Development of novel machine learning algorithms for registration of point clouds and tracking surgical tools

这种合作关系有两个重点，以解决全膝关节置换术（TKA）中注册和跟踪的两个首要问题。注册和跟踪将确定切割大腿骨的位置和角度，这是一个无法撤销的动作。因此，这两项努力都至关重要。推力1将侧重于鲁棒点云配准、配准精度分类和个性化点选择策略。这一推力也将在两个目标上进行。在第一个目标中，我们首先为具有数学挑战性的配准问题开发新的配准技术。然后，我们将开发能够自动评估注册准确性的新方法，以便只有在注册准确的情况下才能进行该程序。在第二个目标中，我们将开发一种针对每个骨骼的个性化采样策略，以便通过选择很少的点来获得最佳的配准结果。在这两个目标中，我们将开发使用传统机器学习（ML）和深度学习（DL）算法的新技术。推力2将专注于手术工具的鲁棒跟踪，并自动检测跟踪是否不准确。这一推力将在两个具体目标中进行，其中第一个目标侧重于经典图像处理和ML技术，第二个目标侧重于深度学习。除了提高跟踪精度外，我们提出的方法将自动检测是否有一个工具没有被准确跟踪，以便外科医生可以清洁受影响的工具。这种研究伙伴关系可以促进新技术的发展，使TKA的定位和登记更加准确和可靠，有可能改善手术结果。它还将培训六名高素质人员，与一家加拿大公司密切合作。

This partnership has two thrusts to solve two overarching problems of registration and tracking in total knee arthroplasty (TKA). Registration and tracking will determine the location and angle for cutting the femur bone, an action that cannot be undone. As such, both thrusts are of critical importance. Thrust 1 will focus on robust point cloud registration, classification of the accuracy of the registration, and personalized point selection strategy. This thrust will also be carried out in two aims. In the first aim, we first develop novel registration techniques for the mathematically challenging problem of registration. We will then develop novel methods that can automatically assess the accuracy of the registration, so that the procedure can only proceed if registration is accurate. In the second aim, we will develop a novel sampling strategy that is personalized to each bone, so that optimal registration results can be achieved with the selection of few points. In both aims of this thrust, we will develop novel techniques that use both traditional Machine Learning (ML) and Deep Learning (DL) algorithms. Thrust 2 will focus on robust tracking of surgical tools, and automatically detecting if tracking is inaccurate. This thrust will be carried out in two specific aims, wherein the first aim focuses on classical image processing and ML techniques, and the second aim focuses on DL. In addition to improving the tracking accuracy, our proposed method will automatically detect if one of the tools is not being accurately tracked, so that the surgeon can clean the affected tool.This research partnership can lead to development of novel technologies that enable more accurate and reliable localization and registration in TKA, potentially improving the outcome of the surgery. It will also lead to training of six highly qualified personnel (HQP) working in close collaboration with a Canadian company.