Applied Research Methods: BC Functional MRI PDF

Eva Berlot [email protected] Applied Research Methods: BC Functional MRI Contents 1. What is (f)MRI: spinning protons → maps of brain activity Underlying physics and neurophysiology 2. The fMRI experiment: research question → experiment Experimental des...

Eva Berlot [email protected] Applied Research Methods: BC Functional MRI Contents 1. What is (f)MRI: spinning protons → maps of brain activity Underlying physics and neurophysiology 2. The fMRI experiment: research question → experiment Experimental designs (and their shortcomings) 3. Interpreting fMRI data: brain activity → shiny images Data analysis: possibilities and pitfalls [many slides by Marius Peelen, Surya Gayet, Jody Culham] MRI vs fMRI MRI studies brain Functional MRI (fMRI) anatomy studies brain function fMRI Quizz 1. fMRI has a particularly good … ~ 1-3 sec A. Temporal resolution B. Spatial resolution C. Both ~ 1-3 mm D. None of the above 2. Which concept is unrelated to fMRI? A. Blood oxygen-level dependent contrast B. Hemodynamic response function C. Event-related potentials D. General linear model EEG MRI An MRI examination described in a few steps: 1. The participant is placed in a magnet, 2. A brief radio wave (pulse) is sent in (at resonance frequency), 3. The participant emits a signal, which is measured … 4. to form an image of the tissue emitting the signal. Magnetic Resonance Imaging: Scanner 1 1. Static magnetic field 2. Radiofrequency coil 3 3. Gradient coils 2  Physics of MRI Scanner is large electromagnetic coil, made superconductive by cooling (liquid helium, -261 ºC) Very strong static magnetic field (e.g., 3 Tesla, earth = 0.0001 T) !!!  Physics of MRI Protons can be found The proton spins A proton behaves like in the nuclei of atoms around its axis. a tiny bar magnet with (e.g., hydrogen), and A moving electric north and south poles. have a positive charge (or current) electric charge. produces a small magnetic field.  Most human tissue is water-based, amount of water in each type of tissue varies  Physics of MRI longitudinal axis z Longitudinal Magnetic field magnetization Mz y Transverse transverse magnetization plane Mxy x Hydrogen protons align with the magnetic field (protons drawn obviously not to scale!)  Physics of MRI longitudinal axis z Longitudinal magnetization Mz Magnetic field y Transverse transverse magnetization plane Mxy x Apply radiofrequency pulse! 90 RF Pulse  Physics of MRI Video schematic Measure during recovery period longitudinal axis Longitudinal z magnetization z z Mz y y y Transverse transverse magnetization plane Mxy x x x Before Immediately after Long after 90° pulse 90° pulse 90° pulse Measure radio waves as protons gradually return to original configuration within the magnetic field Physics of MRI Lower intensity Higher intensity T1: time needed to return to the aligned state (longitudinal relaxation). Differs for grey matter, white matter, cerebrospinal fluid. T2: measures how quickly the protons give off energy as they recover to equilibrium (transverse relaxation) Physics of MRI We can measure human brains Lower intensity Higher intensity T1: time needed to return to the aligned state (longitudinal relaxation). Differs for grey matter, white matter, cerebrospinal fluid. We can measure lots of other things too Andy Ellison / Inside images MRI vs. fMRI MRI studies brain Functional MRI (fMRI) anatomy studies brain function Terminology of fMRI in research papers subjects sessions runs volume slices voxels Functional imaging: measuring brain activity Brain uses 20% of the total oxygen consumption of the body When neural activity increases, blood flow to these neurons increases to meet the metabolic demands (i.e. oxygen supply). fMRI is sensitive to the amount of oxygen in the blood. Functional MRI (fMRI) studies brain function  Functional imaging: measuring brain activity Network of blood vessels carries glucose and oxygen. Hemoglobin (protein) brings glucose and oxygen to neurons. Hemoglobin without oxygen: deoxyhemoglobin. Deoxyhemoglobin distorts local magnetic fields. Measure the distortion in the magnetic field → concentration of deoxyhemoglobin → indirect measure of brain activity.  Functional imaging: measuring brain activity So when brain area is active, we have: Increased neuronal activity Increased oxygen metabolism (DOH goes up) But why does fMRI activity not show a DECREASE then? Because of over-compensation of blood flow, delivering oxygen Blood Oxygenation Level-Dependent (BOLD) signal 3 (% signal change) BOLD Response 2 1 0 Time Stimulus Blood Oxygenation Level-Dependent (BOLD) signal 3 (% signal change) BOLD Response 2 1 Initial Dip 0 Time Stimulus Blood Oxygenation Level-Dependent (BOLD) signal Positive BOLD response 3 (% signal change) BOLD Response 2 Overshoot Post-stimulus 1 Initial Undershoot Dip 0 Time Stimulus  Blood Oxygenation Level-Dependent (BOLD) signal Change in BOLD over time: Hemodynamic Response Function (HRF) Responsiveness is rather slow: 1 - 4 seconds  Blood Oxygenation Level-Dependent (BOLD) signal BOLD is recorded in slices, acquired repeatedly. fMRI BOLD signal slices in one TR (Time to Repeat)  Anatomical MRI (T1) vs. functional MRI (T2) MRI fMRI low resolution high resolution (1 mm) (2 or 3 mm) One 3D volume “T1-weighted image” Series of 3D volumes (i.e., 4D data) (e.g., every 2 sec, for 5 min) (Liquids are dark, dense solids are white) “T2/T2*-weighted