Research On Ultrasonic Speckle Tracking Algorithm And Elastography Application Based On Feature Points

Ultrasound elastography has been widely used in clinical medical diagnosis with its low price and clear strain imaging advantages over the past 30 years,which is of great significance for the development of modern medicine.The quality of ultrasound elastography image directly depends on the accuracy of the displacement estimation of the pre and post compression ultrasound signal,Higher displacement estimation accuracy algorithm has always been important researched in the field of signal tracking.A new algorithm based on feature point is proposed in this paper.In order to implement the algorithm for high-precision tracking of two-dimensional ultrasonic signals,the paper has accomplished the following two aspects of research work.1.Feature point method selection for ultrasonic signals:Because the ultrasonic signal is filled with noise,it is different from ordinary image,in order to obtain a feature point method that is more suitable for ultrasonic signals,(Features from Accelerated Segment Test,FAST)and(Histogram of Oriented Gridients,HOG)feature point algoritlums are firstly applied to simulated ultrasound signals.The experimental results on the signal data show that they have poor accuracy of displacement estimation compared with the two traditional fast(Normalized Cross-Correlation,NCC)algorithms,Then,we use the error analysis method to obtain the reason that FAST and HOG feature point algorithms are not suitable for the ultrasound signal and select the(Scale-Invariant Feature Transform,SIFT)algorithm suitable for the ultrasonic signal.Finally,the SIFT algorithm is applied to realize the displacement tracking of two-dimensional ultrasonic signals.2.The new high-precision displacement tracking algorithm is proposed and implemented:This paper proposes a new ultrasonic speckle displacement tracking algorithm based on the SIFT feature method combined with bilinear motion model.The algorithm first uses the SIFT algorithm to obtain the feature points of the local area in two frame ultrasonic signals.Then the feature points are matched to generate matching pairs in the two frame signals.Then analysis if the number of matching pairs exceeds the threshold.If the number of matching pairs exceeds the threshold,the displacement of the signal is estimated by the average difference between the lateral and axial displacements of the matching pairs.If it is less than the threshold,the bilinear motion model method is used to estimate the displacement of the signal.Experiments show that the SIFT feature is combined with bilinear motion model algorithm in this paper.It's the lateral strain SNRe is improved by 17.6%compared with the sub-sampling fast NCC algorithm based on two-dimensional quadratic polynomial fitting.The axial strain SNRe is increased by up to 80.5%,the highest lateral strain contrast noise ratio increased by 4 times,and the highest in the axial strain contrast noise ratio is improved by 50%.Through computer digital phantom experiments,the strain calculation accuracy of this algorithm is indeed better than the two sub-sampling fast NCC algorithms and bilinear motion model algorithm.In summary,the algorithm in this paper is more accurate than the two sub-sampling fast NCC algorithms and the bilinear motion model algorithm in terms of lateral and axial displacement estimation or lateral and axial strain calculations.