Three-dimensional Motion Tracking Of Beating Heart Based On Endoscope

Minimally invasive surgery has developed rapidly in recent years.However,due to the rapid movement of the heart and other reasons,this type of surgery is still very chalenging,so the work of three-dimensional motion tracking of the heart becomes very important.In addition,in the beating heart motion tracking based on stereo endoscope,the tracking process is often interrupted by specular reflection,occlusion and other dynamic interference.Wel-designed predictions can provide missing visual information when the tracking fails and resume normal tracking after the interference has ended.Based on the above problems,the three-dimensional motion tracking and motion prediction of the heart are studied in detail in this paper.The main research results are as follows:1)Aiming at the problem of low accuracy and low speed of three-dimensional motion tracking of heart based on traditional model,a three-dimensional motion tracking based on improved Low-rank Statistical Shape Model(LSM)is proposed.The thin plate spline model is first decoupled into shape and position components,and the decoupled model is used to pre-reconstruct the previous multi-frame region of interest.The feature shape is then learned from the pre-reconstructed results using Principal Component Analysis(PCA).Based on this,an LSM model is constructed,and the subsequent frames are reconstructed by using the model.The 3D tracking problem is transformed into the model parameter optimization problem of each frame,and the parameters are solved iteratively using the Efficient Second-order Minimization(ESM)algorithm.2)The above method for performing three-dimensional motion tracking of the heart based on the LSM model has some disadvantages,such as being insensitive to large noise points,unable to eliminate height anomaly values during reconstruction,and having low computational efficiency and flexibility.Aiming at these defects,a three-dimensional motion tracking method based on improved Incremental Sparse and Low-rank Matrix Decomposition(ISLMD)is proposed.The algorithm decomposes the pose data matrix into a low rank pose data matrix and a sparse spike noise matrix.The former can be used to accurately reconstruct subsequence frames.The accurate low rank pose data matrix obtained by the decomposition is used to model the heart deformation.In addition,adding a penalty term on the vision-based 3D motion tracking problem constrains the three-dimensional shape of the heart so that its deformation can be controlled within a reasonable range.The experimental results show that the proposed method has improved in robustness,accuracy and speed compared to the LSM method.3)Cardiac motion prediction is studied for the failure of cardiac motion tracking.Cardiac motion prediction based on Iterative Optimal Sinusoidal Filtering(IOSF)and cardiac motion prediction based on Long-term and Short-Term Memory(LSTM)networks are proposed.The former method uses the dual-time-varying fourier series to model the motion of the point of interest,then uses the dual kalman filter to estimate the frequency parameters and fourier coefficients of the model separately,and finally performs motion prediction.The latter method applies predictions based on LSTM networks to the heart.Through experiments comparison,it is found that the prediction error of the latter method in the phantom heart and the invivo heart are reduced by 50% and 80% respectively,which is better than the former.