Image Segmentation

Non-deep Learning based Segmentation Approaches

We have worked on a number of different general methods for image segmentation based on convex relaxations ( [zach09a] [zach09b] ). We have also devised segmentation methods for the segmentations of fiber bundles from diffusion weighted images ( [niethammer09] ) and to extract connectivity information from diffusion tensor images ( [niethammer10] ). For the segmentation of biological structures it is often know a-priori how large a particular object should be. We have therefore explored a segmentation method which can impose constraints on the segmentation area ( [niethammer13] ).

Multi-atlas Segmentation

We developed multi-atlas segmentation approaches for knee cartilage ( [shan14] ) and have most recently explored deep-learning approaches to predict the trustworthiness of individual atlases for multi-atlas segmentation approaches ( [ding19] ).

Image Segmentation via Deep Learning

As one of our main research directions is image registration, we are exploring image segmentation approaches based on deep-learning which can benefit from our fast deep-learning registration approaches. This has led us so far, for example, to an approach for joint segmentation and registration ( [xu19] ) as well as to a segmentation approach based on data augmentation at test time ( [shen20] ).

References

zach09a
zach09b
niethammer09
niethammer10
niethammer13
shan14
ding19
xu19
shen20

Marc Niethammer
Marc Niethammer
Professor of Computer Science

My research interests include image registration, image segmentation, shape analysis, machine learning, and biomedical applications.

Related