FINDING THE MICRO IN THE MACRO USING ULTRA- HIGH RESOLUTION MR IMAGING

Transcription

1 OHBM Educational Course Why it all comes back to Anatomy, Vancouver, Canada, June 25, 2017 FINDING THE MICRO IN THE MACRO USING ULTRA- HIGH RESOLUTION MR IMAGING 7T Rainer Goebel Maastricht Brain Imaging Center (M-BIC), Dept. of Cognitive Neuroscience, Maastricht University & National Institute of Neuroscience (NIN) of the Royal Netherlands Academy of Arts and Sciences, Amsterdam, The Netherlands & CEO Brain Innovation

2 Overview From Areas and Networks to Columns and Layers - Understanding visual perception and cognition at multiple scales - Towards mesoscopic-scale neuroimaging at 7 Tesla and beyond Cracking Coding Principles at Columnar Level - Mapping columnar-level features using sub-millimeter fmri - The challenge ahead: Unraveling unknown feature codes - Mesoscopic correlates of visual cognition: Perceptual switches Extraction of Information from Laminar Profiles - Separating bottom-up from top-down information flow - Applications to attention and predictive coding tasks Summary and Conclusions

3 Multi-Scale Functional Organization of the Human Brain Macroscopic Level: Specialized Areas and Large-Scale Networks Sensory input Behavior Brain What are elementary mental/neural functional components? Understand categorical representations in modules of the mind and how they are embedded in areas of the brain (e.g. face area vs house/place area). How do basic components interact and unfold over time? Understand communication between brain areas, i.e. direction of information flow (anatomical and effective connectivity).

4 Multi-Scale Functional Organization of the Human Brain Mesoscopic Level: Features Coded Within Specialized Areas Sensory input Behavior Brain What are elementary mental/neural functional components? Understand alphabet of basic features within specialized brain areas (e.g. face area) and how specific words (e.g. individual faces) are encoded as distributed patterns across mapped features. How do basic components interact and develop over time? Understand interactions between features, e.g. how complex feature codes in higher areas emerge from simpler features in lower areas.

5 Investigating the Brain at Macroscopic Level: Areas & Networks Where pathway (spatial attention, motion, orientation in space ) What pathway (object recognition) Goebel et al. (2012) The Visual System. In G.Paxinos,& J.K. Mai (Eds). The Human Nervous System 2nd edition.

6 Selected Mid-Level Visual Areas Mapped in Individual Brain Mapping specialised areas in individual brains is a prerequisite to investigate internal functional organisation using sub-millimetre fmri LH RH V5/hMT Motion EBA Bodies LOC Objects V4 Colors FFA Faces PPA Places VWFA Words RH LH

7 Retinotopic Mapping of Early Visual Areas using the Population Receptive Field (prf) Estimation Technique (based on Dumoulin & Wandell, 2008) Mapping specialised areas in individual brains is a prerequisite to investigate internal functional organisation using sub-millimetre fmri Goebel R (2015). Functional organization of primary visual cortex. In: A. Toga (Ed). Brain Mapping: An Encyclopedic Reference.

8 Investigating the Brain at Multiple Levels of Organization Conventional (<= 3T) MRI Ultra-high field (UHF) (>= 7T) MRI Courtesy Alard Roebroeck

9 De Martino, Moerel, van de Moortele, Ugurbil, Goebel, Yacoub, Formisano (2013). Spatial organization of frequency preference and selectivity in the human inferior colliculus. Nature Communications, 4, Tonotopic Maps in the Inferior 7 T UHF fmri reveals a feature map in a small structure that shows only unspecific response at conventional resolution

10 Features at Mesoscopic Scale: Neurons and Columns Prime example: Orientation selectivity in primary visual cortex Responses of most V1 neurons are sharply tuned for the orientation of a stimulus in a small region of the visual field Discovered by Hubel & Wiesel during microelectrode recordings (Nobel Prize 1981) They observed that neurons with similar response preference cluster in cortical columns This was later more clearly revealed by studies using optical imaging showing that orientation selective V1 neurons are organized into a topographic map of orientation preference => Resolution of columns sufficient to get (coarse) measure of features! Layers

