Study On Beamspace Virtual Subarrays Eigenspace-Based Minimum Variance Beamforming

Medical ultrasound imaging technology has the advantages of real-time,non-invasive,non ionizing radiation and low cost.As an important medical imaging method,it is widely used in clinical diagnosis.In ultrasonic imaging system,excellent beamforming technology can effectively improve the quality of ultrasonic image from the aspects of contrast and resolution.Traditional ultrasonic imaging uses low complexity time-delay superposition method to reconstruct the ultrasonic image,but it only simply adds the echo signals,which can not distinguish the desired signal and clutter,resulting in the generation of low-quality images with wide main lobe and high side lobe.In recent years,adaptive beamforming technology based on echo signal characteristics has been proved to be able to generate high-quality beams with narrow main lobe and low side lobe,which has become a research hotspot in the field of ultrasonic imaging.In this paper,based on the existing adaptive beamforming algorithm,we further study its improvement method and design a new beamforming algorithm to improve the image quality.The main research work and achievements are summarized as follows:Firstly,this paper introduces the traditional delay accumulation beamforming technology and adaptive beamforming technology.Due to the strong correlation of ultrasonic signal,the traditional beamforming algorithm has poor robustness.Smoothing technology solves the problem of strong correlation of ultrasonic signal to a certain extent,but it brings aperture loss.Then,aiming at the problem of aperture loss in traditional smoothing techniques,a virtual subarrays eigenspace-based minimum variance(VS-ESBMV)beamforming algorithm based on virtual subarrays is designed.On the one hand,the algorithm can effectively avoid the loss of array aperture and improve the spatial resolution of ultrasonic image;On the other hand,the algorithm ensures the decorrelation ability of the ultrasonic echo signal,and effectively improves the robustness of the adaptive beamforming algorithm.However,the application of virtual subarrays will increase the dimension of the ultrasonic echo signal matrix,thus increasing the computation of matrix inversion and eigenvalue decomposition.Therefore,a beamspace virtual subarrays eigenspace-based minimum variance(BVS-ESBMV)beamforming algorithm is proposed.In this method,beamforming in element domain is converted to beam domain.By constructing an appropriate conversion matrix,the beam with the largest contribution to the ultrasonic image is selected in the beam domain for beamforming,so as to reduce the computational complexity of the whole beamforming algorithm and improve its portability.Finally,the beamformer of long short term memory networks(LSTM)is designed.Neural network technology has proved its ability to fit the nonlinear relationship between input and output,and has also been successfully applied to prediction and classification in reality.In this paper,a LSTM beamformer is built for the ultrasonic signal.Based on the minimum variance beamforming algorithm in feature space,the weight vector of the ultrasonic signal received by the array element is calculated,and the data obtained by the ESBMV algorithm is input into the LSTM network for training;The trained LSTM beamformer generates weight coefficients for different ultrasonic echo signals and performs beamforming.The final imaging effect is comparable to that of the ESBMV algorithm,and even the near-field resolution is better than that of the ESBMV algorithm.To sum up,this paper improves the sub array smoothing algorithm of adaptive beamforming to solve the disadvantage of array loss in ultrasonic imaging,and further improves the robustness of beamforming algorithm and the image quality of ultrasonic imaging.In addition,the beamforming method based on virtual subarray is transformed into beam domain,which effectively solves the problem of increasing the amount of computation caused by the increase of dimension of virtual subarray,and improves the portability of the algorithm.In particular,a beamformer based on neural network with good beamforming effect is designed,which has important academic significance and research value for the development of medical ultrasound in the direction of artificial intelligence.