Qu’est-ce que l’IRM cérébrale A quoi sert-elle, Quelles sont ses limitations? Oury Monchi, Ph.D. Centre de Recherche, Institut Universitaire de Gériatrie de Montréal & Université de Montréal 1

Plan du cours 12 conférences de 3 heures 5 travaux pratiques de 3 heures devant ordinateurs 1 examen devant ordinateur (30%) 1 examen écrit (70%) (3 crédit)

Conférences I A quoi sert l’IRM cérébrale? (maintenant!) 2. Introduction aux contrastes d’IRM. Dr. Rick Hoge (16 janvier) 3. Reconstructions d’images. Dr. Rick Hoge (23 janvier) 4. Devis expérimentaux, hypothèses, software de présentation (30 janvier) 6. Introduction à l’analyse de l’IRMf et aux méthodes de manipulation des images. (13 février) 5. Vérifications des données, Prétraitement (6 février)

Conférences II Mise en œuvre des analyses et applications. avec Mathieu Desrosier) (20 février) Normalisation des données. (27 fevrier) Méthodes d’IRM anatomique : VBM, DTI et MT. Dr. Thomas Jubault (12 mars) Connectivité fonctionnelle et anatomique. Dr. Keith Worsley (19 mars) Examples d’applications en neurosciences cognitives (26 Mars) Session No 12 : Fusions de données et applications cliniques. Dr. Claude Kaufmann (2 Avril) Examen final écrit (70%): Mercredi 16 avril de 13.30 à 16.30

Ateliers informatiques (Thomas Jubault, Ph. D et Claudine Gauthier, M Ateliers informatiques (Thomas Jubault, Ph.D et Claudine Gauthier, M.Sc.) 7 février : Vérification des données et pré-traitement 14 février: Modèle Linéaire Générale 21 février: Moyennage et Normalisation 13 mars: Visualisations et tables de résultats 20 mars: Introduction à d’autres méthodes 27 mars: Examen pratique (données à analyser)

Techniques d’IRM I. Imagerie par Résonance Magnétique (IRM) A. Etudes anatomiques B. Etudes Fonctionnelles C. Etudes Physiologiques

Principes de base de l’IRM Aimant: Champ magnétique très puissant (1 à 7T) et homogène qui va inciter les protons d’hydrogène à s’aligner. Champ magnétique de la terre 0.00005T! Bobine de radiofréquence: envoie une impulsion à la fréquence de résonance de l’hydrogène. Après être entrer en état de résonance ces protons vont revenir à leur état de base à une vitesse différente suivant le tissue dans lequel il se trouve. Ceci générera un contraste de type T1 Bobine de gradients: le signal généré par la RF ne nous donnent pas d’information spatiale en temps que tel, ce sont les bobines de gradients alignées sur trois axes (x, y, z) qui nous permettent de le faire.

IRM: Principes de Base Spins des protons d’Hydrogène

IRM: Principes de Base Spins des protons dans le champ statique B0

IRM: Principes de Base Effets de radiofréquences en résonance

IRM: Principes de Base Combinaison de radiofréquences et gradients = localisation spatiale des coupes de l’objet

IRM: Principes de Base Temps de relaxation des spins (T1 et T2

IRM: Principes de Base Temps de relaxation de T1 et T2

IRM: Principes de Base

IRM: Principes de Base Gradients X, Y, Z, Shim.

IRM: Principes de Base Sécurité!!!!

Research programs: Innovations R N Q /U N F Research programs: Innovations MRI Methods

Anatomical MRI (T1)

Voxel Based Morphometry Voxel based morphometry (VBM) is a neuroimaging analysis technique that allows investigation of focal differences in brain volume. Traditionally, brain volume is measured by drawing regions of interest (ROIs) and calculating the volume enclosed. However, this is time consuming and can only provide measures of large areas. Smaller differences in volume may be overlooked. VBM registers every brain to a template, which gets rid of most of the large differences in brain anatomy among people. Then the brain images are smoothed so that each voxel represents the average of itself and its neighbors. Finally, volume is compared across brains at every voxel.

