← ML Research Wiki / 1505.04597

U-Net: Convolutional Networks for Biomedical Image Segmentation

Olaf Ronneberger Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany, Philipp Fischer Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany, Thomas Brox Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany, Olaf Ronneberger Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany, Philipp Fischer Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany, Thomas Brox Computer Science Department and BIOSS Centre for Biological Signalling Studies University of Freiburg Germany (2015)

Paper Information
arXiv ID
1505.04597
Venue
International Conference on Medical Image Computing and Computer-Assisted Intervention
Domain
computational biology
SOTA Claim
Yes
Code
Available
Reproducibility
8/10
Contents

Abstract

There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at

Summary

The paper "U-Net: Convolutional Networks for Biomedical Image Segmentation" presents a novel neural network architecture designed for biomedical image segmentation, focusing on efficient use of small training datasets through extensive data augmentation. The U-Net architecture includes a contracting path for context capture and a symmetric expanding path for precise localization, enabling end-to-end training from few images. The authors demonstrate the network's effectiveness by outperforming existing methods in the ISBI challenges for neuron segmentation and cell tracking, achieving high accuracy and fast processing times on a modern GPU. The U-Net architecture showcases potential applications across various biomedical tasks, supported by the availability of a Caffe-based implementation and pre-trained models.

Methods

This paper employs the following methods:

  • Convolutional Networks

Models Used

  • U-Net

Datasets

The following datasets were used in this research:

  • ISBI challenge for neuronal segmentation
  • ISBI cell tracking challenge 2015
  • "PhC-U373" and "DIC-HeLa" datasets

Evaluation Metrics

  • Warping Error
  • Rand Error
  • Intersection over Union (IOU)

Results

  • Outperformed prior best method on the ISBI challenge for neuronal structure segmentation
  • Achieved significant victories in the ISBI cell tracking challenge 2015 with large margins
  • Segmentation of 512x512 images completed in less than a second on a recent GPU

Limitations

The authors identified the following limitations:

  • Require large amounts of training data for typical deep networks
  • Existing solutions may involve slow processing due to patch-based training strategies
  • Trade-offs between localization accuracy and context usage in traditional methods

Technical Requirements

  • Number of GPUs: 1
  • GPU Type: NVidia Titan GPU (6 GB)

Keywords

U-Net Biomedial image segmentation Convolutional neural network Data augmentation

Papers Using Similar Methods

External Resources

Funding: Not specified
References: 18
Influential Citations: 9659