Cancer screening programmes improved by AI

Catching signs of disease early is the best way to stop cancer...
07 May 2024

Interview with 

Gerald Lip, University of Aberdeen


A doctor coming out of a computer screen


‘Prevention being better than cure’ is not a new idea, and so it’s no surprise that, with cancer being the leading cause of death in the UK, spotting signs of the disease early is a crucial part of the plan to improve patient outcomes. So, how can we do that, and how are scientists trying to harness new technologies to stay one step ahead of cancer? Gerald Lip is the clinical director of breast screening in the north east of Scotland and a senior lecturer at the University of Aberdeen…

Gerald - Screening is the process where we look at healthy people, people who have no symptoms, and we test them to see if they have any disease. For instance, in breast screening, women will come, they don't have any breast lumps, they don't have any problems, they're between the age of 50-70, and every three years they will come and have a mammogram done. Then, we look for any signals on the map to see if there's a change. Similarly, in bowel cancer screening, when you turn 50, you get a pack and then you can test your stool with that pack to check and see if there's any blood that's shed by tumours.

Chris - When I went to medical school, one of the documents that was given to us as students said that a healthy person is someone who's been inadequately screened. I don't know where that fits into the framework you just mentioned, but when we do this, what's the purpose of doing it? Is it just to pick up disease early?

Gerald - If you can pick up something small before it becomes something big, then you can treat something and you have a much better chance of survival as well. Again, in my main field of breast cancer, if you can catch something before it's the size of a 10p, so less than 15 millimetres, with the current treatments and the advances in all the medications that we can use nowadays, you have a 95% chance of survival. If a cancer is presenting much later, that means it's grown into size or is starting to invade the tissue and spread elsewhere in your body, your chances of surviving are less. Likely, you might need more extensive surgery, you might need more chemotherapy or more radiotherapy.

Chris - Some cynics, though, say that when we aggressively screen for things, what we do is turn healthy people into people with a label of disease and they just live with the disease for longer, but they die at the same point they would've done?

Gerald - Depends on the age you are as well as the type of disease that you have. You could be diagnosed with a very slow growing breast cancer, as an example, that would take maybe 10 years before it even doubles in size or spreads anywhere else. This is something called the overdiagnosis debate where you could be labelled with cancer, you have to pay more insurance, you may have problems with travel, but this is the thing that probably won't kill you in the end. You might then die instead of heart disease or something else. Whereas, on the other side, you can have the aggressive cancers and if you catch them earlier, before they spread, then you will have a very long and full life. So that's the balance of screening that we want to do to increase our ability to pick up the cancers that matter and not the ones that will not harm you in the long run.

Chris - We've only got a limited number of screening programmes for disease at the moment though, haven't we? We've got, for instance, breast cancer cervical screening, there's no prostate cancer screening for what is ostensibly the most common cancer in men and a killer of many men. So why are there some things which are screened for but not others?

Gerald - I think that comes down to the accuracy of the screening test. So, for bowel cancer, where they have the Faecal Occult Blood Test, that's shown to have a very high level sensitivity. You have confidence that when you're calling back you have a good chance of seeing a tumour or lesion. The same with breast cancer screening: you'd have to screen a thousand people to find 6-8 cancers and, as such, you need a test that's quite good. Mammography is about 90 to 95% effective. Cervical cancer screening, and diabetic screening is another example, these are tests that have a high sensitivity and specificity. For prostate cancer, at the moment, the PSA, which is the biochemical blood test that's used, they can be inaccurate. As a result you may do more harm than good in having unnecessary surgery, unnecessary worry or biopsies. The UK National Screening Committee is the coordinating body that evaluates screening programmes. They look at all the evidence and say, 'This can now be a national screening programme. There's enough evidence here that this test is the right test and it can find as many conditions as possible.'

Chris - Presumably this is a moving thing in the sense that as new technologies emerge and new practice emerges, it enables us to do it better and therefore what might have been regarded as impractical before is now regarded as a reasonable target.

Gerald - That's correct, yes. I think there's now trials looking at low dose CT for lung cancer screening. You do a health economics analysis to see if the costs of it justifies providing a screening programme. Similarly, with cervical cancer screening, when the evidence of the HPV vaccine came in, that provided evidence of how you balance the screening programmes. So those are the ranges of criteria that you're dealing with: what we can afford to provide and the benefit to the population as well.

Chris - And artificial intelligence - one of the words of the year over the last 12 months and is barely off the front page - that also has cropped up in association with us becoming better and more agile at managing cancer. A lot of times I've noticed.

Gerald - Yes, and it's taking the world by storm. So I've worked with the previous technology, which is Computer Aided Detection (CAD), and what it does is it shows circles around areas that are of worry. With CAD, you could have up to 20 or 30 circles on an image, and that's not really very useful because that's pretty much the whole image covered by circles everywhere. But the artificial intelligence programmes are trained on hundreds and thousands or millions of images and it says, this is what cancer looks like, this is what cancer doesn't look like. Radiologists are the doctors who look at the mammograms, who look at CT scans. There's going to be a 30% decrease in the workforce in the next few years. The Covid catch up, where we had to catch up with all these cases, and the increase in activity and work in hospitals means that the medical workforce is under increasing pressure. In screening, you're looking at a lot of normal cases to find the few abnormal cases, so you could actually use AI to automate the normal cases. We project from the study that we did called Gemini in Aberdeen that you could automate probably about 30% of normal mammogram reading. So instead of 20,000 mammograms a year, I'd be reading 15,000 mammograms a year, and that extra 5,000 mammograms I wouldn't have to read would then free us up to do the harder cases or the more complex mammograms to try and find the cancers in this small group.


Add a comment