University of Turku Human Neuroscience Master's Thesis Guide: TBMC4001, 40 ECTS, Brain Imaging, Behaviour and UTUGradu

Quick answer: what is the University of Turku Human Neuroscience thesis route?

The current University of Turku Master’s Degree Programme in Human Neuroscience is a 120 ECTS, two-year Master of Science programme organised by the Turku Brain and Mind Center (TBMC) within the Faculty of Medicine, with collaboration across Medicine, Social Sciences and Technology. The controlling Peppi programme is MDPHN2427 / programme 98833. The exact thesis is TBMC4001 Master’s Thesis in Human Neuroscience, 40 ECTS, an Advanced Studies course graded 0–5. The thesis plan must be accepted by the TBMC board before hands-on work starts. A separate 5 ECTS Human Neuroscience Seminars module also exists, but those credits are outside the 40 ECTS thesis and must not be double-counted.

1. Start from the exact 120 ECTS Peppi structure

Peppi is unusually clean for this programme: the root object is exactly 120 ECTS. It contains 103 ECTS Major Subject Studies in Human Neuroscience and 17 ECTS elective courses. The 103-credit major contains Human Neuroscience Seminars 5 ECTS, Brain Imaging Methods 15, Data Analysis 10, Clinical Neuroscience 23, Behavioral Methods 10 and TBMC4001 thesis 40. This arithmetic matters because the thesis and seminar module are both already inside the 120 ECTS degree.

2. TBMC4001 is the exact current thesis course

TBMC4001 is a 40 ECTS Advanced Studies course assessed on the 0–5 scale. Its purpose is not merely to produce a long document. Students learn rigorous scientific practice, writing, presentation and the methodology used in the research group. The public programme page likewise describes a research-plan, seminar, practical-research and report process based on behavioural and/or brain-imaging data.

3. The TBMC-board plan gate comes before hands-on work

At the beginning of the thesis, the student creates a thesis plan with the supervisors. Once submitted, the plan must be accepted by the TBMC board before hands-on work begins. Do not replace this with the approval workflow of Biomedical Imaging, Drug Discovery and Development or another Faculty programme. Before the gate, settle the question, dataset or acquisition plan, supervision, ethics route, data access, safety controls and analysis strategy.

4. Plan for an approximately eight-month thesis process

TBMC4001 estimates approximately eight months for combined hands-on research and writing. The course also identifies lectures, assignments, presentation work, reading and a thesis-plan component. Treat this as an integrated research period, not a final-semester writing sprint. Build time for ethics or permit decisions, recruitment or access delays, pilot work, failed scans, preprocessing, quality control, analysis, interpretation and revision.

5. The separate Human Neuroscience Seminars module is 5 ECTS

Peppi places a 5 ECTS Human Neuroscience Seminars module beside the thesis inside the 103 ECTS major. It resolves into TBMC0999 Tutorial Group 2 ECTS, TBMC0003 Career Seminar 1 ECTS and TBMC0002 Human Neuroscience Symposium 2 ECTS. The Symposium requires students to present results in a scientific meeting. These are separate degree credits and are not hidden inside TBMC4001.

6. Do not invent a separate extra thesis-seminar credit

The public programme page says Human Neuroscience Seminars include lab, thesis and career-related seminars, while the thesis description mentions participation in a research seminar. Peppi nevertheless resolves the credited seminar block into Tutorial Group, Career Seminar and Symposium. The safe interpretation is 40 ECTS thesis + separate 5 ECTS seminar module, not 40 ECTS plus an additional unlisted thesis-seminar credit.

7. Know the supervision and examination structure

At least one supervisor must hold a PhD. The thesis has at least two examiners, both with PhDs. One examiner is an expert outside the research group and the supervisor acts as the other examiner. The TBMC board appoints examiners under Faculty authorization, and the head of Clinical Department decides acceptance and the final grade on the basis of examiner opinions. This structure matters when planning independence and conflicts of interest.

8. Human Neuroscience combines brain and behavioural evidence

The programme trains students to measure human brain function and cognition/performance. The environment includes PET, MRI/fMRI, EEG, TMS and neurophysiology, while behavioural courses address cognition, perception, language, emotion and social behaviour. A strong thesis should identify the evidence level represented by each result instead of collapsing every signal into a generic claim about “the brain.”

