NeuroscienceResearch Methods

How to Measure Neuroplasticity

What brain imaging, electrical recordings, stimulation, and behavioral experiments can reveal about lasting change, and how to connect that change to useful outcomes.

By

A wooden pegboard and a dish of pegs on a laboratory table beside a notebook
Illustrative setup for a manual dexterity task. The measurement depends on a defined protocol and comparison.

Imagine that a person spends several weeks learning a difficult skill. Their movements become more precise. A brain scan changes. An electrical response becomes stronger. Which of these observations tells us that neuroplasticity occurred? Which tells us that the training was useful?

The answer depends on what changed, how it was measured, and what would have happened without the training. A striking brain image can leave the most important question unresolved: can this person now do something that matters, and can they still do it later?

Evaluating claimed neuroplasticity benefits requires connecting experience, persistent neural change, and independently assessed behavior. Different experiments can establish different parts of that chain. A well-designed study states which part it is testing.

This article examines established research frameworks, physical measurement methods, and behavioral approaches. It draws on published studies of human learning and neural adaptation. The study designs and numerical examples proposed here are illustrations, not results from a new experiment.

Define the change before choosing the instrument #

Neuroplasticity describes the nervous system’s capacity to change its organization or functioning in response to experience and other influences. The relevant changes can involve synaptic efficacy, cellular structure, myelination, or the organization of activity across circuits. Human imaging usually observes aggregate signals that can have several biological explanations. Zatorre, Fields, and Johansen-Berg describe this gap between the structural changes visible in imaging and their possible cellular origins. [1]

An operational definition should name a population, an experience, a neural property, a timescale, and a comparison. For example: does a specified motor-learning program produce a reproducible change in a prespecified motor-system measure that remains detectable after practice stops, relative to an equally credible comparison program?

That question is narrower than whether a person’s brain became “more plastic.” It is also much easier to investigate. A training-induced change, a person’s capacity to undergo a future change, and the practical benefit of a change are different research targets.

Four claims that require different evidence
Claim Evidence to seek What remains unresolved
A state changed A neural or physiological signal differs during or shortly after an activity. Whether the difference outlasts arousal, fatigue, task demands, or the measurement session.
A neural property changed A controlled longitudinal difference in a defined structural or functional measure. The cellular mechanism and whether the change benefits the person.
A capability improved Better independent performance on a relevant, reliable assessment after a meaningful delay. Which neural mechanisms produced the improvement.
Plasticity capacity changed A different response to a standardized learning or induction challenge, with adequate repeatability and controls. Whether that response predicts learning in another circuit, task, or setting.

The timescale belongs in the claim. A physiological aftereffect lasting an hour can be a legitimate finding about short-term plasticity. A claim about a skill useful next month requires evidence over that interval. There is no single waiting period that makes every measurement a test of durable change.

The direction also needs interpretation. Stronger activity, larger volume, and more connectivity are not universal definitions of improvement. An efficient strategy might require less effort. A persistent habit might be unhelpful. The benefit must be defined through the intended capability and the person’s circumstances.

Frameworks that make the question measurable #

Several established frameworks help organize a study, but they serve different purposes. A theory can suggest what should cause change. A measurement framework can specify what a score means. A reporting standard can make an analysis reproducible. Combining their roles into one “neuroplasticity framework” would conceal important decisions.

Experience-dependent plasticity and the demands of learning #

Kleim and Jones organized principles of experience-dependent neural plasticity around factors including specificity, repetition, intensity, timing, salience, transfer, and interference. Their framework was developed in the context of rehabilitation after brain damage. It helps formulate questions about training dose and task design. It does not provide a device or a numerical plasticity score, and its clinical context matters when extending it to healthy learners. [2]

Lövdén and colleagues proposed a framework for adult cognitive plasticity in which a sustained mismatch between environmental demands and available functional capacity can provoke adaptation. They distinguish flexible use of an existing repertoire from changes in the capacity supporting that repertoire. This is a theoretical account of when adaptation might be induced, rather than a universal test for detecting it. [3]

For a study of adaptive training, these ideas motivate a concrete prediction: a program that keeps demands appropriately challenging might produce a different learning trajectory from fixed practice. The experiment would need to record actual difficulty, success, practice dose, and persistence. “More difficult” alone is an incomplete intervention specification.

RDoC and BEST: specify the construct and validate the signal #

The National Institute of Mental Health’s Research Domain Criteria, or RDoC, organizes research constructs across units of analysis such as circuits, physiology, behavior, and self-report. Its original purpose is research on mental disorders. Its useful contribution here is the discipline of studying a defined function at several levels. Selecting a construct such as working memory still leaves the researcher responsible for choosing and validating the tasks and neural measures. RDoC is neither a diagnostic system nor a measure of plasticity. [4]

The FDA-NIH BEST resource supplies another useful distinction. Analytical validation concerns whether a measurement adequately measures what it is intended to measure. Clinical validation concerns whether the biomarker measures or predicts the relevant clinical concept. The required evidence depends on the intended context of use. [5]

Applied to learning research, this means that a headset can record a physiological signal accurately while its proprietary “learning readiness” score remains unvalidated. A signal that changes after training might be a response marker without being a substitute for a meaningful outcome. Using this validation logic does not confer regulatory qualification on an educational measure.

