The Golgi apparatus (GA) is an organelle found in eukaryotic cells.
Due to its complex organization and until recently lack of appropriate approaches GA is still rather poorly researched.
Knowledge of its 3D organization and distribution inside complex cells would aid to further understand the role of GA.
However, manual segmentation of large volumes is very time consuming and its quality depends on the ability of the human annotator.
In the thesis, an approach for automatic segmentation of GAs in electron microscopy volumetric data is proposed.
The proposed pipeline consists of a neural network trained on roughly annotated data, active contours for a more precise segmentation, and filtering of false segmentations.
To our knowledge, this is the first approach that segments complex GAs in volumetric data automatically.
The use of roughly annotated ground truth saved 80% of the time needed for manual annotation of the input data.
The method was evaluated on a volumetric dataset and it showed promising results - it was able to annotate a wast majority of GAs with 89% sensitivity and 99% specificity.