ARTIFICIAL INTELLIGENCE (AI) tech is constantly making humans look a bit basic, as is the case with DeepMind's AI tech that can detect eye diseases as accurately as doctors.
The Google-Alphabet-owned AI wrangler worked with Moorfields Eye Hospital and University College London to create a deep learning system that identifies common eye diseases by analysing 3D scans of patient's eyes. After that, it then recommends treatment for those with ocular afflictions.
Trained on some 15,000 optical coherence tomography scans - essentially 3D pictures of an eye created through bouncing near-infrared light off an eye's interior surfaces - the AI system was able to make the same diagnosis of some 50 eye diseases as human doctors 94 per cent of the time.
To have this positive level of accuracy the system uses two neural networks. One called a segmentation network was used to convert the scans into a 3D tissue map to show the layers of an eye and pinpoint the location of a disease. The system was trained on 877 scans, manually segmented by ophthalmologists, meaning it had expert information to work with rather than infer pattern by itself.
The second network works on classification, whereby it analyses the 3D tissue map and works out what diseases may be present in the scanned eye.
Separating the networks is unusual and normally they'd be combined and the system would simply spit out a diagnosis based n the data it was fed. But one of the problems with smart algorithms and systems is working out why they came to a decision, given the constantly changing nature of machine learning code. As such, separating the networks allows eye healthcare specialists to peruse the tissue map and spot how the AI system came to its final diagnosis.
Furthermore, the AI doesn't just give a single diagnosis, rather it provides a collection of them with how confident it feels each diagnosis is the correct one. This acts a safeguard for clinicians to check the results the AI has come up with and ensure a correct diagnosis in the mix.
DeepMind has worked with NHS hospitals before, to some controversy, in order to use smart tech to speed up diagnosis and help make the sifting through data an easier process for time-strapped doctors.
But this AI system does more than just diagnosis, as it also provides a triage service to decide which patients need care the most urgently. In a packed hospital, such a system could help rushed clinicians with managing the delivery of care.
"The number of eye scans we're performing is growing at a pace much faster than human experts are able to interpret them. The AI technology we're developing is designed to prioritise patients who need to be seen and treated urgently by a doctor or eye care professional," said Pearse Keane, an ophthalmologist at Moorfields Eye Hospital and a clinician-scientist at UCL, who co-authored of a research paper on DeepMind's work published in Nature Medicine.
"If we can diagnose and treat eye conditions early, it gives us the best chance of saving people's sight. With further research it could lead to greater consistency and quality of care for patients with eye problems in the future."
Currently, the AI systems is just a research project and will need further testing to firm up its capabilities, After that it'll need to go through clinical trials and meet the approvals of healthcare regulations before it can be put to work in eye hospitals.
But DeepMind's overall goal with the eye AI and other parts of its healthcare-related work is to push AI tech into the NHS to make life easier for doctors and other healthcare professionals when it comes to the diagnosis and delivery of care for patients.
This may require people to part with their medical data, which is a touchy subject of privacy for some despite the fact we live in a society where spilling your guts on Twitter or photograph and share online every meal you gobble.
But at the cost of some medical privacy, DeepMind could create systems that ease the massive burden put on NHS staff and infrastructure. Such tech could be the key to saving the NHS back from potential implosion; that, or killer robots are set loose to thin out the population - all valid options in our opinion. µ
Put a Ring-Con on it
We know. We're as surprised as you are
It's available across all major UK networks