@@ -24,7 +24,7 @@ Quantitative characterization of 3D images involves measuring the geometric and
If advantageous, we use machine learning and deep learning for segmentation, for example, when analyzing cardiac tissue as in [Reichardt et al., 2021](https://elifesciences.org/articles/71359). To minimize the need for human input, we research in active learning.
### Dictionary-based segmentation
Our dictionary-based segmentation [Dahl et al., 2020](https://ieeexplore.ieee.org/document/9150995) is based on machine learning principles. Both the the dictionary and the associated labels are learned from data. This enables automatic image segmentation, fast enough for real-time interaction. An outcome of this is the Insegt segmentation tool ([project page](https://github.com/vedranaa/InSegt)). InSegt has been used for numerous projects, and in particular for analysing microstructure of composite materials ([Wang et al. 2021](https://www.sciencedirect.com/science/article/pii/S0266353821002852?via%3Dihub)) but also compound bee eye ([Tichit et al. 2022](https://bmczool.biomedcentral.com/articles/10.1186/s40850-021-00101-w)).
Our dictionary-based segmentation [Dahl et al., 2020](https://ieeexplore.ieee.org/document/9150995) is based on machine learning principles. Both the the dictionary and the associated labels are learned from data. This enables automatic image segmentation, fast enough for real-time interaction. An outcome of this is the Insegt segmentation tool ([code repository](https://github.com/vedranaa/InSegt)). InSegt has been used for numerous projects, and in particular for analysing microstructure of composite materials ([Wang et al. 2021](https://www.sciencedirect.com/science/article/pii/S0266353821002852?via%3Dihub)) but also compound bee eye ([Tichit et al. 2022](https://bmczool.biomedcentral.com/articles/10.1186/s40850-021-00101-w)).
<divalign="center">
<imgtitle="Interactive segmentation of an X-ray CT slice of a bee eye."src="images/segmentation.gif"alt="Interactive segmentation segmentation of a X-ray CT slice of a bee eye"width="600"/>
We have developed novel methods for graph-based segmentation ([Jeppesen et al. 2020](https://ieeexplore.ieee.org/document/9156301) and [Jensen et al. 2020](https://ieeexplore.ieee.org/document/9151036)), implemented efficient graph-based algorithms ([Jeppesen et al. 2021](https://ieeexplore.ieee.org/document/9710633)) and benchmarked existing graph-based algorithms ([Jensen et al. 2022](https://ieeexplore.ieee.org/abstract/document/9763394)). Our methods have been used for analysing large 3D data, for example from samples of peripheral nerves ([Dahlin et al. 2020](https://www.nature.com/articles/s41598-020-64430-5)) and muscle fibres ([Pingel et al. 2022](https://www.nature.com/articles/s41598-022-21741-z)).
We have developed novel methods for graph-based segmentation ([Jeppesen et al. 2020](https://ieeexplore.ieee.org/document/9156301) and [Jensen et al. 2020](https://ieeexplore.ieee.org/document/9151036)), implemented efficient graph-based algorithms ([Jeppesen et al. 2021](https://ieeexplore.ieee.org/document/9710633)) and benchmarked existing graph-based algorithms ([Jensen et al. 2022](https://ieeexplore.ieee.org/abstract/document/9763394)). Our findings are collected in [project page](https://patmjen.github.io/maxflow_review/). Our methods have been used for analysing large 3D data, for example from samples of peripheral nerves ([Dahlin et al. 2020](https://www.nature.com/articles/s41598-020-64430-5)) and muscle fibres ([Pingel et al. 2022](https://www.nature.com/articles/s41598-022-21741-z)).
<divalign="center">
<imgtitle="Non-overlapping surfaces for segmenting foam"src="images/nos_wh.png"alt="Non-overlapping surfaces for segmenting foam"width="600"/>
