Thesis topic

Uncertainty-Aware Stenosis Segmentation in Digital Subtraction Angiography using Probabilistic Deep Learning

Probabilistic deep learning for segmentation uncertainty in angiography.

Uncertainty-Aware Stenosis Segmentation in Digital Subtraction Angiography using Probabilistic Deep Learning

Uncertainty-Aware Stenosis Segmentation in Digital Subtraction Angiography using Probabilistic Deep Learning

Overview

  • Stenosis segmentation is essential for optimal stent choice.
  • Expert opinions vary depending on experience, education, and skill level.
  • Probabilistic segmentation can support quantifying uncertainty in stenosis segmentation.
  • Relevant methods include Probabilistic U-Net and follow-up work on probabilistic segmentation ( arXiv:1907.01949 ).
  • Quantify expert label uncertainty based on multi-expert labeled data.

Skills

  • Python programming experience with Keras or PyTorch, scikit-learn, pandas, and NumPy
  • Deep learning knowledge, especially CNNs and U-Net
  • Image processing knowledge
  • Basic medical or anatomy knowledge is a plus

Tasks

  • Literature research
  • Data preparation and preprocessing pipeline implementation
  • Probabilistic segmentation and/or detection model implementation, training, and evaluation

Contact

Daniel Wulff

d.wulff@uni-rostock.de