Space and cancer

Astronomical techniques could help analyse cancer samples

Today our researchers announce the results of an exciting project bringing together two unlikely scientific bedfellows – astronomy and pathology.

Back in 2010, Dr Raza Ali and his team at our Cancer Research UK Cambridge Institute joined forces with the University of Cambridge’s Institute of Astronomy to focus their techniques for scanning the night sky onto the challenge of spotting rogue cancer cells.

Spotting key differences between tumour samples holds the key to understanding why some cancers progress more quickly than others, and why patients respond differently to treatments. But despite increased automation in many areas, this kind of analysis still largely relies on expert pathologists looking down the microscope at tumour samples and scoring the presence of particular protein molecules.

Regular readers of the blog will know that Cancer Research UK recently launched the world’s first citizen science cancer project Cell Slider to help break this data bottleneck. But for some time, scientists have also been looking at ways to automate the process and train machines to do the hard work. Today, our scientists published their latest progress towards this goal.

Writing in the British Journal of Cancer, Dr Ali and his team show how the automated techniques the astronomers use to analyse deep sky images can also detect subtle differences in protein levels between healthy and cancerous breast cells.

They used the automated technique to measure the levels of three different proteins in tumour samples from more than 2,000 patients – high levels of these proteins are linked to more aggressive cancers.

Then they compared the results to the same task done by a pathologist looking down a microscope to score the samples.

Reassuringly, the automated technique seems to be just as accurate as the human version but – crucially – is many times quicker, analysing up to 4,000 individual images a day. The team now plans to carry out a much larger study to confirm the accuracy of the test, using samples from more than 20,000 breast cancer patients.

This new development is exciting and could change the way that samples are analysed in the future, if the technique passes muster in the larger study. This in turn could speed up the discovery of more effective treatments for breast cancer by allowing researchers to analyse huge amounts of data in record time.

But there are still hurdles to be overcome. One of the issues with using machines to analyse this data is that they’re simply not as good as pathologists at identifying the presence of cancer cells in the first place. We also don’t yet know if they are able to pick up trickier-to-spot molecules in certain parts of the cell. And computer analysis techniques may not be sensitive or specific enough to reliably score certain markers.

Until these barriers can be reliably overcome, this process urgently needs the human touch. And this is where you come in.

Robots vs humans – CellSlider needs your help

We recently teamed up with Citizen Science Alliance to launch CellSlider – the first ever public cancer research project that you can do from the comfort of your own home.

CellSlider uses similar data to that analysed by the Cambridge team, but employs the eyes and brains of thousands of people rather than a super-computer.

So far, our army of helpers have made hundreds of thousands of classifications of cancerous cells, cutting the time needed for this kind of analysis from 18 months to just three.

So the jury’s still out on whether machines can truly beat humans in the world of pathology – as it stands, computers aren’t quite ready to be a replacement for the intelligence and intuition of thousands of human minds.

But while our researchers are still faced with a huge backlog of unique tumour samples that hold the key to saving more lives – we’ll keep trying as many ways as possible to get through the data faster. Help us make it sooner by joining in at



Ali H.R. et al. (2013). Astronomical algorithms for automated analysis of tissue protein expression in breast cancer., British journal of cancer, PMID: