News & Perspective

New artificial intelligence system to predict conversion to wet age-related macular degeneration

Ophthalmology
2 months ago, OP Editor

Exudative ‘wet’ age-related macular degeneration (exAMD) is a chronic eye disorder that causes visual deterioration in the central vision area. Patients with exAMD in one eye often develop exAMD in the fellow eye. Symptoms of exAMD usually appear suddenly and worsen rapidly, making the time point of conversion to exAMD for the fellow eye a critical moment to intervene.1 Moorfields Eye Hospital, in partnership with Google Health, DeepMind and University College London, published an article on an artificial intelligence (AI) system that can achieve better prediction of imminent conversion to exAMD in patient’s fellow eye.

exAMD is generally caused by abnormal blood vessels leaking fluid or blood into the macula that results in blurred vision or blindness.1 Once exAMD develops, sight is rapidly lost and often cannot be fully restored by current therapies.2 As patients with exAMD in one eye often develop exAMD in the fellow eye, the early detection and treatment of exAMD before conversion of the fellow eye is critical in reducing vision loss and, in some cases, recovering vision.1 Although preventive strategies are being studied, robust prediction methods of identifying exAMD onset before conversion are needed to prevent the development of the disease in the fellow eye.2

An AI system is one such method that can help predict whether a fellow eye will convert to exAMD. By using the optical coherence tomography (OCT) scans, sensitivity and specificity of the AI system can be verified through comparing the model’s performance with the clinical assessments of retinal specialists and optometrists.2 In a study by Yim et al., the AI system was adopted to predict the conversion to exAMD based on an interpretable tissue segmentation of the OCT and the raw OCT itself. A deep learning (DL) segmentation model was then used to output a 3D anatomical and the pathological tissue segmentation helped predict the risk of conversion to exAMD in the fellow eye within 6 months.2

This was a retrospective and consecutive cohort study on AI between June 2012 and June 2017 (n=2,795) that consisted of 62% female with an average age of 78.8 years having the first eye presentation of exAMD.2 The primary investigations were the sensitivity and specificity of the prediction and the number of false-positive outcomes. Further investigation on the automatic segmentations of clinically relevant tissue types were also included to identify early eye changes and high-risk subgroups.2

Based on the results, the AI system achieved 80% per-volumetric-scan sensitivity at 55% specificity, and 34% sensitivity at 90% specificity.2 Positive and false-positive cases of exAMD were identified, at 78% and 56% of the time, respectively.3 Comparing to clinical expert assessment, the AI system outperformed the majority of experts in predicting the exAMD conversion within 6 months even when the experts had additional access to patients’ previous OCT records.2 By automatically segmenting clinically relevant tissue types by volume, the AI system also provided an efficient quantitative method to stratify clinical subgroups and outperformed manual segmentation methods like the drusen or the hyper-reflective foci method alone.2

The limitations of this study could be that the AI system was trained by demographic data from Moorfields Eye Hospital only, and the multifactorial aspect of exAMD may not be fully evaluated and represented in the study results.2 Some exAMD cases were also not included due to the missing of subtle early signs of conversion or the patient was asymptomatic.2

To conclude, the AI system was demonstrated to enable earlier identification of high-risk exAMD patients and provided guidance to preventative treatments. Through automatic segmentation, new possibilities are now made available to retinal specialists and optometrists for further exAMD researches.2 Dr. Pearse Keane, consultant ophthalmologist at Moorfields Eye Hospital, commented, ‘’In about 30%-40% of exAMD patients who are receiving injections in one eye, we can now predict with very high specificity whether they will develop wet age-related macular degeneration in their good eye within the next 6 months.’’4

  1. Wet macular degeneration. Mayo Clinic. https://www.mayoclinic.org/diseases-conditions/wet-macular-degeneration/symptoms-causes/syc-20351107. Accessed July 20, 2020.
  2. Yim, J et al. Predicting conversion to wet age-related macular degeneration using deep learning. Nat Med 2020; 26: 892–899.
  3. DeepMind AI system ‘detects risk of developing serious eye condition’. Digital Health. https://www.digitalhealth.net/2020/06/deepmind-ai-system-detects-risk-of-developing-serious-eye-condition/. Accessed July 20, 2020.
  4. Predicting conversion to wet age-related macular degeneration using deep learning. Optometry Today. https://www.aop.org.uk/ot/science-and-vision/technology/2020/07/02/predicting-conversion-to-wet-agerelated-macular-degeneration-using-deep-learning. Accessed July 20, 2020.

Tags

Menu Section