11 Columnar- And Laminar-Level Imaging with fmri at 7 + Tesla When a Quantitative Improvement of Spatial Resolution Turns into a Qualitative Change Individual neurons code features but they are too small to be detected with highresolution human fmri. If neurons would be distributed randomly, ultrahigh field imaging would provide no qualitative improvement. If neurons cluster into functional units, we might be able to reveal fine-grained neuron-like representations at the columnar level. There is indeed substantial evidence that many areas of the cortex are organized in vertically extending columns that contain neurons with rather similar response profiles. Specialized brain area Neurons (feature detectors) 3T voxel size 7T + voxel size Pial surface Layers White/gray matter boundary column size: ~ mm

12 High-Resolution fmri Reveals Orientation Columns in V1 Human - fmri (SE, 7T) Monkey - Optical Imaging Yacoub, Harel, Ugurbil (2008) Proc Natl Acad Sci USA, 105, Mapping of the (larger) ocular dominance columns had already been reported earlier (e.g. Cheng et al., 2001; Goodyear and Menon, 2001; Yacoub et al., 2007) but this spin echo (SE) EPI study was the first study revealing detailed maps of the much smaller orientation columns!

13 High-Resolution fmri Reveals Orientation Columns in V1 High-resolution UHF SE fmri provides the unique opportunity to investigate these basic computational units in the human brain. Columns have been imaged non-invasively in the human primary visual cortex (V1) located within flat calcarine sulci in selected subjects. A single thick slice with high in-plane resolution (0.5 mm) were prescribed to anatomically identified calcarine sulcus due to limitations of inner-volume SE-EPI. 3 cm Yacoub, Harel, Ugurbil (2008) Proc Natl Acad Sci USA, 105,

14 Data Analysis Strategies: To Pool or Not To Pool? Pial surface Lamina Profiles Information (MVPA) No explicit columnar features White/gray matter boundary No pooling - Every voxel interpreted! Mapping columnar features No Layer Separation Mapping columns across depth High signal to noise required Kemper, De Martino, Emmerling, Yacoub, Goebel (2017). High-resolution data analysis strategies for mesoscale human functional MRI at 7 and 9.4 Tesla. Neuroimage.

15 Columnar-Level Features At Different Relative Cortical Depth Levels The classical model of a cortical column assumes a nearly perfect vertical penetration through the cortex. It has been, however, shown that in areas of monkey IT cortex columns do show strong irregularities across different layers (e.g. Keiji Tanaka, 2011) To reveal how feature codes change across cortical laminae, it is important to map the topography of features at different relative cortical depth levels within specialized areas. We developed two methods to sample topographic information at different cortical depth levels: 1) based on reconstructed cortex meshes (see also Polimeni et al., 2010), and 2) with a novel regular-grid sampling technique. Ideal columnar organization Pial surface More realistic columnar organization White/gray matter boundary

16 The Cortex Modelled as a 3D Structure (7+ Tesla)

17 Isotropic High-Resolution Scans and Grids for Advanced Analysis High-resolution cortical depth analyses using standard mesh approach vs Cartesian grids Kemper, De Martino, Emmerling, Yacoub, Goebel (2017). High-resolution data analysis strategies for mesoscale human functional MRI at 7 and 9.4 Tesla. Neuroimage.

18 Layer Sampling - The Book Principle Bok (1929). A cortical cross section depicting six cytoarchitectonic layers. The volume fraction of a segment is constant across the whole layer. This is possible because the thickness of the layer changes to compensate the curvature.

19 Isotropic High-Resolution Scans and Grids for Advanced Analysis 2D regular grid sampling using equi-volume model (Bok, 1925, Waehnert et al., 2014) Layers (voxels between depth grids) can be filled in voxel space Kemper, De Martino, Emmerling, Yacoub, Goebel (2017). High-resolution data analysis strategies for mesoscale human functional MRI at 7 and 9.4 Tesla. Neuroimage.

20 Isotropic High-Resolution Scans and Grids for Advanced Analysis Kemper, De Martino, Emmerling, Yacoub, Goebel (2017). High-resolution data analysis strategies for mesoscale human functional MRI at 7 and 9.4 Tesla. Neuroimage.

