Department of Computer Science, Peking University, Beijing, China.
Deepwise AI Lab, Beijing, China.
Int J Comput Assist Radiol Surg. 2020 Nov;15(11):1817-1824. doi: 10.1007/s11548-020-02267-z. Epub 2020 Oct 12.
Automatic surgical skill assessment is an emerging field beneficial to both efficiency and quality of surgical education and practice. Prior works largely evaluate skills on elementary tasks performed in the simulation laboratory, which cannot fully reflect the variety of intraoperative circumstances in the real operating room. In this paper, we attempt to fill this gap by expanding surgical skill assessment onto a clinical dataset including fifty-seven in vivo surgeries.
To tackle the workflow and device constraints in the clinical setting, we propose a robust and non-interruptive surrogate for surgical skills, namely the clearness of operating field (COF), which shows strong correlation with overall skills and high inter-annotator consistency on our clinical data. Then, an automatic model based on neural networks is developed to regress surgical skills through the surrogate of COF using only video as input.
The automatic model achieves 0.595 Spearman's correlation with the ground truth of overall technical skill, which even exceeds the human performance of junior surgeons. Moreover, an exploratory study is conducted to validate the skill predictions against the clinical outcomes of patients.
Our results demonstrate that the surrogate of COF is promising and the approach is potentially applicable to clinical practice.
自动手术技能评估是一个新兴领域，有利于提高手术教育和实践的效率和质量。先前的工作主要评估在模拟实验室中进行的基本任务上的技能，这些任务无法完全反映真实手术室中的各种手术情况。在本文中，我们试图通过将手术技能评估扩展到包括五十七例体内手术的临床数据集来填补这一空白。
为了解决临床环境中的工作流程和设备限制，我们提出了一种稳健且非中断的手术技能替代指标，即手术视野清晰度（COF），该指标在我们的临床数据上与整体技能具有很强的相关性，并且具有较高的注释者间一致性。然后，我们开发了一种基于神经网络的自动模型，通过仅使用视频作为输入，通过 COF 的替代指标来回归手术技能。
自动模型与整体技术技能的真实值之间的 Spearman 相关系数达到 0.595，甚至超过了初级外科医生的人类表现。此外，还进行了一项探索性研究，以验证针对患者临床结果的技能预测。
我们的结果表明，COF 的替代指标很有前途，该方法有可能适用于临床实践。