Before collecting data, write down the signal, the processing procedure, the interpretation, and the decision it is supposed to support. If the decision is whether someone can perform a task independently, validate the signal against that outcome in the intended population. A correlation found while developing the score needs confirmation in new data.

ICF: connect body function to activity and participation #

The World Health Organization’s International Classification of Functioning, Disability and Health, or ICF, separates body functions and structures from activities and participation, while accounting for environmental context. This distinction is particularly useful in rehabilitation. A physiological change, better performance on a laboratory task, and greater independence in everyday life are related but separate outcomes. [6]

Consider a person practicing hand movements. A motor-cortex response might change. A standardized dexterity task might improve. Preparing a meal might still be difficult because of fatigue, pain, equipment, or the home environment. A research program concerned with everyday benefit needs an outcome at that level.

The corresponding lesson for education is to measure the capability where it is supposed to become useful. A laboratory task can offer precise control. A field assessment can reveal whether the capability survives the demands of its intended setting. Each answers a different part of the question.

What physical devices can measure #

The first decision is the biological question. Instruments differ in the signals they capture, the spatial and temporal scales they resolve, and the conditions under which a participant can be measured. The following overview is a map of research options. It is not a ranking of how much “plasticity” each device can detect.

Measurement methods and their main interpretive limits
Method Primary observation Useful research question Main limit
Structural and diffusion MRI Anatomical image properties and patterns of water diffusion. Does a defined tissue-related measure change with training? Image-derived measures have multiple biological and technical determinants.
Functional MRI Blood-oxygen-level-dependent changes associated with neural activity. Does task recruitment or a defined activity pattern change? An indirect hemodynamic signal depends on task and physiological conditions.
EEG and MEG Electrical potentials or magnetic fields associated with population activity. Does the timing or organization of a response change? Source ambiguity, artifacts, and state dependence complicate interpretation.
Functional near-infrared spectroscopy Changes in oxygenated and deoxygenated hemoglobin near the cortical surface. Do cortical hemodynamic responses change in a more mobile task? Scalp physiology, motion, and limited depth must be addressed.
TMS with EMG A muscle response to stimulation of the motor system. Does a standardized challenge alter corticospinal responsiveness? The response is circuit- and protocol-dependent, with substantial variability.
Magnetic resonance spectroscopy and PET Selected metabolites or binding of a specified molecular tracer. Does a targeted neurochemical or molecular measure change? Specialized proxies require their own biological interpretation and validation.

Eye tracking, pupil size, heart rate, skin conductance, and movement sensors can add useful information about what happened during a task. They can help characterize attention, effort, physiological state, or behavior. Their inclusion does not by itself turn a study into a direct measurement of neural remodeling.

Structural and diffusion MRI: changes in tissue-related measures #

Structural magnetic resonance imaging can estimate regional volume or cortical thickness. Diffusion MRI measures how water displacement varies across tissue and can support models of white-matter organization. These measurements are valuable because they can be repeated in living people and related to a learning history. Their biological interpretation remains indirect. [1]

In a longitudinal juggling study, Scholz and colleagues reported training-related changes in a diffusion measure in white matter beneath the intraparietal sulcus. This established a measurable association between a controlled training experience and a tissue-sensitive imaging signal. It did not provide a count of new axons or a direct measurement of new myelin. [7]

Jones, Knösche, and Turner explain why terms such as “white matter integrity” can overstate what diffusion measures establish. Fractional anisotropy, a common summary of directional diffusion, can reflect several aspects of tissue geometry. A higher value has no universal interpretation as healthier or better connected tissue. [8]

The time course is equally instructive. Wenger and colleagues repeatedly imaged adults practicing left-handed writing and drawing. Motor-cortex volume estimates initially expanded and later partially renormalized while proficiency continued to improve. The study suggests that a single before-and-after contrast can miss a changing trajectory. It does not show that every skill follows the same sequence or that MRI directly observes synaptic pruning. [9]

A structural study therefore needs consistent acquisition and processing, careful motion and image-quality checks, and a comparison group measured on the same schedule. A null result at the final visit could mean no detectable effect at that time. It could also miss an earlier change. Additional scans should be chosen to test a temporal hypothesis, rather than added simply to produce more images.

Functional MRI: changes in recruitment and representation #

Functional MRI commonly uses the blood-oxygen-level-dependent, or BOLD, signal. In foundational work combining electrophysiology and imaging in macaque visual cortex, Logothetis and colleagues showed why BOLD should be interpreted as an indirect signal related to local neural processing. It is not a direct meter of neurons firing or synapses strengthening. [10]

A learning experiment might ask whether a prespecified region responds differently, whether patterns of activity distinguish learned categories more clearly, or whether coordination between regions changes. Each requires a defined contrast and an analysis chosen before inspecting the effect. Correlated activity between regions does not establish a new anatomical connection.

Task difficulty creates a subtle problem. After training, the same task may require less effort. Matching the task exactly preserves the stimulus comparison but changes its subjective demand. Adjusting difficulty to equalize performance changes the stimulus comparison. A strong design can include both fixed-difficulty trials and separately analyzed performance-matched trials, explaining what each estimates.

Less activation with equal or better performance could be consistent with more economical processing. It could also reflect a changed strategy or another physiological difference. More activation could reflect greater recruitment or greater difficulty. Neither direction alone establishes improvement. Predictions about behavior and the neural contrast should be stated together.

