Data Science
Can AI support breast cancer screening?
The challenge
Healthcare and radiology are facing substantial challenges due to high workload, a shortfall of clinical radiologists, and an ageing population. For breast cancer screening, all women in the UK between 50 and 70 years are invited for mammograms every 3 years. This results in approximately 2 million mammogram examinations being performed. Only a small percentage of these mammograms contain signs of potential disease.
Artificial intelligence (AI) could support breast screening by performing a similar task to the human experts: examining mammograms and highlighting potential areas of suspicion. It could replace one of the two human readers, reducing human workload. It could also help detect cancers which would otherwise be missed.
The research
Despite the promise of AI, the UK National Screening Committee does not recommend the use of AI in the NHS breast screening programme. They highlighted that both the quality and quantity of the evidence base was insufficient. Previous work focused on evaluating AI for breast cancer screening. As part of the Industrial Centre for Artificial Intelligence Research in Digital Diagnostics (iCAIRD) programme, we worked with colleagues from the NHS, academia and industry to assess whether the AI algorithm Mia, developed by Kheiron Medical Technologies (since procured by DeepHealth), could support breast screening for Scottish women in NHS Grampian.
The results
In a retrospective study of 55,000 cases, AI performance was affected by the mammography equipment and software running on the equipment. Tuning to the local setting and equipment was needed to return performance to expected levels.
The retrospective evaluation highlighted both strengths and limitations of the AI, enabling the subsequent prospective study (GEMINI: Grampian’s Evaluation of Mia in an Innovative National breast screening Initiative) to safely exploit its full capability. GEMINI employed both live AI integration and workflow simulations to model 17 different ways AI could be used in breast screening. With the help of AI, 11 additional cancers were detected (10.4% increase). The primary AI workflow evaluated would also reduce workload by up to 31%. Other workflows show different benefits, allowing for adaptation to local healthcare needs.
The impact
The results from the retrospective study have been cited in international guidelines for developing, purchasing, implementing, and monitoring AI in radiology, helping set the standard for assessing and implementing AI for breast screening. A follow-up multicentre trial (start date: autumn 2025) aims to generate sufficient evidence to determine whether AI can be safely and effectively implemented into UK breast screening. The trial will assess multiple AI algorithms and include a total of 660,000 women from across the UK.
Lead
Read the paper
The study, Impact of different mammography systems on artificial intelligence performance in breast cancer screening, is published in Radiology:Artificial Intelligence.
