July 31, 2020
Use a second set of eyes to improve breast screening accuracy
With double reading, two radiologists read the same mammogram. It is standard practice in Australia and Europe but not widely used in Canada, although certain clinics do perform some double reading.
What double reading looks like in practice
Double reading affects abnormal call rates (ACRs) differently depending on how it is used. If the second reader reviews only normal screens, double reading will likely increase ACRs by identifying more abnormalities. If the second reader focuses just on abnormal scans, double reading will likely decrease ACRs by identifying more abnormalities as benign. Given resource constraints in Canada, double reading only cases being considered for recall is the most realistic option.
One Australian clinic uses double reading widely, even calling on a third reader in case of disagreement. One provincial breast screening centre in Canada has screens read by a senior mammography technologist and a radiologist for quality assurance.
AI innovations in breast screening
Artificial Intelligence (AI) within breast cancer screening has the potential to increase efficiency and accuracy with breast cancer diagnosis and it can be incorporated within double reading practices. Given the increasing use of AI in cancer care, it is important to consider challenges and opportunities for this practice and how this may play a role in the future application of double reading. AI has a great deal of potential to act as a second read; however, it has not been developed for this purpose yet. Breast screening programs need to begin to think about the implications of AI in the future. The Canadian Association of Radiologists will be developing an AI product through 2020-2021, which may be used by programs as they explore this approach.
How to put this approach into action
Following these steps will help breast cancer screening programs implement this approach:
- In the short term, coordinate a nationally supported pilot project to assess the feasibility and effectiveness of double reading for potential recalls.
- Over time, assess technologies (such as artificial intelligence) that could be used to double-read screens.
Challenges to overcome
- Lack of clear evidence supporting double reading
- Difficulty of double reading and potential delays in delivering test results where there are not enough radiologists
- Potential for technology advances (such as artificial intelligence) to make double reading less relevant
More approaches on improving abnormal call rates in breast cancer screening with mammography