image” (Active regions are white, passive regions are black)  Recap: What is (f)MRI Structural MRI: Brain anatomy Functional MRI: Amount of oxygen in the blood, dynamic Advantages – No radioactivity or X-rays (not harmful) – Very high spatial resolution Disadvantages – Temporal resolution low (slow hemodynamic response) – Certain areas distorted (near cavities) – Very sensitive to motion artefacts (need to lie still) – Narrow tube (claustrophobia) – Screening necessary regarding metal/magnetic field – It’s loud!!! T1/MPRAGE T2/DWI Contents 1. What is (f)MRI: spinning protons → maps of brain activity Underlying physics and neurophysiology 2. The fMRI experiment: research question → experiment Experimental designs (and their shortcomings) 3. Interpreting fMRI data: brain activity → shiny images Data analysis: possibilities and pitfalls  Experimental design: presentation patterns Block design: experimental conditions grouped in alternating blocks.  Experimental design Signal change in 1 voxel time “Blobs” depict the statistical > outcomes of such comparisons for each voxel. <  Experimental design: presentation patterns Block design: experimental conditions grouped in alternating blocks. Sensitive design: strong activation, more statistical power. But no flexibility: only long-lasting, generalized responses. Event-related design: stimuli are presented as individual events. More flexibility, sorting of trials after the experiment is possible (e.g., memory experiment: correctly or incorrectly remembered images). But each event is drowned in other events: less statistical power. Brain always uses oxygen, also at rest Subtraction method Posner & Raichle, Images of Mind Measuring time of mental processes Subtraction method Derive time needed for cognitive processes from reaction times. Key aspect of (behavioral) experiments in the cognitive sciences. F.C. Donders Dutch physiologist ? 1818-1889 Classical example How long does it take your brain to … (1) Distinguish between colors (red / green) (2) Select a motor response (left / right hand)  Measuring time of mental processes Task 1: Simple reaction time Press button when you see a light See light Press button Task 2: Discrimination reaction time Press button when light is green (but not when red) See light See which color Press button Task 3: Choice reaction time Press left button when light is green and right button when light is red See light See which color Select button Press button  Measuring time of mental processes: subtraction logic T2 See light See which color Select button Press button - - T1 See light See which color Press button = = “How much longer Select button does it take to ALSO choose the response?”  Measuring time of mental processes: subtraction logic T2 See light See which color Press button - - T1 See light Press button = = “How much longer See which color does it take to ALSO determine the color?” Key assumption: these processes are INDEPENDENT (they do not interact). Weakness substraction method: assumption of ‘pure insertion’ Assumption: adding a component to a process (task) does not affect the other components of the process. This assumption is often violated. - =? Alternatives subtraction method It’s not always easy/possible to find a PERFECT baseline condition. Factorial design Manipulate multiple factors to avoid the ‘pure insertion’ problems. Example question: why can’t you tickle yourself? H: Tactile sensation is reduced for self-produced touch, because the brain can predict the tactile sensation from the motor command (‘efference copy’). Factorial design From Ward (Chapter 4, p.59) Factorial design From Ward (Chapter 4, p.59) Factorial design Is response to touch greater when produced externally vs when produced by self? Interaction analysis: (C – D) - (A – B) C From Ward (Chapter 4, p.59) Replicated more recently Kilteni & Ehrsson, 2020 Experimental designs Subtraction method Pure insertion. Factorial design Manipulate multiple variables at once. Parametric design Continuous variables instead of categorical differences. Parametric design Associations between brain activity & function, not just differences. Passively listening to speech with varying speeds. Parametric design Associations between brain activity & function, not just differences. Passively listening to speech with varying speeds. Experimental designs Subtraction method Pure insertion. Factorial design Manipulate multiple variables at once. Parametric design Continuous variables instead of categorical differences. Repetition suppression Repeating a stimulus characteristics reduces the neural response.  Repetition suppression Premise: repeated stimulation reduces the neural response. Approach: Present (part of) a stimulus twice. Measure the neural activity. Locate regions that show a reduced response to the second occurrence of the stimulus (characteristic). → TT BOLD change Repetition suppression Brain areas that are sensitive to this repetition are involved in processing the stimulus (characteristics).  