Voxel Based Morphometry Brenneis et al., 2004 JNNP

Anatomy: vascular studies

Anatomie: Tenseurs de diffusion Étude de la connectivite anatomique a b c f e g

Principes de base de l’IRMf On connait une relation entre l’activité cérébale et le taux d’hémoglobine désoxygéné dans le sang Début des années 90 il a été découvert qu’une séquence d’impulsions produites par l’IRM pourrait mesurer le taux d’hémoglobine désoxygéné (Thulborn et al.; Ogawa et al.) Ceci a donné naissance au Blood Oxygenation Level Dependent (BOLD) fMRI ou T2* qui nous donne une mesure indirecte de l’activité cérébrale.

Experimental Design Blocked vs. event-related Source: Buckner 1998

Experimental design Block design compare long periods (e.g., 16 sec) of one condition with long periods of another traditional approach most statistically powerful approach less dependent on how well you model the hemodynamic response Event-related design compare brief trials (e.g., 1 sec) of one condition with brief of another very new (since ~1997) approach less statistically powerful but has many advantages trials can either be well-spaced to allow the MR signal to return to baseline between trials (e.g., 12+ seconds between trials) or closely spaced (e.g., every 2 sec)

Preprocessing

Modeling the expected response (fmridesign)

Modeling the data (GLM) (From Dr. J. Armony)

(From Dr. J. Armony)

Connectivité fonctionnelle et effective

IRMf: Principes de Base Variations de la réponse hémodynamique

Toni et al. (2002) Cerebral Cortex

Physiological Studies: Spectroscopy

Physiological Studies: Spectroscopy

Imagerie Optique La technique est basé sur l’émission d’un faisceau lumineux dans le cerveau à des fréquences proches de l’infra-rouge L’absorption de ce faisceau nous donne de l’information sur l’oxygénation et la désoxygénation du sang similairement à l’IRMf. La diffusion de ce signal nous donnent de l’information spatiale

Imagerie optique

Imagerie optique: activation Moteur Langage

Imagerie optique: Épilepsie Diagnostic de l’épilepsie

Science sans conscience n’est que ruine de l’âme! (Francois Rabelais) Une éxperience sans question ou hypothèse ne sert pas à grand chose et peut être couteuse! L’important c’est la question, si l’IRMf peut y répondre, il faut savoir faire des dessins éxpérimentaux les plus appropriés

Prof. Martha Farah, Interview in JOCN

Prof. Martha Farah, Interview in JOCN

Prof. Martha Farah, Interview in JOCN

Prof. Martha Farah, Interview in JOCN

IRMf chez différentes populations Attention: Différences d’activité neuronal, ou différences dans le métabolisme de désoxyhémoglobine Heureusement certains chercheurs essaient de répondre à cette question, p.ex. Dr. Rick Hoge au CRIUGM

Voici l’aire de l’éternuement! Mauvaise étude 1! X = 12 2.5 5 T-stat Caudate Voici l’aire de l’éternuement!

functional MRI: Voice recognition Belin, et al. (2000) Nature

Voici le réseau de l’attention! Mauvaise étude 2! Voici le réseau de l’attention!

sélection ou exécution d’une nouvelle action Théorie proposée Niveaux Monitoring/ association Comparaison/ Sélection Association stimulus/action Organisation dans la mémoire de travail CORTEX Préfrontal Dorsal 9, 46 Ventral 47/12 Postérieur int 6, 8, 44 Planification, sélection ou exécution d’une nouvelle action Caudé dorsal Caudé ventral STRIATUM Putamen

Montreal Card Sorting Task, Étude I Retrieval w/o shift Cue card Retrieval w/ shift vs Prédictions avec changement: CPF-VL+ Noyau caudé sans changement: CPF-VL, PAS de striatum Monchi et al., Ann. Neurol., 2006

Montreal Card Sorting Task Changement de règle continu Matching according to colour Matching according to number Condition contrôle Prédictions Changement de règle continu: CPF-VL, PAS de striatum

IRMf MCST: Contrôles en santé VL-PFC No striatum Cue Card VS Retrieval NO shift Control VL-PFC Caudate 3 7 T-stat Cue Card Retrieval WITH shift Control VS Significant VLPFC occurs in all active conditions vs control VL-PFC No striatum Continuous shift Control VS X = 18 Y = -4 Monchi et al. Feb 2006, Annals of Neurology

fMRI MCST: Healthy Controls 3 5 T-stat X = 12 Caudate Putamen Cue Card Cue Card VS Retrieval NO shift Retrieval WITH shift X = 12 2.5 5 T-stat Caudate Cue Card VS In order to further separate out the role of the caudate nucleus we also performed the following subtractions Which implies ?? That is not shif per se but the planning of the novel action that is important. Retrieval WITH shift Continuous shift Le noyau caudé n’est pas particulièrement impliqué dans le changement de règle en soi, mais dans la planification d’une nouvelle action. Monchi et al. Feb 2006, Annals of Neurology

