1. ©AP Diffusion Tensor Imaging and Fiber Tractography: Analysis Options in Pediatric Neuroimaging Research Avner Meoded, Thierry A.G.M. Huisman, Andrea Poretti Section of Pediatric Neuroradiology, Division of Pediatric Radiology, Russell H Morgan Department of Radiology and Radiological Science, The Johns Hopkins University School of Medicine, Baltimore, MD ASNR 53rd Annual Meeting, Chicago, April 25-30, 2015 eEdE-181

2. ©AP Disclosure We have nothing to disclose No relevant financial relations interfering with our presentation

3. ©AP Learning objectives 1.To identify the principles of Diffusion Tensor Imaging (DTI) and Fiber Tractography (FT) 2.To recognize the significance of the DTI scalars including fractional anisotropy (FA), mean (MD), axial (AD) and radial (RD) diffusivity and their changes in different conditions 3.To describe different analytic approaches of DTI data and their advantages and disadvantages

4. ©AP Outline 1.Basic knowledge about DTI + FT 2.Qualitative DTI/FT analysis 3.Quantitative analysis: i.Region of interest (ROI) based analysis ii.Atlas-based analysis iii.Voxel-based analysis iv.Tract-based spatial statistics 4.Structural connectome Principles Examples

5. ©AP Diffusion tensor imaging (DTI) Diffusion = in all directions: isotropic Diffusion ≠ in all directions: anisotropic Characterization of 3D shape of water diffusion

6. ©AP Characterization of diffusion 1.v1, v2, v3 = eigenvectors  describes the orientation of the principal coordinate axes 2.λ 1, λ 2, λ 3 = eigenvalues  define the shape of the ellipsoid z x y Measure diffusion along various directions (≥ 6) λ1λ1 λ2λ2 λ3λ3

7. ©AP Fractional anisotropy (FA) FA = degree of anisotropic diffusion Rotationally invariant scalar that measures the fraction of the tensor that can be assigned to anisotropic diffusion

8. ©AP Fractional anisotropy (FA) FA Isotropic diffusion = low FA = dark; anisotropic diffusion = high FA = bright

9. ©AP Directionally encoded color map Color-coding = information about the principal direction of diffusion: 1.red = predominant left-right anisotropic diffusion 2.green = predominant anterior-posterior anisotropic diffusion 3.blue = predominant superior-inferior anisotropic diffusion

10. ©AP Fractional anisotropy (FA) Axon Fiber mixture Myelination FA increaseFA decrease LESS MORE LESS MORE Mori S and Zhang J, Neuron, 2006 anisotropic diffusion isotropic diffusion

11. ©AP Axial and radial diffusivity Axial diffusivity (AD) Rate of diffusion in the direction that is parallel to the main direction of diffusion AD = λ 1 Radial diffusivity (RD) Rate of diffusion in the direction that is perpendicular to the main direction of diffusion RD = (λ 2 + λ 3 )/2 White matter characteristicsAxial diffusivityRadial diffusivity Increased myelinationHighLow Dense axonal packingUnaffectedLow Large axonal diameterHighLow Axonal degenerationLowHigh DemyelinationUnaffectedHigh Feldman HM et al, J Dev Behav Pediatr, 2010

12. ©AP Deterministic fiber tractography (FT) Post-processing technique  reconstruction of the course of major fibers in the brain (several algorithms) Assumption: each voxel contains a single, coherently oriented bundle of axons

13. ©AP Deterministic FT: Limitations 1.Crossing, kissing, merging, or diverging fibers 2.High dependency on FA and angle thresholds  Tracking of pathways that do not exist (false positive) or ineffective tracking of existing pathways (false negative)  Interpretation of FT results requires knowledge of brain anatomy

14. ©AP Crossing fibers 1.In regions with crossing, kissing, merging, or diverging fibers = more than one population of fibers 2.New techniques needed with a higher number of diffusion directions, and higher b-values (HARDI, DSI)  longer acquisition time Jbabdi S and Johansen-Berg H, Brain Connect, 2011

15. ©AP Threshold dependency FA>20 + Angle>30⁰FA>20 + Angle>45⁰FA>20 + Angle>70⁰

16. ©AP Qualitative analysis Morphological information Based on directionally encoded color maps and FT Knowledge of normal DTI images is mandatory Application: e.g. brain malformations to better understand the pathogenesis

17. ©AP Joubert syndrome Molar tooth signNormal anatomy Poretti A et al, AJNR, 2011

