The last few years have seen a number of applications whereby artificial intelligence makes smart diagnoses based upon medical scans.  The latest comes via research from Duke University, which interprets retinal images to identify Alzheimer’s disease.

The system looks at the retinal structure and blood vessels in images of the inside of an eye that is correlated with various cognitive changes.  The researchers believe that their findings show that machine learning can potentially provide a non-invasive means of detecting Alzheimer’s.

“Diagnosing Alzheimer’s disease often relies on symptoms and cognitive testing,” the researchers say. “Additional tests to confirm the diagnosis are invasive, expensive, and carry some risk. Having a more accessible method to identify Alzheimer’s could help patients in many ways, including improving diagnostic precision, allowing entry into clinical trials earlier in the disease course, and planning for necessary lifestyle adjustments.”

Blood vessel density

The researchers built on earlier work that had identified that changes in the blood vessel density in the retina corresponded with changes in cognition.  As the density of the capillary network in the center of the macula decreased, this was strongly linked with Alzheimer’s.

They used this insight to train a machine learning algorithm, with four types of retinal images used as inputs to train the machine to discern the key differences between each image.  In all, scans from 159 people were used, with 123 of them healthy individuals, and 36 from people who were known to have Alzheimer’s.

“We tested several different approaches, but our best-performing model combined retinal images with clinical patient data,” the researchers say. “Our CNN differentiated patients with symptomatic Alzheimer’s disease from cognitively healthy participants in an independent test group.”

Next steps

The researchers go on to say that the next step will be to ensure that a diverse group of participants is used to ensure that the models are able to detect Alzheimer’s in all racial groups, as well as in a variety of more complex scenarios, such as in patients who also have diabetes or glaucoma, both of which also alter the retinal and vascular structures.

“We believe additional training using images from a larger, more diverse population with known confounders will improve the model’s performance,” the researchers say.

Additional work will also help them to understand how effective the AI-based system is compared to existing methods of diagnosing the disease.  These often include very expensive and invasive neuroimaging, or cerebral spinal fluid tests.

“Links between Alzheimer’s disease and retinal changes—coupled with non-invasive, cost-effective, and widely available retinal imaging platforms—position multimodal retinal image analysis combined with artificial intelligence as an attractive additional tool, or potentially even an alternative, for predicting the diagnosis of Alzheimer’s,” the researchers conclude