Enabling technologies for quantitative performance assessment in simulated training of image-guided needle interventions

在加拿大，数百万人在超声波引导下进行微创针头干预，如注射、活组织检查和各种局部癌症治疗。取代侵入性开放手术，这些手术的主要目标是利用实时超声成像引导，精确地将手术针头导航到目标，同时避免周围正常组织受到附带损伤。虽然这些程序在专业医生手中可能看起来很简单，但它们往往是复杂而微妙的“小手术”，需要掌握困难的技能并遵守复杂的程序工作流程。传统上，新手医生在资深专家医生的监督下学习这些对人类的干预措施。除了使患者暴露在潜在的错误中外，这种典型的学徒情景从根本上限制了对技能和能力进行客观评估的机会。必须以实际和负担得起的方式，通过对受训人员的进展进行频繁和客观的评估，通过严格跟踪来确保其医学能力。出于患者安全、效率和学员舒适度的考虑，最佳做法是在模拟环境中进行多个阶段的练习，将患者换成人体模型，然后只有熟练掌握技术的学员才能在实际的患者护理中继续他们的培训。在目前的医学模拟器中，医生手中的手术器械上安装了各种物理传感器，以测量它们的位置、速度和加速度，然后使用这些测量的数值分析来推断受训者相对于专家医生基准的技能水平。然而，使用物理传感器有根本的局限性：它们改变了受训者的感知和学习过程，也许更重要的是，在没有物理传感器的真实临床环境中，不可能衡量受训者如何从模拟过渡到人类。此外，目前的“任务训练师”模拟器不能提供关于总体程序能力的信息，这可以说是允许受训者承担病人护理责任的最重要的衡量标准。我的计划的目标是克服这些障碍，开发使能技术，通过使用非侵入性多通道视频流(例如，网络摄像头、超声波)来计算程序能力的量化指标，从而消除培训环境中的物理传感器，并向学员提供建设性的可操作反馈。拟议计划的预期贡献将是开发方法和软件工具，以支持外科技能和能力的定量评估的跨学科研究。为新型医疗培训系统的发展提供信息的工程和计算机科学研究有可能提高从业者的能力，从而对医疗保健产生积极影响。

Ultrasound-guided minimally invasive needle interventions, such as injections, biopsies and a variety of localized cancer therapies, are performed by the millions in Canada. Replacing invasive open surgeries, the primary goals in these procedures are to precisely navigate a surgical needle to the target while sparing surrounding normal tissues from collateral damage, using real-time ultrasound imaging guidance. While these procedures may look straightforward in the hands of expert physicians, they are often complex and delicate "mini surgeries" that require the mastery of difficult skillsets and compliance to complex procedural workflows. Traditionally, novice physicians learn these interventions on humans under the supervision of senior expert physicians. In addition to exposing patients to potential errors, this classic apprenticeship scenario fundamentally limits the opportunity for objective assessment of skill and competence. Competency in medicine must be assured by rigorous tracking through frequent and objective assessment of the trainee's progress, in a manner that is practical and affordable. For reasons of patient safety, efficiency and learner comfort, multiple stages of practice are best done in simulated environments, where the patient is replaced with mannequins, and then only technically proficient learners may continue their training in actual patient care. In current medical simulators, a variety of physical sensors are attached to the surgical instruments in the physician's hands, to measure their position, speed, and acceleration, and then numerical analysis of these measurements is used to infer the trainee's level of skill with respect to expert physician benchmarks. There are, however, fundamental limitations in using physical sensors: they alter the trainee's perception and learning process, and perhaps more importantly, it is impossible to measure how trainees transition from simulation to humans in a real clinical setting where there are no physical sensors. Moreover, current "task trainer" simulators are unable to inform about overall procedural competence, arguably the most important metric for admitting a trainee to the responsibilities of patient care. The objective of my program is to overcome these barriers, by developing enabling technologies that eliminate physical sensors from the training environment by using non-intrusive multichannel video streams (e.g., webcams, ultrasound) to compute quantitative metrics of procedural competence and to provide constructive actionable feedback to learners. The intended contribution of the proposed program will be the development of methods and software tools in support of cross-disciplinary research on quantitative assessment of surgical skill and competence. Engineering and computer science research that informs the development of novel medical training systems has the potential to improve the competence of practitioners and, as a result, positively impact healthcare.