1. 1 Brain Atlases Epidaure-LONI Associated teams 2002-2004 X. Pennec, N. Ayache, A. Pitiot, V. Arsigny, P. Fillard P. Thompson, A. Toga, J. Annese EPIDAURE Project 2004, route des Lucioles B.P. 93 06902 Sophia Antipolis Cedex (France) LONI Laboratory of Neuro Imaging UCLA California, USA

2. 2 Epidaure LONI Associated teams on Brain Atlases Functional and structural analysis of the human brain o High variability of the shape of structures o High variability of functions localization o High number of relevant variables o (age, sex, pathologies…) Building Statistical atlases requires o Powerful algorithms o Large datasets (several hundreds) Complementarity of the teams o Epidaure: Methodology of image registration/segmentation o LONI: neuroanatomical experts, development/exploitation of large international databases EPIDAURE LONI Atlas MRIHistology

3. 3 Brain Atlas: Original objectives Capitalize on the join expertise for more powerful atlases o Algorithms Registration / segmentation tools o Databases MRI, Spect, Histology o Evaluation / validation of the methodologies Exchange researchers o PhD Co-supervision (A. Pitiot) o Annual visits P. Thompson, A. Toga, N. Ayache, H. Delingette, X. Pennec o Workshop on Brain Atlases summer school on Computational Anatomy EPIDAURE LONI EPIDAURE

4. 4 LONI at UCLA The LONI team TO BE DEVELOPPED The coordinator: Paul Thompson (To be summurized) l Batchelor 1991, Master 1993, Oxford Univ. l PhD in Neuroscience 1998, UCLA. l Assistant professor at UCLA since 1994 o Publications l 200 refereed publication (1996-2004) (Nature Neuroscience, Nature genetics, TMI, MedIA, J. Neurosciences, NeuroImage…) l Assoc. editor of Human Brain Mapping, Editorial board of MedIA… o Research interests: l Human brain mapping (mathematical and computer intensive methods, building of atlases, encoding variability…) l Brain pathologies (Alzheimer, Schizaphrenia, neuro-oncology) l Barin development (pediatric images)

5. 5 Quick History Scientific collaboration initiated in 1999 o Evaluation of histological slices registration (Rat brain images fro LONI) l S. Ourselin et al., “Reconstructing a 3D Structure from Serial Histological Sections”, IVC 2000. o 2 Chapters in “Brain Warping”, A.W. Toga editor, 2003 l J.-P. Thirion, “Diffusing Models and Applications” l G. Subsol, “Crest Lines for Curve Based Warping” o Co-supervision of Alain Pitiot’s PhD (started sept. 2000) l Alain Pitiot, “Segmentation automatique de structures cerebrales s’appuyant sur des connaissances explicites”, Ecole des Mines de Paris, Nov. 2003. Leveraging by the asociated team program (2002-2003) o 5 peer reviewed publications co-authored by both teams o Software exchanges (10 UCLA publication co-authored by A. Pitiot) o Data exchanges (Arsigny, prize at MICCAI, Media) New collaborations on brain variability modeling (summer 2003-2005++) o PhD Vincent Arsigny: matching sulcal lines (summer 2003, in preparation) o PhD Pierre Fillard (started sept 2004): article IPMI’05, MICCAI’05

6. 6 Presentation Overview To be completed

7. 7 PhD Alain Pitiot Automatic segmentation of brain structure using explicit knowledge Automated segmentation system maximum a priori knowledge deformable models explicit information Hybrid MRI/histology atlas MRI: in vivo, macroscopic histology: post mortem, microscopic  3-D reconstruction 3-D MRI with superimposed segmented structures 3-D histological volume

8. 8 Composite segmentation system neural texture filtering piecewise affine registration knowledge-driven segmentation learnt shape models

9. 9 Neural Texture Filtering Texture classification neural classifier  texture maps Hybrid neural architecture hybrid neural architecture classification results

10. 10 Statistical Shape Modeling Objective statistical shape analysis  PCA on shapes Learning approach to reparameterization introduction of explicit knowledge  shape distance matrix  observed transport shape measure  shape modes of variation

