Breast Cancer Screening
Academic Lead: Dr Jorge Cardoso
Clinical Lead: Dr Keshthra Satchithananda
Clinical Area: Oncology, Mammography
Partner: Thames Mammography
Breast cancer claims approximately 500,000 deaths worldwide each year. The UK’s National Breast Screening Programme seeks to detect breast cancer early on to improve health outcomes.
While attendance of the Programme is relatively high (71%), it relies on a two-dimensional (2D) Full-field Digital Mammography (FFDM) method, which fails to detect up to 20% of cancers and reports a number of false positives, placing an unnecessary burden on the NHS to retest needlessly.
Digital Breast Tomosynthesis (DBT), which reconstructs x-rays to create three-dimensional (3D) images of the breasts, provides a clearer image with more sensitive differentiation between healthy and abnormal tissue. However, while it could address the above issues, it takes radiologists up to twice as long to interpret the results compared to 2D scans.
With the NHS expecting a 50% shortfall of radiologists over the next five years it needs a smarter and more time efficient workflow to maintain its quality of service.
In partnership with Thames Mammography, we are creating a prototype AI-enabled clinical workflow that can interpret both FFDM and DBT images and automatically detect abnormal results.
This application will improve early diagnosis of breast cancers and reduce the time spent interpreting results to free clinicians for other complex tasks.
Our AI tool is based on deep learning algorithms that learn to detect malignant and benign features from annotated scans. Drawing on state of the art object detection research, the prototype will leverage both FFDM and DBT imaging data in the NHS to provide greater reliability in its assessment compared to one image type alone.
The system assigns a risk score to suspicious objects in the scans and calculate the probability of cancer in each patient. High-risk cases can be prioritised within the workflow for immediate specialist assessment, while low risk cases could avoid the need for human review at all.
Our novel system will provide a more reliable detection of breast cancer, enable health services to optimise the use of highly trained but scarce staff resources and reduce the economic burden on the NHS.