Automatic Liver and Tumor Segmentation

Liver tumor resections, living donor liver transplantations, catheter-based liver interventions and other image-guided liver therapies incorporate information derived from the liver and tumor segmentation in the decision-making process. Automatic extraction of liver and tumors from the planning images would not only speed-up and standardize the therapy planning process but would also allow for determination of additional volume- and texture-based information. What is more, automated analysis of baseline and follow-up scans would facilitate a more reliable therapy response classification.



Liver and liver tumors can differ greatly in shape and appearance across different patients, lesion types and modalities. Moreover, many liver imaging protocols involve contrast agents, which further increase the variability. These factors make the task of automatic liver and tumor segmentation a challenging one.



Fraunhofer MEVIS developed within the SIRTOP project ( a custom solution for liver and tumor segmentation in CT and MR images based on recent advances in the deep learning technology. Our two-stage approach involves multiple fully convolutional neural networks (FCNs) coupled with conventional classifiers trained with hand-crafted features. Our approach delivers state-of-the-art segmentation quality which was confirmed by the 3rd place in the second round of the Liver and Tumor Segmentation Challenge LiTS 2017 (




  • Deep learning based automatic segmentation method
  • 3rd place in the second round of the LiTS challenge
Automatic liver and tumor segmentation.
© Fraunhofer MEVIS
Automatic liver and tumor segmentation based on analysis with a deep learning network.