AI Reads ECGs In Under Two Seconds. Beats The Human Eye. The Training Data Did The Heavy Lifting, Naturally.

Researchers trained an AI on millions of routine ECG recordings to extract diagnostic signals beyond what a human cardiologist can visually detect. The system identifies heart disease in under two seconds, potentially replacing a wait of several months for a traditional heart ultrasound scan. The work was led by Prof Fu Siong Ng, a professor of cardiology at Imperial College London.

This illustrates a principle I like to call latent signal extraction. A standard ECG has contained more information than any clinician could parse for roughly a century. The AI does not invent new data. It surfaces patterns already present but perceptually invisible to human observers. The mechanism is pattern recognition at a scale no biological brain can match.

Prof Fu Siong Ng and colleagues at Imperial College London developed the tool using training data from millions of patient ECGs. Their published results suggest the system could fast-track high-risk patients for earlier treatment.