9. Build a measurement chain before choosing statistics

Write the evidence chain from participant and task through acquisition, preprocessing, QC, feature extraction, modelling and final figure. For behavioural work, include stimulus, task, timing, scoring and exclusions. For imaging or EEG, include acquisition protocol, reconstruction or referencing, preprocessing, artefact handling and statistical model. A result is only as interpretable as the transformations connecting raw measurement to the endpoint.

10. EEG and ERP require signal-processing provenance

TBMC1013 covers EEG signal generation, artefact rejection, filtering, ERP experiments, collection and analysis. Record reference scheme, filters, bad-channel handling, epoch windows, baseline correction, trial rejection and participant exclusion. If settings change during troubleshooting, version them. A clean-looking ERP is not sufficient evidence unless the processing route and usable trial counts are transparent.

11. Separate trial counts from participant counts

Hundreds of EEG trials from ten participants do not create hundreds of independent people. Multiple sensors, epochs or time points from one person are repeated observations. Report participant-level denominator and lower-level observation counts separately. If inference is at participant level, the model must respect that hierarchy.

12. MRI/fMRI measurement depends on acquisition and preprocessing

TBMC1012 covers MRI physics, image formation, reconstruction, contrast, artefacts, safety and post-processing. Structural MRI, fMRI, spectroscopy and diffusion imaging are not interchangeable. Record scanner and sequence parameters that affect the result, document motion and artefacts, preserve preprocessing versions and explain transformations such as spatial normalisation or smoothing when they matter.

13. Motion is a scientific variable, not only a technical nuisance

Motion can differ by age, diagnosis, task difficulty or participant group. If high-motion sessions are removed, report who was lost and why. If motion regressors or censoring are used, describe them. A group difference that is also a motion difference needs sensitivity analysis before it is interpreted as neural.

14. PET signals need tracer and modelling context

PET is a quantitative imaging method using positron-emitting tracers to measure biochemical processes in living subjects. State which tracer and biological process are relevant, what model or outcome quantity is used and what assumptions connect the signal to the claim. A PET image is not simply a map of “brain activity,” and a research PET association is not automatically a diagnostic test.

15. PET creates a radiation and governance boundary

The PET curriculum explicitly involves radioactive tracers in tracer quantities. A thesis using PET should separate scientific analysis responsibility from radiotracer production, exposure control and clinical or research authorization. Students should work within the approved facility and protocol. Coursework does not itself authorize independent clinical radiation procedures.

16. TMS can support perturbational inference, but controls matter

TBMC5003 covers TMS safety, motor-threshold measurements, paired-pulse paradigms, online/offline perturbation, repetitive stimulation and clinical examples. If a thesis uses TMS to argue causal contribution, explain protocol, target, timing, sham/control and sensory or expectancy confounds. TMS can provide stronger causal leverage than correlation, but only within the design.

17. Clinical neurophysiology is not automatically clinical diagnosis

TBMC2007 covers EEG/video-EEG, ENMG, evoked responses, TMS-related recordings, sensory testing and sleep methods, including electrical safety and artefact recognition. A research recording can use clinically established technology without constituting diagnosis. State whether the dataset is experimental, retrospective clinical material or generated within a clinical workflow.

18. Behavioural measures need construct validity

Reaction time, accuracy, eye movement, rating, recall, recognition, task performance and self-report measure different things. Define what the outcome represents and how it is scored. Practice, fatigue, order, attention and exclusions can alter performance. Do not use “cognition” as if all behavioural outcomes were interchangeable.

19. Brain-behaviour correlations do not prove causal direction

A correlation between an imaging feature and behavioural score can be meaningful, but it does not establish which variable causes the other or whether a third factor explains both. Report effect size, uncertainty, sample size and plausible alternatives. Causal language requires causal leverage from the design.

20. High-dimensional neuroimaging needs multiplicity control

Neuroimaging can involve thousands of voxels, vertices, channels, time points or features. Searching many locations creates a multiple-comparison problem. State the correction or modelling strategy, search space and whether hypotheses were pre-specified or exploratory. A visually striking cluster is not enough without defensible inference.

21. Quality control belongs inside the scientific method

TBMC1019 explicitly teaches QC of structural, functional and diffusion data and outputs. Preserve QC outcomes, not only successful images. Define what causes a scan or processing output to fail. Log exclusions and inspect whether they differ systematically between groups. A dataset can become biased if difficult participants disappear without documentation.

22. Pipeline choice can change the answer

The neuroimaging curriculum explicitly notes that tools and approaches can affect processing and statistical results. Record software, container, script and parameter versions when they matter. If the project changes pipeline, do not silently pool outputs. Check whether the change alters headline findings or explain why comparability is defensible.

