Teaching Assistant
Paper ID: 94


  • CNN
  • Segmentation

Abstract - In this paper, we present a fully automatic brain tumor segmentation method based on Deep Neural Networks (DNNs). The proposed networks are tailored to glioblastomas (both low and high grade) pictured in MR images. By their very nature, these tumors can appear anywhere in the brain and have almost any kind of shape, size, and contrast. These reasons motivate our exploration of a machine learning solution that exploits a flexible, high capacity DNN while being extremely efficient. Here, we give a description of different model choices that we’ve found to be necessary for obtaining competitive performance. We explore in particular different architectures based on Convolutional Neural Networks (CNN), i.e. DNNs specifically adapted to image data. We present a novel CNN architecture which differs from those traditionally used in computer vision. Our CNN exploits both local features as well as more global contextual features simultaneously. Also, different from most traditional uses of CNNs, our networks use a final layer that is a convolutional implementation of a fully connected layer which allows a 40 fold speed up. We also describe a 2-phase training procedure that allows us to tackle difficulties related to the imbalance of tumor labels. Finally, we explore a cascade architecture in which the output of a basic CNN is treated as an additional source of information for a subsequent CNN. Results reported on the 2013 BRATS test dataset reveal that our architecture improves over the currently published state-of-the-art while being over 30 times faster.

Paper - https://arxiv.org/pdf/1505.03540v3.pdf

Dataset - https://www.smir.ch/BRATS/Start2015 (Request Dataset using University Email. It’ll be accepted in about 2 working days)

Problem Statement -

  1. The paper is implemented on BRATS 2013 dataset. Implement the paper using the BRATS 2015 dataset (as given above). Clearly mention the details of the data set, like the number of samples, input features, etc. Report all these through necessary graphs in the notebook. Also, include the final processed data set used for training and testing in the submission.
  2. Implement the InputCascadeCNN architecture, as described in the paper (Fig. 3 (a)) using the Keras API of Tensorflow.
  3. Show the segmented image for 10 samples from the test set. Also report the precision, recall, and dice score for all three tumor categories (in the notebook).
  4. Bonus: Train the LocalCascadeCNN architecture (Fig. 3 (b)) and perform a comparative analysis with InputCascadeCNN in terms of metrics defined in point number 3.

Few Resources to Understand Key Concepts -

  1. Understand CNNs - http://cs231n.github.io/convolutional-networks/, http://cs231n.github.io/understanding-cnn/, https://towardsdatascience.com/understanding-cnn-convolutional-neural-network-69fd626ee7d4
  2. Intuitive Explanation of ConvNets - https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

(Note - These resources are just suggestive and not compulsory to go through. Feel free to explore and understand in your own way)