Royal Free to study AI tool for breast cancer screening

The Royal Free London NHS Foundation Trust is to investigate whether artificial intelligence (AI) could be as effective at detecting breast cancer as currently-used screening methods – and therefore speed up detection rates and reduce patient waiting times.

When a woman has a mammogram, the images are assessed by two radiologists. The use of two pairs of eyes increases the rate of cancer detection and keeps recall rates low, but it is resource-intensive – a problem exacerbated by a shortage of radiologists.

To see if AI could do the job as well, the Royal Free London will use Mia (mammography intelligent assessment) from Kheiron Medical Technologies, to evaluate mammograms carried out by the North London Breast Screening Service (NLBSS). Each mammogram will have already been evaluated by two radiologists, making it possible to compare Mia’s accuracy with the accuracy of current standard practice. Any woman whose mammograms have been identified by the radiologists as abnormal will have already been referred for further investigations or treatment. All the patient data used in the study will be anonymised.

Mia uses deep learning for continual improvement

The more mammograms Mia analyses, the more it will learn, so it will continually improve its ability to detect cancerous or pre-cancerous cells. The tool has been built by using a form of deep learning called convoluted neural networks, which involved being trained on three million images.

Mia works by analysing the image and suggesting that either no further action is needed or that further investigation is required. If it is found to be effective, it could be implemented nationwide, eventually replacing the need for an evaluation by a second radiologist. This would reduce the current backlogs in the system and alleviate the problems caused by the shortage of radiologists.

NLBSS, situated in Edgware Community Hospital, is one of the largest screening services in the country, screening 50,000 women a year. It is one 14 hospital sites across the country participating in the study, which has been funded by the AI in Health and Care Award.

Dr William Teh, a consultant radiologist at the Royal Free London, said: “The study will provide evidence to help assess whether the use of AI in this instance could be a viable option. It’s an opportunity to analyse and assess the differences between human and machine made decisions and to learn whether AI has the potential power to positively transform clinical practice in real-world screening populations.”