1) INTRO TO fMRI

How does MRI measure brain function? Why are neuropsychological cases useful for mapping cognitive functions to the brain?

• Brain damage can cause a specific cognitive deficit.

• Researchers identify the damaged area and investigate its association with the lost function.

How does MRI measure brain function? Why do researchers also need to study healthy participants?

Neuropsychological cases provide valuable evidence, but researchers also want to investigate brain activity in healthy people while they perform specific cognitive tasks.

• This allows researchers to examine brain function without relying on existing brain damage.

How does MRI measure brain function? How can intracranial EEG and brain stimulation investigate brain function, and what is their limitation?

PROS

• These techniques can measure or manipulate brain activity directly.

• They can be used in animals and in humans undergoing brain surgery

CONS

• highly invasive and are not suitable for routine studies of healthy participants.

How does MRI measure brain function? Why is MRI useful for brain mapping?

• Allows researchers to investigate the living human brain non-invasively

• can repeatedly measure brain signals while participants perform tasks, helping researchers investigate which brain regions respond to particular conditions.

How does MRI measure brain function? What is the difference between structural MRI and functional MRI?

• Structural MRI: Provides anatomical images showing brain structure, including grey matter, white matter and cerebrospinal fluid.

• Functional MRI (fMRI): Repeatedly measures changes in the MR signal over time to investigate brain activity indirectly.

How does MRI measure brain function? What happens during Step 1 of the MRI process?

• A strong magnetic field, called B₀, is applied.

• Hydrogen protons, which are initially randomly oriented, become aligned with the magnetic field.

How does MRI measure brain function? What happens during Step 2 of the MRI process?

• A short radio-frequency (RF) pulse supplies energy to the protons, tipping them away from alignment with B₀.

• Their magnetic moments precess around the field.

• When the RF pulse stops, the protons relax and generate a detectable MR signal.

How does MRI measure brain function? What happens during Step 2 of the MRI process?

• The scanner detects MR signals and uses spatial encoding to determine their origin.

• The signals are reconstructed into an image showing differences in signal intensity across the brain.

How does MRI measure brain function? Why do different tissues appear differently in structural MRI?

• Different tissue types have different hydrogen densities and magnetic relaxation properties

• These affect the strength of the detected signal, allowing structures such as grey matter, white matter and cerebrospinal fluid to be distinguished.

How does MRI measure brain function? How does structural MRI acquisition differ from fMRI acquisition?

• Structural MRI collects multiple slices to create a brain volume, typically taking approximately 5–10 minutes per volume

• fMRI repeatedly collects multiple slices to create multiple brain volumes over time, typically taking approximately 1–3 seconds per volume.

How does MRI measure brain function? What is a voxel

A three-dimensional pixel and the basic spatial unit of a brain image

How does MRI measure brain function? How does voxel size affect image resolution

• Smaller voxels provide finer spatial detail

• Larger voxels provide coarser anatomical detail

→ Voxel size affects how precisely researchers can localise a signal.

How does MRI measure brain function? What is the trade-off between temporal and spatial resolution in fMRI?

• Faster acquisition improves temporal resolution because brain volumes can be collected more frequently.

• However, faster acquisition may come at the cost of spatial resolution and anatomical precision, for example when larger voxels are used.

How does MRI measure brain function? What does fMRI measure

changes in the blood-oxygen-level-dependent (BOLD) MR signal, rather than measuring neuronal firing directly.

How does MRI measure brain function? why is BOLD signal considered an indirect measure of brain activity?

Neural activity triggers physiological changes in local blood flow, blood volume and oxygenation, which influence the signal detected by the scanner

How does MRI measure brain function? Why does the brain's blood supply matter for fMRI?

Although the brain represents a relatively small proportion of total body mass, it receives ~20% of the body's blood, oxygen and glucose.

• Its high metabolic requirements are closely linked to the blood-flow changes that fMRI measures.

How does MRI measure brain function? What physiological sequence typically produces an increase in the BOLD signal?

Increased neural activity leads to increased local blood flow.

How does MRI measure brain function? What does the BOLD signal depend on

differences in the magnetic properties of oxygenated and deoxygenated blood, as well as changes in blood flow and blood volume.

How does MRI measure brain function? How do oxygenated and deoxygenated blood differ magnetically?

• Oxygenated blood: Diamagnetic; weakly repelled by a magnetic field.

• Deoxygenated blood: Paramagnetic; attracted to a magnetic field.

  • → Deoxygenated haemoglobin alters the local magnetic field and affects the MR signal

How does MRI measure brain function? What happens to local blood vessels when neural activity increases?

Local blood vessels dilate, increasing blood flow and typically blood volume.

