This article is more than 5 years old. Information may no longer be current.
It is our job to harness AI in medicine
Click Here to Manage Email Alerts
A presentation at Digestive Disease Week concluded that, “AI detects polyps better than colonoscopists.” Another study in Annals of Oncology online reported, “AI beats dermatologists on skin lesion images.” On Medscape, Linda Brookes asked, “Diabetic retinopathy: Can artificial intelligence provide a better way to detect disease?”
It’s understandable the medical community feels a bit conflicted about artificial intelligence (AI). After all, isn’t it enough contending with arbitrary hurdles imposed by insurers, complying with ever-changing governmental mandates and dealing with the EHR debacle every day? Do doctors now face the prospect of being replaced by a machine? This is an unsettling thought for all of medicine, but perhaps even more for optometry as we churn out a record number of graduates each year. In an era of exponentially increasing technology, we all appreciate what AI brings to the table, but its disruptive nature is a little uncomfortable.
It was just a matter of time before AI found its way into our world. Health care deals in large volumes of data and strives for a reasonable level of probability but also very much relies on instinct and experience. While we’ve made tremendous strides in recent years (think evidence-based medicine and white papers), we are still not an exacting science. To a great degree, AI is really all about exacting science. Given a large enough data set and a robust neural network, a machine can be trained to become exceptionally adept at identifying colon polyps, skin lesions and diabetic retinopathy. Simply put, AI has the potential to change the practice of medicine to the exacting science of medicine. If this is the case, is it time we all hang up our 90-D lenses and binocular indirect microscopes? Well, maybe not so fast.
At the recent Human Intelligence and Artificial Intelligence in Medicine Symposium at Stanford University, experts from around the world debated the readiness of AI in medicine. While the consensus was positive, there was a recurring sentiment of proceeding cautiously. Even though the FDA has already approved a few AI apps (stroke, bone health and diabetic retinopathy detection), there was the obligatory call for more trials and clinical data.
But perhaps even more importantly is the concern that AI is still a machine-based discipline. Simply put, you can teach a machine to analyze vast amounts of data and detect patterns to answer a specific question. However, because they are still machines, they lack insight and the ability to understand why, the kind of insight and understanding we rely on every day. While AI has tremendous potential, there is no guarantee it will always provide us with the best clinical guidance. And that is precisely where we come in. Harnessing and regulating AI in medicine will be a monumental task, well beyond the FDA’s resources and requiring a concerted effort from all of us.
In this month’s issue, our feature article, “AI advances vision science, identifies disease patterns faster” (pages 1, 6 and 7) provides a great overview of this emerging discipline. I’m sure you’ll find our expert contributors’ comments insightful and relevant. I’m also sure you’ll recognize a common theme ... AI is here to complement us, but not replace our clinical acumen. At least, not yet.
- References:
- Brookes L. Diabetic retinopathy: Can artificial intelligence provide a better way to detect disease? Medscape. www.medscape.com/viewarticle/897547. Posted June 7, 2018.
- Harrison P. AI beats dermatologists on skin lesion images. Medscape. www.medscape.com/viewarticle/897302. Posted May 29, 2018.
- Salamon M. AI detects polyps better than colonoscopists. Medscape. www.medscape.com/viewarticle/897530. Posted June 3, 2018.