[论文解读] An AI Framework for Microanastomosis Motion Assessment

Yan Meng, Eduardo Torres-Rodríguez+6 more|arXiv (Cornell University)|Jan 28, 2026

Surgical Simulation and Training被引用 0

一句话总结

该论文提出一个端到端的AI框架，用于通过基于YOLO的检测、DeepSORT跟踪、工具尖端定位以及基于NOMAT的技能分类，对微吻合器器械操作进行自动评估，达到较高的检测准确度并与专家评分有良好一致性。

ABSTRACT

Proficiency in microanastomosis is a fundamental competency across multiple microsurgical disciplines. These procedures demand exceptional precision and refined technical skills, making effective, standardized assessment methods essential. Traditionally, the evaluation of microsurgical techniques has relied heavily on the subjective judgment of expert raters. They are inherently constrained by limitations such as inter-rater variability, lack of standardized evaluation criteria, susceptibility to cognitive bias, and the time-intensive nature of manual review. These shortcomings underscore the urgent need for an objective, reliable, and automated system capable of assessing microsurgical performance with consistency and scalability. To bridge this gap, we propose a novel AI framework for the automated assessment of microanastomosis instrument handling skills. The system integrates four core components: (1) an instrument detection module based on the You Only Look Once (YOLO) architecture; (2) an instrument tracking module developed from Deep Simple Online and Realtime Tracking (DeepSORT); (3) an instrument tip localization module employing shape descriptors; and (4) a supervised classification module trained on expert-labeled data to evaluate instrument handling proficiency. Experimental results demonstrate the effectiveness of the framework, achieving an instrument detection precision of 97%, with a mean Average Precision (mAP) of 96%, measured by Intersection over Union (IoU) thresholds ranging from 50% to 95% (mAP50-95).

研究动机与目标

• 解决神经外科训练中客观、可扩展的显微手术技能评估需求。
• 开发一个端到端的AI系统，用于对微吻合术中的器械操作进行自动评估。
• 整合鲁棒的检测、跟踪、尖端定位和专家标注的技能分类。
• 提供适于培训场景的实时、硬件高效的评估。

提出的方法

• 使用定制的YOLOv11模型进行器械检测，具备每帧两个器械、不同类型的约束。
• 混合式器械跟踪，优先采用YOLO检测而非DeepSORT，并保持一致的ID以提高时间鲁棒性。
• 通过多点形状描述符实现器械尖端定位，与参考描述符进行余弦相似度匹配。
• 在专家标注的NOMAT基础评估上进行监督学习的分类模块，以分级器械操作熟练度。
• 从器械尖端轨迹中提取运动特征（速度、加速度、跃度），采用基于TSFresh的可扩展特征表示，并用梯度提升进行技能分类。

实验结果

研究问题

• RQ1 在微吻合术视频中，利用经微手术器械变体扩展的YOLO检测器，检测精度能达到何种水平？
• RQ2 集成跟踪与尖端定位是否能提供可靠的连续运动学数据用于客观技能评估？
• RQ3 在NOMAT标签上训练的监督模型在不同技能类别下能否较好复现专家评估？
• RQ4 所提出框架在训练环境中的实时计算性能如何？

主要发现

Class	No. of Images	Total Instances	Precision	Recall	mAP50	mAP50–95	Recovery	Correction
all	4174	9417	0.969	0.966	0.989	0.906	0.987	0.906
scissors_c	398	398	0.981	0.960	0.992	0.976	0.992	0.864
scissors_s	70	70	0.921	1	0.995	0.979	0.995	0.872
needle_driver_c	2861	2861	0.989	0.975	0.993	0.973	0.989	0.902
needle_driver_s	3669	3779	0.986	0.975	0.992	0.965	0.994	0.895
needle	2309	2309	0.971	0.918	0.972	0.872	0.963	0.997

• 总体器械检测精度为0.969，mAP50为0.987，mAP50–95为0.906。
• 平均实时处理速度为每秒29.7帧。
• 总体技能分类准确率为0.87，覆盖Poor、Moderate与Good三个类别。
• 按类别的技能指标显示Moderate和Good类别的表现优于Poor，指出标注变异性与数据集规模问题。
• 跟踪改进带来对YOLO错误分类或漏检的高恢复率（98.7%）与纠正率（90.6%）。
• 器械尖端定位在各帧保持较高的空间精度，便于后续运动分析。