Effect of latent space distribution on the segmentation of images with multiple annotations

Ishaan Bhat10000-0002-4398-018X, Josien P.W. Pluim2, Max A. Viergever1, Hugo J. Kuijf1
1: Image Sciences Institute, University Medical Center Utrecht, The Netherlands, 2: Department of Biomedical Engineering, Eindhoven University of Technology, The Netherlands
Publication date: 2023/04/30
https://doi.org/10.59275/j.melba.2023-18ae
PDF · Code · arXiv

Abstract

We propose the Generalized Probabilistic U-Net, which extends the Probabilistic U-Net by allowing more general forms of the Gaussian distribution as the latent space distribution that can better approximate the uncertainty in the reference segmentations. We study the effect the choice of latent space distribution has on capturing the variation in the reference segmentations for lung tumors and white matter hyperintensities in the brain. We show that the choice of distribution affects the sample diversity of the predictions and their overlap with respect to the reference segmentations. We have made our implementation available at https://github.com/ishaanb92/GeneralizedProbabilisticUNet

Keywords

Deep learning · Image segmentation · Uncertainty estimation · Bayesian machine learning

Bibtex @article{melba:2023:005:bhat, title = "Effect of latent space distribution on the segmentation of images with multiple annotations", author = "Bhat, Ishaan and Pluim, Josien P.W. and Viergever, Max A. and Kuijf, Hugo J.", journal = "Machine Learning for Biomedical Imaging", volume = "2", issue = "UNSURE2022 special issue", year = "2023", pages = "151--171", issn = "2766-905X", doi = "https://doi.org/10.59275/j.melba.2023-18ae", url = "https://melba-journal.org/2023:005" }
RISTY - JOUR AU - Bhat, Ishaan AU - Pluim, Josien P.W. AU - Viergever, Max A. AU - Kuijf, Hugo J. PY - 2023 TI - Effect of latent space distribution on the segmentation of images with multiple annotations T2 - Machine Learning for Biomedical Imaging VL - 2 IS - UNSURE2022 special issue SP - 151 EP - 171 SN - 2766-905X DO - https://doi.org/10.59275/j.melba.2023-18ae UR - https://melba-journal.org/2023:005 ER -
2023:005 cover