21 High-Resolution 2D Grid Sampling At Multiple Cortical Depth Levels Precisely aligned functional data is sampled at each 2D grid point Sampled data can be directly visualized in 2D grid space and as 3D volumes A 2D grid coordinate refers to the same vertical unit across levels ( column ) Distance, area and volume values can be easily calculated 9 depth grids

22 Localizing hmt + /V5 at 3T: Moving vs Static Contrast Can we map features in hmt with 7T fmri? V5 Flowfield Stationary dots V1/V2 05/30/12 V5 V5 V1/V2

23 Mapping Axis-of-Motion Columns in hmt/v5 at Different Cortical Depth Levels using High-Resolution Grid Sampling Zimmermann, Goebel, De Martino, Adriani, Van de Moortele, Feinberg, Chaimov, Shmuel, Ugurbil, Yacoub (2011). PLoS One, 6(12), e28716.

24 Measuring Binocular Disparity Tuning in hmt Background Emmerling, Frost, Goebel (2016). work in progress.

25 Two Joint Maps of Disparity and DoM Tuning in monkey MT General Background Disparity tuning has been shown in Macaque Area MT (DeAngelis & Newsome 1999) Emmerling, Frost, Goebel (2016). work in progress.

26 Mesoscopic Disparity and AoM Tuning in hmt Design Stimulation Emmerling, Frost, Goebel (2016). work in progress.

27 Mesoscopic Disparity Tuning in hmt Results Disp -2 Disp -1 Disp 1 Disp Emmerling, Frost, Goebel (2016). work in progress.

28 Simultaneous Disparity and Motion Tuning in hmt Mapping Results Disparity conditions Direction conditions right left up down Emmerling, Frost, Goebel (2016). work in progress.

29 Color and Disparity Selective Columns in V2 / V3 Repeating columnar stripes have been observed in histological variations of cytochrome oxidase (CO) levels. Thin and thick stripes of dark CO staining reportedly respond selectively to stimulus variations in color and binocular disparity. Relatively large color-selective stripes could be revealed with GE EPI with a resolution of 1 mm x 1 mm x 1 mm Nasr, Polimeni, Tootell (2016). JNS, 36,

30 Color and Disparity Selective Columns in V2 / V3 Nasr, Polimeni, Tootell (2016). JNS, 36,

31 Columnar Coding in Primary Auditory Cortex De Martino, Moerel, Ugurbil, Yacoub, Goebel, Formisano (2015). PNAS, 112,

32 The Challenge Ahead: Unraveling Unknown Feature Codes Columnar-level features in LOC, VWFA, FFA.. Distributed coding of shapes across columns in monkey IT (e.g. Tanaka, 1996) Goal: Unraveling columnar-level feature representations in mid-level and higher-level areas of the visual hierarchy. In V1-V3, A1, V5 features were known! It is a much more challenging task to map features that are hitherto unknown. Cracking the columnar-level code involves not only high-end technology (7T +, GRASE) but also smart experimentation. Areas that are targeted in our lab: LOC, letter area, VWFA, OFA, FFA, and invariance transformation across areas of the ventral stream. After establishing columnar-level fmri, applications targeting changes of columnar organisations during development and learning can be investigated.

33 Going Beyond Feature Mapping: Using Mesoscopic 7T + fmri to Study Human Cognition Being able to separate fmri responses from different columnar-level features and cortical layers opens the possibility to relate cognitive phenomena like attention, expectation, working memory, imagery and awareness to the human mesoscopic scale for the first time providing substantially increased explanatory power for testing and creating detailed cognitive theories of the mind.

34 Towards Mesoscopic Neural Correlates of Consciousness Ambiguous Motion Quartett Stimulus with two squares alternating same positions: Horizontal / vertical motion happens in the brain (apparent motion) New model-based motion localiser for hmt, see Schneider, Marquardt, De Martino, Goebel, poster #2183 this meeting.

35 Raw Preprocessed Data - Horizontal vs Vertical Motion Goebel R, Schneider M, Ugurbil K, De Martino F, Yacoub E.(2017). in preparation.

36 Mesoscopic Neural Correlates of Consciousness Goebel R, Schneider M, Ugurbil K, De Martino F, Yacoub E.(2017). in preparation.

37 Mesoscopic Neural Correlates of Consciousness Goebel R, Schneider M, Ugurbil K, De Martino F, Yacoub E.(2017). in preparation.

