Case Study: U-Net TensorFlow with INS, Rapperswil

In collaboration with the Institute for Networked Solutions INS we used XamFlow and TensorFlow U-Net to automate prostate segmentation.



  • Prostate MR images from MRI T2 acquisitions

  • 5251 slices of 181 patients

  • Captures with various machines using multiple protocols

    • Promise12

    • Decathlon

    • Siemens-3T

    • GE-1.5T

    • NCI-2013

  • Preprocessed to normalize all the data

  • Augmented using various methods


  • Martin Stypinski, ProSeg – Increasing Robustness in ProstateSegmentation with Deep Learning, Master Thesis