Automated Fetal Brain Segmentation from 2D MRI Slices for Motion Correction
1 like2,248 views
This document describes a method for automated fetal brain segmentation from 2D MRI slices in order to perform motion correction. The method uses box detection algorithms like MSER and SIFT to detect the brain region in each slice. It then trains a random forest classifier on brain and non-brain patches to perform brain extraction. Finally, it uses a conditional random field for motion correction across slices to generate a 3D volume with less artifacts from fetal movement. The results showed the proposed method produced motion-corrected volumes of diagnostic quality in 85% of test cases.
Automated Fetal Brain Segmentation from 2D MRI Slices for Motion Correction
- 1. Automated Fetal Brain Segmentation from 2D MRI Slices for Motion Correction Kevin Keraudren, Maria Murgasova, Vanessa Kyriakopoulou, Christina Malamateniou, Mary Rutherford, Bernhard Kainz, Jo Hajnal & Daniel Rueckert October 7th, 2014
- 2. Imaging the fetus using MRI As the fetuse moves the data is acquired as stacks of misaligned 2D slices 2/16
- 3. 3/16
- 4. 4/16
- 5. Motion correction of fetal brain MRI orthogonal stacks of 2D slices 3D volume 5/16
- 6. Box detection MSER Size filtering Bag-of-SIFT RANSAC MSER regions Filtering by size and Bag-of-Words 6/16
- 7. Box detection Brain and non-brain patches are learnt from the detected box and its MSER regions 7/16
- 8. Brain extraction Random Forest classifier and Conditional Random Field 8/16
- 9. Motion correction Adapted from Kuklisova-Murgasova et al., 2012. 9/16
- 10. Motion correction Conditional Random Field 10/16
- 11. Results Automated segmentation A B C Manual segmentation A 42 3 0 B 2 8 5 C 0 1 5 A. no evident artefacts from motion B. minor artifacts but can still be used for diagnosis C. cannot be used for clinical diagnosis 11/16
- 12. Results “In 85% of the tested cases, our proposed method produced a motion corrected volume of a relevant quality for clinical diagnosis.” NeuroImage, 2014. 12/16
- 13. Results Tighter mask, less artifacts and brain already extracted 13/16
- 14. Code 14/16
- 15. Code 15/16