Individual interpretation also requires repeatability. Elliott and colleagues’ meta-analysis found low average test-retest reliability for common measures of individual differences in regional task-related activation. Their meta-analysis excluded intervention studies designed to assess change. It cautions against assuming that a robust group contrast also provides a dependable score for ranking people. It does not establish the reliability of every measure of change after training. [11]

EEG and MEG: when neural responses change #

Electroencephalography records voltage differences at the scalp. Magnetoencephalography records magnetic fields associated with neural currents. Both can resolve the timing of population responses on a millisecond scale. Researchers can examine responses to particular events, oscillatory activity, and patterns of coordination. The COBIDAS MEEG recommendations address acquisition, analysis, artifact handling, and reporting choices that make those comparisons interpretable. [12]

An EEG response averaged around repeated events is called an event-related potential. Teyler and colleagues reported that patterned visual stimulation could increase a component of the visual evoked response. In follow-up experiments that reduced intervening test stimulation, an increase remained detectable at 60 minutes. The testing schedule affected its persistence. They interpreted the response as an LTP-like phenomenon. Long-term potentiation, or LTP, refers to persistent strengthening of synaptic transmission in its cellular research context. A scalp response resembling aspects of LTP does not directly measure an individual synapse. [13]

The assay remains an active research area. Dias and colleagues combined a reassessment, new experiments, and meta-analyses and raised concerns about the reliability and validity of sensory stimulation procedures for inducing these changes. More recently, Galuba and colleagues compared visual-evoked-potential modulation protocols as candidates for plasticity biomarkers. That work explores how to improve an assay. It does not establish a general-purpose test of a person’s learning potential. [14] [15]

For a learning study, the practical question might be whether a specified response to an untrained probe changes after training and remains different at follow-up. Eye movements, muscle activity, electrode placement, alertness, and the number of usable trials need attention. A stronger rhythm during a stimulating experience could be a state effect. A repeatable delayed response under standardized conditions supports a different, narrower inference.

The measurement itself may also alter the system. A stimulation procedure designed to induce an aftereffect is an intervention, and repeated sensory probes are additional experiences. Their dose and schedule belong in the experiment.

fNIRS: cortical hemodynamics in more natural tasks #

Functional near-infrared spectroscopy, or fNIRS, uses light to estimate changes in oxygenated and deoxygenated hemoglobin. Its portability can make some movement, interaction, and learning studies easier to conduct than in an MRI scanner. Its accessible measurements are primarily superficial, and the recorded signal can include substantial contributions from scalp tissue and systemic physiology. [16]

Yücel and colleagues’ publication guidelines emphasize acquisition, processing, and reporting choices that make fNIRS findings interpretable. Short source-detector separations can help characterize superficial signals for appropriate correction. Motion handling, physiological confounds, sensor placement, and the statistical model also matter. A change in oxygenation during a demanding task cannot automatically be assigned to cortical learning. [16]

A suitable study would specify a cortical hypothesis, verify signal quality in the actual task, and repeat comparable measurements after training. Portability makes additional questions possible. It does not remove the need for control conditions or turn every everyday setting into a clean measurement environment.

TMS and EMG: probe a circuit’s response to a challenge #

Transcranial magnetic stimulation can perturb cortical activity. When used to stimulate the motor system, electromyography can record a motor-evoked potential in a muscle. Researchers can examine how that response changes under a standardized protocol. This is an input-output assay of the tested system, with contributions from more than a single synapse.

Stefan and colleagues demonstrated lasting changes in motor-system responses after paired associative stimulation, which combined peripheral nerve stimulation with TMS. In the duration experiment, responses remained elevated throughout 30 minutes of observation in 11 participants. Two were followed for 60 minutes, with elevated responses persisting in both. The timing-dependent effects offered a way to investigate associative, LTP-like plasticity in humans. The result concerns a specific physiological preparation. It does not establish how much unrelated knowledge a participant can learn. [17]

Hamada and colleagues found substantial variation in responses to theta-burst stimulation. Responses were associated with a physiological marker interpreted as reflecting which interneuron networks the stimulation recruited. This supports a possible explanation for the variation, rather than a direct measurement of each person’s intrinsic plasticity. A weak aftereffect needs interpretation in the context of the tested protocol and circuit. [18]

These methods belong in appropriately equipped research or clinical settings with trained staff, participant screening, and an approved protocol. Their scientific value is a controlled perturbation and a defined response. Using stimulation to induce change and using stimulation to measure responsiveness should be distinguished in the study design.

MRS and PET: more targeted biological questions #

Magnetic resonance spectroscopy can estimate concentrations of selected metabolites within a sampled region. Depending on the acquisition, research may examine compounds such as glutamate. A regional concentration estimate is different from the amount released at a particular synapse or the strength of a particular connection. Acquisition quality, tissue composition, and the analysis model constrain the inference. [19]

Positron emission tomography uses radiotracers to investigate specified molecular targets. Finnema and colleagues demonstrated imaging with a tracer binding to synaptic vesicle glycoprotein 2A, or SV2A, as an approach to estimating synaptic density in living humans. Their study combined validation in a baboon with initial imaging in healthy people and patients with temporal lobe epilepsy. It did not test synapse formation after learning. Tracer binding remains a proxy with its own assumptions and cannot identify which synapses encode a newly learned skill. [20]

PET also involves radiotracer administration and ionizing radiation, with specialized facilities and oversight. These methods can be valuable when the hypothesis specifically concerns a molecular or neurochemical process. Adding them to a learning study without such a hypothesis would create expense and complexity without necessarily clarifying the outcome.

