MRI Brain Tumor Segmentation and Patient Survival Prediction Using Random Forests and Fully Convolutional Networks (original) (raw)

Abstract

In this paper, we propose a novel learning based method for automated segmentation of brain tumor in multimodal MRI images. The machine learned features from fully convolutional neural network (FCN) and hand-designed texton features are used to classify the MRI image voxels. The score map with pixelwise predictions is used as a feature map which is learned from multimodal MRI training dataset using the FCN. The learned features are then applied to random forests to classify each MRI image voxel into normal brain tissues and different parts of tumor. The method was evaluated on BRATS 2013 challenge dataset. The results show that the application of the random forest classifier to multimodal MRI images using machine-learned features based on FCN and hand-designed features based on textons provides promising segmentations. The Dice overlap measure for automatic brain tumor segmentation against ground truth is 0.88, 080 and 0.73 for complete tumor, core and enhancing tumor, respectively.

Key takeaways

sparkles

AI

  1. The proposed method achieved a Dice score of 0.88 for complete tumor segmentation.
  2. The study utilized a hybrid approach combining Fully Convolutional Networks (FCN) and Random Forests (RF).
  3. Texton-based features improved segmentation accuracy by considering voxel neighborhood and spatial dependencies.
  4. The model was evaluated on the BRATS 2013 dataset with 30 training and 10 test cases.
  5. The method ranks 5th on the VSD website for segmentation accuracy against other top methodologies.

Loading...

Loading Preview

Sorry, preview is currently unavailable. You can download the paper by clicking the button above.

References (19)

  1. Patel, M.R., Tse, V.: Diagnosis and staging of brain tumors. Semin Roentgenol. 39, 347-360 (2004).
  2. Gordillo, N., Montseny, E., Sobrevilla, P.: State of the art survey on MRI brain tumor segmentation. Magn Reson Imaging. 31, 1426-1438 (2013).
  3. Menze, B.H., Jakab, A., et.al.: The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS). IEEE Transactions on Medical Imaging. 34, 1993-2024 (2015).
  4. Bauer, S., Wiest, R., Nolte, L.-P., Reyes, M.: A survey of MRI-based medical image analysis for brain tumor studies. Phys Med Biol. 58, R97-129 (2013).
  5. Hamamci, A., Kucuk, N., Karaman, K., Engin, K., Unal, G.: Tumor-Cut: Segmentation of Brain Tu- mors on Contrast Enhanced MR Images for Radiosurgery Applications. IEEE Transactions on Medical Imaging. 31, 790-804 (2012).
  6. Kleesiek, J., Biller, A., Urban, G., Köthe, U., Bendszus, M., Hamprecht, F.: ilastik for Multi-modal Brain Tumor Segmentation.
  7. Subbanna, N.K., Precup, D., Collins, D.L., Arbel, T.: Hierarchical Probabilistic Gabor and MRF Seg- mentation of Brain Tumours in MRI Volumes. In: Mori, K., Sakuma, I., Sato, Y., Barillot, C., and Navab, N. (eds.) Medical Image Computing and Computer-Assisted Intervention -MICCAI 2013. pp. 751-758. Springer Berlin Heidelberg (2013).
  8. Pinto, A., Pereira, S., Correia, H., Oliveira, J., Rasteiro, D.M.L.D., Silva, C.A.: Brain Tumour Segmen- tation based on Extremely Randomized Forest with high-level features. In: 2015 37th Annual Interna- tional Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). pp. 3037-3040 (2015).
  9. Gotz, M., Weber, C., Blocher, J., Stieltjes, B., Meinzer, H., Maier-Hein, K.: Extremely randomized trees based brain tumor segmentation. In: Proceeding of BRATS Challenge-MICCAI. pp. 006-011 (2014).
  10. Pereira, S., A., Alves, V., Silva, C.A.: Brain Tumor Segmentation Using Convolutional Neural Networks in MRI Images. IEEE Transactions on Medical Imaging. 35, 1240-1251 (2016).
  11. Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., Pal, C., Jodoin, P.-M., Larochelle, H.: Brain tumor segmentation with Deep Neural Networks. Medical Image Analysis. 35, 18-31 (2017).
  12. Kistler, M., Bonaretti, S., Pfahrer, M., Niklaus, R., Büchler, P.: The virtual skeleton database: an open access repository for biomedical research and collaboration. J. Med. Internet Res. 15, e245 (2013).
  13. BRATS :: The Virtual Skeleton Database Project, https://www.smir.ch/BRATS/Start2013.
  14. Long, J., Shelhamer, E., Darrell, T.: Fully Convolutional Networks for Semantic Segmentation. Pre- sented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2015).
  15. Simonyan, K., Zisserman, A.: Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556 [cs]. (2014).
  16. Liaw, A., Wiener, M.: Classification and regression by randomForest. 2, 18-22 (2002).
  17. Vedaldi, A., Lenc, K.: MatConvNet: Convolutional Neural Networks for MATLAB. In: Proceedings of the 23rd ACM International Conference on Multimedia. pp. 689-692. ACM, New York, NY, USA (2015).
  18. Taormina, R.: MATLAB_ExtraTrees -File Exchange -MATLAB Central, http://uk.math- works.com/matlabcentral/fileexchange/47372-rtaormina-matlab-extratrees.
  19. Tustison, N.J., Shrinidhi, K.L., Wintermark, M., Durst, C.R., Kandel, B.M., Gee, J.C., Grossman, M.C., Avants, B.B.: Optimal Symmetric Multimodal Templates and Concatenated Random Forests for Su- pervised Brain Tumor Segmentation (Simplified) with ANTsR. Neuroinform. 13, 209-225 (2014).

FAQs

sparkles

AI

What impact do Gabor filters have on texton feature extraction for tumor segmentation?add

The study reveals that Gabor filters significantly enhance texture description in images, improving segmentation accuracy due to their ability to capture local features effectively.

How does the proposed hybrid method improve segmentation accuracy compared to existing techniques?add

The hybrid approach combining FCN features and texton characteristics achieved a Dice score of 0.88, outperforming conventional methods in core tumor segmentation on the BRATS 2013 dataset.

What were the optimal parameters chosen for the Random Forest classifier in this research?add

Optimally, the Random Forest classifier used 50 trees with a depth of 15 and a feature subset size of 7, yielding improved classification results.

In what way did the proposed method address the limitations of Fully Convolutional Networks?add

By integrating texton-based features, the method compensated for the lack of spatial regularization in FCNs, thus refining segmentation boundaries and considering voxel neighborhoods.

What preprocessing steps enhance the efficacy of MRI image analysis in this study?add

Preprocessing involved intensity normalization by excluding extreme values and matching histograms, which standardized input for improved model training and segmentation outcomes.