• The resulting change in blood oxygenation usually increases the BOLD signal.

How does MRI measure brain function? What is the haemodynamic response function (HRF)?

describes the characteristic time course of the BOLD response following neural activity.

• → links the relatively rapid neural response to the slower physiological changes in blood flow and oxygenation measured by fMRI.

How does MRI measure brain function? When does the BOLD response typically peak

Peak: ~5s after stimulus onset

How does MRI measure brain function? Why is the BOLD response peak described as sluggish

changes in blood flow, blood volume, and oxygenation develop more slowly than neuronal activity

How does MRI measure brain function? What information can be extracted from each fMRI voxel?

• time series of BOLD signal values from each voxel.

• Increases and decreases in the signal over time can then be analysed to investigate how that voxel responds during different conditions.

Understand task-related fMRI What is task-related fMRI

when participants perfrom a task while brain volumes are repeatedly collected

Understand task-related fMRI task-related fMRI use

Researchers analyse the resulting BOLD time series to identify voxels whose signals are associated with a particular task or cognitive process.

Understand task-related fMRI What is the logic of subtraction in task-related fMRI?

• Researchers compare two conditions that ideally differ in only one cognitive dimension.

• Subtracting the estimated activity associated with one condition from that of the other aims to isolate the activity associated with the process that differs between the conditions.

Understand task-related fMRI Why must the two comparison tasks be cognitively similar?

Performing a task involves multiple cognitive processes.

• If the comparison task shares these processes, their contributions can be controlled for, making it easier to identify activity associated with the specific process of interest.

• If the tasks differ in several ways, the contrast becomes harder to interpret.

Understand task-related fMRI What cognitive processes are involved in a face-matching task?

• Low-level visual processing.

• Face processing.

• Visual attention.

• Discrimination and matching judgements.

• Motor preparation.

• Motor response.

→ An activation map for the task may reflect all these processes, not just face processing.

Understand task-related fMRI Why might researchers compare a face-matching task with a house-matching task?

• Both tasks involve visual input, attention, discrimination, motor preparation and responding.

• However, face matching involves face processing, whereas house matching involves processing houses.

• Comparing the conditions aims to identify activity associated with the difference between them, including face-related processing.

Understand task-related fMRI What is a contrast map

shows the estimated difference in activity between two experimental conditions at each voxel

Understand task-related fMRI contrast map example

a face-matching minus house-matching contrast highlights voxels with greater estimated responses to face matching than house matching.

Understand task-related fMRI Why can subtraction produce misleading conclusions?

The two tasks may not be perfectly matched.

• may differ in difficulty, familiarity, attention or other cognitive demands

• the contrast may reflect several processes rather than only the intended process.

• The subtraction method also cannot guarantee that the remaining activity is unique to the cognitive function of interest.

Understand task-related fMRI What is the General Linear Model (GLM)

a statistical method for estimating how strongly a task predicts the BOLD time series at each voxel.

Understand task-related fMRI What is GLM used for in task-related fMRI?

to identify brain regions whose activity is associated with an experimental condition.

Understand task-related fMRI What is a task regressor

represents the timing of an experimental condition

Understand task-related fMRI How is HRF used to construct a task regresser

• Researchers use their knowledge of the HRF to model the expected BOLD response to that condition.

• They then compare this predicted time series with the observed BOLD signal in each voxel.

Understand task-related fMRI How does the GLM determine whether a voxel responds to a task?

The model estimates how well the task-related predictor explains the observed BOLD time series.

• If the signal changes in a way that corresponds to the predicted response, the voxel may show a substantial task-related beta coefficient.

Understand task-related fMRI What does the beta coefficient represent in a GLM?

estimates the strength and direction of the relationship between a predictor and the BOLD signal in a voxel.

• larger positive beta = stronger positive task-related response relative to the model's reference or baseline.

• negative beta = negative relationship relative to that reference.

Understand task-related fMRI How are beta values used to investigate brain activation?

• A beta estimate is calculated for each predictor in each voxel.

• Researchers can compare beta values across conditions using a contrast, then conduct statistical tests to determine whether the estimated differences are sufficiently reliable.

Understand task-related fMRI What is the difference between a beta coefficient and a statistical significance value?

• beta coefficient = estimates the size and direction of a modelled effect.

• statistical test = evaluates how strongly the data support an effect relative to its uncertainty.

  • A large beta is not automatically statistically significant, and a statistically significant result does not necessarily represent a large effect.

Understand task-related fMRI How are task-related fMRI results typically analysed across a group of participants?

• Obtain a contrast image for each participant.

• Conduct a one-sample statistical test at each voxel across participants.

