Introduction to Diffusion Tractography In vivo tracer Manganese Tractography ex vivo Tim B. Dyrby, PhD Danish Research Centre for Magnetic Resonance (DRCMR) Copenhagen University Hospital Hvidovre Denmark www.drcmr.dk Mini pig brain Ex vivo DTI Hippocampal layers
Tractography IFOF Brain network analysis Subject 1 Subject 2 Courtesy Nina Reislev
Suggested readings: Jones et al., 2013: White matter integrity, fiber count, and other fallacies: The do's and don'ts of diffusion MRI, NIMG Jbabdi & Johansen-Berg, 2011: Tractography: Where Do We Go from Here?, Brain Conn Tractography To be a good tractographer: - 1) is a matter of training and experience.... This you will NOT learn from this lecture - 2) It is essential to have basic insight into and understand the tractography method and the relation to underlying axonal pathways. that you will learn from this lecture. The learning objectives are: To understand how the true underlying axonal projections look like To get insight into the underlying basic pipeline of tractography To learn the potential impact of using different seeding strategies To be aware of the sources of ambiguity, pros and cons of tractography important for interpretation and analysis.
Outline The projection of an axonal bundle the truth The black box of the Tractography pipeline Seeding strategies; free and constrained tracking Understanding tractography Potential and limitations Example :: Anatomical Connectivity Mapping (ACM) Summary
Tracing techniques Klinger dissection Extracting the highways Invasive tracer labeling Detailed single axonal projections SLF III Martino et al., 2010, Cortex Schmahmann and Pandya, 2006 (book)
Tissue microstructure Axons in white matter Electron microscopy (EM) study - Monkey corpus callosum www.neurophilosophy.wordpress.com Small (unmyelinated 0.4 μm) to large (myelinated>1 μm) Density: ~781.300-3 mill per mm 2 diffusion Lamanti & Rakic,1990, J. Comp. Neurol.
The physiological meaning of brain networks Conduction speed and delays Axon diameters identified from BDA tracer; Innocenti et al., 2013, CerCor
The physiological meaning of brain networks Conduction speed and delays But today: We assume a brain connection as being binary connected or not (segmentation) We try to express connection probabilities between A B via by counting streamlines Human Tractography Monkey Tracer Catani et al., 2008, Cortex Schmahmann and Pandya, 2006 (book)
Outline The projection of an axonal bundle the truth The black box of the Tractography pipeline Seeding strategies free and constrained tracking Understanding tractography Potential and limitations Example :: Anatomical Connectivity Mapping (ACM) Summary
Definition of tractography A framework to non-invasively estimate and visualise fibre pathways in the brain Deterministic A B Tractography Probabilistic Parker and Alexander (2005)
Deterministic tractography Processing stages Scanning Fibre reconstruction Fibre tracking (Streamlining)
Probabilistic tractography Processing stages Scanning Fibre reconstruction Fibre orientated uncertainty Fibre tracking (Streamlining) Outputs a Probabilistic Index of Connectivity (PICo) map describing the confidence that a voxel is connected to the seed. Parker & Alexander, 2005
Drawing a streamline Scanning Fibre reconstruction Fibre tracking (Streamlining) Scanning Fibre reconstruction Fibre orientated uncertainty Fibre tracking (Streamlining)
Drawing a streamline : methods Fibre Assignment by Continuous Tracking - FACT The streamline follows a path parallel to the principal eigenvector until the end of the voxel. ( then it takes the new direction from the new voxel, etc.) The step size is not fixed This technique works well with low curvature regions FACT is used in DTI-Studio and TrackVis (default algorithm) Mori et al. 1999, Ann Neurol
Drawing a streamline : methods Interpolated Streamline Tractography General Description : At each step a new direction is interpolated taking into account the surrounding eigenvectors New Position = Old + Step x Position Size Principal Eigenvector With the Interpolated Streamline approach the step size is fixed and is smaller than voxel dimensions. Sub-voxel Interpolation is necessary because we need a new direction in each new position (Interpolation can be done both on Raw Data or Tensor Elements) Different numerical methods exist (Euler, Midpoint, 4th Order Runge Kutta, TEND) Euler-like approaches are used in Brainvisa, ExploreDTI, Trackvis, etc. (Image from Mori S, van Zijl PCM. NMR Biomed. 2002; 15: 468-480 )
