Enhanced complex local frequency elastography method for tumor viscoelastic shear modulus reconstruction. (October 2020)

• Record Type: Journal Article
• Authors: Hu, Liangliang; Shan, Xiang
• Is Part Of: Computer methods and programs in biomedicine. Volume 195(2020)
• Publication Date: 2020-10
• Subjects: Inverse problem -- Magnetic resonance elastography -- Shear modulus -- Viscoelasticity -- Tumor; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2020.105605
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: Proposed a highly sensitive elastography method for the identification of small tumors. Introduced a convolutional neural network to enhance robustness of viscoelastic reconstruction. Subfeatures of neural network were set manually to avoid the complex learning process. Proposed method outperformed the MRE Wave software by the Mayo Clinic. Abstract: Background and objectives: The Mayo Clinic provides a magnetic resonance (MR) elastography software named MRE Wave, which uses the conventional local frequency elastography (LFE) method. However, MRE Wave is unable to supply complex viscoelasticity maps for elastography. We sought to improve the local frequency estimation algorithm used in LFE, which we refer to as the Enhanced Complex Local Frequency Elastography (EC-LFE) algorithm. Methods: The proposed algorithm uses wave equations under the hypotheses of being linear, isotropic, and locally homogeneous. Two 2D simulation models were used to investigate the accuracy and sensitivity of the EC-LFE algorithm for detecting small tumors. The corresponding statistical parameters were the relative root mean square (RMS) error and contrast-to-noise ratio (CNR). EC-LFE was investigated with two different parameter sets, one with an optimally chosen parameter ξ (EC-LFE Adj, for short) and the other with ξ = 0 (EC-LFE0). We compared the MRE Wave and the EC-LFE using series signal-to-noise (SNR) wave data. Results: The elasticity RMS error of the MRE Wave software was about 1%, and that of the EC-LFE0 and EC-LFE Adj were about 0.2%. The elasticity standard deviation of the MRE Wave software was about 3% of the mean value, and those of the EC-LFE0 and EC-LFE Adj were about 1% of the mean value. The elasticity CNR value of EC-LFE0 reached 1.93 times that of the MRE Wave in the region of small tumors (less than 10-point sampling). The viscosity RMS errors of the EC-LFE0 could be less than 5%. Conclusion: Compared to conventional methods, the EC-LFE was more accurate and sensitive for small tumor detection and exhibited higher noise immunity. The improved algorithm output more parameters and outperformed than the MRE Wave, thereby rendering them more suitable for clinical applications.