A contextual detector of surgical tools in laparoscopic videos using deep learning.
A contextual detector of surgical tools in laparoscopic videos using deep learning.
复制标题
DOI：
10.1007/s00464-021-08336-x
复制
发表时间：
2022-01
期刊：
影响因子：
--
通讯作者：
Devarajan V
中科院分区：
文献类型：
--
作者：
Namazi B;Sankaranarayanan G;Devarajan V
腹腔镜术的复杂性需要专门的培训和评估。分析手术过程中的流媒体视频可能会改善外科教育。除其他外，使用自动化工具检测系统可以极大地减少这种分析的单调和成本。我们提出了一种新的多标签分类器，称为LapTool-Net，用于检测腹腔镜视频的每一帧中是否存在手术工具。LapTool-Net的新奇之处在于利用了不同工具的使用与工具和任务之间的关联，即工具使用的上下文。为了达到这一目标，利用工具共现的模式，设计了一种端到端训练的基于递归卷积神经网络的多标签分类器的决策策略。在后处理步骤中，通过使用RNN对长期任务的顺序进行建模来修正预测。LapTool-Net使用公开可用的腹腔镜胆囊切除术数据集，即M2CAI16和Cholec80进行训练。对于M2CAI16，在线和离线模式下的准确匹配准确率分别为80.95%和81.84%，每类F1得分分别为88.29%和90.53%。对于Cholec80，线上准确率为93.10%，离线准确率为96.11%，F1评分准确率分别为85.77%和91.92%。结果表明，LapTool-Net显著优于最先进的方法，即使使用较少的训练样本和较浅的体系结构。我们的上下文感知模型不需要专家的领域特定知识，并且简单的体系结构可以潜在地改进所有现有的方法。
The complexity of laparoscopy requires special training and assessment. Analyzing the streaming videos during the surgery can potentially improve surgical education. The tedium and cost of such an analysis can be dramatically reduced using an automated tool detection system, among other things. We propose a new multilabel classifier, called LapTool-Net to detect the presence of surgical tools in each frame of a laparoscopic video. The novelty of LapTool-Net is the exploitation of the correlations among the usage of different tools and, the tools and tasks - i.e., the context of the tools’ usage. Towards this goal, the pattern in the co-occurrence of the tools is utilized for designing a decision policy for the multilabel classifier based on a Recurrent Convolutional Neural Network (RCNN), which is trained in an end-to-end manner. In the post-processing step, the predictions are corrected by modeling the long-term tasks’ order with an RNN. LapTool-Net was trained using publicly available datasets of laparoscopic cholecystectomy, viz., M2CAI16 and Cholec80. For M2CAI16, our exact match accuracy (when all the tools in one frame are predicted correctly) in online and offline modes were 80.95% and 81.84% with per-class F1-score of 88.29% and 90.53%. For Cholec80, the accuracies were 85.77% and 91.92% with F1-scores if 93.10% and 96.11% for online and offline respectively. The results show LapTool-Net outperformed state-of-the-art methods significantly, even while using fewer training samples and a shallower architecture. Our context-aware model does not require expert’s domain-specific knowledge and the simple architecture can potentially improve all existing methods.
登录
查看更多内容
DOI：
10.1007/s11548-016-1393-4
发表时间：
2016-06
影响因子：
3
作者：
Du X;Allan M;Dore A;Ourselin S;Hawkes D;Kelly JD;Stoyanov D
通讯作者：
Stoyanov D
影响因子：
6
作者：
Charte, Francisco;Rivera, Antonio J.;Herrera, Francisco
通讯作者：
Herrera, Francisco
DOI：
10.1007/s11548-019-01958-6
发表时间：
2019-06-01
影响因子：
3
作者：
Nwoye, Chinedu Innocent;Mutter, Didier;Padoy, Nicolas
通讯作者：
Padoy, Nicolas
影响因子：
10.6
作者：
Jin, Yueming;Dou, Qi;Heng, Pheng-Ann
通讯作者：
Heng, Pheng-Ann
DOI：
10.1007/s00464-012-2237-0
发表时间：
2012-09-01
影响因子：
3.1
作者：
Perrenot, Cyril;Perez, Manuela;Hubert, Jacques
通讯作者：
Hubert, Jacques