An Introduction to the Use of Artificial Intelligence in Ophthalmology

  • Post author:Zaid Alsafi, Sara Fatima Memon, Ammar Mohamed Yusuf
  • DOIDOI:10.48089/jfo7689090
  • Reader Impact RatingImpact Rating: 9.59 / 10 from 165 reader votes.

Zaid Alsafi1, Sara Fatima Memon2, Ammar Mohamed Yusuf3

1 The Hillingdon Hospitals NHS Foundation Trust

2 The Royal Bournemouth Hospital

3 Western Eye Hospital Imperial College Healthcare NHS Trust

Introduction

Artificial intelligence (AI) is a discipline within computer science which aims to develop tools capable of undertaking tasks that usually require human intellect. Coined by the Computer Scientist John McCarthy in 1956, AI has been likened to the industrial revolution (1). It has played a pivotal role in the growth of many key industries, ranging from the financial sector to manufacturing. Seamlessly integrated into our daily lives; it is used to recommend products when shopping online, suggest films based on our preferences when binge-watching Amazon Prime and recognise speech when calling an automated call centre (2).

The potential of AI has not gone unnoticed within healthcare and a race to develop efficient tools to optimise patient care has begun, with the recent pandemic hastening the need for digital transformation. From a logistical perspective, AI has been used to send reminders to patients about vaccinations, organise follow up appointments and recognise adverse drug reactions in polypharmacy (3). Clinically, AI has shown promise in image analysis. It has been successful in detecting tuberculosis on Chest X-rays, skin cancer from photographs and metastases from CT scans (4).  Ophthalmology is an image-rich speciality lending itself well to AI, with a variety of well-defined conditions where a ‘spot-diagnosis’ can be made on multimodal imaging.

Applications in Ophthalmology

Diabetic Retinopathy

Timely recognition using methods such as slit-lamp biomicroscopy and retinal imaging, referral, and treatment of diabetic retinopathy is vital to minimising visual loss. For this reason, screening programs performed by ophthalmologists and other allied healthcare professionals exist worldwide. Recent studies have shown that novel AI algorithms can match, and in some cases, outperform clinicians when detecting diabetic retinopathy from fundus photographs, achieving a sensitivity of 94% and a specificity of 98% (5). The implementation of these algorithms were also shown to be more cost effective.

Glaucoma

Glaucoma can result in significant morbidity due to irreversible sight loss. With an insidious onset in chronic forms, early detection and treatment is of paramount importance. Using fundus photographs, Li et al (6) developed an algorithm that can detect glaucomatous optic neuropathy with a sensitivity of 95.6% and specificity of 92%. Algorithms have also been used to detected visual field loss during the early stages of glaucoma and could predict visual field loss 5 years into the future with a high degree of accuracy (7).

Age-related macular degeneration

Age related macular degeneration (AMD) is a significant cause of visual impairment in elderly populations and is expected to affect 300 million individuals within the next 2 decades (7). Models have not only been able to grade the severity of AMD using fundus photography (sensitivity of 97.8% and specificity of 97.4%), but also predict disease progression (7). AI has also been used to predict outcomes and dosage requirements of Anti–VEGF therapy which could optimize the use of resources (7).

Paediatric ophthalmology

Whilst most advances in AI have been on adult ophthalmic conditions, paediatric ophthalmology is a field which can hugely benefit from AI. Given the developmental capabilities, inability to communicate effectively and unpredictable behaviour of young children, reliable and objective investigations are key. Brown et al (8) developed a model to detect plus disease in retinopathy of prematurity (ROP) from fundus images, focusing on vessel tortuosity. Their model demonstrated a better accuracy in detecting plus disease (sensitivity 93%, specificity 94%) than seasoned ophthalmologists.

Other Ophthalmic Conditions

These are but a few examples of AI’s utility. It has been used to grade and diagnose cataracts, identify populations that are at risk of developing congenital cataracts, diagnose anterior segment disease, distinguish between benign and malignant disease, and predict the suitability for refractive surgery (7).

