Uncertainty-Based Segmentation of Myocardial Infarction Areas on Cardiac MR Images
- Resource Type
- Authors
- Camarasa, Robin; Faure, Alexis; Crozier, Thomas; Bos, Daniel; de Bruijne, Marleen; Puyol Anton, Esther; Pop, Mihaela; Sermesant, Maxime; Campello, Victor; Lalande, Alain; Lekadir, Karim; Suinesiaputra, Avan; Camara, Oscar; Young, Alistair
- Source
- Statistical Atlases and Computational Models of the Heart. MandMs and EMIDEC Challenges-11th International Workshop, STACOM 2020, Held in Conjunction with MICCAI 2020, Revised Selected Papers, 12592, 385-391
Statistical Atlases and Computational Models of the Heart. M&Ms and EMIDEC Challenges ISBN: 9783030681067
M&Ms and EMIDEC/STACOM@MICCAI
- Subject
- business.industry
Computer science
ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION
medicine.disease
Pipeline (software)
Task (project management)
medicine
Computer vision
Segmentation
Myocardial infarction
Artificial intelligence
Mr images
Cardiac magnetic resonance
business
Focus (optics)
Image resolution
- Language
- English
- ISSN
- 0302-9743
Every segmentation task is uncertain due to image resolution, artefacts, annotation protocol etc. Propagating those uncertainties in a segmentation pipeline can improve the segmentation. This article aims to assess if segmentation can benefit from uncertainty of an auxiliary unsupervised task - the reconstruction of the input image. This auxillary task could help the network focus on rare examples that are otherwise poorly segmented. The method was applied to segmentation of myocardial infarction areas on cardiac magnetic resonance images.