Skip to main content Skip to navigation

Theme 8 Projects

Show all news items

Use of Artificial Intelligence in Breast Cancer Screening: Rapid Review and Evidence Map

Leads: Dr Lena Al-Khudairy, Dr Amy Grove, Prof Aileen Clarke, Dr Karoline Freeman (Public Health), Dr Angela Noufaily (Meths)

The National Screening Committee commissioned this study.

Dates: 01/09/2020 – 31/03/2021

Background:

Breast cancer is the most common cancer in women in the UK. The UK NHS Breast cancer screening programme (NHSBSP) aims to reduce breast cancer-related morbidity and mortality through the earlier detection and treatment of disease. In the current programme, women aged 50-70 years are invited to breast cancer screening every three years which involves taking x-rays of each breast. The interpretation of the images is carried out by two readers, with each reader making a decision about whether the image appears normal or if a woman should be recalled for further assessment. Arbitration is employed where the readers do not agree on whether a women should be recalled. Recent data from the NHSBSP indicated that in the period from 2018 to 2019 1.82 million women attended breast screening. Of those 3.8% (84,559) of women were referred for further assessment, with 15,285 cases of breast cancer diagnosed. Approximately 6,000 additional interval cancers are diagnosed annually, which are cancers detected symptomatically in the interval between screening rounds. Of these, an estimated 20% were present but not detected during screening. Type of cancer detected is important, because some types are more aggressive so there is likely to be greater potential mortality benefit from detection, whilst others are more associated with over-diagnosis harm. Artificial intelligence (AI) for image recognition is now being applied to the healthcare sector. There are numerous potential benefits of AI within breast screening programmes including greater efficacy and efficiency in clinical care by reducing the workload of the second reader and increasing the detection rate. Artificial intelligence could alter the spectrum of disease detected at breast screening. This could alter the balance of benefits and harms of breast screening.

Policy and Practice Partners:

National Screening Committee.

Co-Funding Partners:

TBC.

Aims and Objectives:

In response to a request from the National Screening Committee to synthesise the evidence on the use of deep learning AI algorithms (assistive or stand-alone) in breast cancer screening in terms of test accuracy and clinical utility outcomes.

Methods:

A rapid review and evidence map. The database searches were undertaken on 9 September 2020 and considered studies published since 2010.

Main Results:

Ten studies were included in the evidence synthesis. The review found that:

  • There is no direct evidence on how AI may affect accuracy if integrated into UK breast screening practice so we do not know what impact AI would have if introduced into the NHS Breast screening programme.
  • There are no good quality studies in the UK, and designing such studies might be a challenge.
  • Most studies are small and biased, and used test sets of mammograms.
  • We do not know how good AI is at finding different types of breast cancer or at finding breast cancers in different groups of women (for example different ethnic groups).
  • AI has a lot of potential, it might be as accurate as human readers, and might reduce the workload of staff in the NHS breast screening programme. AI might reduce the number of cancer missed at screening, and it might reduce the number of women called back for further tests when they do not have cancer. These things are all very uncertain at the moment as the quality of evidence is very low.

Conclusions:

The current evidence is a long way from the quality and quantity required for implementation of AI into clinical practice of breast screening programmes.

Implications for Implementation:

It is premature to encourage assimilation of AI in breast screening programmes.