In a remarkable example of interdisciplinary teamwork, astronomers are helping cancer researchers use computerized stargazing algorithms developed for spotting distant galaxies to identify biomarkers in tumors to determine how aggressive they are.

The teams, from Cancer Research UK Cambridge Institute, and the Department of Oncology and the Institute of Astronomy at the University of Cambridge in the UK, describe how they adapted the astronomers’ image analysis algorithms and tested them on 2,000 breast tumors in a study published online on 19 February in the British Journal of Cancer.

The current method of analyzing tumor aggressiveness relies on skilled pathologists looking down microscopes to spot subtle differences in staining of tumor samples. A computerized approach could speed up this process quite significantly.

The astronomers’ image analysis algorithms were originally developed to help them automatically pick out indistinct objects in the night sky. At one time this was also a laborious manual exercise, performed with the help of powerful telescopes.

The techniques are not unlike those used in immunohistochemistry (IHC), where pathologists gaze down microscopes to pick out subtle differences in staining of tumor cells to distinguish different proteins.

Lead author Dr Raza Ali, a pathology fellow from the Cancer Research UK Cambridge Institute, says in a statement:

“We’ve exploited the natural overlap between the techniques astronomers use to analyse deep sky images from the largest telescopes and the need to pinpoint subtle differences in the staining of tumour samples down the microscope.”

Co-author Nicholas Walton, from Cambridge’s Institute of Astronomy, says:

“It’s great that our image analysis software, which was originally developed to help, for instance, track down planets harbouring life outside of our Solar system, is now also being used to help improve the outlook for cancer patients, much closer to home.”

To test the adapted algorithms, the researchers used them to assess levels of three proteins (ER, BCL2, and HER2) that are known biomarkers for aggressive cancer, in samples from more than 2,000 breast cancer patients.

Each sample went through two assessments: one using manual image analysis with pathologists looking down microscopes, and the other where a computer, equipped with the adapted algorithms, analyzed the images automatically.

When they compared the results, the teams found the computer was just as accurate as the manual system:

“Automated scores showed excellent concordance with manual scores for the unsupervised assignment of cases to ‘positive’ or ‘negative’ categories with agreement rates of up to 96%,” write the authors.

Plus, the added advantage is that the computerized system was much faster.

“The results have been even better than we’d hoped,” says Ali, “with our new automated approach performing with accuracy comparable to the time-consuming task of scoring images manually, after only relatively minor adjustments to the formula.”

The researchers now plan to do a much larger international study using samples from more than 20,000 breast cancer patients. This will help to refine the new approach.

Senior author Carlos Caldas, a professor who is also with the Cancer Research UK Cambridge Institute says:

“Modern techniques are giving us some of the first insights into the key genes and proteins important in predicting the success or failure of different cancer treatments. But before these can be applied in the clinic, their usefulness needs to be verified in hundreds or sometimes thousands of tumour samples.”

Caldas says the new methods are already helping them analyze up to 4,000 images a day.

In January, researchers at the University of Oslo in Norway, reported a new study that found 3D mammography (tomosynthesis), used in conjunction with traditional imaging, increased invasive breast cancer detection by 27%.

Written by Catharine Paddock PhD