Ultrasound Decompression for Large Field-of-View Reconstructions

Abstract

Tracked medical ultrasound allows for cost-effective and radiation-free imaging of anatomy featuring a very high spatial resolution. To overcome the limitations of the small field-of-view, sonographers can acquire multiple adjacent sweeps and compound them into a single volumetric representation. However, due to the inherent local and non-uniform compression of the underlying anatomy (caused by the ultrasound probe) the adjacent sweeps often exhibit poor alignment and discontinuities. We propose a novel decompression model to compensate for probe pressure related artifacts. It incorporates domain knowledge of the global acquisition pattern for regularization and allows for seamless stitching of multiple overlapping 3D freehand sweeps into one volume. The resulting extended field-of-view visualization provides the clinician with spatial context so that the relationship between individual features are easier to understand. Our experiments show that the resulting extended field-of-view reconstructions have a superior image quality in terms of alignment and continuity of the visible anatomy compared to the original acquisitions. Comparison to ground truth MRI data demonstrates the plausibility of our non-rigid decompression model.