Selected engagements are summarised at a high level. Further detail is available under appropriate confidentiality.
AI Evaluation — Frontier AI research organisation
Frontier large language models are increasingly used by pathologists and clinicians to support diagnostic reasoning. Identifying subtle factual errors, unsafe reasoning patterns and gaps in differential diagnosis at scale requires consultant-level review, not generic medical sign-off.
Pathology Nucleus consultants evaluated more than 1,000 clinical scenarios and model responses across three programmes, applying structured evaluation criteria across factual accuracy, diagnostic reasoning, safety and completeness. Failure modes were categorised and fed back to the development team.
Structured, reproducible feedback on model behaviour in pathology and adjacent clinical reasoning contexts, contributing to ongoing development of AI tools intended to support pathologists in clinical care.
Dataset Annotation — Global pharmaceutical company (oncology)
AI models for digital histopathology require expert ground-truth labels at scale, with reproducibility good enough to support translational and regulatory use. Inconsistent annotation is a leading source of model misbehaviour downstream.
Consultant-led annotation and validation across 1,314 whole-slide images, spanning regional tumour segmentation and single-cell annotation for oncology biomarker programmes. Annotation protocols were defined with a pathologist-in-the-loop QC layer, including review of model outputs for misclassification, boundary errors and diagnostic drift.
Expert-labelled datasets and validation reports supporting AI biomarker development workflows, with reproducibility metrics and structured documentation aligned to translational and regulatory expectations.
Central Review & Translational Biomarker — Top 10 pharmaceutical company (immuno-oncology)
Companion diagnostic frameworks for immuno-oncology require reproducible scoring and independent central review at regulatory standard, with clear adjudication of discordant cases.
Reproducible PD-L1 scoring framework and consultant-led central review. Cases scored to defined criteria with adjudication on discordant calls and full audit documentation.
Central review supporting successful EMA approval of an immunotherapy programme.
Digital Diagnostics & Structured Reporting — UK national healthcare system
Clinical pathology reporting is often unstructured and variable across pathologists and sites. Without standardised criteria and executable logic, diagnostic outputs are difficult to audit, replicate or integrate into digital workflows — limiting both clinical reliability and the potential for automation.
Pathology Nucleus consultants defined diagnostic histological criteria and translated them into structured, executable algorithms — including scoring logic, thresholds and decision rules aligned with established international frameworks such as the Banff criteria. Template-driven data capture was designed to automatically generate structured diagnostic reports from pathologist input.
Deployed in clinical practice with measurable impact: diagnostic concordance improved from 80% to 95%, reporting time per case reduced by over 50%, and adequate turnaround times raised from 70% to over 95% of cases through purpose-built workflow tools, specimen triage and test-ordering protocols.
We engage with additional pharmaceutical, biotechnology and AI partners under confidentiality. To discuss whether we have worked on something relevant to your project, get in touch.
Get in touch