38 Lamina-Specific Functional 7T+ Sub-millimeter fmri can reveal a coarse representation of cortical laminae allowing to investigate layer-specific interactions between brain areas. Importantly, lamina profile measurements might help to separate bottomup from top-down information flow. adapted from Bastos et al. (2012) Furthermore, imaging laminar profiles opens new avenues to study spatial and feature-based attention effects with the prediction that supragranular layers exhibit increased activation during top-down attention and imagery. Laminar imaging may also help to test predictive coding theories.

39 Layer-Specific Attention 7T in Auditory Cortex De Martino, Moerel, Ugurbil, Yacoub, Goebel, Formisano (2015). PNAS, 112,

40 Kanizsa Illusion Selective Activation of Deep Layers in V1 Kok, Norris, De Lange (2016). Current Biology, 26,

41 Kanizsa Illusion Selective Activation of Deep Layers in V1 Kok, Norris, De Lange (2016). Current Biology, 26,

42 Context Decoding in V1 - Cross-Condition Generalization (SEM) n=6 Smith F & Muckli, L (2010). PNAS.

43 Layer-Specific MVPC 7 T Subj.1 V1 Classifier)performance)(%)) 80) 60) 40) 20) WM Average) GM 90) 74) 58) 42) 16) 10) 33%) Cor;cal)depth)(%)) Blood vessels excluded Subj. 2 S1)FF) S2) S3) S4) s)<)0.05) S1)FB) S2) S3) S4) s)<)0.05) WM GM Muckli L, De Martino F, Vizioli L, Petro LS, Smith SW, Ugurbil K, Goebel R, Yacoub E (2015). Current Biology.

44 Selected Research Techniques for in-vivo Human and Animal Neuroscience Human 3T (GE EPI) Human 7T + (SE EPI, GRASE) Human EEG/MEG Spatial resolution Temporal resolution Coverage mesoscopic level of columns/layers - but lamina profiles + meso whole-brain difficult to achieve Non-Invasive Macaque optical imaging Macaque electrode recordings Mesoscopic-level UHF fmri creates a bridge between human and animal research, especially to invasive optical imaging. After establishing mesoscopic spatial resolution, UHF fmri, research can be extended to more human-specific cortical areas. Despite laminar profiles, temporal resolution remains a severe limitation of fmri.

45 Comparing field dependent human fmri with electrophysiology and computational modeling of neural networks De Martino et al. (2017). The impact of ultra-high field MRI on cognitive and computational neuroimaging. Neuroimage.

46 Summary and Conclusions To better understand brain anatomy and function, multiple levels of brain organization need to be integrated. Sub-millimeter ultra-high field (f)mri is an important new tool to bridge macro- and mesoscopic scales as well as human and animal research. Recent fmri experiments show that it is possible to map known columnar-level representations in specialised brain areas (V1, hmt) using ultra-high field fmri and spin-echo based MR pulse sequences. It remains a challenge to crack the functional code for areas where the alphabet of features is hitherto unknown such as the face areas. Revealing (columnar) feature codes in specialized brain areas at mesoscopic scale has the potential to provide important new insights in the neural substrate of human perception and cognition. Feature codes and laminar profiles in multiple brain areas will lead to a deeper understanding of how visual perception and cognition emerge from feature representations and their interactions in the brain.

47 Marian Schneider (UM, NL) Jan Zimmermann (NYU, US) Federico de Martino (UM, NL) Valentin Kemper (UM, NL) Mario Senden (UM, NL) Thomas Emmerling (UM, NL) Elia Formisano (UM, NL) Alard Roebroeck (UM, NL) Francesco Gentile (NIN, NL) Nienke van Atteveldt (VU, NL) Essa Yacoub (CMRR, USA) Kamil Ugurbil (CMRR, USA) van de Moortele (CMRR, USA) Gregor Adriany (CMRR, USA) Lars Muckli (Glasgow, UK) David Feinberg (Berkeley, USA) Miguel Castelo-Branco (Coimbra, PT) Joel Reithler (UM, NL) Judith Peters (UM, NL) Fabrizio Esposito (U Salerno, IT) Martin Frost (UM, NL) Acknowledgements