Hemodynamic neuroimaging Flashcards
What is the history behind hemodynamic neuroimaging?
• Temperature of brain goes up during mental exercise
• Case study: auditory noise made by blood flow in an arteriovenous malformation correlated with effortful visual processing
• Hemodynamic imaging: neuroscience hype since the 1990’s
• Hemodynamic imaging to assess neural activity still not without criticism
What are 3 criticisms of hemodynamic neuroimaging?
• Indirect measure of neural activity
• Complex relationship between neural activity & hemodynamics
• Complex = cannot be trusted?
• Deep understanding allows discriminating (in)valid ways of using it
• Complexity of methods and analysis
What are 3 criticisms of hemodynamic neuroimaging?
• Indirect measure of neural activity
• Complex relationship between neural activity & hemodynamics
• Complex = cannot be trusted?
• Deep understanding allows discriminating (in)valid ways of using it
• Complexity of methods and analysis
What is the relation between hemodynamics and neural activity?
• Neurons require continuous supply of glucose and oxygen to function –> blood circulation
• Blood comes in through arteries and arterioles
• Exchange of glucose and oxygen in capillaries
• Oxygen removed from hemoglobin = deoxyhemoglobin
• To venules and to larger veins to leave the brain
• Energy = adenosine triphosphate (ATP), from glucose
• Neural activity –> hemodynamic response function (HRF)
What happens when neural activity occurs?
• Slightly delayed local increase in oxygen and glucose consumption
• Ratio oxygenated and deoxygenated hemoglobin (blood oxygenation) decreases
• Signal through neurovascular coupling mechanism, triggering increase in supply of blood
• Accompanied by marked increase in blood oxygenation
• Peak increase in blood oxygenation several seconds after initial oxygen consumption
• Blood volume and oxygenation decay again (negative overshoot to below baseline levels)
• Expand across larger territory than region of neural activity
What are the 3 components of HRF?
• Initial dip: Decrease in blood oxygenation and measured signal
• Primary (strongest) response: Influx oxygenated blood –> strong increase in signal
• Negative overshoot: signal decreases
What are 3 factors that influence hemodynamic signal?
• Which process and parameter dominates the measurement
• E.g. Blood volume instead of oxygenation: no initial dip
• Whether measurement very near to site of neuronal activity or average across a larger area
• Across larger area:
• No initial dip
• Only positive peak and negative overshoot
• Additivity assumption: in case of multiple stimuli, total hemodynamic
response (HR) is sum of HRF’s to individual stimuli
What is the relation between HR and electrical potential changes?
• Measure action potentials = measure output of a neuron
• Measure HR of a region, not sure whether it represents overall action potential output of that region
• Situations possible where energy consumption increases, while output of neuron stays the same
• Inhibitory input, energy consumption increases but output decreases or stays the same
What does the HR represent?
• Relation HR and other electrophysiological measures
• Multi-Unit Activity (MUA): number of action potentials
• Local Field Potentials (LFP): synaptic input of neurons (slow changes in post-synaptic membrane potential)
• Typical situation = everything correlates: HR, MUA, LFP
• When partially dissociated, e.g. through long stimulation: HR (in fMRI) slightly more correlated with LFP than with MUA
What is fMRI?
• Blood-oxygenation-level dependent (BOLD) signal
• Deoxygenated hemoglobin: magnetic momentum (paramagnetic)
• Alters spin-spin interactions –> faster T2 decay
• Increase in oxygenation –> increased fMRI signal
• More macroscopic side effects of paramagnetic particles:
• Field inhomogeneity
• Tissue susceptibility
–> Total dephasing = T2* decay
What is the relevance of fMRI?
• Decade of the brain
• Has pinpointed the neural basis of a range of mental processes
• Beyond mere localization of function
• Effects sometimes overestimated
• Group differences often not consistent enough between subjects to allow prediction at individual subject level
What is PET?
• 1980’s: dominant hemodynamic imaging method
• Unique contribution relative to fMRI:
• Measuring metabolism
• Detection of biomarkers and neurotransmitter concentrations
• Positron emission: involves injection of radioactive tracers
• Injection not of isolated isotope, but attached to a molecule with specific biological action
• Molecule and site of injection determines spread of the tracer
• Radionuclides: short half-life = positron emission decay
How does PET work?
• Positron is positively charged –> interacts with negatively charged electron –> annihilation
• Pair of photons travelling in opposite directions
• Detected by photo-sensitive tubes or diodes
• Original position of annihilation localized along a straight line
• 2 such photons have to be detected at the same time (coincidence detection)
• Production of radionuclides near the PET requires a cyclotron
How can we use PET to measure neural activity?
