Oxford ML researcher applying causal generative modelling and counterfactuals to medical imaging and trustworthy AI.
3
papers here
2
reviewed by Evolve
0
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—
Not yet scored: the snapshot lacks the publication-count and output-volume metrics this axis measures, so Evolve cannot currently quantify research output; only three 2026 catalogue papers are available.
Impact—
Not yet scored: citation counts, h-index and influence metrics are absent from the provided data, so this axis — which would measure community uptake of the work — cannot be reported.
Focus0/100
Scores 0 on focus because the field distribution is split exactly 50/50 between two labels — 'machine learning — medical image analysis (causal generative modelling)' and 'machine learning for medical image analysis'. In practice the two labels describe one coherent specialisation, so the zero reflects label fragmentation across closely adjacent fields rather than genuinely scattered interests; all three catalogue papers sit in medical image analysis.
Collaboration39/100
Scores 39 on collaboration, based on 6 distinct coauthors in the snapshot — a compact network. Recurring collaborators include Yasin Ibrahim, Robin J. Evans, Hermione Warr, Harry Anthony, Lilli J Freischem and Daniel R McGowan, spanning causal modelling and clinical PET/CT work; on all three catalogue papers the researcher appears in the last author position, consistent with a senior or supervisory role.
Activity—
Not yet scored: the snapshot provides no yearly publication-frequency data beyond the 2026 catalogue entries, so this axis — which would measure ongoing research tempo — remains unavailable.
Reproducibility55/100
Scores 55 on reproducibility (reproMean 55), computed from 2 papers with reproducibility assessments; both reviewed papers carry a 'medium' reproducibility readiness. Evolve notes convincing, broad experiments in the counterfactual augmentation work but a single architecture and thin statistics, and validation limited to synthetically injected biases in the marginalisation framework.
Research summary
Konstantinos Kamnitsas is a machine learning researcher at the University of Oxford whose recent work, as captured in the Evolve catalog, centres on causal generative modelling for medical image analysis. All three catalogued papers are 2026 arXiv preprints on which Kamnitsas appears as last author — a position typically associated with senior or supervisory responsibility — though none lists him as corresponding author. Two form a coherent causal programme: a counterfactual generation framework that uses structural causal models to condition medical image synthesis for dataset augmentation and bias mitigation, and a Counterfactual Marginalisation framework that turns SCM-based counterfactuals into metrics for auditing classifier robustness to nuisance variables. Evolve's reviewers judge both promising: the augmentation work shows broad, convincing experiments albeit on a single architecture with thin statistics, while the marginalisation metrics are validated only on synthetically injected biases. A third paper compares domain-specific and large language model approaches to error detection in PET/CT radiology reports. The bibliometric snapshot is narrow: six distinct coauthors, a 50/50 field split between two adjacent labels, a collaboration score of 39 and a reproducibility score of 55 across two assessed papers. Output, impact and activity metrics are not yet available, so this profile should be read as an early-stage view of an active causal-medical-imaging agenda rather than a complete career picture.
Collaboration pattern
The snapshot records 6 distinct coauthors, giving a collaboration score of 39. The recurring group — Yasin Ibrahim, Robin J. Evans, Hermione Warr, Harry Anthony, Lilli J Freischem and Daniel R McGowan — pairs causal/ML collaborators with clinical partners, matching the mix of causal generative papers and a PET/CT radiology study. Kamnitsas takes the last author position on all three catalogued papers, suggesting a supervisory role within a small, stable Oxford-centred team.
Research keywords
Highlights
Counterfactual medical images from structural causal models for dataset augmentation
Counterfactual marginalisation: auditing robustness to nuisance variables
Domain-specific models versus LLMs for PET/CT report error detection
A third 2026 preprint compares domain-specific and large language model approaches for detecting errors in PET/CT radiology reports; it is catalogued but not yet reviewed by Evolve, so no verdict is available.
Timeline
- 2026Publishes (last author) 'To do(x) or not to do(x)', using structural-causal-model counterfactuals for medical image dataset augmentation (arXiv).
- 2026Releases the Counterfactual Marginalisation framework for evaluating robustness to nuisance variables (arXiv).
- 2026Co-authors a comparison of domain-specific versus large language model approaches for PET/CT radiology report error detection (arXiv).
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 derived from the platform's bibliometric snapshot only; output, impact and activity axes are missing from the current data.
- Coverage is limited to three 2026 arXiv preprints in the Evolve catalog; earlier publications, citations and career facts are outside this profile's scope.
- Profile is name-resolved from public bibliographic data (ORCID 0000-0003-3281-6509); verify identity before reuse.
- One catalogued paper (PET/CT report error detection) has not yet been reviewed by Evolve and lacks keywords and a reproducibility assessment.