Brain tumors are abnormal formations of mass that apply pressure to the surrounding tissues, causing several health problems such as unexplained nausea, seizures, personality changes or even death. Brain lesion segmentation is a critical application of computer vision to the biomedical image analysis. The difficulty is derived from the great variance between instances, and the high computational cost of processing three dimensional data.
We introduce a neural network for brain tumor semantic segmentation that parses their internal structures and is capable of processing volumetric data from multiple MRI modalities simultaneously. As a result, the method is able to learn from small training datasets. We develop an architecture that has four parallel pathways with residual connections. It receives patches from images with different spatial resolutions and analyzes them independently. The results are then combined using fully-connected layers to obtain a semantic segmentation of the brain tumor. We evaluated our method using the 2017 BraTS Challenge dataset, reaching average dice coefficients of 89%, 88 and 86% over the training, validation and test images, respectively.
Figure 1 and 2. Qualitative results.

Brain Tumor Segmentation and Parsing on MRIs using Multiresolution Neural Networks

LS. Castillo, LA. Daza, LC. Rivera, P. Arbeláez

Brain Lesion workshop of the medical Image Computing and Computer assisted Interventions Conference, 2017

Volumetric multimodality neural network for brain tumor segmentation

L.S. Castillo, L.A. Daza, L.C. Rivera and P. Arbeláez

13th International Conference on Medical Information Processing and Analysis (SIPAIM), 2017

Method

Figure 3. Proposed Architecture. The kernels of the convolutions in the three pathways are 33 and no padding was made in those operations. The input of the 3 paths are centered in the same voxel, but the medium resolution and low patches are obtained from downsampled versions of the image by factors of 3 and 5, respectively

Results

Table 1.

Method

  •  
  • Deepmedic
  • Ours
  • Dice

    Enh.    Wh.     Cor.
  • 0.69    0.86    0.68
  • 0.71    0.88    0.68
  • Sensitivity

    Enh.    Wh.     Cor.
  • 0.72    0.86    0.64
  • 0.72    0.86    0.68
  • Specificity

    Enh.    Wh.     Cor.
  • 0.99    0.99    0.99
  • 0.99    0.99    0.99
  • Hausdorff

    Enh.    Wh.     Cor.
  • 10.1    25.0    17.5
  • 6.12    9.63    11.4
  • Table 2.

    Dataset

  •  
  • Train
  • Val
  • Test
  • Dice

    Enh.    Wh.     Cor.
  • 0.74    0.89    0.87
  • 0.71    0.88    0.68
  • 0.65    0.86    0.67
  • Sensitivity

    Enh.    Wh.     Cor.
  • 0.83    0.91    0.89
  • 0.72    0.86    0.68
  • -    -    -
  • Specificity

    Enh.    Wh.     Cor.
  • 0.99    0.99    0.99
  • 0.99    0.99    0.99
  • -    -    -
  • Hausdorff

    Enh.    Wh.     Cor.
  • 5.85    15.9    11.2
  • 6.12    9.63    11.4
  • 51.7    10.4    36.2
  • Citation

    @proceeding{sipaim2017Brats,
    author = {Laura Silvana Castillo and Laura Alexandra Daza and Luis Carlos Rivera and Pablo Arbeláez},
    title = {Volumetric multimodality neural network for brain tumor segmentation},
    journal = {Proc.SPIE},
    volume = {10572},
    pages = {10572 - 10572 - 8},
    year = {2017},
    doi = {10.1117/12.2285942},
    URL = {https://doi.org/10.1117/12.2285942}
    }
    @InProceedings{Brats2017,
    author={Laura Silvana Castillo and Laura Alexandra Daza and Luis Carlos Rivera and Pablo Arbeláez},
    editor="Crimi, Alessandro and Bakas, Spyridon
    and Kuijf, Hugo and Menze, Bjoern and Reyes, Mauricio",
    title="Brain Tumor Segmentation and Parsing on MRIs Using Multiresolution Neural Networks",
    booktitle="Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries",
    year="2018",
    publisher="Springer International Publishing",
    address="Cham",
    pages="332--343"
    }
    @InProceedings{proceedings2017,
    author={Laura Silvana Castillo and Laura Alexandra Daza and Luis Carlos Rivera and Pablo Arbeláez},
    booktitle="20th International Conference on Medical Image Computing and Computer Assisted Intervention 2017 (MICCAI 2017)",
    address="BraTS 2017 Challenge proceedings",
    pages="34--41"
    }
    • Address:
      Cra. 1 E No. 19A - 40. 111711, Bogotá, Colombia - Mario Laserna Building - School of Engineering - Universidad de Los Andes
    • Phone:
      [571] 332 4327, 332 4328, 332 4329