• Oxygen-15
• Short half-life of 2 minutes
• Distribution: linear relationship to incoming blood volume
• Total amount of oxygen in a brain region: indication of local neural activity
• Because of over-supply of oxygenated blood following neural activity
• Typical PET experiment
• Low number of conditions (4-8)
• Conditions are typically tested in blocks of around 1 minute
• Often only 2 blocks per condition
• In between blocks: short waiting period with new injection
What are 2 advantages and 3 disadvantages of PET?
+ Ability to measure blood volume quantitatively
+ Confronted with less unknown parameters when we try to relate measured signal to neural activity
- Injecting radionuclides
• Need for cyclotron
• Health risks of radioactivity - Poorer spatial resolution, about 1 cm
• Combination with (simultaneous) MRI helps to some extent - Poorer temporal resolution: minutes
What are 4 unique contributions of PET?
• Metabolism: tracer fluorine-18 (half live 110 minutes) attached to glucose = fluorodeoxyglucose (FDG)
• Diagnose cancer
• Diagnose brain diseases
• Hypometabolism of temporoparietal region in Alzheimer’s disease
• 11C Pittsburgh B PET for beta-amyloid deposits
• Target specific neurotransmitter system
• Tracer attached to molecule with concentration related to activity of one specific neurotransmitter
• Dopamine: 6-(18F)-fluoro-L-DOPA
What should we do before we start an experiment?
Think before you start an experiment
• Hard work
• Is it worth the effort?
• Know the methods and the theories
• Formulate relevant hypotheses
• Design study to provide evidence for or against some hypothesis
• Expensive machines do not think for you
How do we determine which conditions to include in our experiment?
• Subtraction method: comparison of two conditions that only differ in one mental process of interest
• Related to behavioural method of Donders: mental chronometry
• fMRI: subtract brain activation
• Multiple extensions:
• Multiple conditions, compared pairwise
• Factorial design
• Parametric design
What is the assumption of additivity?
• Also known as pure insertion
• The addition of a new mental process does not affect other processes
• If not correct, then comparison between conditions is confounded
• Empirically, violations of this assumption can be visible through interaction effects in the case of a factorial design
What are 3 ways to present the conditions?
• HRF peak delayed with 6s, total duration more than 12s from onset before at baseline again –> overlapping HRFs
• Option A: ISI of 16s; –> Inefficient!
• Option B: ISI of 2s; signal reaches asymptotic level –> low sensitivity for differences between conditions
• Option C: Block design: trials are blocked per condition
What is the block design?
• Alternate blocks of different conditions
• Within blocks: strong HRF because of additivity signal
• After a block: signal goes down again (possibly to baseline) before next block
• Condition-associated ups and downs in BOLD signal are large
• Large power and sensitivity to detect changes in BOLD signal
• Very efficient
• Many trials can be presented per unit of time
• No long waiting time between trials
What are 3 drawbacks of the block design?
• Predictability of conditions (tendency to prepare)
• Unwanted cognitive processes could confound the one process assumed to differ between conditions
• Can be boring to participants in comparison to less predictable situation
• Impossible for some experimental questions
• Impossible to estimate single-trial response function
• Effect that was elicited by one trial
• Possible if one makes strong assumptions about shape of HRF and additivity
• These assumptions are often invalid, e.g. exact form of HRF differs between regions
How does the block design work in practice? (5)
• Block length
• Short: 6-12s
• Intermediate: 12-21s
• Long: up to 30s or more
• Can vary between blocks
• Period of rest between blocks = rest interval
• Allows signal to go back to baseline
• Increases onset time asynchrony of successive blocks –> increases signal changes
• Less presentations of conditions of interest
• Only if interested in estimating the response to a block of trials
• Counterbalancing condition order to avoid that signal of condition X is predominantly biased by the condition that always precedes it
• One-back counterbalancing
• Number of blocks: 16-20 intermediate-length blocks sufficient for large effects; more needed for smaller effects
• Period of continuous data acquisition = runs
• Typically duration of 4-10 minutes
• Each condition in each run, for at least 2 blocks if possible
What is the event related design?
• Slow event-related design: long ISI (with or without jitter)
• Ideal for estimating HRF function by the event-related response
• Inefficient use of time and boring
What is the Rapid counterbalanced event-related design?
• ISI = 0 or not longer than the trial duration
• Conditions alternate in pseudo-random order: each condition follows each condition an equal number of times (counterbalancing)
• Peak in a difference of signal between conditions when particular condition occurs frequently in short period of time
• Reasonable sensitivity and power
• > Alternating design and slow design
• < Block design
• Estimation possible when one of the conditions = rest
• < Slow design
• Trial order similar to that of most behavioral experiments
Is there dichotomy between block and event related designs?