18. ©AP DTI in Joubert syndrome Joubert syndrome Control Poretti A et al, AJNR, 2007 superior cerebellar peduncles Absence of decussation of the superior cerebellar peduncles (“red dot”)

19. ©AP FT in Joubert syndrome (JS) Joubert syndromeControl Poretti A et al, AJNR, 2007 Absence of decussation of the superior cerebellar peduncles (SCP) and corticospinal tracts (CST)  JS = axonal guidance disorder SCP decussation CST decussation

20. ©AP ROI-based analysis Manual = ROIs are manually placed Selected anatomical regions = hypothesis driven

21. ©AP ROI-based analysis Advantages 1.Accurate especially for subjects with large anatomical changes 2.Low number of regions = higher statistical power and possibility to depict small differences Disadvantages 1.Reproducibility  multiple measurements + ICC calculation 2.Time consuming (?) 3.Hard to include the entire (3D) anatomical structure 4.Not applicable without hypothesis 5.Differences missed in regions not included in the analysis ICC: Intra Class Coefficient

22. ©AP ROI-based analysis: DTI of the brainstem in achondroplasia Skeletal dysplasia with narrowing of cranio-cervical junction + foramen magnum Goal = To study the microstructural integrity of brainstem white matter tracts in children with achondroplasia compared to age-matched controls Bosemani T, Meoded A, Huisman TA, Poretti A et al, Dev Med Child Neurol, 2014

23. ©AP ROI-based analysis: DTI in achondroplasia 7 ROIs; FA, MD, AD, RD 2 analysis by first reader, 1 analysis by second reader  inter-/intra-rater reliability Comparison patients  controls Correlation with: –Severity of CCJ narrowing –Neurological findings Bosemani T, Meoded A, Huisman TA, Poretti A et al, Dev Med Child Neurol, 2014

24. ©AP ROI-based analysis: DTI in achondroplasia Lower brainstem: ↓ FA + ↑ MD and RD in patients compared to controls ↑ MD+RD+AD in bilateral CST/MCP = white matter injury not limited to lower brainstem ↓ FA in lower brainstem  ↑ CCJ narrowing  anatomical proximity lower brainstem  CCJ No correlation DTI  clinical findings Bosemani T, Meoded A, Huisman TA, Poretti A et al, Dev Med Child Neurol, 2014

25. ©AP Atlas- + voxel-based analysis No underlying theory about where we might find group differences  global approach 1.Atlas-based analysis = automated segmentation of the brain into 170-180 well defined anatomical areas 2.Voxel-based analysis = analysis of each voxel within the brain (>100.000)

26. ©AP Atlas- + voxel-based analysis Yoshida S et al, Pediatr Radiol, 2013

27. ©AP Normalization process Standardization of info about location of brain structures Transformation to an age- matched template Two steps: 1.Linear = low degree of freedom transformation, good for aligning images within subject 2.Non-linear = high degree of freedom transformation (LDDMM) Yoshida S et al, Pediatr Radiol, 2013

28. ©AP LDDMM Large Deformation Diffeomorphic Metric Mapping Properties of LDDMM: –Highly elastic: important for normalization of brains with atrophy and/or ventriculomegaly –Preservation of topology –Reciprocal transformation: allows ABA + VBA

29. ©AP Atlas- + voxel-based analysis Atlas-based analysis  Advantages: 1.“Limited” number of anatomical information (170-180 regions) 2.Evaluation of native images  Disadvantages: 1.Information about anatomical localization lower than VBA 2.Less sensitive for very small lesions Voxel-based analysis  Advantages: 1.Unbiased whole brain analysis 2.Sensitive for small lesions 3.Specific locations of significant group differences/correlations is automatically shown  Disadvantages: 1.High amount of information  low statistical power, correction for multiple comparison needed 2.Accuracy of voxel-to-voxel alignment across subjects is low 3.Not sensitive for widespread lesions

30. ©AP ABA + VBA in children after hemispherectomy Hemispherectomy = surgical therapy for intractable seizures arising from a single cerebral hemisphere Goal = to study the changes of DTI metrics in the white matter regions of the remaining hemisphere (model for brain plasticity?) Meoded A, Huisman TA, Poretti A et al, in preparation

31. ©AP ABA + VBA in children after hemispherectomy n=19 –Congenital etiology (n=11) –Acquired etiology (n=8) ABA + VBA of the remaining cerebral hemisphere Measurement of FA, MD, AD and RD Meoded A, Huisman TA, Poretti A et al, in preparation VBAABA