Paramètres Nécessaires lors de la Publication d’Articles en IRMf

Experimental Design Design specification Number of blocks, trials or experimental units per session and/or subject. Length of each trial, and interval between trials If variable interval, report the mean and range of ISIs and how they were distributed. Blocked designs: Number of blocks Length of blocks For event-related fMRI, was efficiency optimization used, and if so, how? For mixed designs, report correlation between block and event regressors

Experimental Design Task specification What were subjects asked to do? What were the stimuli? Did specific stimuli repeat across trials?

Human subjects Details on subject sample Number of subjects Age (mean and range) Handedness Number of males/female Additional inclusion/exclusion criteria, if any If any subjects were scanned but then rejected from analysis after data collection, state how many and reasons for rejection For group comparisons, what variables were equated across groups Ethics approval state which IRB approved the protocol

Data Acquisition Image properties - As acquired MRI system: Manufacturer, field strength (in Tesla), model name MRI acquisition: Number of experimental sessions and volumes acquired per session Pulse sequence type (gradient/spin echo, EPI/spiral) Field of view, matrix size, slice thickness, interslice skip Acquisition orientation (axial, sagittal, coronal, oblique; if axials co-planar w/ AC-PC, the volume coverage in terms of Z in mm) Whole brain? if not, state area of acquisition Order of acquisition of slices (sequential or interleaved) TE/TR/flip angle

Data Acquisition Pre-processing: General Specify order of preprocessing operations Slice-timing correction minimally, software version; ideally, order and type of interpolation used and reference slice Motion correction software version (major and minor version numbers) Interpolation method (also ideally, image similarity metric and optimization method) Motion-susceptibility correction?

Data Acquisition Pre-processing: Intersubject registration Intersubject registration method used. Software version Transformation model Linear - Number of parameters Nonlinear - Nature of deformation and number of parameters (E.g. in AIR, a polynomial order is specified; in SPM, a DCT basis size is specified, 3x2x3). - Non-linear regularization? (E.g. in SPM, e.g. "a little"). Crucial for fluid-deformation methods Interpolation method

Data Acquisition Pre-processing: Intersubject registration… Object Image information. (Image used to determine transformation to atlas) Anatomical MRI? Image properties (see above). co-planar with functional acquisition? Functional acquisition co-registered to anatomical? if so, how? Segmented grey image? Functional image (single or mean)

Data Acquisition Pre-processing: Intersubject registration… Atlas/target information Brain image template space, name, modality and resolution. (E.g. "SPM2's MNI, T1 2x2x2"; "SPM2's MNI Gray Matter template 2x2x2") Coordinate space? Typically MNI, Talairach, or MNI converted to Talairach If MNI converted to Talairach, what method? E.g. Brett's mni2tal? How were anatomical locations (e.g. Brodmann areas) determined? (e.g. paper atlas, Talairach Daemon, manual inspection of individuals' anatomy, etc.)

Data Acquisition Pre-processing: Smoothing What size smoothing kernel? What type of kernel (especially if non-Gaussian, or adaptive). Is smoothing done separately at 1st and 2nd levels?

Statistical Modeling General issues For novel methods that are not described in detail in a separate paper, provide explicit description of method either in the text or as an appendix Intrasubject fMRI Modeling Info Statistical model and software version used (e.g. Multiple regression model fit with SPM2, updates as of xx/xx/xx; or FSL release 3.3). Block or event-related model Hemodynamic response function (HRF) assumed or estimated? If HRF used, which (e.g. SPM's canonical dual-gamma HRF; SPM's gamma basis; Gamma HRF of Glover).

Statistical Modeling Intrasubject fMRI Modeling Info… Additional regressors used (e.g. motion, behavioral covariates) Drift modeling (e.g. DCT with cut off of X seconds; cubic polynomial) Autocorrelation modeling (e.g. for SPM2, 'Approximate AR(1) autocorrelation estimated at omnibus F-significant voxels (P<0.001), then pooled over whole brain'; for FSL, 'Regularized autocorrelation function estimated at each voxel').

