docet

Key Points

• Increased amounts of data and improved processing power have allowed the development of new machine learning systems.
• A variant of machine learning is deep learning and is based on artificial neural networks which mimic to some extent the human brain in that it can learn to take an input, process it and give a predicted output which can be compared with the actual output (such as an image).
• Expert human input can allow the machine to get better in its predictive accuracy.
• Predicted eye disease prevalence mean that early detection is not going to be possible with current screening and assessment protocols.
• AI will be able to assist the clinician to make the best decision at the earliest opportunity.
• Pegasus was designed by giving it many images of known conditions and it supports two main modalities; standard digital images and high resolution OCT scans.
• The system can assess images for signs of disease at the macula or throughout the retina, as with diabetic retinopathy. For the latter, it can grade retinopathy and also identify individual lesions such as microaneurysms and haemorrhages. OCT data can be analysed in seconds.
• AI systems have potential in telemedicine projects where technicians in the field may gather data for further analysis by a centrally located clinician.
• Such AI systems assist but will never replace eye care professionals. Decisions about care and referrals will always need human input.