Sources of ambiguity Fibre reconstruction and streamlining Added ambiguity Scanning Fibre reconstruction Fibre orientated uncertainty Fibre tracking (Streamlining)
Tractography Fibre reconstruction: Crossing fibres Image resolution matters Angular resolution Sequence parameters used really important but often not considered Schmahmann and Pandya, 2006 (book)
Tractography Fibre reconstruction and streamlining Single fibre reconstruction Multi-fibre reconstruction Fibre model: Single or multiple fibre methods? From Behrens et al. 2007 (Neuroimage) Behrens et al., 2007, NIMG
Example :: Tractography Single versus multi-fibre A straight tract: The corticospinal tract Same seed ROI Deterministic tractography Single-fibre: DTI Multi-fibre: CSD
Example :: Tractography Single versus multi-fibre A complex tract: Callosal projections Same seed ROI Deterministic tractography Single-fibre: DTI Multi-fibre: CSD
Example :: Tractography Deterministic versus probabilistic A complex tract: Callosal projections Same seed ROI Deterministic - CSD Take-home message: Probabilistic - CSD 1) For tractography, multi-fibre reconstruction methods should always be preferred 2) How well you sample a pathway depends on tractography method and seeding strategy
Tractography General settings Stopping criteria: - FA values - very low e.g. 0.15-0.2 (avoid using FA; DTI is a single fibre model) - Max. Curvature angle between the old and new direction typ. 80 degrees - Brain mask - e.g. GM θ Take-home message: Check your toolbox; how it works and default settings Streamline rejections (Not to include): - Streamline length to max or min length - Streamline looping Parameter settings: - Number of streamlines to send out WARNING: Tractography toolboxes have different definitions Always check - Step size Updating streamline drawing using sub-voxel steps e.g. 1/5-1/10 - Tracking method (normally not optional)
Outline The projection of an axonal bundle the truth The black box of the Tractography pipeline Seeding strategies; free and constrained tracking Understanding tractography Potential and limitations Example :: Anatomical Connectivity Mapping (ACM) Summary
Seeding and constraining tracking strategies Seeding based : Free tracking: - Seeding ROI from where to start (bidirectional) Constrained tractography: - Target ROI Streamlines to be investigated (comb. w. terminate ROI) - Waypoint ROI Streamlines must project through - Exclusive ROI Streamlines must NOT project through
Seeding and constraining tracking strategies Seeding based : Free tracking: - Seeding ROI from where to start (bidirectional) Constrained tractography: - Target ROI Streamlines to be investigated (comb. w. terminate ROI) - Waypoint ROI Streamlines must project through - Exclusive ROI Streamlines must NOT project through
Seeding and constraining tracking strategies Seeding based : Free tracking: - Seeding ROI from where to start (bidirectional) Constrained tractography: - Target ROI Streamlines to be investigated (comb. w. terminate ROI) - Waypoint ROI Streamlines must project through - Exclusive ROI Streamlines must NOT project through
Seeding and constraining tracking strategies Seeding based : Free tracking: - Seeding ROI from where to start (bidirectional) Constrained tractography: - Target ROI Streamlines to be investigated (comb. w. terminate ROI) - Waypoint ROI Streamlines must project through - Exclusive ROI Streamlines must NOT project through
Seeding and constraining tracking strategies Seeding based : Free tracking: - Seeding ROI from where to start (bidirectional) Take-home message: When delineating a pathway by constraining tractography The ROIs should normally not impact the shape of the segmented tract Constrained tractography: - Target ROI Streamlines to be investigated (comb. w. terminate ROI) - Waypoint ROI Streamlines must project through - Exclusive ROI Streamlines must NOT project through Whole-brain deterministic filtering: - AND ROI Streamlines must project through - NOT ROI Streamlines must NOT project through
