View source for Publications/boutry.20.brainles

{{Publication
| published = true
| date = 2020-06-01
| authors = Nicolas Boutry, Joseph Chazalon, Élodie Puybareau, Guillaume Tochon, Hugues Talbot, Thierry Géraud
| title = Using Separated Inputs for Multimodal Brain Tumor Segmentation with 3D U-Net-like Architectures
| booktitle = Proceedings of the 4th International Workshop, BrainLes 2019, Held in Conjunction with MICCAI 2019
| editors = A Crimi, S Bakas
| volume = 11992
| series = Lecture Notes in Computer Science
| pages = 187 to 199
| publisher = Springer
| abstract = The work presented in this paper addresses the MICCAI BraTS 2019 challenge devoted to brain tumor segmentation using mag- netic resonance images. For each task of the challenge, we proposed and submitted for evaluation an original method. For the tumor segmentation task (Task 1)our convolutional neural network is based on a variant of the U-Net architecture of Ronneberger et al. with two modifications: first, we separate the four convolution parts to decorrelate the weights corresponding to each modality, and second, we provide volumes of size 240 * 240 * 3 as inputs in these convolution parts. This way, we profit of the 3D aspect of the input signal, and we do not use the same weights for separate inputs. For the overall survival task (Task 2), we compute explainable features and use a kernel PCA embedding followed by a Random Forest classifier to build a predictor with very few training samples. For the uncertainty estimation task (Task 3), we introduce and compare lightweight methods based on simple principles which can be applied to any segmentation approach. The overall performance of each of our contribution is honorable given the low computational requirements they have both for training and testing.
| lrdepaper = http://www.lrde.epita.fr/dload/papers/boutry.20.brainles.pdf
| lrdeprojects = Olena
| lrdenewsdate = 2020-06-01
| type = inproceedings
| id = boutry.20.brainles
| identifier = doi:10.1007/978-3-030-46640-4_18
| bibtex = 
@InProceedings<nowiki>{</nowiki>	  boutry.20.brainles,
  author	= <nowiki>{</nowiki>Nicolas Boutry and Joseph Chazalon and \'Elodie Puybareau
		  and Guillaume Tochon and Hugues Talbot and Thierry G\'eraud<nowiki>}</nowiki>,
  title		= <nowiki>{</nowiki>Using Separated Inputs for Multimodal Brain Tumor
		  Segmentation with <nowiki>{</nowiki>3D<nowiki>}</nowiki> <nowiki>{</nowiki>U-Net<nowiki>}</nowiki>-like Architectures<nowiki>}</nowiki>,
  booktitle	= <nowiki>{</nowiki>Proceedings of the 4th International Workshop, BrainLes
		  2019, Held in Conjunction with MICCAI 2019<nowiki>}</nowiki>,
  year		= 2019,
  month		= may,
  editor	= <nowiki>{</nowiki>A. Crimi and S. Bakas<nowiki>}</nowiki>,
  volume	= <nowiki>{</nowiki>11992<nowiki>}</nowiki>,
  series	= <nowiki>{</nowiki>Lecture Notes in Computer Science<nowiki>}</nowiki>,
  pages		= <nowiki>{</nowiki>187--199<nowiki>}</nowiki>,
  publisher	= <nowiki>{</nowiki>Springer<nowiki>}</nowiki>,
  doi		= <nowiki>{</nowiki>10.1007/978-3-030-46640-4_18<nowiki>}</nowiki>,
  abstract	= <nowiki>{</nowiki>The work presented in this paper addresses the MICCAI
		  BraTS 2019 challenge devoted to brain tumor segmentation
		  using mag- netic resonance images. For each task of the
		  challenge, we proposed and submitted for evaluation an
		  original method. For the tumor segmentation task (Task 1),
		  our convolutional neural network is based on a variant of
		  the U-Net architecture of Ronneberger et al. with two
		  modifications: first, we separate the four convolution
		  parts to decorrelate the weights corresponding to each
		  modality, and second, we provide volumes of size 240 * 240
		  * 3 as inputs in these convolution parts. This way, we
		  profit of the 3D aspect of the input signal, and we do not
		  use the same weights for separate inputs. For the overall
		  survival task (Task 2), we compute explainable features and
		  use a kernel PCA embedding followed by a Random Forest
		  classifier to build a predictor with very few training
		  samples. For the uncertainty estimation task (Task 3), we
		  introduce and compare lightweight methods based on simple
		  principles which can be applied to any segmentation
		  approach. The overall performance of each of our
		  contribution is honorable given the low computational
		  requirements they have both for training and testing.<nowiki>}</nowiki>
<nowiki>}</nowiki>

}}