Google DeepMind's New Model Beats Doctors at Early Cancer Detection
In a landmark clinical trial, DeepMind's medical AI correctly identified early-stage lung cancer in 94% of cases — outperforming radiologists by a significant margin.
A study published in Nature Medicine this week details results from a three-year clinical trial in which Google DeepMind's diagnostic AI was pitted against senior radiologists across six UK hospitals. The AI detected early-stage lung cancer with 94.3% accuracy — compared to a human average of 88.1%.
How the Model Works
The model was trained on over 2.4 million anonymised CT scans and uses a multi-scale attention mechanism to flag regions of interest at both the macro level (overall lung structure) and micro level (individual nodule shape and density). It outputs not just a positive or negative result, but a confidence score and a highlighted region for clinician review.
Critically, the AI reduced false negatives by 12% — the type of error where a cancer is missed entirely. False positives also dropped slightly, meaning fewer patients are sent for unnecessary biopsies.
The NHS has announced a pilot programme to integrate the tool into routine screening workflows starting later this year.