23. Keep manual intervention visible

Manual reorientation, ROI editing, artefact marking, bad-channel selection or quality rating can introduce analyst judgement. Document who performed the step, what criteria were used and whether raters were blinded when relevant. If reasonable alternative settings change the conclusion, show that sensitivity.

24. Statistical analysis starts with the independent unit

TKO_7093 covers data handling, exploration, testing and suitable method choice. Before choosing a test, define the independent unit. Multiple scans from one person, trials in one session or derived features from one image do not automatically increase independent participants. Report missingness and exclusions before the final model.

25. Machine-learning validation must be participant-safe

TKO_3103 explicitly covers model selection, holdout validation and cross-validation. If each participant contributes many images, windows or trials, keep correlated observations from one participant in the same model-development partition. Feature selection, normalisation and hyperparameter tuning must be performed without using final test outcomes.

26. AI accuracy is not brain-mechanism evidence

TBMC5009 emphasises AI applications, interpretability and biological plausibility. A classifier can predict labels using nuisance structure, acquisition differences or confounded demographics. High accuracy alone does not prove that the model discovered a neural mechanism. Explain what the model predicts, what information it saw and how explanations were validated.

27. Retrospective performance is not prospective clinical utility

Human Neuroscience can use patient or dementia-imaging datasets. A retrospective model or group comparison can be clinically interesting without being ready for patient care. Diagnostic or prognostic claims require appropriate reference standards, patient selection, calibration and validation, ideally prospectively. Keep research classification, clinical association and decision support separate.

28. Human research ethics depend on the actual design

University of Turku distinguishes human-sciences ethical review from medical-research assessment. The correct route depends on participants, procedures, intervention, data source and legal context. A non-invasive label does not mean “no ethics.” Resolve the applicable route before recruitment, acquisition or other hands-on work covered by the project.

29. Consent and risk communication must match the protocol

Participants should be told what is actually being done: behavioural tasks, MRI, EEG, TMS, PET, reuse of clinical data or another procedure. Information should cover relevant burdens, storage, secondary use and withdrawal. For TMS, PET or MRI, method-specific safety and eligibility should be handled through approved facility procedures.

30. Brain data can remain identifiable

Names are not the only identifiers. Imaging headers, acquisition dates, facial anatomy in structural MRI, rare diagnoses, clinical variables and demographic combinations can enable linkage. De-identification should cover visible labels and metadata. If a re-identification key exists, keep it separately controlled. Pseudonymised data remain sensitive research data.

31. Research permits are conditional and source-specific

A University research permit may be relevant for University units, staff, students or datasets, while hospital, company or collaborating-group data can have separate access procedures. A permit does not replace consent, ethical review or lawful processing. Map each approval to the cohort, source and setting before the TBMC-board plan gate.

32. Company collaboration requires an independence plan

The programme allows collaboration with research groups or companies. Document who owns data, code or devices, what can appear in the public thesis, confidentiality boundaries and whether negative findings may be reported. A partner’s preferred interpretation is not independent scientific evidence. Resolve disclosure rules early enough to keep the assessed thesis defensible.

33. AI use does not transfer scholarly responsibility

AI can support code, language or exploratory analysis where current rules allow, but the student remains responsible for sources, accuracy, confidentiality and reproducibility. Do not upload identifiable brain images, clinical records or proprietary company data to uncontrolled services. Re-run AI-generated code and verify citations against original sources.

34. Preserve a participant-to-figure provenance chain

For headline results, trace final figure back to participant/session, acquisition protocol, raw or minimally processed source, preprocessing version, QC outcome, model and plotting script. Protected raw data do not need to be public, but the research team should be able to reconstruct the result. Failed scans and excluded sessions remain part of the audit trail.

A second audit layer should examine modality-specific provenance. For EEG, preserve channel montage, reference, filter and epoch definitions. For MRI/fMRI, preserve scanner, sequence, reconstruction, motion handling and preprocessing version. For PET, preserve tracer, acquisition timing, reconstruction and kinetic or reference-region model where relevant. For TMS, preserve device, coil, target definition, thresholding procedure, intensity and timing. These details are not decorative metadata. They define what the numerical result actually means and whether sessions or participants are comparable.

When a thesis combines modalities, document the alignment step explicitly. A behavioural score collected on one day, an MRI scan on another and an EEG task on a third are not automatically simultaneous measurements of the same state. Explain participant matching, session interval, task equivalence, spatial registration or temporal alignment as applicable. If one modality has more missing cases than another, report the multimodal analysis sample separately rather than silently dropping participants.

