通常基于传统的4D-CT采集，其仅考虑每个切片1-2个呼吸周期，并且当患者不规则呼吸时易受运动伪影的影响。我们建议，表面引导放射治疗（SGRT），使用3-3D立体相机吊舱跟踪患者表面上的预定义感兴趣区域，以及实时荧光透视kV成像，可以实时无创预测肿瘤的位置，以允许动态肿瘤跟踪。为了便于预测模型，我们的目标是开发一个患者特定的4D-CT呼吸运动模型，可以在一分钟的扫描，通过与体积CT扫描仪成像几乎消除运动伪影。该扫描仪可以在上级/下级方向上成像16 cm，每0.28 s同时提供肿瘤和皮肤表面的3D-CT图像。使用回顾性CT重建方法，我们可以将图像采集时间减少到0.1秒。我们的目标是使用生物力学建模和深度学习的组合来建立皮肤表面运动和内部肿瘤运动之间的相关性。在对人类癌症患者进行测试之前，将在呼吸运动模型和临床前模型上对这些技术进行验证。容积4D-CT、SGRT、实时kV成像以及结合深度学习和肺生物力学特性的新型呼吸运动模型的组合将为非侵入性动态肿瘤跟踪提供一种新方法。

Managing respiratory-induced tumour motion during radiation therapy still poses a major challenge in ensuring the intended radiation dose is delivered to the tumour while ensuring that nearby critical organ radiation dose limits are not exceeded. Real-time tumour tracking during radiation therapy represents a near ideal motion management strategy as it can allow for a conformal dose distribution to the target by eliminating treatment margins while the patient breathes freely with minimal beam delivery interruption. Successful implementation of real-time tumour tracking requires accurate tumour position identification, tumour motion prediction that also accounts for time delay in beam positioning response, beam repositioning technique, and an accurate 4D dose calculation model. Current methods that can perform dynamic tumour tracking involve implantation of multiple markers in or near the moving tumour. Unfortunately, the benefits of these techniques may be outweighed by the cost of the markers and implanting procedure, potential toxicities associated with the procedure, including excessive bleeding or pneumothorax, potential treatment delays, and marker migration. The use of external surrogates as a means of predicting tumour position have been proposed but have two major limitations. The first is that most external surrogates are single blocks placed at arbitrary positions that may not be well correlated with the internal motion. Second, breathing motion models are typically based on conventional 4D-CT acquisition which only consider 1-2 breathing cycles per slice and are susceptible to motion artifacts when patients breathe irregularly. We propose that surface-guided radiation therapy (SGRT) that uses 3-3D stereo camera pods to track a predefined region of interest on a patient's surface, together with real-time fluoroscopic kV imaging, can non-invasively predict the position of the tumour in real-time to allow for dynamic tumour tracking. To facilitate the prediction model, we aim to develop a patient-specific 4D-CT breathing motion model that can be acquired over one minute of scanning with near eliminated motion artifacts by imaging with a volumetric CT scanner. This scanner can image 16cm in the superior/inferior direction providing 3D-CT images of the tumour and skin surface, simultaneously every 0.28s. With retrospective CT reconstruction methods, we can decrease the image acquisition time to 0.1s. We aim to use a combination of biomechanical modeling and deep learning to establish a correlation between the skin surface motion and internal tumour motion. Validation of these techniques will be performed on respiratory motion phantoms and a preclinical model before testing on human cancer patients. The combination of volumetric 4D-CT, SGRT, real-time kV imaging, and a novel breathing motion model that combines deep learning and lung biomechanical properties will provide a novel approach to non-invasive dynamic tumour tracking.