Performance comparison of algorithms for estimating the strain of soft tissue from ultrasound

作     者：Li, XF Huang, LH Zhang, GX Freiheit, T 

作者机构：Tianjin Univ State Key Lab Precise Measurement Technol & Instr Tianjin 300072 Peoples R China Univ Calgary Dept Mech & Mfg Engn Calgary AB Canada 

出 版 物：《INSTRUMENTATION SCIENCE & TECHNOLOGY》 (仪器科学与技术)

年 卷 期：2005年第33卷第6期

页      面：673-689页

核心收录：

学科分类：07[理学] 0804[工学-仪器科学与技术] 0703[理学-化学] 

基　　金：China National Science Foundation, (50375108) Tianjin Science Foundation, (033601611) 

主　　题：soft tissue ultrasound strain correlation algorithm 

摘      要：A specially designed device is used to link ultrasound and force transducers by mounting them in series. The combined transducers can simultaneously sense the deformation and compression force in the soft tissue. In this system, the method for obtaining the exact strain of the soft tissue during compression is very important. In this paper, six different algorithms to estimate the strain of the tissue are compared: Normalized Correlation (NC), Direct Correlation (DC), Sum Absolute Differences (SAD), Sum Squared Differences (SSD), Hybrid-sign Correlation (HC), and Maximum algorithm. In order to evaluate the results of the six algorithms, a calibration system is setup in which the grating transducer and ultrasound sensor are connected in series. The strain of the soft tissue analyzed by six methods is compared. The results show that the estimating accuracy of tissue strain by the NC, DC, and SSD algorithms are completely the same, and there is no distinct difference in estimating accuracy between the SAD and NC (or DC and SSD) algorithms(alpha = 0.05). The standard deviation and maximal deviation of the delay time obtained by the HC and Maximum algorithms is unstable, and the delay time estimated by these two algorithms also shows false peaks. The time to run the SAD algorithm is more than the time to run the DC and SSD algorithms (alpha = 0.05), while there is no distinct difference in the time to run the DC and SSD algorithms. For systems with limited hardware resources, the DC and SSD algorithms are suitable.