AI is no stranger to blowing minds, but its latest party trick is straight out of a sci-fi mystery novel: it can tell your gender just by scanning a photo of your retina—something even expert ophthalmologists can’t manage. And the kicker? We still don’t fully understand how it does it.
In a new study from University Malaysia Computer Science & Engineering and Al-Azhar University, researchers trained a deep learning model on over 20,000 fundus eye scans (images of the back of the eye). The model—a variation of the popular Xception neural network—hit an astonishing 96.83% accuracy rate in guessing a person’s gender from a retinal image. Precision and recall were both nearly perfect.
MORE: Scientists Have Just Found A ‘New Colour’ That’s Never Been Seen Until Now
DISCOVER: Are UV Lamps Bad For Your Nails: Here’s What The Science Says
This would be impressive under any circumstances, but it becomes even more intriguing when you realise that human experts looking at the same scans can’t spot any meaningful anatomical difference between male and female retinas. The retina’s blood vessels, optic disc, and macula all look pretty much the same across genders—at least to the naked eye. So what is the AI seeing that we aren’t?
That’s where things get complicated.
Image @Helariuss via Pinterest
MORE: 5 Tips for Instant Skin Glow from a Scientist’s Perspective
DISCOVER: Scientists Just Built Ridiculously Cute Tiny VR Headsets for Mice
The researchers attempted to reverse-engineer the model’s logic using visualisation tools, but even then, the AI only highlighted vague regions—such as blood vessels and optic discs—without offering any clear explanation. The truth is, deep learning models often operate like a black box: incredibly good at making predictions, but notoriously bad at explaining themselves.
And while gender recognition might sound like a fun party trick, it actually hints at something deeper. If AI can pick up on imperceptible gender signals in the retina, it could also detect equally subtle indicators of disease—before symptoms become obvious to clinicians. Retinal scans are already used to spot conditions like diabetes and Alzheimer’s. Add AI into the mix, and you’ve got a powerful new diagnostic tool—if we can figure out how to interpret it.
MORE: The Science Behind Coperni’s Ultra-Light, Space-Age Tote
DISCOVER: The Squalene Found in Beauty Products Endangers Sharks Says New Scientific Study
There’s also a challenge here for medical ethics and AI development. Doctors can’t rely on models they don’t understand, especially in clinical environments. So, this study highlights a growing need for “explainable AI”—systems that can justify their decisions in a way that humans can understand and trust.
As for what’s next? The researchers say their model is a proof of concept, but it raises big questions. Could these hidden retinal patterns be linked to hormones, genetics, or even early disease markers? And what else might AI spot in our biology—secrets that we don’t even know we’re hiding?
One thing’s for sure: AI isn’t just looking at the world differently—it’s seeing things we didn’t even know were there.
Leave a Comment