A key concept in image segmentation is surface determination, where the boundary between one region and another is accurately captured. Greyscale information stored in the scan is used to determine the location of these boundaries. The exact procedure is highly variable depending image type and quality, as well as the subject type and other factors. The processes can also vary from highly manual to semi-automated or even fully automated segmentation that can incorporate elements of machine learning.
Segmentation is frequently made easier by image pre-processing steps, which involve filtering the images to remove noise and scanning artefacts, or to enhance contrast.
In Synopsys Simpleware software, a suite of image processing tools is available for efficient segmentation of 3D images. Pre-processing tools and intelligent time-saving options help to efficiently obtain accurate segmentation from even very challenging scans, for example:
- Using interpolation approaches to segment large volumes with greatly reduced manually labelled regions
- Automatically adjusting surface positions to correct for noise, artefacts or other segmentation inaccuracies
More recently, Simpleware software has introduced automated segmentation capabilities with Artificial Intelligence (AI) technology using Machine Learning (ML). These segmentation solutions learn, from examples of good segmentation, how to produce similarly high quality result in new cases. This strategy has already been applied to common anatomies of interest in the orthopaedic field like the knee and hip.
For high throughput applications, Machine Learning approaches are inarguably the future of segmentation, where an automated, robust, and fast segmentation process can replace those requiring more user interaction, making more time available for high-value tasks.