The amount of muscle that people have, so when we talk about people having lost muscle we talk about the term sarcopenia, has been well described as being a prognostic factor in the general population. So as people get older they tend to lose muscle mass and those who have less muscle tend to live less long. So that has been well recognised for a long time. But the conventional measure of muscle mass normally involves… in healthy people involves using something called a DEXA scan but in most cancer patients has been done using CT scans of the body and looking at muscles around the lumbar spine.

When we think about our brain tumour population, most of those patients don’t have CT scans and they certainly don’t have regular ongoing CT scans. Because, of course, if you’ve got a brain tumour we image up here. So our real question was can we look at the amount of muscle you’ve got up here, can we reliably measure how much muscle you’ve got up there, and can we see whether that relates to your outcome?

So what we did is we just used a set of local patients, so you take their imaging, you take a slice at about this level here from their MRI scan, we draw around the temporalis muscle, and then essentially you just teach a computer to draw around the temporalis muscle. With modern deep learning techniques you just need to give it lots of examples so you draw around lots of temporalis muscles and then you give them to a computer. It turns out that the computer can draw around the temporalis muscle pretty well. So if you look at the results we present in the paper, we’re getting accuracy rates of better than 95%. So a computer can draw around the temporalis muscle and then once you draw around it you can measure how big it is.

So you can say how much temporalis muscle a group of patients has. But then the second bit is to say how well does muscle mass predict survival? To do that you have to take into account other things. So, for example, age is important in people with many brain tumours, as are some molecular markers and things like that.

So we got some fairly simple baseline data on those patients and then we included muscle mass measured by the computer into the model. We showed that the amount of muscle that you’ve got also significantly impacts on survival. So that’s really interesting because we’ve shown not only can a computer draw around the muscle but when you do that and therefore measure the amount of muscle, the amount of muscle you’ve got helps predict survival. It’s not the only thing that predicts survival but it does contribute to predicting survival.

What were the results of this study?

The key results are really twofold. First of all, you can train a deep learning system to measure muscle mass at the temporalis muscle in people’s heads using routinely acquired MRIs. So these are MRI scans that our patients are having anyway as part of their treatment, you don’t have to do anything special about the MRI, and you can teach a computer system to draw around the temporalis muscle and therefore measure how much muscle is there.

So that’s an interesting technical result, it’s not earth-shattering, it’s not very surprising. No-one had done it before but lots of other people have done deep learning for other things. So it’s nice to show but it’s not very surprising. But what’s much more useful from a clinical perspective is that when you do that, the amount of muscle that you measure is then a prognostic factor in adult patients with glioblastoma and it contributes to predicting how long they live for. That’s quite an important finding because, remember, we’re taking routinely acquired scans that patients are having as part of their standard treatment, we’re putting them through a computer system and although training the computer system took quite a long time, actually running it takes less than ten seconds. So you put them through the computer and it gives you more information about that patient and more information about their prognosis. And that’s a useful thing.

How can this study impact the future of cancer care and prognosis?

There are two bits, really. The first thing is the better we understand prognostic predictions, the better conversations we can have with patients and their families. So that’s important. But the second thing is that once you understand what drives prognosis, of course that allows you a route to intervention. So we know, for example, that patients who have more of their tumour resected live longer so the obvious reaction to that is therefore we should try and take out more of the tumour. If you look at changes in surgical practice over the last twenty years, that’s what’s happened.

We absolutely need to do some validation work, which we’re doing at the moment, but if it turns out that even using multiple other cohorts the amount of muscle does indeed predict survival, and I think there’s enough evidence from other cancers to make that plausible, of course that opens up something really interesting which is, if muscle mass predicts survival then does doing stuff that increases muscle mass, will that improve patient survival? So you could talk about doing exercise programmes, weight training, stuff like that. We’ve already got some ideas for what that might look like but potentially it’s a really interesting idea and a really interesting intervention. So if you really can show that muscle mass is a significant predictor, then the next step is to say if we do things that improve muscle mass, is that going to improve survival? Of course, that’s an open question but it gives you a route into doing a clinical trial.

Is there anything you would like to add?

The final thing, I think, is it’s really nice to see the NCRI recognising the work that we did around sarcopenia and brain tumours. It was led by a junior investigator and it’s about brain tumours, both of those things often don’t get a great deal of attention. So it’s really nice to see a junior investigator doing a piece of work in brain tumours which then gets some recognition from the NCRI. Part of my job is to bang the drum for brain tumour research, so doing research into brain tumours is good and is important and is worthwhile and this is a nice way of demonstrating that.