What are the benefits and challenges of AI for glaucoma detection in the developing world?
Click Here to Manage Email Alerts
Click here to read the Cover Story, "MIGS breaks through barriers in developing world."
The benefits
Glaucoma is a major blinding disease. The onset is hidden, and early diagnosis is difficult. Cost-effective screening tools are lacking, and often we diagnose the disease when the optic nerve damage is irreversible.
Using AI, we can accelerate the management of glaucoma, including diagnosis and prediction, screening and monitoring. AI’s diagnostic capabilities in identifying complex conditions frequently exceed those of human specialists. A case in point is an Indian study in which AI successfully identified 94% of genuine glaucoma cases, significantly more than the 60% identified by human experts.
Early prediction of disease progression offers the opportunity to locate the individuals who need early interventions and allows physicians to develop rational treatment plans, saving money and avoiding unnecessary treatments and treatment side effects.
Currently available AI-based fundus imaging tools perform well in detecting glaucoma suspects with high sensitivity and specificity. Some of them were successfully used in screening campaigns in sub-Saharan Africa. Our team has also developed the first bimodal AI system based on visual field and OCT.
Glaucoma diagnosis through AI algorithms introduces a uniform approach, reducing the inconsistencies inherent in human decision-making. This uniformity is especially beneficial in regions where glaucoma diagnostic training is limited and there is a critical shortage of glaucoma specialists.
The combination of smartphone and AI technology for screening and monitoring has great potential for glaucoma management, particularly in the rural areas of developing countries where the burden of disease is most significant. Researchers have already developed smartphone-based stereoscopic fundus cameras, and our team has developed a multimodal visual field-based deep learning algorithm, iGlaucoma, a smartphone, cloud-based glaucoma detection tool.
The use of portable technologies is a powerful helper in automated screening and monitoring. With a single smartphone and AI, we can implement glaucoma telemedicine programs and contribute to reducing inequalities in glaucoma screening, making it more widely available and possibly cost-effective.
Xiulan Zhang, MD, PhD, is a professor of ophthalmology at Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, China.
The challenges
AI holds the potential to bring groundbreaking shifts in the health care paradigms of glaucoma, but there are many challenges related to both the development and the deployment of AI algorithms.
First is the absence of standardized diagnostic criteria. Even in studies with the same aim, such as diagnosing glaucoma based on fundus photography, different criteria are used to establish what is glaucoma and what is glaucoma suspect. The lack of universally accepted, objective standards for glaucoma diagnosis makes the application of AI in its detection and diagnosis challenging. In addition, most of the algorithms are developed based on homogeneous populations. When these models are tested on data from another country, the performance may get worse. For example, models trained on people of Japanese descent perform poorly on people of African or European descent.
Regulatory hurdles are, at this stage, another significant limitation. The existing regulatory frameworks are not applicable to a technology that is intrinsically and by its nature dynamic and fast-changing, and new adaptive approaches are needed. The implementation of health care AI also poses serious privacy challenges. Ensuring the confidentiality and security of patient data is paramount, especially when dealing with AI systems that require comprehensive training data sets. Adhering to local data protection regulations and maintaining privacy are significant hurdles.
In order to deploy AI tools in real-world practice, gaining the confidence of health care providers and patients in AI systems is vital. This is particularly challenging in regions where the integration of AI into health care is still in its infancy and skepticism regarding the accuracy of AI diagnoses persists. In addition, the effective deployment of AI for glaucoma detection demands stable internet connectivity, suitable digital infrastructure and compatible diagnostic tools, which are often unavailable in many developing countries. Considerable investments will be needed, and time and resources will be required to train health care professionals to use these tools proficiently.
Fei Li, MD, PhD, is with Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, China.
Collapse
Read more about
Click Here to Manage Email Alerts