1 of 32

Data Sharing in EU

August 2025

Dr Cyril Pernet, PhD

cyril.pernet@nru.dk

@cyrilrpernet.bsky.social

2 of 32

Article 1(2) “The protection of natural persons in relation to the processing of personal data is a fundamental right.”

Recital 26 data-protection principles do not apply to anonymous information, but they do apply to pseudonymised data, because re-identification remains reasonably likely.

3 of 32

Meet Dr Commūnicāre

  • Dr. Commūnicāre is a medical doctor and a brain imager. She has two problems. Let’s help her out.

4 of 32

Problem number 1

One of her patients has a rare condition. She wants to send the DICOM file, with the patient’s name and details, to her colleague in France.

👉 Hands up: how many think this is allowed under GDPR?

👉 And how many think she must first get explicit patient consent?

5 of 32

What does the law say:

GDPR Article 6(1)(d) → processing is lawful if “necessary in order to protect the vital interests of the data subject or of another natural person.”

GDPR Article 6(1)(e) → processing is lawful if “necessary for the performance of a task carried out in the public interest or in the exercise of official authority vested in the controller.”

GDPR Article 9(2)(h) → explicitly allows processing of special categories of data (including health data) when “necessary for the purposes of preventive or occupational medicine, medical diagnosis, the provision of health or social care or treatment, or the management of health or social care systems …”

6 of 32

We don’t need to help her

Medical professionals don’t need consent to share personal health data with other professionals if it’s necessary for diagnosis or treatment. That’s exactly Dr. Commūnicāre’s case. She is doing her job.

  • The transfer within the EU (to another doctor bound by medical confidentiality and GDPR) is lawful and does not require consent.

Art 1(3): shall not restrict or prohibit the free movement of personal data within the Union

7 of 32

What is Privacy?

Data privacy: A set of principles and governance rules that define who may access which data and under what conditions. (technical)

Information privacy: “The claim of individuals … to determine for themselves when, how, and to what extent information about them is communicated to others.” (Westin, 1967 - philosophical)

8 of 32

Problem number 2

“Now, she wants to share a BIDS dataset of all her patients with other scientists.

👉 removing all ID = pseudonymization – can she share openly?

(data privacy issue)

9 of 32

  • concrete means
  • cost and the know-how
  • likelihood

Access to freely available datasets like post-codes, census, etc

Gaussian-copula modelling from (incomplete) metadata - not attempt to re-identify here, estimate likelihood of success given the data

  • Julia and Python (free)
  • Good stats, data science knowledge

> 95%

Re-identification and singling-out subjects

Metadata in neuroimaging !!

10 of 32

blurred

zero-ed

Re-identification from structural MRI

Loss of potential information!! defacing is likely not a solution anyway

  • Removing facial features degrades perf of image analysis method
  • De-identification [..] impact on structural brain measures
  • Defacing biases manual and automated quality assessments

11 of 32

Re-identification from PET

12 of 32

Re-identification from f-MRI

13 of 32

Re-identification from MEEG

good discriminability, stability over time, and tightly correlated with molecular expression/genes

14 of 32

  • concrete means
  • cost and the know-how
  • likelihood

Refacing

Face matching

GANS, CNN, training is hard

Access to freely available datasets photos like Facebook

- Reconstructing 3D MRI is mega easy

- read the doc to use automated face-recognition software

80-90%

(Schwarz, et al. 2019)

Re-identification from connectome

Access to multiple datasets and possibly original data

Imaging and connectomic knowledge, HPC - infer subject across sets, task performance, possibly ID if access to some original data

~ 90%

Re-identification from time series

Access to metadata outside the dataset to link genetic/molecular markers

Time-series and Spectral analysis knowledge

??

Re-identification in neuroimaging

15 of 32

Neuroimaging data types under GDPR

Adapted from Poldrack & Gorgolewski (2014) https://www.nature.com/articles/nn.3818

Potential for reuse

Raw (+ clinical) + processed data

Personal

Raw data

Subject level summary data

Anonymous

/ Personal ?? lot of work to do

Anonymous

Group level summary data

Personal

Pseudonymization is a process – there is only two types of data

16 of 32

Problem number 2

“Now, she wants to share a BIDS dataset of all her patients with other scientists.

�👉 removing all ID = pseudonymization – can she share openly?

