Assessing the critical view of safety in laparoscopic cholecystectomy requires accurate identification and localization of key anatomical structures, reasoning about their geometric relationships to one another, and determining the quality of their exposure. In this work, we propose to capture each of these aspects by modeling the surgical scene with a disentangled latent scene graph representation, which we can then process using a graph neural network. Unlike previous approaches using graph representations, we explicitly encode in our graphs semantic information such as object locations and shapes, class probabilities and visual features. We also incorporate an auxiliary image reconstruction objective to help train the latent graph representations. We demonstrate the value of these components through comprehensive ablation studies and achieve state-of-the-art results for critical view of safety prediction across multiple experimental settings.

翻译：评估腹腔细胞切除的临界安全观点需要准确辨别和定位关键的解剖结构,解释它们之间的几何关系,并确定其暴露质量。在这项工作中,我们提议通过模拟外科手术场景来捕捉其中的每一个方面,用一个分解的潜伏图示来模拟,然后我们可以用一个图形神经网络来处理。与以前使用图示的方法不同,我们明确地将物体位置和形状、等级概率和视觉特征等语义信息编码在我们的图表中。我们还加入了一个辅助图像重建目标,以帮助培训潜伏图示。我们通过全面的通缩研究来展示这些组成部分的价值,并实现对多个实验环境的安全预测的批判性观点的最新结果。