This article is more than 5 years old. Information may no longer be current.
Do you envision ophthalmologists becoming too dependent on artificial intelligence?
Click Here to Manage Email Alerts
Click here to view the Cover Story for this Point/Counter.
Working with the machine
A common reflex reaction to machine learning advances in medicine is concern that AI may one day replace physicians or that physicians become too dependent upon it, deteriorating their clinical skill set as a result.
However, it is important to keep in mind that this technology has been developed to work synergistically with clinicians and to help augment their care. Numerous studies have demonstrated that AI, together with humans, outperforms either component alone in whatever task was being evaluated.
I believe our immediate future with AI in ophthalmology will be one in which it helps decrease task burden on the physician, especially for repetitive tasks in which a human may be more prone to error with fatigue (for example, automated image analysis). In such an environment, there would be human oversight to affirm or dispute the findings of the AI. This would allow more patients than ever before to at least obtain some form of access to care by being screened for potentially blinding diseases earlier, and those at risk will be getting referred to the clinician earlier for timely intervention care.
In the longer run, AI has shown early promise in potentially being able to guide treatment decisions and clinical outcomes. This is the evolving field of precision or predictive medicine. If and when this time comes, it will be important as a field to once again work in tandem with the machine, using it to supplement, rather than supersede, the physician’s clinical acumen.
Ehsan Rahimy, MD, is a vitreoretinal physician and surgeon at Palo Alto Medical Foundation, Palo Alto, California. Disclosure: Rahimy reports no relevant financial disclosures.
Not a detriment to rely on tools
This is a fair question when you consider that a 2016 White House report stated that, “AI holds the potential to be a major driver of economic growth and social progress.” The short answer to the question is yes, especially if it is close to ground clinical truth greater than 97% of the time. Let’s face it. Navigation apps have done away with handwritten directions and the use of road atlases to aid us in getting from point A to point B, largely because they work. If our virtual navigation tools got us lost even 5% of the time, we would not use them. Are we too dependent on navigation tools? This is a semantic matter.
AI shows promise to detect referable diabetic retinopathy, identify the glaucoma-like disc and find visual field progression with a higher degree of accuracy than a clinician. Ophthalmologists will demand validation that these tools actually work, and once that is established, it is reasonable to rely on them. With that said, patients will demand that we articulate their outcomes with the use of these tools and will not accept black box yes or no answers. Essentially, we will have to use the output from an AI algorithm to satisfactorily explain a patient’s disposition. As such, we will seek construct validity within the patient’s source data to further validate AI outcomes. Thus, it will not be a detriment that we will rely on these tools because they will teach us to see high-dimensional features in the patient’s data set that we currently ignore yet contain critical information. Reliance on AI will make us better ophthalmologists and will result in better patient outcomes.
Louis R. Pasquale, MD, FARVO, is professor of ophthalmology at Harvard Medical School and the director of glaucoma service at Massachusetts Eye and Ear, Boston. Disclosure: Pasquale reports he is an unpaid consultant to Visulytix and EyeNexo.