AI for Surgical Robotics 15 map, and the navigation is done based on this map. During surgery, the OR coordinate system must be registered to the preoperative model. This registration is developed based on the system capabilities, for example known landmarks (anatomical or artificial), that are seen both in the CT and in the c-arm images. These landmarks are used for the registration. A different type of registration can be done during real time navigation and by leveraging Simultaneous Localization and Mapping (SLAM) algorithms into this unique setup. Object tracking - An external tracking systemisusedtotrackdifferenthardware objects used in the surgery. Integration of tracking systems is not a trivial task as its limitations and dead spots can impact the performance. Designing and taking into consideration the entire setup is required when using tracking systems. Watch Video Integration & temporal algorithms Temporal understanding - In some cases the understanding of the scene and its different stages is needed. Addressing this task can be done by using an LSTM neural network, which is based on using memory from previous periods as well as data from the current one. Applying this will enable the system to know at what stage the procedure is, and to predict subsequent events. Integration & real time feedback - In some cases the system has real time feedback. This feedback can be used to improve the location of different tools in the scene. Extended Kalman filter is one of the leading solutions to integrate different types of data into a single system and leverage all the available data and their variances. Summary In order to implement a robotic surgery system, a multidisciplinary suite of algorithms is required. A robust solution will need to integrate the different algorithms within a single system, in order to meet the overall precision requirements. For more information and to schedule a call with our engineers, please contact us at contact@rsipvision. com Fig 5. Tool segmentation in surgical video.