Answers are grounded in this paper and cite their sources.
Skilled sonographers have the incredible ability to look at 2D ultrasound images and re-create in their minds a 3D understanding of the anatomy they are investigat- ing. Given the noisy images and the limited context of tomographic slices, creating a 3D model from 2D images is undeniably challenging. In the context of robotic ultrasound systems, building 3D volumes of the tissue conventionally relies on compounding, where images taken at various robot poses fill the voxels in a 3D volume [1], [2], [3]. However compounding uses interpolation to fill the inherent gaps in the 3D volumes [4], and current interpolation methods tend to not generalize well. Interpolation methods often require hand-tuned parameters or are tailored to the specific anatomy. Further, interpolation over large gaps can pro- duce implausible content in the compounded volumes [5]. Finally, storing and manipulating these volumes also demands considerable memory. These fundamental challenges with typical volume reconstruction methods motivated the development of Ultra-NeRF [6], a neural implicit representation for med- ical ultrasound. It estimates tissue properties from multi- view image data and learns an implicit 3D ultrasound volume. The method yields smooth interpolation and physically plausible novel views. Ultra-NeRF has pre- viously only been tested on relatively simple ultrasound images, either generated from simulations or robotically captured from medical phantoms. In order to test the applicability of Ultra-NeRF to real world in-vivo data, we applied the method to ultrasound images captured robotically from live pigs. It was observed that Ultra- NeRF did not capture high-frequency features well, struggled with seamless interpolation between obser- vations and produced artifacts caused due to unstable regression.
No takes yet. Share an insight, caveat, or question.
Bal et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: