RF-based motion estimation using non-rigid image registration techniques: In-silico and in-vivo feasibility

Abstract

US deformation techniques can roughly be divided in block matching (BM) and non-rigid image registration (NRIR). Motion can be extracted from the radio-frequency (RF) signals, from their envelope, or from the B-mode data. RF-based BM is known to outperform B-mode tracking in a small displacement setting, whereas NRIR has only been applied to B-mode data. The aim of this study was to test the feasibility of RF-based NRIR in-silico and in-vivo. First, synthetic 2D images of a phantom with a soft inclusion undergoing an axial compression (0.25%) were simulated. Its performance was assessed by varying the inclusion thickness (range: 2-20 mm in 2 mm steps) and stiffness (resulting strain range: 0.50%-1.50% in 0.25% steps). Both RF and envelope tracking were better at identifying smaller and more subtle inclusions compared to B-mode tracking (down to 8 mm and 6 mm resp.). Furthermore, when tracking the RF instead of their envelope, inclusion borders were more sharply defined (border size 2.57 mm vs 4.88 mm, p<;0.001) and strain errors in the inclusion were lower (0.08% vs 0.10%; p<;0.05). Next, NRIR was used to track the septum of a healthy volunteer from high frame rate US recordings (436 Hz), and compared against a recent RF-based BM method. In-vivo tracking revealed that RF-based BM and RF-based NRIR performed similarly, both producing physiological axial velocity and strain curves. The lateral components could only be estimated using NRIR.