Deep Learning to Improve Diagnosis Must Also Not Do Harm

Both arteritic and nonarteritic anterior ischemic optic neuropathy (AAION and NAION) are conditions that cause sudden vision loss in people over 50 years old. AAION requires immediate treatment with corticosteroids to prevent further vision loss and complications, while NAION does not have any proven acute interventions. Differentiating between these two conditions is important to reduce the risk of vision loss and complications. However, diagnostic errors are common, and access to neuro-ophthalmic experts is limited. The BONSAI Group has developed a deep learning system (DLS) that can classify AAION and NAION based on ophthalmoscopic photographs. The DLS performs better than human experts in image interpretation and could potentially supplement frontline eye care. However, the extent of its benefit is still unclear, and further research is needed to determine its real-world impact. The DLS has limitations, as it can only classify the conditions it has been trained on and may contribute to diagnostic errors if applied to other optic neuropathies. The generalizability of the DLS is also dependent on training data and potential biases. To justify its clinical implementation, the DLS-enhanced workflow needs to be compared to current medical practice. Developing nonhuman or human intelligence-enhancing strategies to improve diagnostic accuracy for neuro-ophthalmic conditions is important due to their impact on visual function and quality of life. The BONSAI Group is commended for their efforts in this area.