Cognitive load
We measure the effort required to review, edit, and sign a report, then reduce unnecessary work without removing physician judgment.
Foundation models for radiology.
Mecha Health is an applied AI lab building models that read medical images and draft radiology reports for physician review.
We work across clinical medicine, machine-learning research, and the deployment details that make these systems useful in practice.
What we build
Systems that turn studies into reviewable reports.
We build models for radiology that read images, reason over findings, and produce clear report drafts for a radiologist to over-read.
A foundation model for radiographs and draft report generation.
Mecha CTA draft foundation-model page for computed tomography.
What we optimize
Three outcomes, measured in practice.
We evaluate each system against the clinical workflow it is meant to improve, then optimize the model and product together.
Turn-around time
We measure time from study opening to signed report and optimize the points in the workflow that delay care.
Clinical accuracy
We measure finding-level fidelity and clinically meaningful discrepancies, with radiologists remaining the final interpreters.
Team
Doctors and ML researchers.
A small technical team building at the intersection of medical imaging, language models, and clinical deployment.
The team brings experience from Cambridge, Imperial, UCL, AstraZeneca, Microsoft, and Y Combinator.
Ahmed Abdulaal
CEO & Co-founder
Imperial-trained physician and UCL Microsoft PhD Scholar. He identified loss of smell as a COVID signal using AI, informing public-health policy; previously worked on AstraZeneca’s foundation-modelling team and has published 20+ peer-reviewed papers.
Nina Montaña Brown
CTO & Co-founder
UCL PhD researcher in AI for surgical vision and navigation, with patented work and publications at NeurIPS, ICLR, and MICCAI. Previously built clinical AI systems across health-tech startups, including safety benchmarks, accelerated deployment, and regulatory documentation.
Ayodeji Ijishakin
COO & Co-founder
UCL medical-imaging ML PhD researcher building diffusion models for interpretable imaging. His work spans NeurIPS, ICLR, and ICML; he also founded the London Founders Club.
Hugo Fry
CSO & Co-founder
Cambridge-trained mathematician and theoretical physicist, MATS scholar, and mechanistic-interpretability researcher. His work applies interpretability techniques to vision models and has been cited by Anthropic and DeepMind.