Gemini_Generated_Image_453ubo453ubo453u

Segmentation of micro-vessels and nerves in digital histology images

by Healthcare

Segmentation of micro-vessels and nerves in digital histology images

Segmentation of micro-vessels and nerves in digital histology images

Segmentation of micro-vessels and nerves in digital histology images

Perineural invasion (PNI), lymphovascular invasion (LVI) and tumor angiogenesis have strong correlation with cancer recurrence, metastasis and poor patient survival. The accurate segmentations of nerves and microvessels can be considered as the preliminary step in objective identification of PNI, LVI and tumor angiogenic analysis in histology images.
We proposed a deep network for simultaneous segmentation of microvessel and nerves in routinely used H&E-stained histology images. The network is designed as an encoder–decoder architecture with embedded feature attention blocks and an uncertainty prediction.
The proposed network uses Xception residual blocks, followed by atrous spatial pyramid pooling for feature extraction at multiple scales. Feature attention blocks are used in the skip connections from encoder to decoder as well as in the decoder up-sampling, which enables the network in focusing on more salient features while making prediction for segmentation. The method is evaluated using 7780 images of size 512 × 512 pixels, extracted from 20 WSIs of oral squamous cell carcinoma tissue at 20× magnification. The ensemble of network outputs at test time is used to obtain a better segmentation result and simultaneous generation of network prediction uncertainty map. The proposed network achieves state-of-the-art results compared to currently used deep neural networks for semantic segmentation (FCN-8, U-Net, Segnet and DeepLabV3+). The proposed network also gives robust segmentation performance when applied to the full digital whole slide image.

Collaborators

Selected Publications

  • M.M. Fraz , S. A. Khurran, S. Graham, M. Shaban, Asif Loya, N.M. Rajpoot , “FABnet: Feature attention based network for simultaneous segmentation of microvessels and nerves in routine histology images of oral cancer”, Neural Computing and Applications, Vol. 20 , No. 1, Nov, 2019. IF: 4.66
  • M.M.Fraz, M.Shaban, S.Graham, S.A.Khurram, N.M.Rajpoot , , “Uncertainty Driven Pooling Network for Microvessel Segmentation in Routine Histology Images”, Proceedings of the COMPAY workshop in 21st International Conference on Medical Image Computing and Computer Assisted Intervention, Sep, 2018, Granada , Spain.