Repetition suppression → → → → BOLD change TT TL TT TL ‘T’ area BOLD change BOLD change  Repetition suppression → → → → BOLD change TT TL TT TL ‘T’ area BOLD change blue area What type of information does this brain area respond to? BOLD change A. Letters (T) B. Colors (blue) C. A conjunction between colors and letters (blue T) D. All visual stimulation  Repetition suppression → → → → BOLD change TT TL TT TL ‘T’ area BOLD change blue area BOLD change blue T area Experimental designs Subtraction method Pure insertion. Factorial design Manipulate multiple variables at once. Parametric design Continuous variables instead of categorical differences. Repetition suppression Repeating a stimulus characteristics reduces the neural response. Allows for isolating single functional components. Naturalistic designs Assessing human brain functioning in rich, natural conditions Inter-subject correlation Assessing how similar responses are across individuals Similarity higher for more ‘meaningful’ stimuli in higher-order regions Hasson et al. (2010) Within-subject encoding models Encoding models Huth et al. (2016) Contents 1. What is (f)MRI: spinning protons → maps of brain activity Underlying physics and neurophysiology 2. The fMRI experiment: research question → experiment Experimental designs (and their shortcomings) 3. Interpreting fMRI data: brain activity → shiny images Data analysis: possibilities and pitfalls  Analysis step 1: Data pre-processing What is pre-processing? Minimizing variability in the data that is not related to the task. Oblivious to the stimulus design (hence: pre-processing) Why pre-processing? fMRI data is very noisy. Change in BOLD contrast is ~2%. Correcting for (head) motion, heartbeat, breathing, etc. Correcting for unwanted changes in the magnetic field, scanner drift. Correcting for / taking into account individual differences in brain anatomy. Etc.  Analysis step 1: Data pre-processing Motion Slice timing correction Raw data Smoothing Normalisation Preprocessed (MNI) Normalized data template image brain  Analysis step 1: Data pre-processing Overview Motion correction: correcting for head motion of subjects. Slice timing: correcting for the timing difference in slice acquisition (within one TR). Normalisation: “correcting” individual differences in anatomy by squeezing scans into a standardized brain. Smoothing: increasing sensitivity to clustered activation patterns. Analysis step 2: fMRI statistics, modelling the BOLD response Goal: find out which voxels react to the experimental manipulation. Two-step approach: 1. “First level” (model within subject). General linear model (GLM): Massive regression analysis applied to EACH voxel of EACH subject. 2. “Second level” (test across subjects). Often: ANOVA / t-test between conditions, across participants, for EACH voxel (or brain region).  Predicting the BOLD response Approach: Derive a prediction of the BOLD response by combining.. 1. The shape of the BOLD response to individual events (HRF). 2. The time points of events (e.g., your experimental manipulations) Predicting the BOLD response Auditory cortex Visual cortex Measured Predicted response response Predicting the BOLD response Auditory cortex Visual cortex Measured Predicted response response Visual cortex Auditory cortex Predicted response Measured response  General linear model Build a model with experimental conditions as regressors to predict measured BOLD signal Predicted signal is a linear combination of regressors. Y: Activity in 1 voxel ε  General linear model Build a model with experimental conditions as regressors to predict measured BOLD signal Predicted signal is a linear combination of regressors. Y: Activity in 1 voxel ε General linear model: parameter estimation Parameter estimation: finding the optimal values of betas βn to find the linear combination of regressors that best explain the data. Y: Activity in 1 voxel BAD FIT (large ε) General linear model: parameter estimation Parameter estimation: finding the optimal values of betas βn to find the linear combination of regressors that best explain the data. Y: Activity in 1 voxel GOOD FIT (small ε)  General linear model: statistical inference 1. We found the optimal parameters for each voxel (beta weights of each experimental condition). 2. Combine these voxels into beta-maps: 3D brain images, comprising the beta values of your condition (e.g., ‘house map’, or ‘face map’). 3. Compare beta-maps of different condition with a statistical test, e.g. contrasting [house > face], or [face > house]. Contrast: β1 (house) > β2 (face). P < 0.001  fMRI results: Region of Interest (ROI) analysis Whole brain analyses ROI-based analyses (Exploratory) Average beta weight Based on atlas BA37 BA17  fMRI results: Region of Interest (ROI) analysis Whole brain analyses (f)ROI-based analyses (Exploratory) Based on contrast FFA: < Average beta weight PPA: > V1: ( )> FFA PPA V1 Signal change in 1 voxel time → Quick recap  First need to perform careful preprocessing  1-st level analysis (for each subject): GLM - find the parameter values (beta weights) for each regressor that allow to best describe the observed data (minimize error). Summarize these in beta-maps for each condition of interest.  Second-level analysis (group): Test for systematicity across individuals Statistical tests comparing different experimental conditions (i.e., between beta-weights). In the whole brain, or in regions of interest (ROIs) Contents 1. What is (f)MRI: spinning protons → maps of brain activity Underlying physics and neurophysiology 2. The fMRI experiment: research question → experiment Experimental designs (and their shortcomings) 3. Interpreting fMRI data: brain activity → shiny images Data analysis: possibilities and pitfalls Dead salmon feels emotions Brain area involved in emotional judgement? p(uncorrected)