Challenges and ethical implications of AI

The decision-making process of an AI model is often referred to as a ‘black box’. Put simply, we have an input and an output but limited knowledge of how the answer was obtained. The quality of AI algorithms is heavily dependent on the quality and quantity of data used to train the model. In short, the model is able to extract features from large data-sets and ‘learn’, subsequently recognising patterns through learning-based predictions when faced with new data-sets. This requires exceedingly large data sets, raising concerns regarding data privacy and ownership at a time where data theft is becoming commonplace.

At present, a significant portion of algorithms require an ophthalmologist to label images and areas indicative of disease (such as drusen). This is not only time consuming and labour intensive, but is dependent of the skill of the ophthalmologist and introduces a degree of human error, susceptible to unintended racial, socioeconomic and gender bias. These issues may ultimately result in mistrust and limit patient engagement with screening programs.

Additionally, datasets are unevenly distributed, with 172 countries comprising 45% of the global population, found to have no ophthalmological data-sets (9). A lack of representative data-sets (overly focussed on glaucoma, diabetic retinopathy and age-related macular degeneration with limited information on rare diseases and ethnic minorities) can result in underdiagnosis and misdiagnosis of pathology, further widening healthcare inequalities. This begs the question: who is responsible for the misdiagnosis of a sight threatening condition, the algorithm, the engineer or the clinician?

Conclusion

As we enter a new era of immersive virtual experiences, artificial intelligence is being implemented in all aspects of life, to enhance human interaction (10). In ophthalmology, current studies have shown acceptable outcomes when detecting disease. We should aim to educate the current and future generation of ophthalmologists to critically appraise these tools. In addition, we should invest in health informatics globally, when taking into account that 89% of the visually impaired live in low to middle income countries (11). This should be combined with stringent tools to ensure standardization and transparency of datasets. Future research should also aim to tackle the ‘black box’ nature of AI algorithms and minimise healthcare inequalities.

References

1.            Shannon CE, McCarthy J. Automata Studies. (AM-34), Volume 34. Princeton University Press; 2016. 297 p.

2.            Basu K, Sinha R, Ong A, Basu T. Artificial Intelligence: How is It Changing Medical Sciences and Its Future? Indian J Dermatol. 2020;65(5):365–70.

3.            Amisha, Malik P, Pathania M, Rathaur VK. Overview of artificial intelligence in medicine. J Fam Med Prim Care. 2019 Jul;8(7):2328–31.

4.            Ting DSW, Pasquale LR, Peng L, Campbell JP, Lee AY, Raman R, et al. Artificial intelligence and deep learning in ophthalmology. Br J Ophthalmol. 2019 Feb;103(2):167–75.

5.            Gargeya R, Leng T. Automated Identification of Diabetic Retinopathy Using Deep Learning. Ophthalmology. 2017 Jul 1;124(7):962–9.

6.            Li Z, He Y, Keel S, Meng W, Chang RT, He M. Efficacy of a Deep Learning System for Detecting Glaucomatous Optic Neuropathy Based on Color Fundus Photographs. Ophthalmology. 2018 Aug 1;125(8):1199–206.

7.            Jin K, Ye J. Artificial intelligence and deep learning in ophthalmology: Current status and future perspectives. Adv Ophthalmol Pract Res. 2022 Nov 1;2(3):100078.

8.            Brown JM, Campbell JP, Beers A, Chang K, Ostmo S, Chan RVP, et al. Automated Diagnosis of Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural Networks. JAMA Ophthalmol. 2018 Jul 1;136(7):803–10.

9.            Khan SM, Liu X, Nath S, Korot E, Faes L, Wagner SK, et al. A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability. Lancet Digit Health. 2021 Jan;3(1):e51–66.

10.          Milmo D, editor DMG. Enter the metaverse: the digital future Mark Zuckerberg is steering us toward. The Guardian [Internet]. 2021 Oct 28 [cited 2022 Nov 8]; Available from: https://www.theguardian.com/technology/2021/oct/28/facebook-mark-zuckerberg-meta-metaverse

11.          Bourne RRA, Flaxman SR, Braithwaite T, Cicinelli MV, Das A, Jonas JB, et al. Magnitude, temporal trends, and projections of the global prevalence of blindness and distance and near vision impairment: a systematic review and meta-analysis. Lancet Glob Health. 2017 Sep;5(9):e888–97.

Leave a Reply