Outline The projection of an axonal bundle the truth The black box of the Tractography pipeline Seeding strategies; free and constrained tracking Understanding tractography Potential and limitations Example :: Anatomical Connectivity Mapping (ACM) Summary
Sources of ambiguity Getting insight into tractography challenges? Added ambiguity Scanning Fibre reconstruction Fibre orientated uncertainty Fibre tracking (Streamlining)
Sources of ambiguity Getting insight into tractography challenges? Added ambiguity Scanning Fibre reconstruction Fibre orientated uncertainty Fibre tracking (Streamlining)
Image resolution has impact Typical clinical in vivo 2.3x2.3x2.3 mm 3 voxels 0.8x0.8x0.8 mm 3 Postmortem Monkey brain 0.4x0.4x0.4 mm 3 0.25x0.25x0.35 mm 3 Dyrby et al., 2011, HBM; Dyrby et al 2014, NIMG
Ex vivo imaging Direct validation of tractography Seed defined from tracer injections Manganese tracer BDA tracer Seed Invasive in vivo neuronal tracer Ex vivo Probabilistic tractography Dyrby et al., 2007, NIMG, Knösche et al., in revision
Tractography validation Reproducibility of tracts across brains :: Not guarantee for the correctness Brain I Brain II Brain III False-negative False-positive RED: Contralateral cortico-cortical fibre BLUE: Corticothalamic fibre ORANGE: Corticonigral fibre WHITE: Intersection of two or more fibres Path-length dependency A termination problem Dyrby et al., 2007, NIMG
Tractography Termination can be challenging Reveley et al., 2015, PNAS Dyrby et al., 2007, NIMG
Tractography challenges Termination in GM structures Limited image resolution Partial volume effect (PVE) Due to PVE the GM appears as GM/WM and tracking risk just skirting around as a FALSE-POSITIVE Seed Take-home message: PVE is a problem for all tractography methods Deterministic as well as for probabilistic REMEMBER : tractography is bidirectional
Tractography challenges Termination in GM structures Limited image resolution Partial volume effect (PVE) Due to PVE the GM appears as GM/WM and tracking risk just skirting around as a FALSE-POSITIVE Seed
Tractography validation Unwanted :: Path-length dependency (PLD) effect Brain I Brain II Brain III False-negative False-positive RED: Contralateral cortico-cortical fibre BLUE: Corticothalamic fibre ORANGE: Corticonigral fibre WHITE: Intersection of two or more fibres Path-length dependency A termination problem Dyrby et al., 2007, NIMG
Probabilistic tractography Path-length dependency (PLD) Lowering the threshold the longer tracts appears shortened Parker & Alexander et al., 2005, Phil. Trans. R. Soc. B
Path-length dependency Tracer versus tractography Tracer - decreasing number of tracer labeled axons from distance 12 9 5 5 8 5 3 3 3
Path-length dependency Tracer versus tractography Tracer - decreasing number of tracer labeled axons from distance Tractography has PLD dominated by diffusion anisotropy locally 12 9 5 5 8 5 3 3 3
Path-length dependency Not a linear effect for probabilistic tractography Tracer - decreasing number of tracer labeled axons from distance Tractography has PLD dominated by diffusion anisotropy locally $
Free streaming With ICE-T Probabilistic tractography Path-length dependency is non-linear Distance Along Canonical Streamline (voxels) Liptrot et al., 2014, PlosOne
Take-home message: The PLD effect is a non-linear effect in probabilistic tractography. It affects the longer pathways. More investigations to understand it s impact Liptrot et al., 2014, PlosOne
Tractography Comments on seeding strategies The selected seeding strategy will impact your result Whole-brain seeding (typical: Deterministic) Local seeding ROI (typical: Probabilistic) Li et al., 2012, HBM
Outline The projection of an axonal bundle the truth The black box of the Tractography pipeline Seeding strategies; free and constrained tracking Understanding tractography Potential and limitations Example :: Anatomical Connectivity Mapping (ACM) Summary
Multiple sclerosis Changing the microstructural environment Courtesy Matthew Liptrot