Neurofeedback: distinguish control of a signal from useful learning #

Neurofeedback gives a participant information about a measured brain signal so they can attempt to regulate it. The signal is part of both the intervention and its assessment. This makes independent outcomes particularly important.

The CRED-nf consensus checklist developed by Ros and colleagues addresses experimental design and reporting for neurofeedback research. It calls attention to controls, feedback specifications, and both brain and behavioral outcomes. Its framework helps distinguish effects of regulating the intended signal from expectations and other features of participation. [21]

A study should ask whether participants learned to regulate the target, whether regulation persists without feedback, and whether an independently assessed capability improves. Better control of the trained display answers only part of that question. A device-generated success score needs validation against the outcome being claimed.

What behavioral science can establish #

Behavioral science can establish that a person learned, retained, generalized, or changed a strategy under specified conditions. It can often do so with greater direct relevance to the intended benefit than a neural measure. It cannot, by itself, identify the cellular mechanism that produced the change.

Soderstrom and Bjork review why successful performance during practice can diverge from lasting learning. Instructions, immediate feedback, repetition, and visible cues can support a response that is unavailable later. Delayed assessment and changed testing conditions help distinguish those possibilities. [22]

Behavioral outcomes to keep separate
Outcome Possible measure Interpretive check
Acquisition A learning curve for accuracy, error, or task completion. Separate task learning from device familiarization and temporary assistance.
Retention Performance after a defined interval without renewed instruction. Specify intervening practice, testing, feedback, and elapsed time.
Transfer Performance on prespecified untrained tasks or contexts. Describe exactly what is shared and what has changed.
Relearning Trials or time needed to recover a prespecified proficiency. Account for baseline skill, test difficulty, and opportunities to remember a strategy.
Everyday function Independent completion of a relevant activity in its intended setting. Record assistance, environmental support, effort, and practical significance.

A battery should reflect the claimed capability. For a perceptual skill, that might involve estimating the smallest difference a person can reliably discriminate. For a motor skill, it might involve precision at a controlled speed. For reasoning, it might involve predictions, explanations, or diagnostic choices scored with explicit criteria. Faster responses accompanied by more errors should not be called an unqualified improvement.

Barnett and Ceci’s transfer taxonomy helps specify distance across dimensions such as knowledge domain, physical and social context, time, and modality. “Two levels removed” has no common scientific meaning until those dimensions are defined. Transfer to a new display of the same task and transfer to an unfamiliar everyday problem should be reported separately. [23]

Use behavioral experiments to separate competing explanations #

A carefully designed behavioral task can reveal more than a total score. In visuomotor adaptation experiments, researchers alter the relationship between a movement and its visible consequence. Taylor, Krakauer, and Ivry used reports of aiming direction to help separate explicit strategy from implicit contributions to adaptation. Similar final accuracy can therefore arise through different processes. [24]

That observation matters for interpreting a successful training session. A person might learn a verbal workaround, recalibrate a sensorimotor prediction, or combine both. Tests that remove feedback, change instructions, or examine aftereffects can help distinguish accounts. The interpretation depends on the task and assumptions, and such tests are additional experiences that may affect subsequent behavior.

Computational models can make these accounts quantitative. A reinforcement-learning model might distinguish a learning rate from how consistently choices follow the estimated value of each option. A fitted parameter is meaningful only if the task can identify it. Wilson and Collins recommend simulation, parameter recovery, model recovery, and checks of model predictions as part of responsible behavioral modeling. [25]

For example, before interpreting a higher fitted learning rate, simulate participants whose learning rate is unchanged but whose choices become less noisy. If the analysis mistakes reduced noise for faster learning, the parameter cannot support the proposed conclusion. This is a test of the measurement model before it becomes an explanation of the person.

Behavioral experiments can thus narrow a mechanism without directly observing neural tissue. Their conclusions should remain at the level the experiment distinguishes.

Separate a group effect from a reliable individual change #

A method that detects a group effect is not automatically suitable for telling one person how much they improved. Hedge, Powell, and Sumner describe a reliability paradox: tasks that produce robust experimental effects can still rank individuals unreliably. Reliability depends partly on the variation between people, so it must be evaluated for the population and use in question. [26]

Longitudinal work introduces another requirement. Reliable scores at each visit do not guarantee reliable change scores. Measurement error at both visits contributes to uncertainty, and practice effects can shift the whole distribution. A method intended to monitor individuals should be evaluated for repeatability of the relevant change, not merely for whether its signal looks plausible.

Repeated baselines help characterize ordinary variation and familiarization, although additional testing is itself an exposure. Equivalent test forms need evidence of comparability. A threshold measured on one set of stimuli may not be interchangeable with a threshold measured on another. The scoring system should retain the same meaning across visits.