32. ©AP ABA + VBA in children after hemispherectomy FA + AD ↓ and MD + RD ↑ = secondary Wallerian degeneration Wallerian degeneration: commissural + association > projection Presurgical DTI changes + postsurgical DTI normalization: acquired > congenital  reorganization, plasticity? Meoded A, Huisman TA, Poretti A et al, in preparation

33. ©AP Tract-Based Spatial Statistics (TBSS) Variation of whole-brain analysis FA images from multiple subjects are aligned to a common space Tract representation including only voxels at the center of tracts common to all The resulting skeleton data for all subjects are analyzed

34. ©AP Tract-Based Spatial Statistics (TBSS) Advantages 1.All locations across the brain are tested without a priori hypotheses 2.Powerful statistical module Disadvantages 1.Skeletons are derived for the entire white matter region and no distinction between white matter structures is made 2.Variations in the periphery of white matter tracts may be missed (not included) 3.Possibile effect of brain atrophy

35. ©AP TBSS in Niemann Pick disease type C Niemann Pick disease type C (NPC) = rare neurometabolic disease Goal = study the microstructural changes in patients with NPC compared to age- matched controls 30-year-old man with NPC

36. ©AP TBSS in Niemann Pick disease type C n=8 Measurement of FA, MD, AD and RD Diffuse reduction in FA and increase in MD, AD and RD in patients compared to age- matched controls Poretti A, Meoded A, Huisman TA et al, in preparation

37. ©AP The human structural connectome Structural connectome = map of the brain's structural connections, rendered as a network Network = set of nodes + edges Nodes: discrete subunits  cortical regions Edges: structural connection between any pairs of gray mater regions Rubinov M and Sporns O, Neuroimage, 2010; Sporns O, Neuroimage, 2013

38. ©AP Structural connectivity 1.Topology analysis: Network metrics  global/regional network organization Information: segregation, integration, small worldness, centrality (Hubs) 2.Network based statistics: Strength of connectivity between brain regions at a pairwise level Subnetworks

39. ©AP Topology analysis Segregation = ability for processing to occur within densely interconnected groups Clustering coefficient = fraction of connections that connect the neighbors of a given node Integration = ease for brain regions to communicate based on route of information flow Small world = well-designed network combining high clustering and short path Centrality = measurement of the influence of a node compared to the rest of the network Betweenness centrality = fraction of short paths between nodes of the network that pass through a given node Hub = region with high degree of connections (high centrality) Assortativity = correlation between the degrees of connected node pairs Rubinov M and Sporns O, Neuroimage, 2010

40. ©AP Structural connectivity matrix Probabilistic tractography : network mode between all ROI AAL template: parcellation of gray matter in 108 structures Connectome reconstruction DTI in native space T1 MPRAGE Linear registration FLIRT Non linear registration NIRT Data registered to MNI space Connectome analysis Network based statistics Topology analysis

41. ©AP Structural connectome in agenesis of the corpus callosum (AgCC) Meoded A, Huisman TA, Poretti A et al, Eur Radiol, 2014 ↑ Cluster coefficient (p=0.005) ↓ Small worldness (p=0.005) ↑ Transitivity (p=0.003) ↓ Assortativity (p=0.03)

42. ©AP Hubs in patients  Network hubs = nodes with high betweenness  Hubs in right insula, precuneus, and lingual gyrus and left insula and precuneus Meoded A, Huisman TA, Poretti A et al, Eur Radiol, 2014

43. ©AP Network based statistics 1.Intrahemispheric left fronto-parietal (p=0.013) 2.Interhemispheric fronto-cerebellar (p=0.021) 3.Intrahemispheric left temporo-occipital (p=0.048) Meoded A, Huisman TA, Poretti A et al, Eur Radiol, 2014 Children with AgCC  virtual callosotomy controls:

44. ©AP Structural connectome in AgCC 1.AgCC = more segregated + less integrated brain connectome 2.No hub in the cerebellum in patients = global connectivity ↓ 3.Highly inter-connected interhemispheric fronto- cerebellar network + two insular hubs = brain plasticity  ↑ interhemispheric flow of information through alternative pathway (anterior commissure?) 4.Lower modularity = connectome reorganization to reduce connection costs at the expense of decreasing integrative capacity? Meoded A, Huisman TA, Poretti A et al, Eur Radiol, 2014

45. ©AP Conclusions 1.DTI/FT = ideal neuroimaging tool to study the pediatric brain ultrastructure and networking 2.Application to several pediatric neurology diseases 3.Qualitative and quantitative analysis 4.Different methods for quantitative analysis 5.Method depends on research question