AI has been all over the news recently – from artificially generated LinkedIn profile pictures to Homer Simpson’s rendition of Ace of Spades – but there are applications for AI being utilised that go far beyond asking ChatGPT to write an episode of Friends. AI technologies are rapidly being developed for diagnosis and treatment of a multitude of diseases in healthcare.
We spoke to Dr Gerald Lip, Clinical Director of Breast Screening at Aberdeen Royal Infirmary, who is heading up the Gemini project – a study looking at the accuracy of AI’s ability to detect breast cancer in mammograms taken at breast screening to assist radiologists in the NHS.
What is the Gemini project?
The Gemini project is looking at how AI can help radiologists to detect small cancers on mammograms taken in breast screening. The images taken during screening appointments are read by two radiologists as normal, who decide based on their analysis whether the patient needs further investigation or can return to routine screening.
After this, the mammograms of the patients who were returned to routine screening are then anonymously uploaded to the AI software (Mia from Kheiron Technologies) for analysis. Mia then evaluates the images and highlights any areas which it thinks require a second opinion. All of the images earmarked by Mia are then reread by the Gemini team – the ultimate goal is to see if AI can detect extra cancers that would otherwise have been missed.
What is the triple helix approach?
The triple helix approach is a partnership between the industry, the academics and the NHS. We are working together to bring these innovations to fruition sooner by collaborating and supporting each other.
We have a few projects (listed on our innovation website) not just on AI but also on drones that can deliver medication and robot porters to carry bedsheets, towels and things around our hospitals (but we’re still using fax machines and pagers)!
How reliable have you found it to be?
We’ve found it to be generally good at picking up cancers almost matching human performance.
It can sometimes flag up some false positives, which is why it still needs a radiologist to interpret its results afterwards. It doesn’t know the patient’s history – it doesn’t know if you’re on HRT or your family history, so humans still remain key in the process.
Can it tell the difference between benign and malignant breast lesions?
Mia sends an opinion to call or not recall. Other software can circle an area of suspicion and give a score out of 100% for what it rates its malignancy as (100 being the most malignant, 0 being completely benign). Some technologies are using heatmaps to distinguish areas of higher and lower suspicion.
Some people might be a little sceptical of having their scans read by AI rather than a human – what would you say to those patients?
For the Gemini project obviously nothing changes in that respect as the scans are still being read by two radiologists. The use of AI is a good thing for patients because it is able to detect cancers that radiologists have missed – AI doesn’t get tired or hungry or have moody days – it is always working at its optimum level! Mia has already picked up extra cancers during the Gemini project that would otherwise have not been flagged, so that those women could undergo treatment while the cancers were in the early stages which dramatically improves outcomes.
The AI technologies used in these kinds of studies have also been shown to be detecting the higher grade cancers – those that grow more rapidly and are more likely to become invasive, which are absolutely vital to be detected early. A cancer under 15mm has a 95%+ survival rate.
How do you see AI like Mia being used in breast screening after the trial?
There is a trial call for a big project involving multiple centres and multiple AI technologies across the UK to see how it could be used in screening. The general opinion is that mammograms in the future could be read by one radiologist and then once by AI.
How do you see AI alleviating the workload of radiologists?
There is a huge shortage of radiologists in the UK which is projected to grow – in the modelling done as part of the Gemini project, outsourcing of scan reading would be hugely reduced and around 40% of the workforce costs could be saved. It would give radiologists more hands on time with patients in clinics and reduces the cognitive load which is directly related to stress levels, improving the life of the radiologist and allowing them to give the best possible care to their patients. A benefit for the radiologist is a benefit for the patient.
Other than potentially detecting missed cancers and improving the workload of the radiology workforce, are there any other benefits of AI being used in breast screening?
One key benefit is how it will improve turnaround times for patients’ results – if a radiologist reads the scan and then the AI immediately reads it, the wait time for results could go from 2 weeks to 3 days. This would mean less time for ladies anxiously awaiting their results, and the ability to start treatment earlier where necessary. Wait times for assessment for symptomatic patients could also be reduced as a result.
How easy would it be to implement in screening centres across the UK?
In Scotland we use a paperless reporting system so we are actually AI ready for as soon as we have the green light. In England, there are adaptations required for the NBSS (English Breast Screening System) which will take some time but it will likely be ready by the time the national trial starts.
How do you see this technology being used in the wider NHS?
First of all I think we need to gain the trust of radiologists – we need to look at whether it’s worth it, what the health economics are, but there is obviously a workforce shortage which AI can hugely impact in a positive way. Increasing quality of diagnosis in multiple fields is something that could be beneficial across the board – while we think we’re excellent as humans, we do make mistakes. Having a second tool will only help us and make everyone safer. It can only improve the quality of care.