A worked example of reliable change #

Jacobson and Truax proposed a reliable change index to ask whether an individual’s observed change exceeds what would be expected from measurement error. Under a simple model with independent, equal measurement errors, the standard error of the difference is the baseline standard deviation multiplied by the square root of twice one minus reliability. The index divides the observed change by that standard error. [27]

Hypothetical calculation under a simple reliable-change model
Quantity Illustrative value
Baseline standard deviation in an appropriate reference sample 10 points
Relevant test-retest reliability 0.90
Standard error of the difference 10 × √(2 × (1 − 0.90)) = 4.47 points
One person’s observed gain 8 points
Reliable change index 8 ÷ 4.47 = 1.79

Under the model’s assumptions and an approximately normal error distribution, an absolute index greater than 1.96 would exceed the conventional two-sided 95% threshold. The hypothetical gain above does not meet that criterion. It could still be a real improvement. The measurement is not precise enough to establish it at that threshold.

This illustration assumes comparable scores, an appropriate reliability estimate, and no unmodeled systematic shift. Actual evaluations may need practice-adjusted or regression-based norms, baseline-dependent error, and correction for regression to the mean. Applying the formula mechanically to a ceiling-limited score or an unreliable neural feature would be misleading.

Reliable change also differs from meaningful change. A statistically dependable gain could be too small to matter in daily life. A practically valuable gain could be measured too imprecisely to classify confidently for one individual. Neither threshold identifies a neural mechanism. For individual use, report uncertainty and validate the practical decision the score is meant to support.

Design a study that can connect experience to change #

A useful research program begins with measurement development, then tests the intervention. The pilot asks whether the task behaves sensibly, whether the signal is usable, whether participants can complete the protocol, and whether the proposed outcome is sensitive to the relevant range of ability. The confirmatory study tests a narrower, prespecified claim.

Consider an illustrative study of adult novices learning a visuomotor skill. One group receives adaptive practice, with difficulty adjusted using a predefined rule. An active comparison group receives fixed practice. Both receive comparable scheduled time, contact, equipment, and feedback. The primary question is whether adaptive practice improves delayed precision at a specified movement speed. A secondary question concerns the trajectory of a prespecified motor-region MRI measure.

Both groups may learn and undergo neural change. The randomized contrast estimates the added effect of the adaptive program relative to fixed practice. It cannot establish the total effect of having any practice at all. A different comparison would be needed for that question.

Illustrative schedule for a longitudinal learning study
Stage Behavioral assessment Neural assessment and purpose
Familiarization and repeated baseline Separate device learning from the target skill. Estimate initial precision and ordinary variation. Obtain consistent baseline measurements and quality checks before the assigned program.
Early training Measure the learning trajectory and actual exposure. Sample an early time point if the hypothesis predicts a transient or evolving neural change.
End of four weeks of practice Assess the trained skill and a prespecified untrained variant without coaching. Repeat the same acquisition with the prespecified processing and quality criteria.
Four weeks after practice ends Measure the primary delayed outcome and transfer. Record intervening practice. Test persistence or further evolution of the selected neural measure.

These intervals illustrate a design choice. They are not biological thresholds. The schedule should follow the hypothesized process, the expected learning trajectory, and the real-world retention horizon. A rapidly changing physiological response might need a different schedule from a tissue-related MRI measure.

Randomize after the relevant baseline procedures, conceal allocation until assignment, standardize instructions, and keep outcome scorers unaware of group assignment where possible. Match the comparison to the causal question. A highly engaging program compared with no contact leaves expectations, social interaction, and practice opportunity entangled with the proposed mechanism.

Assessment can contribute to learning. Retrieval experiments by Roediger and Karpicke show that testing can improve later retention. Equivalent test forms reduce memorization of exact answers but can still rehearse the underlying capability. Keeping the assessment schedule identical across groups supports a fair comparison of the two programs under that schedule. A separate randomized assessment schedule can investigate what persists without interim testing. [28]

The same issue applies to periods between visits. Record additional practice and major changes in conditions. Sleep and rest are plausible parts of a learning trajectory, but an association with later improvement does not establish their causal role. What the Brain Does After We Stop Working on a Problem examines that distinction in research on breaks and sleep.

Prespecify the analysis and the precision needed #

Choose one primary outcome and time point, or a clearly defined small set with a plan for multiple testing. Specify the behavioral scoring, neural regions or features, preprocessing, motion thresholds, exclusions, handling of missing visits, and the main statistical contrast before examining group differences. Exploratory analyses can remain useful when identified as exploratory.

For a randomized study with a continuous baseline and follow-up outcome, a baseline-adjusted comparison is often more appropriate than comparing percentage improvements. Vickers and Altman explain the advantages of this approach. A repeated-measures model may be useful for a trajectory question, provided its assumptions and handling of missing data fit the design. [29]

As an invented example, suppose average delayed scores improve by 15 points in the adaptive group and 8 in the comparison group. The descriptive difference in change is 7 points. The result to report is the prespecified between-group estimate with its uncertainty, rather than treating the adaptive group’s full 15-point gain as the program’s unique effect. The adjusted estimate can differ from the simple subtraction.

Sample-size planning should target the smallest effect that would matter or the precision needed for a decision, using defensible assumptions about variability, reliability, repeated measures, and attrition. Lakens describes several ways to justify sample size. A pilot can inform feasibility and measurement, but its estimated effect size may be too uncertain to serve as the sole planning input. [30]

Count participants as the units supporting a claim about people. Thousands of trials, sensor channels, or image voxels from a small number of participants do not create thousands of independent participants. Neural feature selection and predictive-model evaluation should keep training and evaluation data separate at the level required by the intended generalization.

