Using AI as a Neuroimaging Biomarker for Brain Health
AI deep learning predicts biological age from brain MRIs with high accuracy.
Posted September 29, 2022 | Reviewed by Kaja Perina
In neuroscience and medical research, artificial intelligence (AI) deep learning is increasingly being used as a potential biomarker for brain health. A new study by the Max Planck Institute for Human Cognitive and Brain Sciences demonstrate how an AI algorithm can estimate biological age with high accuracy based on brain scan images.
“This bias-free computational approach yields insights into the global nature of brain aging as well as pathomechanisms,” wrote the researchers.
Artificial intelligence trained on magnetic resonance imaging (MRI) that estimate brain age in comparison to chronological age is emerging as a potential new form of a digital biomarker for cognitive decline, lifestyle cardiovascular risk factors, Alzheimer’s disease (AD), dementia, and hypertension according to the researchers who authored the study.
The Max Planck Institute researchers used ensembles of convolutional neural networks (CNNs) with Layer-wise Relevance Propagation (LRP) in order to discover which brain features are involved in brain aging. A convolutional neural network is a type of deep neural network that is somewhat inspired by the biological brain’s visual cortex.
Convolutional neural networks are a feedforward neural network that uses features to classify images. If an artificial neural network contains more than one convolution operation, it is called a convolutional neural network. A convolution is a mathematical operation often used for signal and image processing where two signals or images are used as input in order to produce a third signal or image as the output.
To train the AI algorithm, the researchers used magnetic resonance imaging (MRI) data from a population-based cohort study, the LIFE Adult study, where out of over 10,000 participants, a subset of over 2,630 participants between the ages of 18 to 82 years old had a one-hour MRI recording session.
The researchers reported that their models estimated the age accurately based on single and multiple modalities, as well as both whole-brain images and regionally restricted images. The mean absolute errors were just 3.37–3.86 years.
According to the researchers, the progress and speed of brain aging can now be spotted and analyzed by AI deep learning ensembles from various brain regions and structural MRI modalities.
“Our analysis demonstrates that grey matter changes and atrophies detectable in the cortex, subcortex, cerebellum and brainstem, but also white matter lesions, as well as more global brain shrinkage represented in the larger size of ventricles and sulci drove the age estimates of the models,” the researchers wrote.
The artificial intelligence method used for this study is generalizable, and therefore it can be adapted widely across clinical neuroscience to accelerate novel digital health approaches for personalized medicine in the future.
Copyright © 2022 Cami Rosso All rights reserved.