• Calculate a test statistic, such as a \(t\)-value, and a corresponding \(p\)-value for each voxel.

• Apply a statistical threshold and appropriate multiple-comparisons correction.

• Display significant voxels on an anatomical image to help localise the result.

Understand task-related fMRI Why are task-related fMRI analyses described as massively univariate?

The analysis typically tests each voxel separately.

• Because a brain image may contain approximately 160,000 voxels, this can involve a very large number of statistical tests.

Understand task-related fMRI Why are task-related fMRI analyses described as massively univariate?

The more tests researchers conduct = greater the chance of false-positive findings.

• Some voxels may appear statistically significant even when there is no genuine effect.

Understand task-related fMRI What does an activation map show, and what does it not show?

• shows: voxels that pass a chosen statistical threshold for a particular contrast

• doesnt show:

  • every voxel that contains potentially relevant information,

  • does not prove that significant voxels are exclusively responsible for the cognitive process being studied.

Understand multivariate pattern analysis (MVPA) How does MVPA differ from standard task-related fMRI analysis?

• Standard task-related fMRI = tests individual voxels separately and applies a statistical threshold to identify significant voxels.

• MVPA = examines the distributed pattern of activity across multiple voxels, often within a ROI.

Understand multivariate pattern analysis (MVPA) How does MVPA differ from standard task-related fMRI analysis?

Voxels below the threshold may still contribute useful information to the overall activity pattern.

• MVPA considers the combined pattern across voxels, so a voxel does not need to be individually significant to contribute to distinguishing different mental states or representations.

Understand multivariate pattern analysis (MVPA) What relationship does MVPA investigate?

• Distributed patterns of brain activity.

• Mental states or representations, such as seeing a face, thinking about ice cream or imagining puppies.

The aim is to determine whether different mental states are associated with distinguishable patterns of activity.

Understand multivariate pattern analysis (MVPA) How is an activity pattern within an ROI represented mathematically?

• The activity values across the ROI's voxels can be represented as a vector or a sequence of numbers.

• Each voxel contributes a value to the pattern, allowing researchers to compare patterns associated with different stimuli or conditions.

Understand multivariate pattern analysis (MVPA) How might thinking about ice cream and thinking about puppies produce different patterns?

• Each mental representation may produce a different distribution of activity across voxels.

• These distributions can be converted into numerical vectors and compared to determine whether the patterns are similar or dissimilar.

Understand multivariate pattern analysis (MVPA) What are pattern similarity analysis and representational similarity analysis?

These techniques compare distributed patterns of brain activity to investigate how different stimuli or mental representations are represented in the brain. *

• They assess the similarity or dissimilarity between patterns rather than focusing only on individual voxel activation.

Understand multivariate pattern analysis (MVPA) What did Kriegeskorte et al. (2008) investigate

They investigated patterns of activity in regions of the ventral visual pathway while participants viewed images.

• By comparing patterns elicited by different images, researchers could examine how similarly or differently the brain represented those images.

Understand multivariate pattern analysis (MVPA) What is a representational similarity matrix, and how is it interpreted?

Displays the similarity between pairs of activity patterns.

• High similarity: The patterns elicited by two stimuli are alike.

• Low similarity: The patterns are different.

For example, different faces may elicit similar patterns in the ventral visual pathway, while images of faces and fruit may elicit more dissimilar patterns.

Understand decoding and machine learning in MVPA What is decoding in brain mapping?

asks whether a mental state, stimulus or condition can be predicted from a pattern of brain activity

Understand decoding and machine learning in MVPA decoding example

whether the activity pattern can reveal which person a participant is viewing.

Understand decoding and machine learning in MVPA How can two voxels distinguish between two stimuli?

Each stimulus can be associated with a different combination of activity across the two voxels. For example:

• High activity in voxel 1 and low activity in voxel 2 may be associated with Bob Dylan.

• Low activity in voxel 1 and high activity in voxel 2 may be associated with Tracy Chapman.

The combination of values can distinguish the stimuli, even if neither voxel alone provides a complete representation.

Understand decoding and machine learning in MVPA How can MVPA be combined with machine-learning algorithms?

• Researchers provide a machine-learning classifier with brain activity patterns associated with known stimuli or conditions.

• The classifier learns which combinations of voxel activity distinguish the categories and then predicts categories for new patterns.

Understand decoding and machine learning in MVPA What happens during the training phase of a classifier?

• The classifier receives training data containing activity patterns and their known labels.

• It learns a decision boundary that best separates the categories within the training data.

Understand decoding and machine learning in MVPA training phase of a classifier example

may learn to distinguish activity patterns associated with humans from patterns associated with dogs.

