Machine learning method enables quick analysis of amyloid plaques
In a recent study, NIA-supported researchers at the University of California (UC) demonstrated how a computer system with the ability to learn and improve, known as machine learning, could help analyze amyloid plaques in the human brain. Amyloid plaques are abnormal clumps of protein that accumulate in the brains of people with Alzheimer’s disease, but not all plaques look alike.
As published in Nature Communications, the researchers from both UC San Francisco and UC Davis describe a technique that automates the process of measuring plaques and their different characteristics. This approach could enable larger-scale analysis of brain tissue to help accelerate research on the possible causes of Alzheimer’s and how the disease progresses.
Using 43 healthy and diseased human brain samples donated to the UC Davis Alzheimer’s Disease Center Brain Bank, the researchers taught a computer to detect different types of amyloid plaques within each sample.
To teach the computer, researchers first created a program that allowed for quick annotation of more than 70,000 plaque candidates from half a million close-up images taken from the slides of brain tissue. They then used that database to create what’s known as a convolutional neural network (CNN), which is a computer program with a machine-learning algorithm designed to recognize patterns based on thousands of human-labeled examples.
Next, the team validated the CNN’s performance using a test set of images and found the program was able to discriminate between different types of plaques and could identify abnormalities in blood vessels. The algorithm could process an entire slice of a whole-brain slide with 98.7 percent accuracy.
Many brain banks have archives of slides donated by people with and without dementia, researchers noted. Using a method like this one could help researchers study more samples more quickly. To further the exploration of automated pathology classification, researchers have made the CNN model code and dataset openly available.
This research was supported in part by NIA grant P30AG010129.
Reference: Tang Z, et al. Interpretable classification of Alzheimer’s disease pathologies with a convolutional neural network pipeline. Nature Communications. 2019;10(1):2173. doi: 10.1038/s41467-019-10212-1.