Deep Learning Algorithms Can Detect Subtle Brain Lesions
A convolutional neural network outperformed other methods for detecting white matter lesions on MRI scans.
CC0 Public Domain
In a recent study, deep learning algorithms proved superior to conventional machine learning frameworks for automatically detecting a certain type of brain lesion on magnetic resonance imaging scans.
The brain lesions in question are known as white matter hyperintensities, or WMHs, and they show up as areas of increased brightness on MRI scans. They are associated with a number of neurological disorders and psychiatric illnesses. WMHs are also commonly found on MRIs of older individuals, where they are indicative of age-related cognitive decline.
It is difficult to pick out WMHs on MRI scans visually because they are small and subtle, easily mistaken for imaging artifacts.
In the new study, researchers tested the performance of their own convolutional neural network, a kind of deep learning algorithm, in identifying WMHs on MRI scans. Deep learning is a cutting-edge subset of machine learning in which algorithms create artificial neural networks that can learn and make decisions about data.
Using MRI data from patients with either no or mild cognitive impairment, the researchers compared their convolutional neural network to existing machine learning frameworks, including conventional machine learning algorithms and another type of deep learning algorithm.
Generally, deep learning algorithms performed better than conventional machine learning algorithms because they were able to exclude more MRI artifacts and pathologies that appear similar to WMHs. The researchers’ new convolutional neural network performed best of all. The researchers further improved the performance of the convolutional neural network by adding a new way to incorporate spatial information.
The researchers say the next step is evaluating their convolutional neural network in other kinds of brains, especially those with moderate to abundant vascular pathology such as small vessel disease and strokes. The study was published in the journal Computerized Medical Imaging and Graphics.