DL in MRI

Application of deep learning in MRI

Transcript

1. Deep Learning In medical imaging focusing on MRI LTY,2019 1
2. OVERVIEW ► Machine Learning ► Data Acquisition ► Higher-level applications of DL in MRI ► Challenges 2
3. Machine Learning 3
4. Machine Learning ► Artificial neural networks ► Deep learning ► Convolutional neural networks ► Generative adversarial network 4
5. Artificial neural networks 5
6. Deep learning 6
7. Convolutional neural networks 7
8. Generative adversarial network 8
10. Data Acquisition 10
11. Data Acquisition ► Image reconstruction ► Image restoration ► Image Super-Resolution (SR) ► Image synthesis 11
12. Image reconstruction Visualization results of intermediate steps during the iterations of a reconstruction. (a) Undersampled image by acceleration factor 9 (b) Ground Truth (c-l) Results from intermediate steps 1 to 10 in a reconstruction process CRNN Convolutional Recurrent Neural Networks for Dynamic MR Image Reconstruction
13. Image reconstruction • The comparison of reconstructions on spatial dimension with their error maps. • (a) Ground Truth • (b) Undersampled image by acceleration factor 9 • (c,d) Proposed-B • (e,f) 3D CNN • (g,h) 3D CNN-S • (i,j) k-t FOCUSS • (k,l) k-t SLR
14. Image restoration : Denoising 14 Deep Learning Approaches for Detection and Removal of Ghosting Artifacts in MR Spectroscopy
15. Image SR ► Super Resolution GAN (SRGAN) ► Network ► Discriminator : HR image (T/F) ► Generator : LR→HR 15
16. Image synthesis : DCGAN • Data augmentation • Network • Discriminator : Real image • Generator : Synthetic image
17. Image synthesis : CycleGAN 17 T1 → T2 T2 → T1
18. Higher-level Applications of DL in MRI 18
19. Higher-level Applications of DL in MRI ► Image segmentation ► Prediction 19
20. Image segmentation ► Whole Brain ► Polycystic Kidneys 20
21. SLANT : Whole Brain Segmentation 21 3D Whole Brain Segmentation using Spatially Localized Atlas Network Tiles
22. 22 3D Whole Brain Segmentation using Spatially Localized Atlas Network Tiles
23. Automated Segmentation of Polycystic Kidneys 23 Performance of an Artificial Multi-observer Deep Neural Network for Fully Automated Segmentation of Polycystic Kidneys https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5537093/
24. Prediction ► Brain ► Kidney ► Prostate ► Spine 24
25. Brain ► Brain extraction ► Functional connectomes ► Structural connectomes ► Brain age ► Alzheimer’s disease ► Vascular lesions ► Identification of MRI contrast ► Meningioma ► Glioma ► Multiple sclerosis 25
26. Kidney ► Abdominal organs ► Cyst segmentation ► Renal transplant 26
27. Prostate ► Cancer 27
28. Spine ► Vertebrae labeling ► Intervertebral disc localization ► Lumbal neural forminal stenosis (LNFS) 28
29. DeepSPINE 29
30. LNFS 30 Automated Pathogenesis-Based Diagnosis of Lumbar Neural Foraminal Stenosis via Deep Multiscale Multitask Learning
31. Challenges 31
32. Challenges ► Medical imaging data sets and repositories ► Medical imaging competitions ► Data ► Interpretability 32
33. Data Sets and Repositories ► [TCIA] ► “Large” ► cancer imaging ► [OpenNeuro] ► brain images ► 168 studies ► 4,718 participants ► [UK Biobank] ► 15,000 participants ► [ADNI] ► Alzheimer’s disease neuroimaging ► 2,000 participants 33
34. Competitions ► [Grand Challenges] ► Almost all of challenges ► [Kaggle] 34
35. REFERENCE ► Alexander SelvikvågLundervold, al. An overview of deep learning in medical imaging focusing on MRI, Zeitschrift für Medizinische Physik Volume 29, Issue 2, May 2019, Pages 102-127. 35
36. Thanks! 36 higuma.lu@gmail.com https://github.com/higumalu
37. Q&A 37