Missingness also has a scientific meaning. If participants with greater discomfort or poorer performance are more likely to withdraw or have unusable recordings, the retained sample may be selective. Report usable data and missing visits by group and explain the sensitivity analyses. A small p-value cannot repair a comparison whose observation process has changed.

Combine measures without mistaking correlation for mechanism #

Multiple measures help when they test complementary predictions and have different weaknesses. Behavioral precision and an independently acquired neural measure provide more useful convergence than several closely related scores extracted from the same signal. EEG and fNIRS collected simultaneously can still share motion and task-state confounds.

A plausible mechanistic account should predict a pattern: which neural feature changes, when it changes, which behavior it relates to, and where the relationship should fail. Establishing that time order is more informative than searching every brain feature for a correlation with the final score.

Even if random assignment changes both a neural measure and behavior, their correlation does not prove that the measured neural change caused the behavioral gain. Motivation, exposure, strategy, or other post-assignment factors could influence both. A mediation analysis requires additional assumptions about those relationships. A carefully justified perturbation or a separate mechanism experiment can provide stronger evidence, while still requiring its own controls.

A measurement of the capacity to change needs a further distinction. If a standardized challenge produces a larger aftereffect after training, the study has observed altered responsiveness to that challenge. To claim enhanced learning potential, test whether that response predicts subsequent learning on independently selected tasks, beyond baseline ability and other relevant predictors. Repeat the prediction in new participants.

The most useful result may be a constrained explanation of one capability. A general claim about intelligence would require a broader, validated assessment and evidence that the effect extends beyond the practiced task family.

Where VR can help the experiment #

Virtual reality can make it possible to control spatial relationships, repeat unusual situations, and vary sensory feedback while recording a participant’s actions. Its value in a neuroplasticity study would come from the experiment those capabilities enable. Immersion itself is not a biological measurement.

For example, a researcher could systematically alter the mapping between a hand movement and its virtual consequence, then test retention and generalization after the altered mapping is removed. Another experiment could hold visual content constant while varying whether a learner actively controls an exploration. Each design needs a clear comparison and an outcome that corresponds to the claimed skill.

If the intended benefit concerns the physical world, include an assessment there. A participant might become better at the virtual interface, exploit a display cue, or learn a controller-specific strategy. Transfer should survive the particular change in context that matters to the application.

Concurrent neural recording adds practical constraints. A headset may compete with EEG electrodes or fNIRS sensors for placement. Movement, muscle activity, headset pressure, display timing, and discomfort can affect data quality or performance. Validate the combined setup before interpreting a training effect. For some questions, standardized measurements before and after VR practice will be easier to interpret than recordings during active movement.

A direct test of the value of VR would compare appropriately matched tasks across delivery formats. If the VR condition also changes feedback, practice dose, and learner control, the experiment estimates the effect of that package. Recording those differences makes the result useful for the next design.

Available tools and standards for a reproducible workflow #

Established software can help implement and analyze a study. It cannot supply a valid construct or a causal design automatically. The useful question is which part of the measurement process each tool supports.

Examples of available research infrastructure
Resource Role in a study What researchers still need to establish
PsychoPy and jsPsych Construct behavioral tasks, present stimuli, and collect trial-level responses in supported laboratory or browser workflows. Task validity, stimulus and response timing on the actual hardware, scoring, and data quality.
MNE-Python Process and analyze neurophysiological recordings, including EEG and MEG. Acquisition quality, artifact decisions, appropriate models, and reliability of the selected feature.
fMRIPrep Perform standardized preprocessing of functional MRI and generate quality reports. Inspection of outputs, the final statistical model, and the meaning of the contrast.
Brain Imaging Data Structure (BIDS) Organize neuroimaging and associated data with consistent metadata. Accurate event definitions, complete provenance, appropriate sharing, and valid measurements.
OHBM COBIDAS recommendations Guide design, analysis, and reporting for MRI and electrophysiological research. Application to the specific protocol and transparent documentation of departures.

BIDS was introduced to make the organization and description of neuroimaging datasets more consistent. COBIDAS addresses broader practices in acquisition, analysis, reporting, and sharing. Together they help another researcher understand what was done. Neither certifies that an observed change represents neuroplasticity. [31] [12]

Keep raw observations separate from derived scores. Preserve event timing, calibration records, software versions, processing parameters, and quality decisions. For a behavioral assessment, retain which item was shown, the response, timing, feedback, and relevant assistance. For a neural measure, retain the information needed to reconstruct its acquisition and processing.

Share the protocol, scoring rubric, code, and appropriately de-identified data or controlled-access pathway within consent and research agreements. Participant information and neural recordings require an intentional access policy. The goal is enough transparency to evaluate the claim and repeat the analysis without exposing people unnecessarily.

Interpret the pattern of evidence #

A study should be prepared for disagreement between measurements. Neural and behavioral methods sample different aspects of the process at different resolutions. The table below describes reasonable interpretations to investigate, rather than automatic conclusions from a significant or nonsignificant result.