Participant attrition deserves its own table when it is material. Start from people screened or datasets available, then show exclusions for eligibility, consent, acquisition failure, motion, corrupted files, preprocessing failure and final analysis. If exclusion differs by diagnosis, age, task performance or another study variable, that can bias the final sample. A technically perfect final dataset can still be scientifically misleading if the difficult participants were selectively lost.

Clinical labels also need provenance. A diagnosis from a medical record, a research interview, a screening questionnaire and a self-reported condition are not equivalent reference standards. State who assigned the label, under what criteria and at what time relative to imaging. If diagnostic uncertainty or mixed pathology is expected, discuss it rather than using the label as unquestionable ground truth for an AI or group-comparison model.

Safety and consent should be tied to the exact protocol version. If a sequence, stimulation intensity, tracer procedure or task burden changes after the thesis plan, verify whether the existing approval and participant information still cover the change. A minor technical adjustment can be scientifically harmless yet still require documentation; a substantive procedure change can affect risk, consent or ethics status. Keep a decision log with the supervisor and responsible facility.

35. Keep exploratory and confirmatory analyses separate

Neuroscience datasets invite many analyses. If a finding emerged after inspecting the data, label it exploratory rather than rewriting the story as if pre-specified. Preserve the original question and analysis plan, record protocol changes and use sensitivity analyses where reasonable. Exploratory findings need independent confirmation before strong general claims.

36. Turnitin and UTUGradu are separate final controls

University guidance makes Turnitin part of degree-thesis originality checking. UTUGradu manages submission, examination, approval, publication and archiving. Turnitin does not validate statistics, imaging quality or ethics, while a technically strong thesis still must satisfy originality and submission requirements.

37. Design the public thesis around confidentiality

The assessed thesis enters a library/repository workflow. Restricted visibility is not a substitute for removing identifiable scans, rare-case details, clinical notes or proprietary company information. Decide early what can appear publicly, what must be aggregated or anonymised and what remains only in controlled storage.

38. Final Human Neuroscience checklist

Confirm MDPHN2427 / programme 98833, exact TBMC4001 40 ECTS, 0–5 grading, the TBMC-board thesis-plan gate before hands-on work, and the separate 5 ECTS Human Neuroscience Seminars module. Then verify supervisors/examiners, approvals, modality safety, EEG/MRI/PET/TMS provenance, behavioural construct validity, experimental unit, QC and exclusions, multiplicity, ML leakage, human/clinical claim boundaries, privacy, AI use, figure provenance, Turnitin, UTUGradu and public-thesis confidentiality.

Additional reproducibility controls

A reproducibility table can help before manuscript freeze. For each major dataset, list participant cohort, modality, acquisition version, preprocessing pipeline, primary QC rule, independent unit, primary endpoint, statistical model and allowed inference. This makes it easier to detect when a conclusion has crossed from signal quality to brain association, from brain association to behaviour, or from research evidence to a clinical statement. Preserve negative acquisitions and failed processing outputs in the internal ledger even when they are not displayed in the final article.

When a thesis combines modalities, document alignment explicitly. A behavioural score collected on one day, an MRI scan on another and an EEG task on a third are not automatically simultaneous measurements of the same state. Explain participant matching, session interval, task equivalence, spatial registration or temporal alignment as applicable. If one modality has more missing cases than another, report the multimodal analysis sample separately instead of silently dropping participants.

Participant attrition deserves its own table when material. Start from people screened or datasets available, then show exclusions for eligibility, consent, acquisition failure, motion, corrupted files, preprocessing failure and final analysis. If exclusion differs by diagnosis, age, task performance or another study variable, the final sample can be biased. A technically perfect final dataset can still be misleading if difficult participants were selectively lost.

Clinical labels also need provenance. A diagnosis from a medical record, research interview, screening questionnaire and self-reported condition are not equivalent reference standards. State who assigned the label, under what criteria and at what time relative to imaging. If diagnostic uncertainty or mixed pathology is expected, discuss it rather than using the label as unquestionable ground truth for an AI or group-comparison model.

Safety and consent should be tied to the exact protocol version. If a sequence, stimulation intensity, tracer procedure or task burden changes after the thesis plan, verify whether the existing approval and participant information still cover the change. A minor technical adjustment can require documentation; a substantive procedure change can affect risk, consent or ethics status. Keep a decision log with the supervisor and responsible facility.