Multivariate Autoregressive Model Based Heart Motion Prediction Approach for Beating Heart Surgery

A robotic tool can enable a surgeon to conduct off-pump coronary artery graft bypass surgery on a beating heart. The robotic tool actively alleviates the relative motion between the point of interest (POI) on the heart surface and the surgical tool and allows the surgeon to operate as if the heart were stationary. Since the beating heart's motion is relatively high-band, with nonlinear and nonstationary characteristics, it is difficult to follow. Thus, precise beating heart motion prediction is necessary for the tracking control procedure during the surgery. In the research presented here, we first observe that Electrocardiography (ECG) signal contains the causal phase information on heart motion and non-stationary heart rate dynamic variations. Then, we investigate the relationship between ECG signal and beating heart motion using Granger Causality Analysis, which describes the feasibility of the improved prediction of heart motion. Next, we propose a nonlinear time-varying multivariate vector autoregressive (MVAR) model based adaptive prediction method. In this model, the significant correlation between ECG and heart motion enables the improvement of the prediction of sharp changes in heart motion and the approximation of the motion with sufficient detail. Dual Kalman Filters (DKF) estimate the states and parameters of the model, respectively. Last, we evaluate the proposed algorithm through comparative experiments using the two sets of collected vivo data.