Possible outcomes of a controlled longitudinal study
Observed pattern Supported conclusion and next question
Durable behavioral benefit and a predicted neural change Converging evidence of learning and neural adaptation in the studied setting. Investigate whether the neural change explains the benefit and replicate the pattern.
Durable behavioral benefit without a detectable neural effect Evidence of useful learning on the assessed outcome. The chosen neural measure may be insensitive, mistimed, or unrelated to the mechanism.
Neural change without a behavioral benefit Evidence concerning the specified neural measure, if the comparison and measurement are sound. Practical benefit remains unestablished, and the behavioral test may need scrutiny.
Improvement only during training Evidence of better practice performance. Delayed independence and transfer remain unresolved or unsupported by the current assessment.
Little difference with wide uncertainty An inconclusive comparison. The data remain compatible with a range of effects.
Little difference with a narrow interval inside prespecified equivalence bounds Evidence that effects larger than those bounds are unlikely under the studied conditions. This still concerns the measured outcome, population, and comparison.

The alternatives should influence the next experiment. If behavior improves but a neural measure does not, first examine the measure’s sensitivity and the timing of observation. If only the neural measure changes, test whether the effect relates to any meaningful capability before expanding the intervention’s claims.

A program that improves everyday function through a compensatory strategy can be valuable even if it does not restore a proposed underlying process. Conversely, an intervention can produce a fascinating neural effect without delivering a useful outcome. The purpose of the research determines which result deserves priority.

Choose the evidence that the claim requires #

For an early learning intervention, a disciplined behavioral study may be the most informative starting point: a well-defined skill, a credible comparison, reliable measurement, and delayed testing. Add neural measurement when there is a specific biological question and a method capable of answering it. A research partnership can then supply the instrumentation and expertise that question requires.

For a claim about neural mechanisms, build a longitudinal design around a predicted neural feature and its relationship to behavior. For a claim about an individual’s plasticity capacity, establish repeatability and predictive value in that intended use. For a claim about rehabilitation or everyday benefit, measure activity and participation where the change is supposed to matter.

A useful final report would make five things explicit: the experience that was tested, the neural or behavioral property that changed, the comparison supporting attribution, the interval over which the change persisted, and the circumstances in which the person could use the resulting capability.

The scientific opportunity is to connect those observations carefully. The instrument reveals a signal. The experiment establishes what changed. The follow-up shows whether the change endured. The person’s performance in a meaningful setting shows what it made possible.

References #

The sources below include primary experiments, methodological research, reviews, and official frameworks. Software documentation is linked in the tools section. The proposed study schedule and numerical examples in this article are illustrative.

  1. Zatorre, R. J., Fields, R. D., and Johansen-Berg, H. (2012). Plasticity in gray and white: Neuroimaging changes in brain structure during learning. Nature Neuroscience, 15, 528–536. Read source

    Review of imaging measures and possible cellular mechanisms.

    Back to citation 1 Back to citation 2

  2. Kleim, J. A., and Jones, T. A. (2008). Principles of experience-dependent neural plasticity: Implications for rehabilitation after brain damage. Journal of Speech, Language, and Hearing Research, 51, S225–S239. Read source

    Framework developed in a rehabilitation context.

    Back to citation 1

  3. Lövdén, M., Bäckman, L., Lindenberger, U., Schaefer, S., and Schmiedek, F. (2010). A theoretical framework for the study of adult cognitive plasticity. Psychological Bulletin, 136(4), 659–676. Read source

    Theory of demands, flexibility, and plasticity.

    Back to citation 1

  4. National Institute of Mental Health. About RDoC. Read source

    Official description of the Research Domain Criteria framework.

    Back to citation 1

  5. FDA-NIH Biomarker Working Group. BEST (Biomarkers, EndpointS, and other Tools) Resource: Validation. Read source

    Official definitions of analytical and clinical validation.

    Back to citation 1

  6. World Health Organization. (2002). Towards a Common Language for Functioning, Disability and Health: ICF. Read source

    Official introductory guide explaining body functions and structures, activities, participation, and context.

    Back to citation 1

  7. Scholz, J., Klein, M. C., Behrens, T. E. J., and Johansen-Berg, H. (2009). Training induces changes in white-matter architecture. Nature Neuroscience, 12, 1370–1371. Read source

    Longitudinal diffusion-imaging study of juggling training.

    Back to citation 1

  8. Jones, D. K., Knösche, T. R., and Turner, R. (2013). White matter integrity, fiber count, and other fallacies: The do’s and don’ts of diffusion MRI. NeuroImage, 73, 239–254. Read source

    Methodological analysis of diffusion-imaging interpretations.

    Back to citation 1

  9. Wenger, E., et al. (2017). Repeated structural imaging reveals nonlinear progression of experience-dependent volume changes in human motor cortex. Cerebral Cortex, 27(5), 2911–2925. Read source

    Repeated imaging during motor learning.

    Back to citation 1

  10. Logothetis, N. K., Pauls, J., Augath, M., Trinath, T., and Oeltermann, A. (2001). Neurophysiological investigation of the basis of the fMRI signal. Nature, 412, 150–157. Read source

    Simultaneous electrophysiology and imaging in macaque visual cortex.

    Back to citation 1

  11. Elliott, M. L., et al. (2020). What is the test-retest reliability of common task-functional MRI measures? New empirical evidence and a meta-analysis. Psychological Science, 31(7), 792–806. Read source

    Reliability evidence relevant to individual-difference measurements.

