Ultrasound Data Segmentation in Toroidal Coordinates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ultrasound imaging methods face inefficiencies in memory usage and learning bias due to the acquisition of cone-shaped data, leading to inaccurate segmentation and the need for complex data augmentation in deep learning algorithms.
Innovation Solution
Transforming scan-converted ultrasound volumes from a Cartesian to a Toroidal coordinate system to de-scanned ultrasound volumes, which mimics the acquisition process, reducing empty data and enabling more accurate segmentation using convolutional neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If ultrasound data is stored in rectangular volumes with cone-shaped beam coverage, then the data can be visualized in Cartesian coordinates, but up to 60% of voxels are empty and represent no information, leading to inefficient memory usage
Solution Approach 1:
The patent transforms the ultrasound data from Cartesian coordinates to polar coordinates, changing the dimensional representation to match the cone-shaped beam coverage. This coordinate transformation allows the data to be stored efficiently in a polar volume format, reducing the proportion of empty voxels from up to 60% to a minimal amount, while still enabling visualization in Cartesian coordinates when needed.
2Productivity
If segmentation algorithms are trained on scan-converted ultrasound volumes with cone-shaped data, then the algorithms can process the data as received, but learning bias is introduced due to the position of acquisition cone borders
Solution Approach 1:
Instead of transforming the data to match the acquisition geometry (which would introduce bias), the patent inverts the approach by transforming the segmentation algorithm's input from the standard Cartesian scan-converted format to the original polar acquisition format. This allows the algorithm to process data in its natural acquisition geometry, eliminating learning bias while maintaining processing efficiency.
3Reliability
If data augmentation is applied to create training datasets, then deep learning performance is improved, but complicated transformations are required to preserve the cone shape and realistic aspect
Solution Approach 1:
The patent changes the fundamental parameter of the coordinate system from Cartesian to polar, which simplifies the data augmentation process. In the polar coordinate system, standard geometric transformations like rotation and translation become simpler operations that naturally preserve the cone-shaped beam coverage and realistic appearance of the ultrasound data, eliminating the need for complicated transformations required in Cartesian coordinates.
Data Source
AI summary
A method for segmenting a target anatomy in ultrasound data. Scan-converted ultrasound data is obtained within a scan-converted space in the Cartesian coordinate system. The scan-converted ultrasound data is transformed to de-scanned ultrasound data within a de-scanned space in the Toroidal coordinate system. The de-scanned ultrasound data is an estimate of the ultrasound data as obtained by an original acquisition procedure. A segmentation of a target anatomy can thus be performed on the ultrasound data in the de-scanned space The resulting segmentation data can then be re-scanned back to the Cartesian coordinate system for display with the ultrasound data.