Johns Hopkins Research Points to Increasing Role of Artificial Intelligence in Medical Imaging and Diagnostics

The advent of electronic medical records with large image databases, along with advances in artificial intelligence with deep learning, is offering medical professionals new opportunities to dramatically improve image analysis and disease diagnostics.

Researchers from the Johns Hopkins University Applied Physics Laboratory (APL) in Laurel, Maryland, and collaborators at the Johns Hopkins School of Medicine have developed image analysis and machine-learning tools to detect age-related macular degeneration (AMD). In Nature Medicine, members of the team discuss the potential of such tools to be used clinically and applied to other image-based medical diagnoses as well.

AMD causes lesions that blur the sharp, central vision individuals need for activities such as reading, recognizing faces and driving. It is the number-one cause of blindness for individuals over 50. Typically, much of the vision loss from AMD is irreversible, prompting the need to detect treatable lesions before substantial vision loss has occurred.

In 2015, APL’s Philippe Burlina and colleagues teamed up with the Johns Hopkins Wilmer Eye Institute on ways to automate AMD diagnosis. In recent work published in JAMA Ophthalmology, they demonstrated that machine diagnostics using deep learning can match the performance of human ophthalmologists.

“We’ve been able to show the feasibility of automated fine-grained classification of AMD severity that only highly trained ophthalmologists can achieve,” explained Burlina, a co-principal investigator for the project. “These techniques have the potential to provide individuals with automated grading of images to identify AMD or monitor those individuals with earlier stages of AMD for the onset of the more advanced stages when prompt treatment may be indicated to reduce the risk of blindness.”

The team has also expanded its inquiry to characterize retinal layers in optical coherence tomography, or OCT, which is a noninvasive imaging technique that provides high-resolution, cross-sectional images of the retina, retinal nerve fiber layer and optic nerve head. Such techniques can be used to diagnose other retinal diseases such as diabetic retinopathy but also have the potential to help characterize vascular and neurodegenerative pathologies.

“We were able to show that machines can do as well as humans for diagnosing AMD,” Burlina said. “So now we have started looking at other retinal diseases, and how to combine images with other sources of information — demographics, lifestyle factors such as smoking, and sunlight exposure — to automatically perform prognosis and predict the probability for five-year risk of developing the advanced form of the disease. The end goal is to help clinicians and guide treatment.”

This year, the team expanded its collaboration to include scientists from the Singapore National Eye Center and tested its algorithms on several different Asian ethnic groups, including Malaysian, Indian and Chinese. In “AI for Medical Imaging Goes Deep,” published in the May 2018 issue of Nature Medicine, the team reviews recent work by the research community, examines the potential of AI applied to retinal image diagnostics, and discusses requirements and future work needed to allow translation and deployment of these techniques for clinical and point of care scenarios.

Burlina said the researchers will also work with hospitals in Thailand, Brazil and France to explore how algorithms trained on a database of specific ethnicities, demographics and imaging-capture scenarios can be adapted for different ethnicities and conditions.