Use of AI in Breast Cancer Screening in England Has a Minimal Impact on Greenhouse Gas Emissions and May Increase Diagnostic Accuracy
Author(s)
Amy Swanston, PhD, Emily Archer Goode, PhD, Medha Shrivastava, MSc, John Joseph Bridgewood, MMathStat, Henry James Swales, MMathStat, Lindsay Nicholson, PhD.
Maverex Limited, Newcastle upon Tyne, United Kingdom.
Maverex Limited, Newcastle upon Tyne, United Kingdom.
OBJECTIVES: In England, women aged 50-71 are screened for breast cancer every three years. Screening mammograms miss approximately 20% of breast cancers. Artificial intelligence (AI) tools, such as Mammography Intelligent Assessment, are being trialled to improve early cancer detection. This study evaluated the environmental impact of AI-assisted mammography screening versus the current standard of care (SoC), in women aged ≥45 in England over one year (2023-2024).
METHODS: A life cycle assessment (LCA) was performed using the ReCiPe impact assessment method, with inputs from the ecoinvent database v3.8, and modelled using LCA for Experts. In the SoC pathway (according to published literature), mammogram images are assessed by two human readers, plus a third for arbitration, if required. Real-world baseline data from NHS breast cancer screening statistics in England were applied to two published scenarios; 1. SoC versus two human and one AI reader 2. SoC versus one human and one AI reader. Published values for sensitivity, specificity and arbitration rate informed true/false positive and negative values, which determined referrals for triple assessment (clinical examination, imaging, and biopsy).
RESULTS: In Scenario 1, SoC pathway generated 8,269 tonnes of carbon dioxide equivalents (CO2e). An additional AI reader resulted in 8,277 tonnes CO2e, 8 tonnes difference annually, also equivalent to two times the annual GHG emissions of a single person in the UK. In Scenario 2, replacing one human reader with AI resulted in minimal change, with SoC and the AI-assisted approach generating 7,982 and 7,984 tonnes CO₂e, respectively. Implementing AI in addition to current SoC results in fewer patients missed during screening, improving screening accuracy without substantially increasing the environmental burden of screening, with potential for further impacts downstream.
CONCLUSIONS: Implementing AI assistance in the breast cancer screening programme in England may result in earlier and more accurate cancer detection, whilst having minimal environmental impact.
METHODS: A life cycle assessment (LCA) was performed using the ReCiPe impact assessment method, with inputs from the ecoinvent database v3.8, and modelled using LCA for Experts. In the SoC pathway (according to published literature), mammogram images are assessed by two human readers, plus a third for arbitration, if required. Real-world baseline data from NHS breast cancer screening statistics in England were applied to two published scenarios; 1. SoC versus two human and one AI reader 2. SoC versus one human and one AI reader. Published values for sensitivity, specificity and arbitration rate informed true/false positive and negative values, which determined referrals for triple assessment (clinical examination, imaging, and biopsy).
RESULTS: In Scenario 1, SoC pathway generated 8,269 tonnes of carbon dioxide equivalents (CO2e). An additional AI reader resulted in 8,277 tonnes CO2e, 8 tonnes difference annually, also equivalent to two times the annual GHG emissions of a single person in the UK. In Scenario 2, replacing one human reader with AI resulted in minimal change, with SoC and the AI-assisted approach generating 7,982 and 7,984 tonnes CO₂e, respectively. Implementing AI in addition to current SoC results in fewer patients missed during screening, improving screening accuracy without substantially increasing the environmental burden of screening, with potential for further impacts downstream.
CONCLUSIONS: Implementing AI assistance in the breast cancer screening programme in England may result in earlier and more accurate cancer detection, whilst having minimal environmental impact.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MT45
Topic
Medical Technologies
Topic Subcategory
Digital Health
Disease
Oncology