Automating Cataract Detection and Care Decisions
AI algorithms for image analysis hold promise for the future. Cheryl Guttman Krader reports.
Artificial intelligence (AI) with enhanced computing power and the increasing availability of big data brings new opportunities for developing novel systems and strategies for cataract detection, grading, and management, said Sheila John PhD, at the ARVO 2021 virtual meeting.
“Cataract is the leading cause of preventable blindness. Early detection and treatment can alleviate the suffering of cataract patients and prevent visual impairment from turning into blindness. AI deployed to diagnose cataract would also reduce ophthalmologists’ current workload and potentially enable them to serve more patients,” said Dr Sheila John, Head of Teleophthalmology and E-learning, Sankara Nethralaya, Chennai, India.
“Artificial intelligence is still relatively underexplored in cataract-related clinical practice and services, but it is poised to introduce a paradigm shift in the future. Despite the hype, however, it is important to keep our mind and focus grounded on the ultimate goal of implementation and patient safety.”
In her presentation, Dr John discussed studies of AI-based algorithms for automated cataract assessment using slit-lamp or fundus photographs. Researchers from China utilised deep learning via a residual neural network to create a three-step, sequential AI algorithm for cataract diagnosis and referral. The first step involves capture mode recognition. Next, lens state is diagnosed as normal, cataract, or postoperative intraocular lens. Last, detected cataracts are categorised by type and severity and management, either follow-up or referral for tertiary care, is determined. In their evaluation, the researcher found that the system performed robustly for completing all three tasks.
Recognising the increasing use of retinal imaging for diabetic retinopathy screening in primary care settings, another group of Chinese researchers developed an algorithm for cataract detection and severity grading based on classification of retinal image visibility. With ratings by experienced ophthalmologists serving as the ground truth, the algorithm correctly classified 94% of images for cataract detection and 91% of images for cataract severity.
Dr John noted that retinal imaging of patients who reside in rural communities without access to eye care providers is an opportunistic screening tool for automatic cataract detection as digital images of the lens may be acquired at the same session and transferred to a remote reading centre for automatic cataract detection and grading.
“In the future new AI systems can potentially provide better outreach for cataract screening, especially in rural or less resourced areas. With the increasing availability of ocular imaging modalities, including handheld retinal cameras or slit lamp adapters attached to smart phones, these imaging modalities are increasingly cheaper, easy to use, and can be used by trained technicians.”
Sheila John PhD: email@example.com
Figure 1 Retinal imaging used for automatic detection of diabetic retinopathy and cataract screening.