Smart navigation guidance during cardiac interventions

如今，导尿术是避免心内直视手术的首选方法，因为它减少了患者感染的危险和康复时间。支架置入、主动脉夹层和心脏瓣膜修补术都是导管术的很好的例子。然而，它需要心脏病专家更灵活，因为它涉及到操作呼吸患者血管内的柔性导管。这项研究计划的首要目标是研究新的计算工具，用于心脏介入治疗期间的智能导航指导，从而减轻这一困难。这项研究计划围绕三个具体目标进行阐述。该研究计划的第一个目标是通过双平面X射线血管造影术重建血管结构的3D路线图。由于两个视图对于精确重建并不总是足够的，生成性对抗网络(GAN)将被研究以学习3D/2D关系和综合额外的视图。第二个目标是基于长短期记忆(LSTM)网络开发一种新的运动模型来预测干预期间的心脏和呼吸运动，LSTM网络是一种用于建模时间序列的新型递归神经网络。在X射线血管造影术中使用神经网络进行运动建模以前从未被做过。在运动模型的开发中将考虑患者分组。最后，心脏病专家的手势和任务的建模将通过强化学习(RL)算法进行研究。RL预测的手势将与心脏病专家的手势进行比较，这些手势尚未对导管操作进行评估。这些知识将提高我们对心脏干预期间用户感知的理解，这一主题虽然只发表了几项研究，但对于开发远程或完全自主的心脏导航将非常重要。这项研究计划是创新的，因为它在3D重建、运动建模和人体手势建模方面提出了独特的和基本的工程贡献，这些都是密切相关的。它将在心脏病专家和机器人辅助和/或远程干预之间架起一座重要的桥梁。我的长期愿景是将这些计算工具与机器人导管相结合，形成端到端的智能导航辅助。这将提高操作员的灵活性和协调性，从而更安全地使用导尿管。这也将为使用机器人导管的智能远程导航系统铺平道路，从而提高患者的安全性，减少手术和康复时间，并显著降低加拿大医疗保健系统的医疗成本。这一研究计划将有助于在医学背景下对所有研究级别的HQP进行培训，为受训人员提供可直接转移到对加拿大重要的广泛行业的高度市场化的技能。

Nowadays, catheterization is preferred to avoid open heart surgery because it reduces the danger of infection to patient and duration of convalescence. Stent placement, aortic dissection and heart valve repairs are good examples of catheterization procedures. However, it requires far more dexterity from the cardiologist as it involves manipulating a flexible catheter inside a breathing patient's blood vessels. The overarching objective of this research program is to investigate new computational tools for smart navigation guidance during cardiac interventions, thereby alleviating this difficulty. This research program is articulated around 3 specific objectives. The first objective of this research program aims at reconstructing a 3D roadmap of the vascular structure from biplane X--ray angiography. Since two views are not always sufficient for accurate reconstruction, Generative Adversarial Networks (GAN) will be investigated to learn the 3D/2D relationship and to synthesize additional views. The second objective is to develop a new motion model to predict cardiac and respiratory movements during interventions, based on long short-term memory (LSTM) networks, a new class of recurrent neural networks for modelling time series. Motion modelling using neural networks in X--ray angiography has never been done before. Patient subgroups will be considered in the development of the motion model. Finally, modelling of gestures and tasks of the cardiologist will be investigated through reinforcement learning (RL) algorithms. Predicted gestures by RL, yet to be evaluated for catheter manipulation, will be compared against the gestures performed by the cardiologist. This knowledge will improve our understanding of the user's perception during a cardiac intervention, a topic where only a few studies have been published, but which would be very important for the development of remote or fully autonomous cardiac navigation. This research program is innovative since it proposes unique and fundamental engineering contributions in 3D reconstruction, motion modelling and human gesture modelling, which are closely connected. It will bridge an important gap between the cardiologist and robotic assistance and/or remote intervention. My long -term vision is to combine these computational tools with robotic catheter in an end--to--end smart navigation assistance. This will enhance the operator's dexterity and coordination towards even safer use of catheters. This will also pave the way towards a smart remote navigation system using robotic catheters, and consequently, improve safety for the patient, reduce operating and convalescence times and significantly reduce medical costs for the Canadian health care system. This research program will contribute to the training of HQP at all study levels in a medical context, providing trainees with highly marketable skills that are directly transferable to a broad range of industries of importance to Canada.