11. 11 Knowledge-driven Segmentation Objective segmentation of anatomical structures mix bottom-up constraints with top-down medical knowledge  explicit medical information Rules escape local minima dynamic control meta-rules (error checking) segmentation results

12. 12 Piecewise Affine Registration Motivation registration of 2-D biological images adapted transformation model: piecewise affine specific similarity measure: constrained correlation coefficient Method similarity map hierarchical clustering hybrid elastic/affine interpolation registration results

13. 13 wcci-ijcnn ’03 hbm’02 journal of anatomy ipmi’03 hbm’01,’02,’03 neuroimage 2003 miccai’03 tmi 2002 wbir’03 tmi 2003 Joint Contributions

14. 14 Executive summary 2002-2003 Exchange of researchers o PhD Alain Pitiot: joint supervision and localization (50% each) o Visit of N. Ayache and H. Delingette at UCLA (Dec. 2001) o Visit of P. Thompson at Sophia (Nov 2003, PhD defense) Exchange of software o Reparameterization techniques of Pitiot (IPMI’03) Visual cortex project (Annese HBM’03, Neuroimage 04) o Robust registration software Baladin included in the LONI Pipeline (MAP: Mouse brain Atlas, J. Annese, Visual Cortex, S. Ying, MD, Cerebellum). Exchange of Data o Neuroanatomical expertise o Segmentations of 4 structures in Brain MR (Pitiot MICCAI’03) o Histological data from J. Annese (Pitiot WBIR’03, Arsigny MICCAI’03) o Segmented Brain sulci (V. Arsigny ++)

15. 15 Presentation Overview To be completed

16. 16 A unique database o Acquired during 10 years of participation to international projects o More than 500 subjects o 72 manually delineated sulcal lines (roots of grooves on the surface of the brain cortex) o MRI images o Normal and pathological data o Medical annotations o Potentially new anatomical findings Morphometry of Sucal Lines (summer 2003-2005++)

17. 17 Innovative methods to model the brain variability o Learn local brain variability from sulci o Learn global correlations in variability and link with asymetry o Better constrain inter-subject registration o Correlate this variability with age, pathologies l Understanding of neurological diseases (Alzheimer, schizophrénie...) l Early diagnosis, follow-up studies Morphometry of Sucal Lines (summer 2003-2005++)

18. 18 Objectifs de l’étude Description fine de la variabilité des lignes sulcales grâce à des techniques d ’analyse innovantes Obtention de nouvelles connaissances anatomiques : corrélations morphologiques entre sillons, asymétrie cérébrale... Application : aide au diagnostic, meilleure compréhension de maladies neurologiques (Alzheimer, schizophrénie...) Etiquetage automatique des lignes sulcales Données anatomiques (vert-jaune) et lignes moyennes (rouge) Variance le long des lignes moyennes. Rouge: faible, bleu élevée:

19. 19 Computation of Average Sulci red : mean curve green et yellow : ~80 instances of 72 sulci Alternate minimization of global variance o Dynamic programming to match the mean to instances o Gradient descent to compute the mean curve position Arsigny et al. 2004, to appear Sylvius Fissure

20. 20 Anatomical variability Variance along the mean sulci o Red (low) to blue (high) Arsigny et al. 2004, to appear

21. 21 Extraction of Covariance Tensors Covariance Tensors along Sylvius Fissure Currently: 80 instances of 72 sulci About 1250 tensors Fillard, Pennec, Ayache, Thompson, 2004, to appear Color codes Trace

22. 22 Compressed Tensor Representation Mean sulcal line + 4 covariance matrices o optimize for the 4 most representative tensors o Interpolation in-between, extrapolation outside (removes outliers) Sylvian fissure The 4 most representative tensors. Interpolation from the 4 tensors. Raw estimation

23. 23 Compressed Tensor Representation Representative Tensors (250)Reconstructed Tensors (1250) (Riemannian Interpolation) Fillard-Pennec-Ayache-Thompson 2004, to appear Original Tensors (~ 1250)

24. 24 Variability Tensors Color codes tensor trace Fillard-Pennec-Ayache-Thompson 2004, to appear