No dichotomy between block and event-related
• Distinction = fuzzy
• Intermediate approaches
• E.g. block length of 4-6 s; manipulation of probability of conditions
• Many different designs and extensions developed and used
• Condition-rich event-related design: multi-voxel pattern analysis
• fMRI adaptation – also known as repetition supppresion
What are the tasks and stimuli in the scanner?
• To activate particular mental processes of interest
• Task depends on question
• Comparing conditions different in both stimuli and task violates subtraction method assumptions
• When interested in stimulus effects
• No need for a behavioral response, in contrast to behavioural study
• Possible tasks without a stimulus-related response:
• Passive viewing
• Orthogonal task
• Very hard to avoid criticism about confounding task effects
How can we present the stimuli?
• Few problems for fNIRS
• Most problems for fMRI:
• Visual stimuli: Goggles, projection on a screen, or screen seen through a mirror
• Eye movement control: useful but difficult
• Auditory stimuli: noise of scanner (can be 120 dB!),
interleaved data acquisition
• Verbal responses: interference from scanner noise, confounding head movements during speaking
• MRI-compatible equipment = nonmagnetic and certified for use in a strong magnetic field
How do we go from design to scanning?
• Programming
• Need to synchronize MRI data acquisition with experiment by reading in trigger signal from scanner
• Presenting stimuli is more complicated (e.g., specific hardware)
• Reading behavioural response: no keyboard, other hardware is needed
• Ethical approval takes longer than for behavioral experiment
• Good preparation is critical because scanning is costly
• Keep detailed logbook of everything that happens
What are some considerations for participants?
• Safety issues
• Practice session in dummy/mock scanner for special populations like children
• Handedness
• Gender
• Number
• Depends upon expected effect size and signal to noise ratio
• Prior estimate of power is strongly advised, but not easy to determine
• Variation in literature between only 3-5 participants (when effect is large, can be demonstrated in single subjects, and data collection per subject is intensive) up to hundreds of participants.
What is the role of image preprocessing?
• Pre-processing not or very limited for behavioural data, but very important for imaging data
• Pre-processing is not part of statistics (but can strongly influence stats)
What does the choice of software package depend on? (4)
• Knowledge/background of the researcher
• Flexibility researcher wants
• Potential preference for operating system
• Expertise in lab of researcher
What is the pre-processing step 0?
Quality control
• Continuous point of attention
• Automatization is efficient, but creates large distance between experimenter and data without good quality control
• When starting: cut analysis in small pieces and run step by step
• Large problems often already detected during scanning itself
• After scanning, image files copied to work station but check them
• File names
• File sizes (all runs of equal length should be equal size)
• Inspect some time points
Why to not use data of extensive motion? (3)
• Some displacements cannot be corrected offline
• E.g. abrupt motion during TR àlarge shift of all other slices acquired after
• Motion decreases image quality due to instabilities in fMRI signal
• Alters magnetic field and its inhomogeneities and history of excitation of nuclei
• Takes much longer than actual movement to stabilize again
• Amount or type of motion can be related to experimental conditions
• E.g.: complex movement for a condition –> small head movement
• Hard to dissociate neural activity from motion artefacts/ confounds
How can we do external quality control?
External quality control through transparency and reproducibility
• Questionable research practices and relatively low rate of reproducibility
• In fields where statistics play a big role
• When effect sizes are small
• When many different research groups
• Positive: Many neuroimaging key findings have been documented over and over again (e.g., presence of face-selective regions)
• Negative: Chance of success to replicate a randomly chosen neuroimaging paper will be lower than it should be
• Number from psychology: effect size in replication typically half of effect size in original report
What are some causes of low reproductibility?
• Large degree of flexibility and possibility for exploration in analysis
• Each choice impacts results
• All choices together can have huge influence on final results
• Worst case, and not acceptable: some of the choices influenced by knowing
about the results obtained with these choices –> circular analyses
• Insufficient statistical power
• Due to insufficient number of participants given the typical effect size
• Increases possibility of false negatives
• Increases potential impact of questionable research practices, and thus possibility of false positives
• e.g. decision to take out 1 participant because of “bad data” (based on a priori criterion or on circular reasoning) will have more impact with a lower number of participants
• Decreases trustworthiness of effect size estimates in studies
• Need for a priori power analysis (even though it is difficult, e.g. because effect size is also determined by data quality)
What are 4 solutions for the low reproducibility problem?
• Methods sections as complete and transparent as possible
• In its asymptote very similar to a preregistered study
• Analysis tools and data made available (e.g., Open Science Framework)
• More replication efforts necessary
• More focus upon meta-analytic approaches and innovative tools to do so
• More data sharing