This thesis presents methods for automatically localizing fetal organs in MRI scans. It describes localizing the brain in 2 steps - detecting candidate brain regions using size filtering then further localizing through slice-by-slice segmentation, achieving median error of 5.7mm. For the body, it sequentially localizes organs by normalizing size by gestational age and using steerable image features informed by anatomy, detecting the heart center within 10mm in 90% of cases. This allows fully automated motion correction in over 70% of scans, presenting the first method to fully automatically localize multiple fetal organs beyond just the brain.
5. Fast MRI acquisition methods
MRI data is acquired as stacks of 2D slices
that freeze in-plane motion
but form an incoherent 3D volume.
5
6. Retrospective motion correction
Orthogonal stacks of
misaligned 2D slices
3D volume
Localising fetal organs can be used to initialise motion correction.
B. Kainz et al., “Fast Volume Reconstruction from Motion Corrupted Stacks of 2D Slices,”
in IEEE Transactions on Medical Imaging, 2015.
6
7. Retrospective motion correction
Orthogonal stacks of
misaligned 2D slices
3D volume
Localising fetal organs can be used to initialise motion correction.
B. Kainz et al., “Fast Volume Reconstruction from Motion Corrupted Stacks of 2D Slices,”
in IEEE Transactions on Medical Imaging, 2015.
6
8. Challenges in localising fetal organs
1 Arbitrary orientation of the fetus
2 Variability of surrounding maternal tissues
3 Variability due to fetal growth
7
9. Automated organ localisation
Image registration:
Warp annotated templates to new image
Machine learning:
Learn an abstract model from annotated examples
Implicitly model variability:
age
pose (articulated body)
maternal tissues
8
10. Automated organ localisation
Image registration:
Warp annotated templates to new image
Machine learning:
Learn an abstract model from annotated examples
Implicitly model variability:
age
pose (articulated body)
maternal tissues
8
14. Contributions: brain detection (Chapter 4)
Preselection of candidate brain regions with MSER detection
Filtering by size based on gestational age OFDOFD
BPDBPD
Slice-by-slice approach robust to the presence of motion
K. Keraudren et al., “Localisation of the Brain in Fetal MRI using Bundled SIFT Features,”
in MICCAI, 2013
11
15. Localisation results for the fetal brain (Chapter 4)
Size inferred from gestational age
Median error: 5.7 mm
Improved results compared to Ison et al. (2012):
10 mm reported median error
12
16.
17.
18. Contributions: brain segmentation (Chapter 5)
Label propagation from selected MSER
Brain segmentation integrated with motion correction
Fully automated motion correction
K. Keraudren et al., “Automated Fetal Brain Segmentation from 2D MRI Slices for Motion Correction,”
in NeuroImage, 2014
15
19. Segmentation results for the fetal brain (Chapter 5)
Fully automated motion correction in 85% of cases.
Place holder, place holder, place holder.
16
20. Segmentation results for the fetal brain (Chapter 5)
Improved results compared with the method of Taleb et al. (2013):
Dice score of 93% versus 80%.
16
24. Localising the body of the fetus
Brain largest organ, ellipsoidal shape
Lungs & liver irregular shapes
Motivates 3D approach despite motion corruption
(only coarse localisation)
19
25. Contributions: body detection (Chapter 6)
Size normalisation based on gestational age
24 weeks 30 weeks 38 weeks
Sequential localisation of fetal organs
Image features steered by the fetal anatomy
K. Keraudren et al., “Automated Localization of Fetal Organs in MRI Using Random Forests with
Steerable Features,” in MICCAI, 2015
20
uu
vv
26. Contributions: body detection (Chapter 6)
Size normalisation based on gestational age
24 weeks 30 weeks 38 weeks
Sequential localisation of fetal organs
Image features steered by the fetal anatomy
K. Keraudren et al., “Automated Localization of Fetal Organs in MRI Using Random Forests with
Steerable Features,” in MICCAI, 2015
20
uu
vv
27. Localisation results for the fetal organs (Chapter 6)
24 weeks
29 weeks
37 weeks
Coronal plane Sagittal plane Transverse plane
In 90% of cases, heart center detected with less than 10 mm error
21
28. Localisation results for the fetal organs (Chapter 6)
24 weeks
29 weeks
37 weeks
Coronal plane Sagittal plane Transverse plane
Automated motion correction in 73% of cases 21
32. Conclusion
Automated localisation of fetal organs in MRI:
Brain, heart, lungs & liver
Training one model across all ages
Account for the unknown orientation of the fetus
First method for a fully automated localisation of fetal organs
beyond the brain
Segmentation results enable fully automated motion correction
25