Selective labeling: identifying representative sub-volumes for interactive segmentation

Luengo, Imanol and Basham, Mark and French, Andrew P. (2016) Selective labeling: identifying representative sub-volumes for interactive segmentation. Lecture Notes in Computer Science, 9993 . pp. 17-24. ISSN 0302-9743

Full text not available from this repository.


Automatic segmentation of challenging biomedical volumes with multiple objects is still an open research field. Automatic approaches usually require a large amount of training data to be able to model the complex and often noisy appearance and structure of biological organelles and their boundaries. However, due to the variety of different biological specimens and the large volume sizes of the datasets, training data is costly to produce, error prone and sparsely available. Here, we propose a novel Selective Labeling algorithm to overcome these challenges; an unsupervised sub-volume proposal method that identifies the most representative regions of a volume. This massively-reduced subset of regions are then manually labeled and combined with an active learning procedure to fully segment the volume. Results on a publicly available EM dataset demonstrate the quality of our approach by achieving equivalent segmentation accuracy with only 5 % of the training data.

Item Type: Article
Additional Information: Part of: Patch-based techniques in medical imaging: second International Workshop, Patch-MI 2016, held in conjunction with MICCAI 2016, Athens, Greece, October 17, 2016, proceedings
Keywords: Unsupervised; Sub-volume proposals; Interactive segmentation; Active learning; Affinity clustering; Supervoxels
Schools/Departments: University of Nottingham, UK > Faculty of Science > School of Computer Science
Identification Number:
Depositing User: Eprints, Support
Date Deposited: 11 Aug 2017 07:59
Last Modified: 04 May 2020 18:10

Actions (Archive Staff Only)

Edit View Edit View