Statement by researchers and companies recommends next steps to assess AI tools, which could greatly reduce workload
A roadmap for evaluating and using lessons from trials of artificial intelligence in breast cancer screening has been published by our Fellow Emerita Fiona Gilbert and colleagues in the European Journal of Radiology.
The report follows a conference at Newnham College last November with AI companies and experts, which looked at what should be done to test commercial AI tools.
The report, Overview of trials of artificial intelligence algorithms in breast cancer screening, reviews current evidence and the authors conclude:
- AI studies performed at least comparably in screening mammography to human reading.
- Prospective trials are essential for AI implementation in mammography screening.
They recommend a multicentre multivendor testing platform study with opt-out consent before implementation into a screening programme. This would be a prospective trial which follows-up on patients to check outcomes in relation to AI or human readers.
Further, they suggest that AI thresholds from different vendors should be determined, to state the required number of cases the AI correctly calls positive and the number the AI correctly calls negative.
Professor Gilbert said: “This Newnham Report is an important summary from experts in the field of AI in mammography screening to determine next steps before implementation into clinical practice.
“The potential to reduce workload, address the workforce shortage and redirect human effort to other screening interventions is huge. There is potential to reduce our workload dramatically. The challenge is that the algorithms are not perfect; they miss things and can introduce bias into reading which is why controlled evaluation and prospective trials are required to avoid any unintended consequences.”
- Overview of trials on artificial intelligence algorithms in breast cancer screening – A roadmap for international evaluation and implementation, van Nijnatten et al., European Journal of Radiology, September 2023