The display of MS with conventional MRI T2 weighted (FLAIR) MRI Structural MRI measures Lesions (load and distribution) Active lesions with Gd contrast Brain atrophy (GM/NAWM) T1 weighted MRI The challenge: Weakly correlating with clinical tests
New type of group analysis of MS subjects The Anatomical Connectivity Mapping (ACM) MS subject 1 MS subject 2 MS subject 3
Probabilistic tractography Drawing a streamline Brain connectivity is displayed by drawing a streamline beginning from a SEED region. (Parker et al 2005) A B Probabilistic tractography Probabilistic Parker & Alexander, 2005
Probabilistic tractography Whole-brain probabilistic tractography Anatomical connectivity map (ACM) using ALL voxels within the brain as seed ACM A B Embleton et al., 2007, ISMRM Cercignani et al., 2012 NMR Biomed
Comparison between MS phenotypes SPMS versus RRMS Voxel based t-test statistics within motor projections; FWE corrected Lyksborg et al., 2014, PlosOne
Diffusion Tractography Summary To be a good tractographer is a matter of training and experience. AND to know your tractography method, its potentials and limitations. The true axons from a projections site seem to intermingle with other axonal projections when entering the major pathway highway Such effect is not guaranteed to be visualized with streamline tractography (limited image resolution) SLF III Schmahmann and Pandya, 2006 (book)
Diffusion Tractography Summary To be a good tractographer is a matter of training and experience. AND to know your tractography method, its potentials and limitations. The true axons from a projections site seem to intermingle with other axonal projections when entering the major pathway highway Such effect is not guaranteed to be visualized with streamline tractography (limited image resolution) The tractography pipeline Includes steps from acquisition, post-processing, fibre reconstruction and streamline tracking. - Sources of ambiguities are accumulated - Ensure the quality of your processed data. ALWAYS inspect processed data Added ambiguity Scanning Fibre reconstruction Fibre orientated uncertainty Fibre tracking (Streamlining)
Diffusion Tractography Summary To be a good tractographer is a matter of training and experience. AND to know your tractography method, its potentials and limitations. The true axons from a projections site seem to intermingle with other axonal projections when entering the major pathway highway Such effect is not guaranteed to be visualized with streamline tractography (limited image resolution) The tractography pipeline Includes steps from acquisition, post-processing, fibre reconstruction and streamline tracking. - Sources of ambiguities are accumulated - Ensure the quality of your processed data. ALWAYS inspect processed data Selected seeding strategies have huge impact on extracted pathways. When publishing, remember ALWAYS to describe the seeding strategies in detail
Diffusion Tractography Summary To be a good tractographer is a matter of training and experience. AND to know your tractography method, its potentials and limitations. The true axons from a projections site seem to intermingle with other axonal projections when entering the major pathway highway Such effect is not guaranteed to be visualized with streamline tractography (limited image resolution) The tractography pipeline Includes steps from acquisition, post-processing, fibre reconstruction and streamline tracking. - Sources of ambiguities are accumulated - Ensure the quality of your processed data. ALWAYS inspect processed data Selected seeding strategies have huge impact on extracted pathways. When publishing, remember ALWAYS to describe the seeding strategies in detail Tractography is a unique technique for non-invasive tracing, and the only one. So enjoy its huge potential BUT be realistic about its limitations and be self-critical to your tractography results! Use independent data to support tractography i.e. TMS, tracer knowledge etc.
Acknowledgment Diffusion Imaging Group (DIG), DRCMR (Download data sets :: http://dig.drcmr.dk) In vivo tracer Manganese Illustrations: Puk A. Dyrby Kasper W. Andersen Henrik Lundell Nina L. Reislev Matthew G. Liptrot Flavio Dell Acqua CONNECT Consortium The Lundbeck Foundation The Danish Sclerosis Foundation ICT http://www.brain-connect.eu