Estimation of Kidney's Blood Vessels Deformations for Robot-Assisted Surgery

Surgical robotics may use augmented reality (AR) to provide surgeons with contextual information during operations. Relevant information to show to the surgeon are soft-tissue deformations. However, real-time computation of soft-tissue deformations is challenging with current techniques. A novel approach is proposed using Machine Learning (ML) to estimate, based on organ pose, deformations in the vein and artery. Simulated experiments were conducted using a kidney model from a patient's CT scan in a Simulation Open Framework Architecture (SOFA) platform, and data on blood vessel deformations were collected. Two ML models were trained, a feed-forward neural network and an LSTM, and their performances were compared. Although the results are preliminary, they demonstrate the potential of the proposed approach, particularly given the limited information required to perform the inference.