AI Spots Missed Breast Cancers in Major NHS Trial
In a landmark UK trial, an artificial intelligence tool successfully identified breast cancers in 11 women that had been missed by human radiologists. The study marks the first randomised test of AI in real-world breast screening and shows promising results for improving early detection.
Key Takeaways
- AI tool ‘Mia’ spotted cancers in 11 women missed by two human doctors.
- The AI flagged 81% of cancers, compared to 77% by doctors alone.
- Combined, AI and doctors identified 87% of cancers in over 10,000 scans.
- Experts call for larger trials before widespread NHS adoption.
How the AI Trial Worked
The trial, supported by NHS England, involved over 10,000 women in the national breast screening programme. Currently, two doctors review each mammogram. In this test, the AI tool named Mia acted as an additional reader.
It successfully flagged 81% of cancers, while the human doctors found 77%. When their results were combined, the detection rate rose to 87%.
Expert Reaction and Cautious Optimism
Dr Katharine Halliday, President of the Royal College of Radiologists, stated: “AI-supported mammography reading could, and should, one day become a standard part of the NHS breast screening programme.”
She added that AI has the potential to support clinicians, allowing them to focus more time on patients and complex cases.
However, Dr Halliday and other experts emphasised the need for more testing. Dr Ben Glocker from Imperial College London called the results “very promising” but noted the trial’s relatively small scale. “We need larger trials to understand the potential and the limitations,” he said.
The Scale of the Challenge and the AI’s Training
Breast cancer remains the most common cancer among women in the UK, with approximately 55,000 new cases diagnosed annually. The AI tool Mia, developed by Kheiron Medical Technologies, was trained on a massive dataset of over 6 million historical mammograms from across the globe to recognise subtle signs of cancer.
The trial was conducted at NHS Grampian, with analysis from the University of Aberdeen and NHS Lothian’s Edinburgh Breast Unit. The full findings are published in the journal Radiology.