Understand decoding and machine learning in MVPA What is a decision boundary in machine learning?

separates regions of the model's feature space associated with different predicted categories.

• A new data point is classified according to which side of the boundary it falls on.

Understand decoding and machine learning in MVPA What is a feature space, and how does its dimensionality relate to voxels?

represents observations using their measured features.

• In an MVPA example using 100 voxels, each observation can be represented in a 100-dimensional space, with each dimension corresponding to one voxel's activity value.

Understand decoding and machine learning in MVPA What happens during the testing phase of a classifier?

• New, previously unseen fMRI data are supplied to the trained classifier.

• The classifier predicts the category for each pattern,

• the predictions are compared with the correct labels to evaluate its performance.

Understand decoding and machine learning in MVPA Why might researchers divide their data into 80% training data and 20% testing data?

The training data are used to learn the classification rule, while the testing data evaluate whether that rule generalises to new observations. Separating these datasets helps prevent researchers from overestimating performance based on data the classifier has already learned from.

Understand decoding and machine learning in MVPA What is classifier accuracy?

he proportion of predictions that are correct.

Understand decoding and machine learning in MVPA What does above-chance classifier accuracy suggest

• suggests that the measured activity patterns contain information useful for distinguishing the categories. does not prove that the brain region uniquely represents those categories or that the region is necessary for the cognitive process.

Understand decoding and machine learning in MVPA What doesn’t above-chance classifier accuracy prove

if the brain region uniquely represents those categories or that the region is necessary for the cognitive process.

Understand decoding and machine learning in MVPA Why can two measures of brain activation tell researchers different things?

The overall magnitude of activity and the pattern of activity across voxels are different measures.

• A region may show increased average activity when participants view faces, while a classifier may also distinguish individual faces using distributed patterns.

• Average activation and pattern-based information are therefore not interchangeable.

Understand encoding models and stimulus reconstruction What is an encoding model

predicts brain activity from properties of a stimulus.

Understand encoding models and stimulus reconstruction encoding vs decoding model

• Encoding: stimulus features → predicted brain activity.

• Decoding: brain activity → predicted stimulus or condition.

Understand encoding models and stimulus reconstruction How can researchers use fMRI activity to reconstruct a video?

• Researchers collect fMRI data while participants watch videos

• data is used to learn systematic relationships between visual stimulus features and brain activity.

• They then apply the learned model to activity recorded while participants watch a new video to generate an approximate reconstruction of the visual content.

Understand encoding models and stimulus reconstruction What does video reconstruction suggest about the relationship between brain activity and visual stimuli?

If a model can reconstruct aspects of a video from brain activity, this suggests that the measured activity contains systematic information related to the visual input. It does not mean the reconstruction is a direct recording of the participant's subjective experience; its quality depends on the model, training data and measured signals.

What is resting-state functional connectivity?

examines statistical relationships between spontaneous BOLD signal fluctuations in different brain regions while a participant is not performing a specific task.

What does functional connectivity between two brain regions mean?

the regions' activity signals show a statistical relationship, often measured by correlation.

• This may suggest coordinated activity, but it does not by itself demonstrate a direct anatomical connection or a causal relationship.

What are the key dos and don'ts when interpreting task-related fMRI results?

• Check whether the comparison tasks are well matched.

• Interpret contrasts in relation to the specific conditions tested.

• Consider statistical significance and multiple-comparisons correction.

• Distinguish the size of an estimated effect from its statistical significance.

• Remember that fMRI measures the BOLD response indirectly.

Don't:

• Assume a significant voxel is exclusively responsible for a cognitive function.

• Assume subtraction perfectly isolates one process if tasks differ in other ways.

• Treat an activation map as a complete representation of all relevant brain information.

• Interpret a correlation between brain activity and a task as proof of causation.

• Assume that voxels below a statistical threshold contain no useful information.

What is the big question: What are we mapping when we do brain mapping?

We are mapping relationships between brain signals and cognitive processes, mental states or representations. The interpretation depends on the method:

• Structural MRI: Maps brain anatomy.

• Task-related fMRI: Identifies voxels whose BOLD responses differ between experimental conditions.

• MVPA and representational similarity analysis: Investigate information contained in distributed activity patterns and how representations resemble one another.

• Decoding: Tests whether mental states or stimuli can be predicted from brain activity.

• Encoding: Tests whether stimulus properties can predict brain activity.

• Resting-state functional connectivity: Maps statistical relationships between spontaneous activity in different regions.

The central caution: Mapping a relationship between a brain signal and a cognitive process is not the same as proving that a particular region uniquely performs that process. Brain mapping results must be interpreted in light of the measurement method, experimental design, statistical analysis and alternative explanations.