A recent study found that the diagnostic performance of an artificial intelligence (AI) software platform was comparable to European optometrists and ophthalmologists when diagnosing glaucoma. The researchers believe that deep learning-based AI systems could provide specialist-level accuracy for screening and detecting glaucomatous optic neuropathy, potentially reduce the number of false positives and provide such analysis at very low costs.

The retrospective study compared the performance of AI software and 208 British optometrists and 243 European ophthalmologists (glaucoma specialists) in detecting glaucomatous optic neuropathy from 94 scanned stereoscopic photographic slides.

They found no statistically significant difference between the performance of the deep learning system and the practitioners. The software detected glaucomatous optic neuropathy with an accuracy of 83.4%, while optometrists and ophthalmologists had 80% and 80.5% accuracy, respectively. The AI system also had a higher intra-observer agreement than optometrists or ophthalmologists.

Researchers found that the AI software was fast at grading images, taking seven seconds to grade each image (including the upload time). They think it is likely that the AI would be faster than humans at analyzing the stereoscopic optic disc images, which could help reduce time and expenses for screening programs. The researchers note that further clinical and economic analysis is necessary before considering the integration of AI systems into real-world clinical practices.