Insights into the behaviour of multi-task deep neural networks for medical image segmentation.

Lukasz Tomasz Bienias, Juan José Rubio Guillamón, Line Hagner Nielsen, Tommy Sonne Alstrøm

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

136 Downloads (Pure)

Abstract

Glandular morphology is used by pathologists to assess the malignancy of different adenocarcinomas. This process involves conducting gland segmentation task. The common approach in specialised domains, such as medical imaging, is to design complex architectures in a multi-task learning setup. Generally, these approaches rely on substantial postprocessing efforts. Moreover, a predominant notion is that general purpose models are not suitable for gland instance segmentation. We analyse the behaviour of two architectures: SA-FCN and Mask R-CNN. We compare the impact of post-processing on the final predictive results and the performance of generic and specific models for the gland segmentation problem. Our results highlight the dependency of post-processing on tailored models as well as comparable results when using a generic model. Thus, in the interest of time, it is worth considering to use and improve generic models as opposed to design complex architectures when tackling new domains.

Original languageEnglish
Title of host publicationProceedings of 2019 IEEE International Workshop on Machine Learning for Signal Processing.
Number of pages6
PublisherIEEE
Publication date2019
ISBN (Print)978-1-7281-0824-7
DOIs
Publication statusPublished - 2019
Event2019 IEEE International Workshop on Machine Learning for Signal Processing - University of Pittsburgh, Pittsburgh, United States
Duration: 13 Oct 201916 Oct 2019

Conference

Conference2019 IEEE International Workshop on Machine Learning for Signal Processing
LocationUniversity of Pittsburgh
CountryUnited States
CityPittsburgh
Period13/10/201916/10/2019

Keywords

  • Model optimisation
  • Multi-task learning
  • Image segmentation
  • Deep learning
  • Convolutional Neural Networks (CNNs)

Cite this

Bienias, L. T., Rubio Guillamón, J. J., Nielsen, L. H., & Alstrøm, T. S. (2019). Insights into the behaviour of multi-task deep neural networks for medical image segmentation. In Proceedings of 2019 IEEE International Workshop on Machine Learning for Signal Processing. IEEE. https://doi.org/10.1109/MLSP.2019.8918753