25. 25 Asymmetry Measure Color Codes Distance between “symmetric” tensors

26. 26 Full Brain extrapolation of the variability Color code: principal eigenvector (red: left- right, green: posterior-anterior, blue: inferior-superior) Color code: trace Anterior view Fillard-Pennec-Thompson- Ayache 2004, to appear

27. 27 Color code: principal eigenvector Color code: trace Full Brain extrapolation of the variability Fillard-Pennec-Thompson- Ayache 2004, to appear

28. 28 Scientific results 2004 Leveraging the Theory o General framework for computing on tensor fields o Interpolation, diffusion, filtering… “Side results” in Diffusion tensor imaging o Regularization for fiber tracts estimation o Registration,... Variability of the brain o Learn Variability from Large Group Studies o Statistical Comparisons between Groups o Exploit Variability to Improve Inter-Subject Registration

29. 29 Executive summary 2004 Important scientific results IPAM summer school on Computational Anatomy o 1 week, July 2004, organized by P. Thompson + 1 week on functional brain imaging o Participation of N. Ayache (Invited speaker), X. Pennec, P. Fillard, and members of Odyssee (O. Faugeras, 2 PhDs). Tutorial at MICCAI 2004 on evolving processes in Med. Images o Organized by N. Ayache, P. Thompson speaker o Visit of P. Thompson at Sophia afterward

30. 30 Work plan for 2005 ++ Modeling the brain variability (PhD P. Fillard) o Validation of the developed models o Learn the full Green’s function Cov(x,y) for all x,y o Theoretical tools already available Better constrains inter-subject registration (PhD V. Arsigny) o Non stationnarity OK (R. Stefanescu) o Extend to long distnace correlations Investigate GRID aspects (PhD. T. Glatard) o LONI Pipeline, BIRN Summer school on Computation Anatomy in Sophia-Antipolis in 2006

31. 31 Planned budget (2005) Salaries o PhD Fellowship “complement” for P. Fillard o Visit P. Thompson at Sophia Missions o Long stay of P. Fillard at LONI A budget is planned at LONI for complementing the salary o Conferences + visits at UCLA: l IPMI’05 (Glenwood Springs, Colorado): X. Pennec + P. Fillard l MICCAI’05 (Palm Springs, CA): N. Ayache, V. Arsigny, X. Pennec

32. 32 Conclusion Very active collaboration (software, data, publications) Access to a unique database (acquision cost > 1 M$) Potentially new findings in neuro-anatomy and in some brain pathologies

33. 33 Full Brain Interpolation Color code: principal eigenvectorColor code: trace Superior view

34. 34 Volume Extrapolation Color codes Trace (~1 Million of Tensors) Fillard-Pennec-Thompson- Ayache 2004, to appear

35. 35 Variability Tensors Axial Coronal Fillard-Pennec-Thompson- Ayache 2004, to appear

36. 36 Variability Tensors Sagittal 1 Fillard-Pennec-Thompson- Ayache 2004, to appear

37. 37 Variability Tensors Sagittal 2 Fillard-Pennec-Thompson- Ayache 2004, to appear

38. 38 Extracting a model of variability step by step: 1 - 36 pairs of sulci with 80 instances of each. 2 - Covariance Tensors calculated along the mean sulcal lines. 3 - Covariance Tensors alone. 4 – Tensors retained for the modeling (250) 5 – Riemannian interpolation along the mean sulcal lines (1250)

39. 39 Executive Summary Doctorat en co-tutelle de Alain Pitiot (a passé 50% de son temps dans chaque équipe) Soutenance prévue fin Novembre 2003, Visite et séminaire de Paul Thompson (Loni) à Sophia à cette occasion Nouvelle Collaboration : thèse de Vincent Arsigny qui se prolongera en 2004 et plus Visites de N. Ayache et H. Delingette au Loni. Invitation de N. Ayache en 2004 Nombreuses publications co-signées par les deux équipes Epidaure: plusieurs logiciels d’analyse transférés au Loni. Loni: Mise à disposition de bases de données exceptionnelles sur le cerveau + expertise neuroanatomie