More Collaboration Needed to Unlock AI’s Potential

ChatGPT co-founder Sam Altman recently said he doesn’t expect AI to cause the mass layoffs he once expected. AI development in general has stalled—but in ophthalmology, it continues.

Advanced AI-driven imaging is frequently used in keratitis diagnosis and treatment, but is this application as effective as possible? Jesper Hjortdal MD believes AI has potential but is not being used effectively, largely due to confusion over what it is and what it requires to work effectively.

“AI is a broader term for creating systems that can perform tasks that normally require human intelligence. Machine learning is a subset in which computers create a model that learns from patterns from data (like images) instead of being explicitly programmed with fixed rules,” Dr Hjortdal said.

“I looked at 10 studies of the use of AI in vivo confocal microscopy in keratitis. They report it is difficult to find high-quality images. They can include unrepresentative training sets with insufficient diversity, limited data sets to train machine learning models, and a lack of social acceptance and trust in the results.”

If the machine learning models ophthalmologists create do not have high-quality data to input, then it is not surprising there is dissatisfaction with AI, Dr Hjortdal said. The solution, he said, is to expand data sets, foster improved collaboration within ophthalmology, and develop user-friendly AI tools.

Dr Hjortdal pointed to a review examining machine learning AI in classifying infectious keratitis into clinically meaningful categories. The review included 37 studies, three of which came from the United States and the remaining 34 from across Asia, and the machine learning model was tasked with distinguishing bacterial versus fungal infection.

The review found that in studies focused on bacterial versus fungal classifications, area under the receiver operating characteristic (AUROC) values ranged from 0.81 to 0.89, and accuracy ranged from 71% to 93%. This, Dr Hjortdal said, emphasises how AI shows “significant potential to improve pathogen detection in infectious keratitis, enhancing diagnostic accuracy and accessibility.¹

“The conclusion is we must prioritise multicentre validation and create standardised methodologies to ensure reproducibility. Then we can begin to think ‘where are the needs for AI in patients with keratitis?’” Dr Hjortdal said.

“If you can just take an iPhone-based image, send it somewhere, and find out if it is infectious or non-infectious, that’s one key aspect. In primary eye care departments without access to polymerase chain reaction and microbial culture testing, this could be very useful.”

To ensure AI-driven machine learning reaches its full potential in ophthalmology, Dr Hjortdal said rigorous data validation is required, as are standardised performance metrics and open data reporting. The ultimate key is to ensure AI training is provided, as implementation requires “interoperability with healthcare systems and compliance with regulations.”

Dr Hjortdal presented at the EuCornea annual congress in Porto, Portugal.

Jesper Hjortdal MD is a clinical professor and senior consultant at the department of ophthalmology at Aarhus University Hospital in Denmark and a former president of EuCornea. jesper.hjortdal@clin.au.dk

1. Assaf JF, et al. Ophthalmol Sci, 2025 Jun 19; 5(6): 100861.