Mecha Research
Foundation models for radiology.
Technical reports, evaluations, and research notes.
Locating Any Pathology Anywhere in Any Medical Image
A modality-agnostic approach to grounding free-text findings in XR, CT, and MRI.
Read articlePublications
Archive
We Taught a Fly to Read X-Rays
Training a connectome-constrained network to classify pleural effusion on chest X-rays using sparse retinal inputs.
Qualitative Case Studies for Assessing Foundation Models
Most metrics used to evaluate machine learning models are quantitative, but qualitative assessment is crucial.
Maps, Misconceptions, and the Making of Modern Foundation Models
A deep dive into foundation models, their training, and blindsides.
Mecha Health Raises $4.1M Seed Round
We're building next-generation foundation models to enable superhuman radiologists.
Mecha Net v0.2 — Reaching Human Performance
Performance update for our Chest X-ray report generation model.
Why radiologists make mistakes
On the reasons radiologists make mistakes and how to improve report quality.
Interpreting a radiological foundation model
Exploring human-interpretable features within a radiological foundation model.
Mecha Net v0.1
Our first performance update for Chest X-ray report generation.
Evaluating automated radiology reports
On assessing the quality and accuracy of automatically generated radiology reports.