IT West Ltd

From the blog

News from IT West

Diabetes Eye Disease Diagnosing System Needs No Doctors

An AI system for diagnosing eye disease caused by diabetes that has been approved for use in the US works autonomously and doesn’t need a doctor to interpret its results.

New Way To Solve Old Problem

Diabetic retinopathy, a leading cause of blindness among adults, is caused by high blood sugar levels damaging the blood vessels of the light-sensitive tissue at the back of the eye / the retina. The condition affects up to eight out of 10 people who have had diabetes for 10 years or more.

Given the extent of the problem, Google and DeepMind are reported to have been working on building machine-learning algorithms for detecting diabetic retinopathy for some time.

The new AI-based device from Iowa diagnostics company IDx LLC is the first FDA-approved AI system for diagnosing this particular eye disease.

No Doctors Required For Diagnosis

The system can be used to spot the disease i.e. signs of mild diabetic retinopathy in scans of people’s retinas. This would normally be a job that would require human input, and as such, the new device is a first in eye care.

Although the system can diagnose the disease on its own, and therefore, doesn’t require a doctor’s input for diagnosis, it cannot recommend treatment plans, as this requires human doctors.

How Does It Work?

The system uses two convolutional neural networks.

The first one studies and analyses the image quality of retinal scans, from this it can determine if the focus, colour balance, and exposure are good enough to pass the photos to the diagnostic algorithm.

The second stage / network looks for common signs of damage related to the disease e.g. haemorrhages from burst blood vessels which may be caused by unstable blood sugar levels.

From these processes, the system is able to make a diagnosis.

How Accurate Is It?

Given the complicated nature of the medical condition, the accuracy of the system has been tested (using 900 subjects) in terms of its sensitivity, specificity and imageability. The device is reported to have scored 87% sensitivity i.e. identifying patients who have a mild version of the condition, 90% cent specificity i.e. indentifying those with no eye damage, and 96% imageability i.e. a high enough quality of image was generated to achieve a diagnosis.

What Does This Mean For Your Business?

AI is being incorporated in more value adding and innovative ways to solve many problems across all industries and sectors, and as such, represents an opportunity for those businesses developing devices and systems with an AI element.

Not only does this device perform an important part of a service that hitherto required expert human input, it also frees up time that the human expert would have spent on diagnosis, thereby allowing valuable medical resources to be extended and allocated elsewhere. This demonstrates how AI can add value, save time / costs, and allow more leverage to be gained from existing services.

We already trust devices / machinery to handle many important aspects of medical care, and with this in mind, there should be no real reason to mistrust the accuracy and fitness for purpose of this system, particularly given that it has been tested, and that there will be human input at the treatment plan stage that may help to spot any errors.

AI in medical care represents an important step into the future that could bring some incredible benefits.