I’d say, no – imaging data remains personal (+ clinical data) and thus she needs to share under restricted access

👉 sharing with others – does she need consent for that?

(information privacy issue)

17 of 32

GDRP - again

Article 6(1)(e) – processing is lawful if “necessary for the performance of a task carried out in the public interest”. In most EU countries, scientific research by healthcare institutions and universities is considered such a task.

Article 9(2)(j) – allows processing of special categories (health data) if “necessary for archiving purposes in the public interest, scientific or historical research purposes or statistical purposes … subject to Article 89(1) safeguards.”

Article 89(1) – requires safeguards like pseudonymization, minimization, and governance.

  • consent is not strictly required for sharing scientific research

(consent to be in a study obviously ≠ consent to share)

18 of 32

GDRP - again

Article 5(1)(b) – Purpose limitation: Data collected for one purpose can be further processed for scientific research purposes, provided Article 89 safeguards are applied.

Article 5(1)(e) – Storage limitation: Data may be stored longer “for archiving purposes in the public interest, scientific or historical research purposes or statistical purposes” (again subject to safeguards in Art. 89).

🡪 Researchers have the flexibility on purpose limitation, storage and certain data-subject rights; in exchange, they must demonstrate state-of-the-art safeguards that keep data from becoming a vector of re-identification or discrimination.

19 of 32

Declaration of Taipei (2016)

Ethical use of health data beyond current studies.�Importance of transparency, patient autonomy, and trust.�→ Key idea: Patients’ right to control future use of their data.

Dynamic consent is a pretty tough ask, but at least we need to inform people on sharing, the consequences of sharing, and the ability to revoke access to people’s data.

20 of 32

Problem number 2

“Now, she wants to share a BIDS dataset of all her patients with other scientists.

�👉 removing all ID = pseudonymization – can she share openly?

I’d say, no – imaging data remains personal (+ clinical data) and thus she needs to share under restricted access

👉 sharing with others – does she need consent for that?

I’d say, at least information must be given about the sharing process, consequences, and ability to withdraw data + come up with a firm data sharing policy (~ license)

21 of 32

Explain how it will be shared

Explain re-usage, any limitations?

Consequences of sharing

For how long are data kept

How can people remove their data if they want to

If you want a consent to share …

22 of 32

A solution for sharing

PublicnEUro.eu

  • Hosts and publishes (DOI) BIDS datasets, making them findable and accessible.
  • Public Data: making a dataset information public (open metadata, controlled access for download).
  • Options from fully open (e.g. animal) to fully closed (embargo period for consortia) with ‘all’ intermediate options of control.

23 of 32

What does it looks like?

24 of 32

What does it looks like?

Let’s start by looking at an Open Dataset (CCBY)

  • No Need to be registered
  • Click and download
  • PN000001

25 of 32

What about other datasets?

Each dataset comes with it’s own rules -- i.e. you tell us how you need your data to be legally shared and the platform accommodates your need.

26 of 32

What about other datasets?

Each dataset comes with it’s own rules -- i.e. you tell us how you need your data to be legally shared and the platform accommodates your need.

→ not only you can browse but it lists all sorts of metadata

27 of 32

What about other datasets?

Each dataset comes with it’s own rules -- i.e. you tell us how you need your data to be legally shared and the platform accommodates your need.

→ not only you can browse but it lists all sorts of metadata

28 of 32

What about other datasets?

Each dataset comes with it’s own rules -- i.e. you tell us how you need your data to be legally shared and the platform accommodates your need.

→ not only you can browse but it lists all sorts of metadata

PN000004 approve DUA online and download

29 of 32

What about other datasets?

PN000002 sign paperwork and re-upload, await approval, receive the link to download

(login as wam from Home, USA)

→ Note the SSC is something related to the institution, some layer/DPO/HoD has to approve this (and then works for all users)

30 of 32

You contract us

(yes there is a small fee for service and hosting)

31 of 32

Sharing on PublicnEUro for Dr Commūnicāre

  • Data hosting (Accessible)
  • Metadata enhancement and searchability (Findable)
  • Digital Object Identifier (Citable)
  • Unload the administrative burden
  • data access is managed for her
  • She can comply with her institution requirement (subcontract signing of DUA and SCC / oversight)

32 of 32

Thank you for your attention: let’s free imaging data!