Artificial Intelligence (AI) systems for the classification of mammography images have been developed and are beginning to be trialled and deployed in a breast cancer screening setting with encouraging results.

It is reasonable to think that these systems could be useful in the context of symptomatic breast clinic. However, these systems developed in the screening setting have unknown performance in the context of symptomatic breast clinic.

It is therefore important to test the performance of these systems in this alternative context.

This study will use retrospective data, from where it is possible to determine ground truth outcomes with greater confidence, accessing relatively large volumes of data with less patient burden when compared to prospective studies. This important cohort of patients has been less investigated to date, mainly because symptomatic data is typically more difficult to curate than screening data where key data is methodically prospectively collected.

The proposed work will be carried out in collaboration with a selected AI vendor and local clinical teams to define optimal use case scenarios for the symptomatic breast clinic.

Patients with breast symptoms are referred from primary care to symptomatic breast clinics, often under the two-week-wait cancer pathway. Clinicians assess the patient's breast symptoms by looking at the patient's personal and family history of cancer, conducting a physical examination, and referring the patient for imaging as required.

Ultrasound and / or mammography are typically performed and reported by the imaging team at the same visit, with biopsy performed when indicated. This service is an important part of cancer care provision, with approximately half of the breast cancers diagnosed presenting via the symptomatic service rather than identified at screening.

It is important to note that cancers diagnosed symptomatically tend to be larger and more aggressive with worse outcome than those diagnosed via screening. The volume of referrals to the National Health Service (NHS) symptomatic service has risen over the last decade, placing increased pressure on service delivery, in breast imaging.

Artificial Intelligence (AI) systems for the classification of mammography images have been developed and are beginning to be trialled and deployed in a breast cancer screening setting with encouraging results. It is reasonable to think that these systems could be useful in the context of symptomatic breast clinic. However, these systems developed in the screening setting have unknown performance in the context of symptomatic breast clinic. It is therefore important to test the performance of these systems in this alternative context.

Inclusion criteria

• Patients 18 years or older attending symptomatic breast clinic.

• Mammography images, including both full field two-dimensional digital mammography and digital breast tomosynthesis.

• Dates of attendance will be from January 2015* to December 2019 at the lead data collection site. Dates of collection may be different at the other sites depending on local data curation consideration but will be a minimum of 2 years prior to study start to allow determination of ground truth.

  • If any mammography images prior to 2015 should be available at the lead site, these will be collected as well (a maximum of 3). Prior mammograms will also be collected at the other sites if available, depending on local PACS set-up.

Exclusion criteria

• Patients under the age of 18 years.
• Patients on the National data opt out.

Patient attending a symptomatic breast clinic at the recruiting sites