Oxford ML researcher using causal generative modelling and counterfactual inference for medical imaging
3
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
No data yet: this axis would measure the volume and trajectory of research output, such as total publications and output over time. The catalogued record — three 2026 arXiv entries — is too thin to score.
Impact—
No citation or influence data is available yet. This axis would measure citation counts, uptake and scholarly influence of the work; with all catalogued papers being recent preprints, no impact score can be computed.
Focus0/100
Scores 0 on field concentration: the available field distribution splits output evenly — 50% labelled machine learning — medical image analysis (causal generative modelling) and 50% machine learning for medical image analysis. The two labels are adjacent subfields of the same area, so the low score reflects even division across labels rather than scattered interests; the core theme remains causal counterfactual methods for medical imaging.
Collaboration39/100
Scores 39, based on 6 distinct coauthors across the catalogued work. The recurring circle includes Konstantinos Kamnitsas, Robin J. Evans, Hermione Warr, Harry Anthony, Lilli J Freischem and Daniel R McGowan, spanning the causal-modelling and clinical PET/CT sides of the work. Both first-author preprints were written within a small team rather than a large consortium, indicating a compact, supervision-anchored collaboration pattern.
Activity—
Insufficient data: this axis would measure publishing cadence and recent research activity, such as outputs per year and intervals between releases. All three catalogued papers fall in 2026, leaving no multi-year activity pattern to score.
Reproducibility55/100
Scores 55, computed as a reproducibility mean of 55 across 2 assessed papers, each rated medium reproducibility readiness. The two first-author preprints present broad, coherent experiment suites, but Evolve's reviews flag reliance on a single architecture, thin statistical reporting, and validation restricted to synthetically injected biases — documented limitations that currently cap confidence in independent reproduction.
Research summary
Yasin Ibrahim is an early-stage machine learning researcher at the University of Oxford whose catalogued work centres on counterfactual and causal generative methods for medical image analysis. Two 2026 first-author arXiv preprints define the core research line. The first, "To do(x) or not to do(x)", proposes causal conditioning of counterfactual medical images for dataset augmentation, organised through a three-way taxonomy; Evolve's review finds the experiments broad and convincing while noting reliance on a single architecture and thin statistics. The second introduces counterfactual marginalisation, a framework that uses structural causal model counterfactuals to evaluate robustness to nuisance variables and audit classifier bias, rated promising though validated only on synthetically injected biases. A third 2026 co-authored paper compares domain-specific models with large language models for error detection in PET/CT radiology reports, extending the portfolio toward radiology-report NLP. Bibliometrically the profile is compact: 6 distinct coauthors (collaboration score 39), an even 50/50 split across two adjacent field labels (focus score 0), and a reproducibility mean of 55 across 2 assessed papers, both rated medium. Output, impact and activity axes lack sufficient underlying data and are left unscored, so this should be read as an early-stage profile built on a small preprint corpus.
Collaboration pattern
The catalogued work is produced within a compact Oxford circle of 6 distinct coauthors. Frequent partners are Konstantinos Kamnitsas, Robin J. Evans, Hermione Warr, Harry Anthony, Lilli J Freischem and Daniel R McGowan. The two first-author causal-modelling preprints sit alongside a co-authored PET/CT radiology-report study, suggesting a pattern of pairing counterfactual method development with clinically oriented collaborators.
Research keywords
Highlights
Medical image counterfactuals for dataset augmentation
Counterfactual marginalisation for auditing nuisance-variable robustness
Error detection in PET/CT reports: domain-specific models vs LLMs
Co-authored 2026 arXiv study comparing domain-specific models with large language models for detecting errors in PET/CT radiology reports, extending the portfolio toward radiology-report NLP; not yet reviewed by Evolve.
Timeline
- 2026First-author arXiv preprint 'To do(x) or not to do(x)' proposes medical image counterfactuals for dataset augmentation through a three-way taxonomy.
- 2026First-author counterfactual marginalisation framework evaluates robustness to nuisance variables using structural causal model counterfactuals.
- 2026Co-authored comparison of domain-specific models and large language models for error detection in PET/CT radiology reports.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric scores are computed from public bibliometric data; all catalogued papers are 2026 arXiv preprints without peer-review or citation records.
- Output, impact and activity axes have insufficient underlying data and are left unscored rather than estimated.
- The PET/CT radiology-report paper is catalogued without an Evolve review, keywords or reproducibility assessment.
- Profile is name-resolved from public bibliographic data linked to ORCID 0009-0007-5877-1309; no biographical details are asserted.