SMRVIS: Point cloud extraction from 3-D ultrasound for non-destructive testing
This work addresses a domain-specific problem in non-destructive testing using ultrasound, but it appears incremental as it adapts existing methods to a new dataset.
The authors tackled point cloud extraction from 3-D ultrasound volumes by framing it as an image segmentation problem, achieving first place on an external challenge leaderboard.
We propose to formulate point cloud extraction from ultrasound volumes as an image segmentation problem. Through this convenient formulation, a quick prototype exploring various variants of the Residual Network, U-Net, and the Squeeze and Excitation Network was developed and evaluated. This report documents the experimental results compiled using a training dataset of five labeled ultrasound volumes and 84 unlabeled volumes that got completed in a two-week period as part of a submission to the open challenge "3D Surface Mesh Estimation for CVPR workshop on Deep Learning in Ultrasound Image Analysis". Based on external evaluation performed by the challenge's organizers, the framework came first place on the challenge's \href{https://www.cvpr2023-dl-ultrasound.com/}{Leaderboard}. Source code is shared with the research community at a \href{https://github.com/lisatwyw/smrvis}{public repository}.