In the JournalsPerspective

AI-based retinal screening systems may outperform clinicians

Studies have shown that artificial intelligence-based diabetic retinopathy screening algorithms have identified disease at a high rate of sensitivity and specificity, according to a paper recently published by a group of experts.

AI is a branch of computer science aimed at creating intelligent machines. By imitating the neural structure of the central nervous system, artificial neural networks (ANNs) are capable of complementing or substituting some of the visual recognition tasks performed by physicians.

Mimicking the paths of the human brain, they can potentially identify and quantify pathological features, formulate a diagnosis, produce treatment algorithms and, based on the processing of large amounts of data, predict the course of a given disease in a given patient. Several studies have demonstrated that AI in specific fields can already perform equally to or better than practicing clinicians.

AI-based systems have recently been approved for screening diabetic retinopathy (DR), and their performance was shown to be equal to that of a panel of expert ophthalmologists, achieving a high degree of sensitivity and specificity. In April, the first FDA-approved AI system to detect a greater than mild level of DR was launched on the market.

“FDA approval was granted based on a study of 900 patients with diabetes at 10 primary care sites, which resulted in correct identification of a positive findings indicative of DR in 87.4% of individuals and a correct negative result in 89.5%,” the authors wrote.

The largest study on the subject was performed by Ting and colleagues within the Singapore National Diabetic Screening program. The study evaluated the sensitivity and specificity of an AI system for the detection of DR and concomitant diseases in 494,661 images. Sensitivity ranged from 91% to 100% and specificity between 73% and 92%.

“All these examples show that AI-based DR screening algorithms have reached or may even outperform the level of accuracy of clinical experts. DR screening in particular carries enormous potential as support for ophthalmologists, may help to reduce the prevalence of late and cost-intensive disease stages and is likely to pioneer digital medicine applications in the near future and at a large scale,” the authors wrote. – by Michela Cimberle

Reference:

Ting DSW, et al. JAMA. 2017;doi:10.1001/jama.2017.18152.

Disclosure: Schmidt-Erfurth is a consultant for Boehringer Ingelheim, Genentech, Novartis and Roche. Please see the study for the other authors’ financial disclosures.

Studies have shown that artificial intelligence-based diabetic retinopathy screening algorithms have identified disease at a high rate of sensitivity and specificity, according to a paper recently published by a group of experts.

AI is a branch of computer science aimed at creating intelligent machines. By imitating the neural structure of the central nervous system, artificial neural networks (ANNs) are capable of complementing or substituting some of the visual recognition tasks performed by physicians.

Mimicking the paths of the human brain, they can potentially identify and quantify pathological features, formulate a diagnosis, produce treatment algorithms and, based on the processing of large amounts of data, predict the course of a given disease in a given patient. Several studies have demonstrated that AI in specific fields can already perform equally to or better than practicing clinicians.

AI-based systems have recently been approved for screening diabetic retinopathy (DR), and their performance was shown to be equal to that of a panel of expert ophthalmologists, achieving a high degree of sensitivity and specificity. In April, the first FDA-approved AI system to detect a greater than mild level of DR was launched on the market.

“FDA approval was granted based on a study of 900 patients with diabetes at 10 primary care sites, which resulted in correct identification of a positive findings indicative of DR in 87.4% of individuals and a correct negative result in 89.5%,” the authors wrote.

The largest study on the subject was performed by Ting and colleagues within the Singapore National Diabetic Screening program. The study evaluated the sensitivity and specificity of an AI system for the detection of DR and concomitant diseases in 494,661 images. Sensitivity ranged from 91% to 100% and specificity between 73% and 92%.

“All these examples show that AI-based DR screening algorithms have reached or may even outperform the level of accuracy of clinical experts. DR screening in particular carries enormous potential as support for ophthalmologists, may help to reduce the prevalence of late and cost-intensive disease stages and is likely to pioneer digital medicine applications in the near future and at a large scale,” the authors wrote. – by Michela Cimberle

Reference:

Ting DSW, et al. JAMA. 2017;doi:10.1001/jama.2017.18152.

Disclosure: Schmidt-Erfurth is a consultant for Boehringer Ingelheim, Genentech, Novartis and Roche. Please see the study for the other authors’ financial disclosures.

    Perspective
    Mark Swanson

    Mark Swanson

    “Jaw-dropping,” “astonishing” and “game changing” are just a few of the superlatives that have been applied to artificial intelligence as it relates to the technology’s medical capabilities. Image analysis might be the first thought that AI conjures up, but research demonstrating AI’s ability to predict re-hospitalization after acute illness and death from congestive heart failure shows that image analysis is just the tip of the iceberg.

    Progress in Retina and Eye Research is the preeminent journal in the eye field. Papers are by invitation only, and every paper is important. AI is going to change the face of medicine. Optometry is going to be impacted by AI in a profession-altering way. The profession needs to start charting its course for the future now. Watchful waiting isn’t an option.

    • Mark Swanson, OD, FAAO
    • Professor, University of Alabama at Birmingham, Department of Optometry and Vision Science Senior scientist, Comprehensive Center for Healthy Aging, Center for Community Health, Center for Exercise Medicine and Vision Science Research Center Director, Ocular Disease and Low Vision Clinic

    Disclosures: Swanson reported no relevant financial disclosures.