Is this iconic black hole image accurate? New study delivers an independent result

On April 10 2019, the Event Horizon Telescope (EHT) Collaboration revealed the first picture of a black hole. The object is at the centre of a galaxy known as M87, about 55 million light-years away.

The iconic ring structure instantly became one of the most recognised scientific images of the modern era.

It showed a black hole “shadow” outlined by light from swirling hot gas – just as predicted by Einstein’s theory of general relativity. The result supported the long-standing idea that the centres of galaxies house supermassive black holes.

However, scientists have since had some disagreements on whether the details and structure of the ring in the original publication were accurate.

Our new research, published today in Nature, used new techniques to investigate the original datasets. Not only do our results validate the original work, but they also provide a fresh perspective on the ring structure.

Our independent results enhance confidence in one of the most significant modern scientific results. And we bring the image of a black hole millions of light-years away back into focus.

An Earth-sized telescope

Imaging the supermassive black hole at the centre of galaxy M87 is very challenging. It’s similar to photographing a doughnut on the Moon while standing on Earth. To obtain that level of sharpness and detail, you’d need a telescope the size of the entire planet.

To achieve this, the EHT combined signals from radio telescopes across the globe. Doing so enabled the sharpest imaging possible from down here on Earth. The results ushered in a new era of imaging – we could finally see the extreme environments around black holes.

Four independent imaging teams within the EHT Collaboration participated in the original publication. Though all teams produced a ring-like structure, there were some differences in how the black hole’s fine-scale features were represented. Such differences are known to be caused by human and algorithmic biases in the imaging process.

However, one study fundamentally challenged the ring-like nature of the source. Some discrepancies are understandable. They’re bound to happen when combining data from a sparse telescope array into one Earth-sized virtual telescope.

Other factors contribute, too. Each radio telescope gathering the data will also introduce errors, caused by its instruments or simply by Earth’s atmosphere.

Normally, astronomers estimate these errors and correct them during the analysis. But residual errors from this complex process can lead to incorrect outcomes. To address these issues, we developed a new method for combining data, called “closure invariants”.

The standard practice for analysing the data combines pairs of telescope measurements. By contrast, closure invariants combine triplets – three telescope measurements at once.

This not only eliminates those atmospheric and instrumental errors, it also reveals cleaner information about the object’s structure.

The face of a black hole

To make the most of the data from our closure invariants, we designed a machine learning imaging method called GenDIReCT, which produces images consistent with the measured data. To minimise biases, we trained it on terrestrial images like cats, dogs, and cars. Therefore, it had no idea what a black hole looked like.

We also put the model through a large suite of validation tests, which included imaging a wide range of non-ring structures. This included validation on real datasets of other galaxies like Centaurus A.

Once satisfied with its accuracy, we froze the training and testing procedure so human biases couldn’t affect our imaging outcomes.

On December 4 2025, we finally allowed our fully validated imaging method to “look” at the EHT observations of M87.

Across all datasets, the ring structure was consistent. The appearance and diameter of the ring were consistent with a black hole with a mass of 6.5 billion Suns observed nearly face-on – like looking straight down into the eye of a hurricane.

A fresh perspective

When compared directly to images from the EHT collaboration, our result is consistent in diameter and overall orientation.

Naturally, there are some differences in the fine details of ring width and surface brightness along the circumference, but the overall structure matches. This provides a fresh perspective on the iconic black hole ring while also resolving previous discrepancies.

To prove this success didn’t depend solely on advanced machine learning, we also developed a complementary imaging method based on traditional statistical techniques. This entirely different method also validated our structure using the new closure invariant data we produced.

Overall, our work delivers a strong independent validation of the supermassive black hole structure at the centre of galaxy M87.

And it could do more in the future. Our technique could also be useful in other situations where data from multiple sensors are combined to map unknown structures – such as medical imaging or mapping earthquakes.