Surgical GAN: Towards Real-Time Path Planning for Passive Flexible Tools in Soft Tissue Surgery

Yan Zhao,Yuxin Wang,Jianhua Zhang,Shuxiang Guo,Jinxin Cui,Youchun Ma,Xinke Liu,Shunming Hong,Youxiang Li

DOI: https://doi.org/10.2139/ssrn.3708318

2020-01-01

Abstract:Background: As one of the most important preoperative steps, surgical path planning of passive flexible tool in soft tissue encounters a prohibitive challenge, typically in vascular interventional surgery(VIS). The key problem is that unstructured surgical environment and tools’ unpredictable motion is hard to be explicitly modeled. We propose a generative adversarial networks (GAN)-based framework (defined as surigcal GAN) towards automatic guide wire path planning in real time for VIS. Methods: A novel GAN architecture is proposed by combining convolutional neural networks (CNN) and long short term memory networks (LSTM), which extracts and fuses the spatial features in medical images and temporal features of historical tool path as the conditional information. It inputs the surgical state information and continously outputs local future path of guide wire tip. A training dataset (3.5*105 samples) is collected under laboratory conditions with 10 surgeons. Effects of different CNN architectures and path planning length on network performance are investigated. User experiments are conducted with 10 novice surgeons in an operating room. Findings: The results shows significant reduction of path planning accuracy with surgical GAN comparing with baseline networks (from 46.2%-69.78%) and non-rigid registration method (72.94%). Results of user experiments demonstrate overall better task performance with the guidance of planned surgical path. Interpretation: Collectively, surgical GAN can achieve real-time surgical path planning of passive flexible tools in soft tissue surgery, and holds great potential for novice training and robotic VIS autonomy. Funding Statement: The National High-tech R&D Program of China [No.2015AA043202], National Key Research and Development Program of China [No.2017YFB1304400, 2018YFB1306500], National Nature Science Foundation of China [No.62003128], Natural Science Foundation of Hebei Province of China [No.F2020202053], Natural Science Foundation of Tianjin City of China [No.S20YBE154] and Preferential funding for postdoctoral research projects of Hebei Province of China [No.B2020003020]. Declaration of Interests: The authors declare no potential conflicts of interest.