Title : AI in eye care: Prospects and challenges
Abstract:
Introduction: Artificial intelligence (AI) has transformed medicine, with various applications across diagnostics and clinical decision tools that have been applied in many specialties, including ophthalmology. Systems that use AI have been implemented to identify retinal disorders like diabetic changes, drusen, and glaucoma. Nevertheless, as exciting as the prospects of AI can be, it does not come without its risks: bias, generalizability, data privacy, integration into clinical systems, and ethical concerns are common potential challenges.
Examples of Current AI applications in Ophthalmology: Diabetic retinopathy (DR): The use of AI in DR screening has shown impressive sensitivity and specificity in detecting referable cases compared to a panel of experienced ophthalmologists, with a sensitivity as high as 97.5% and 96.1% and a specificity of 93.4% and 93.9% respectively.
Age-related Macular degeneration (AMD): AI has achieved sensitivity and specificity of 88% and 91% respectively, in detecting AMD from fundus photographs, a performance rivalling that of retina specialists.
Public health and tele-ophthalmology: The use of smartphone technology as a user-friendly tele-option can allow large-scale, readily-accessible screening for common culprits of blindness like glaucoma, cataract, DR, and AMD. Challenges facing AI in ophthalmology: Several challenges are facing the use of AI in ophthalmology
Bias and patient safety: AI may perform impressively during training, but give false alarms and miss-diagnoses in the real world, jeopardizing patient safety.
Data privacy and cybersecurity: Breaches in data held in AI systems can lead to leaks of confidential information, the sale of health information for political or cyberwarfare purposes, and potential jeopardy to patients’ safety if hackers manage to access and configure AI algorithms, potentially giving rise to incorrect diagnoses or management plans.
Ownership of health records: The line between who owns patients’ personal health records can become blurred when data is transferred from health institutions to AI companies without their consent.
The ‘’black-box’’ issue: How transparent AI-based clinical decisions have also been debated in literature as ‘’black-box’’ AI-based tools seem to come up with clinical choices that may seem incomprehensible to patients and even Ophthalmologists. Using simpler ‘’white-box’’ tools that utilize decision trees may seem more transparent; however, they are less efficient than the former black-box ones.
Medical Liability: If an AI system makes an error with negative consequences for the patient, who will be held liable? The system itself, the ophthalmologist, the hospital, the software developer, and so forth.
Accessibility: Whether everyone in a population would benefit from AI-based solutions and be able to access them equally remains a significant concern when adopting them.
Conclusion: AI has revolutionized eye care. From public-based applications like screening programs to more sophisticated individualized decision-making tools and prognostic predictions, AI is anchoring its role in shaping the present and future of ophthalmology. Nevertheless, AI also comes with potential vulnerabilities like bias, legal liability, and accessibility. Therefore, future research should focus on mitigating those challenges to ensure the maximum advantage of this collaboration while minimizing its risks.
