Using a Wavelet-Based and Fine-Tuned Convolutional Neural Network for Classification of Breast Density in Mammographic Images

Magnetic resonance elastography (MRE) can visualize the shear wave propagation of in vivo tissues, which can be mapped into viscoelastic properties. No study has measured the biomechanical properties of the PM muscle in vivo using MRE. In this study, we evaluated stiffness values calculated by local frequency estimate (LFE) and algebraic inversion of differential equation (AIDE) in PM-MRE. The PM muscles of 17 healthy male volunteers were scanned in supine position by MRE. The Laplacian-based estimate (LBE) phase wrapped image data were filtered by gaussian-bandpass filter (GBF), and by both directional and GBF. LFE (MREWave) and AIDE wave inversion methods were used to calculate the respective elastograms. The wave interferences were removed by directional filtering, and smooth wave fields were obtained. The stiffness values calculated by LFE of non-DF images were 1.39 ± 0.25 kPa and 1.33 ± 0.22 kPa for right and left PM respectively, whereas for DF images, they were 1.26 ± 0.20 kPa for right PM and 1.18 ± 0.28 kPa for left PM. The stiffness values calculated by AIDE of non-DF images were 0.78 ± 0.10 kPa and 0.78 ± 0.13 kPa for right and left PM respectively, whereas for DF images, they were 0.73 ± 0.12 kPa for right PM and 0.74 ± 0.11 kPa for left PM. There was no statistically significant difference in mean values of stiffness with/without applying directional filter whereas there was a statistically significant difference in mean values of stiffness between LFE and AIDE. Both LFE and AIDE could be used for psoas major MR Elastography.