    Back to citation 1

  12. Pernet, C., et al. (2020). Issues and recommendations from the OHBM COBIDAS MEEG committee for reproducible EEG and MEG research. Nature Neuroscience, 23, 1473–1483. Read source

    Consensus recommendations for reproducible EEG and MEG research.

    Back to citation 1 Back to citation 2

  13. Teyler, T. J., et al. (2005). Long-term potentiation of human visual evoked responses. European Journal of Neuroscience, 21(7), 2045–2050. Read source

    Early human visual-evoked-response potentiation study.

    Back to citation 1

  14. Dias, J. W., McClaskey, C. M., Rumschlag, J. A., and Harris, K. C. (2022). Sensory tetanisation to induce long-term-potentiation-like plasticity: A review and reassessment of the approach. European Journal of Neuroscience, 56(12), 6115–6140. Read source

    New experiments, review, and meta-analyses examining assay reliability and validity.

    Back to citation 1

  15. Galuba, V., et al. (2025). Development of VEP-based biomarkers to assess plasticity states. Translational Psychiatry, 15, 426. Read source

    Comparison of candidate visual-evoked-potential modulation protocols.

    Back to citation 1

  16. Yücel, M. A., et al. (2021). Best practices for fNIRS publications. Neurophotonics, 8(1), 012101. Read source

    Acquisition, analysis, and reporting guidance for fNIRS.

    Back to citation 1 Back to citation 2

  17. Stefan, K., Kunesch, E., Cohen, L. G., Benecke, R., and Classen, J. (2000). Induction of plasticity in the human motor cortex by paired associative stimulation. Brain, 123(3), 572–584. Read source

    Human study of an associative stimulation assay.

    Back to citation 1

  18. Hamada, M., Murase, N., Hasan, A., Balaratnam, M., and Rothwell, J. C. (2013). The role of interneuron networks in driving human motor cortical plasticity. Cerebral Cortex, 23(7), 1593–1605. Read source

    Study of variation in physiological responses to stimulation.

    Back to citation 1

  19. Wilson, M., et al. (2019). Methodological consensus on clinical proton MRS of the brain: Review and recommendations. Magnetic Resonance in Medicine, 82(2), 527–550. Read source

    Consensus guidance for magnetic resonance spectroscopy.

    Back to citation 1

  20. Finnema, S. J., et al. (2016). Imaging synaptic density in the living human brain. Science Translational Medicine, 8(348), 348ra96. Read source

    PET study targeting the synaptic vesicle protein SV2A.

    Back to citation 1

  21. Ros, T., et al. (2020). Consensus on the reporting and experimental design of clinical and cognitive-behavioural neurofeedback studies (CRED-nf checklist). Brain, 143(6), 1674–1685. Read source

    Consensus framework for designing and reporting neurofeedback studies.

    Back to citation 1

  22. Soderstrom, N. C., and Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176–199. Read source

    Distinction between practice performance and durable learning.

    Back to citation 1

  23. Barnett, S. M., and Ceci, S. J. (2002). When and where do we apply what we learn? A taxonomy for far transfer. Psychological Bulletin, 128(4), 612–637. Read source

    Framework for describing transfer across tasks and contexts.

    Back to citation 1

  24. Taylor, J. A., Krakauer, J. W., and Ivry, R. B. (2014). Explicit and implicit contributions to learning in a sensorimotor adaptation task. Journal of Neuroscience, 34(8), 3023–3032. Read source

    Experiment separating contributions to visuomotor adaptation.

    Back to citation 1

  25. Wilson, R. C., and Collins, A. G. E. (2019). Ten simple rules for the computational modeling of behavioral data. eLife, 8, e49547. Read source

    Guidance on simulation, identification, recovery, and model evaluation.

    Back to citation 1

  26. Hedge, C., Powell, G., and Sumner, P. (2018). The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences. Behavior Research Methods, 50, 1166–1186. Read source

    Empirical investigation of behavioral-task reliability.

    Back to citation 1

  27. Jacobson, N. S., and Truax, P. (1991). Clinical significance: A statistical approach to defining meaningful change in psychotherapy research. Journal of Consulting and Clinical Psychology, 59(1), 12–19. Read source

    Reliable change index and distinction from clinical significance.

    Back to citation 1

  28. Roediger, H. L., III, and Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. Read source

    Experiments showing that assessment can affect later learning.

    Back to citation 1

  29. Vickers, A. J., and Altman, D. G. (2001). Analysing controlled trials with baseline and follow up measurements. BMJ, 323, 1123–1124. Read source

    Methods guidance for baseline-adjusted comparisons.

    Back to citation 1

  30. Lakens, D. (2022). Sample size justification. Collabra: Psychology, 8(1), 33267. Read source

    Methodological guidance on effect sizes, precision, and study planning.

    Back to citation 1

  31. Gorgolewski, K. J., et al. (2016). The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments. Scientific Data, 3, 160044. Read source

    Foundational description of BIDS.

    Back to citation 1

Cite this article #

When referencing this article, credit Paul Greenberg and RapidKT and link to the article. Include the section title when directing readers to a particular passage.

Greenberg, P. (2026, October 28). How to Measure Neuroplasticity. RapidKT. https://rapidkt.com/pages/blog/how_to_measure_neuroplasticity

RapidKT Plain Text Website