Statistical Modeling Intrasubject fMRI Modeling Info… Estimation method: OLS, OLS with variance-correction (G-G correction or equivalent), or whitening. Tom Nichols: Is this too hard core? It's what I want to know, but I guess you could argue that given enough detail about the software it could be inferred. Contrast construction. Exactly what terms are subtracted from what. It might be useful to always define abstract names (e.g. AUDSTIM, VISSTIM) instead of underlying psychological concepts.

Statistical Modeling 2-level, modality-generic Modeling Info Statistical model and software version used (e.g. 1-sample t on intrasubject contrast data, SPM2 with updates as of xx/xx/xx). Whether first level intersubject variances are assumed to be homogeneous (SPM & simple summary stat methods: yes; FSL: no). If multiple measurements per subject, method to account for within subject correlation. (e.g. SPM: 'Within-subject variance-covariance matrix estimated at F-significant voxels (P<0.001), then pooled over whole brain'). Jesper Andersson request: Variance correction corresponding to within-subject variance-covariance matrix, so simply some measure of nonsphericity.

Statistical Modeling 3rd-level group difference modeling info Statistical model and software version used (if different from 1/2) (e.g. 2-sample unpaired t on contrast images).

Statistical Inference Inference on Statistic Image (thresholding) Type of search region considered, and the volume in voxels or CC. If not whole brain, how region was found; method for constructing region should be independent of present statistic image. If threshold used for inference and threshold used for visualization in figures is different, clearly state so and list each. All inferences must explicitly state if they are corrected for multiple comparisons, and if so, what method and over what region. If correction is limited to a small volume, the method for selecting the region should be stated explicitly. If no formal multiple comparisons method is used, the inference must be explicitly labeled "uncorrected".

Statistical Inference Inference on Statistic Image (thresholding)… There was some disagreement over this topic in the discussion group. While most of the discussants felt that it should be acceptable to report uncorrected statistics in some cases, there was also widespread feeling that current reporting standards are too loose. The following comments provide a view of the range of opinions: Keith Worsley: I've always advocated doing both i.e. labelling the 'confirmed' ones with a corrected P-value < 0.05, and reporting anything else as 'unconfirmed' or speculative or suggestive etc. (like PCA/ICA - but we're working on this!) - as long as it's clear, I don't see that it matters too much

Statistical Inference Inference on Statistic Image (thresholding)… Tom Nichols: While requiring some sort of correction, as Nature Neuroscience does, is a very practical stance, it seems to temp a fishing for a correction (e.g. find the right SVC until you have significance). Following on Keith's comment, I think it could be best to simply require: all inferences must explicitly state if they are corrected for multiple comparisons, and if so, what method and over what region. If no formal multiple comparisons method is used, the inference must be clearly labeled "uncorrected". This hopefully would put implicit pressure on people to use corrected methods, but if someone wants to report "p<0.001" uncorrected, or, even as Matthew suggested, p<0.01, so be it, if they can convince the reviewers that it's compelling. As long as inferences are clearly labeled, I think people can do what they want. (Liberal inferences will be seen as weak evidence, and that's that.)

Statistical Inference Inference on Statistic Image (thresholding)… Matthew Brett: I absolutely agree we need the freedom to report effects that are weak. For me the point is that using an uncorrected p value as an index of this effect is bad statistics. The uncorrected p value is meaningless as an attempt at some sort of type 1 error control, depending as it does on a large number of factors including smoothing, df, field of view and so on. You p<0.001 is not comparable to someone else's p<0.001. The uncorrected p value then gives a spurious impression of statistical rigor. If an effect is weak, and you want to state that it is close to being significant, then the two options that seem sensible to me are 1) drop the corrected threshold to - say - 0.1, or 2) show the effect size map and argue your case. To me it is unfortunate that lots of young researchers in FMRI seem to believe that uncorrected p<0.001 is in some magical sense 'sort of significant' - and it's such an unreliable rule of thumb, and so confusing for people entering the field, that we should really try and move away from that.

Statistical Inference ROI Analysis How were ROI's defined (e.g., functional versus anatomical localizer)? How was signal extracted within ROI?

Pour obtenir les diapos: Remerciements Jorge Armony, Ph.D. Claudine Gauthier, M.Sc. Rick Hoge, Ph.D. Julien Doyon, Ph.D. Pour obtenir les diapos: http://unfweb.criugm.qc.ca/oury/downloads.html

Référence IRMf recommendée Sinauer Associates Publisher