Seven AI projects from around the world

The World Association of Eye Hospitals conference was hosted by Moorfields Eye Hospital in London on 7 June, with delegates from around the world delivering presentations on the theme of artificial intelligence (AI).

Below OT shares our pick of projects presented at the event.

Detecting diabetic retinopathy in India and Thailand

Senior clinician scientist at DeepMind Health, Dr Alan Karthikesalingam, described several examples of how AI is being applied to image-led medical specialties, such as radiology, pathology and ophthalmology.

Within eye health, he highlighted that Google has developed a deep neural network that reads fundus photographs. The model performs on par with retinal specialists in detecting disease, Dr Karthikesalingam added.

He highlighted that the system is now being used to detect diabetic retinopathy in India and Thailand.

Man or woman?

Google has trained algorithms to identify gender from fundus photographs using UK Biobank data.

Planning treatment for cancer patients

Dr Karthikesalingam shared that an AI system has been developed that can achieve the same level of accuracy as head and neck cancer specialists in delineating organs for radiotherapy.

The task takes the system seconds, where it may take a clinician hours, he added.

Identifying lung cancer

AI that can identify lung cancer with the accuracy of specialist clinicians was also highlighted by Dr Karthikesalingam.

He added that early diagnosis is key when it comes to the disease.

“If it is caught earlier, survival is much higher,” Dr Karthikesalingam emphasised.

Predicting readmission to hospital

Dr Karthikesalingam highlighted that AI is being applied to the data that is generated when a patient enters the hospital system.

“These days when patients come in to hospital they create this digital exhaust,” he shared.

Scientists wondered what would happen if electronic health records were used to predict adverse future adverse events.

An AI system has outperformed existing methods for predicting hospital readmissions and mortality by 10%, Dr Karthikesalingam said.

Screening for glaucoma

Koen Vermeer, director of the Rotterdam Ophthalmic Institute, delivered a presentation on a project to establish a data set that would aid the development of screening for glaucoma using AI.

His team aims to get 100,000 retinal images from an optical retail chain and augment that with 10,000 secondary care images.

This data could then be used to train an algorithm to detect features of disease.

The research group plans on providing open access to the data set to aid progress on AI-based screening for glaucoma.

Predicting glaucoma progression

Rahul Shah, assistant administrator at the Wilmer Eye Institute in the US, detailed several ways that AI is being used within ophthalmology to improve clinical care, education and research at the institute.

One avenue that is being explored is applying AI at the first presentation of glaucoma patients to help predict the trajectory of the patient’s disease.

Mr Shah highlighted that the project aimed to establish whether it was possible to identify those at greatest risk for rapid onset glaucoma from baseline measurements and provide timely interventions.

The AI predictions have so far been 90% accurate across a data set of one million visual field tests.