AI Reveals Asymmetric Brain Aging and Alzheimer's Markers

A recent investigation published in the journal PNAS presents an innovative artificial intelligence technique for mapping the varied aging processes across different areas of the human brain. By conducting a detailed examination of MRI scans, the study offers compelling evidence that the brain's aging trajectory is not uniform, with particular regions displaying more pronounced biological aging in individuals afflicted with Alzheimer's disease.

Healthcare professionals and scientists frequently employ Magnetic Resonance Imaging (MRI), a method that uses potent magnetic fields to generate elaborate visualizations of internal bodily structures, to determine an individual's "brain age." This measurement indicates whether a person's brain appears biologically older or younger than their chronological age. A brain that exhibits characteristics of advanced age is often considered a risk factor for cognitive decline and various neurodegenerative conditions.

Uneven Brain Aging Across Hemispheres and Regions Uncovered by AI

Traditionally, researchers have relied on a singular, overarching brain age metric. However, the human brain is an intricate organ composed of distinct regions, each responsible for specialized functions. For instance, the frontal lobe is critical for complex decision-making, while the hippocampus plays a crucial role in memory formation. Given their differing roles, these areas tend to age at varying rates. A single brain age score can mask these regional disparities, making it more challenging to identify early indicators of localized damage. The study's lead author, Andrei Irimia, an associate professor of gerontology at the USC Leonard Davis School of Gerontology, explained that their previous work focused on global brain age, where AI estimated the brain's apparent age from an MRI. While useful as a summary, this approach condensed the entire brain's complexity into a single number, hindering biological interpretation. To address this, they developed a local brain age model, which not only assesses the overall brain's appearance but also meticulously evaluates the age of each individual part. This model transforms a structural MRI into a spatial map illustrating local brain age patterns.

This pioneering research, spearheaded by Nikhil N. Chaudhari alongside Irimia and their colleagues at the Irimia Lab, sought to construct a more granular map of brain aging. The team aimed to quantify biological aging at an exceptionally detailed level, dissecting the brain into minute three-dimensional units called voxels. This meticulous approach was intended to precisely pinpoint areas of accelerated brain aging and understand the relationship between these specific regional changes and cognitive decline. To develop their localized brain aging map, the scientists trained a deep learning computer model using MRI scans from 14,748 cognitively healthy adults ranging from 19 to 100 years old. Irimia emphasized the importance of scale and interpretability, noting that the model was trained on a vast and diverse dataset, ensuring exposure to significant variations in age, anatomy, and imaging data. The primary objective was not merely to create an AI model with high predictive accuracy but to generate an output that neuroscientists could directly correlate with brain anatomy, thereby bridging AI-based predictions with fundamental questions regarding the biology of human aging.

Implications for Alzheimer's Detection and Future Research

After the training phase, the research team validated the system on a separate cohort of 1,985 individuals from the Alzheimer's Disease Neuroimaging Initiative. This validation group comprised 1,102 cognitively normal adults, 354 individuals with mild cognitive impairment, and 529 diagnosed with Alzheimer's disease. In cognitively healthy adults, the model's estimations deviated by approximately 5.9 years on average, revealing distinct aging patterns across the brain. The frontal and temporal lobes, crucial for higher-order cognitive functions and memory, exhibited a more advanced biological age compared to posterior brain regions, such as the visual processing centers in the occipital lobe. The researchers also conducted a detailed examination of the cerebral cortex, observing that the sulci (deep grooves) appeared biologically older than the gyri (outward-facing ridges). On average, the local brain age gap, representing the difference between estimated biological age and chronological age, was 0.40 years greater in the sulci. Furthermore, the right cerebral hemisphere consistently displayed an older biological age than the left, a pattern that persisted even after adjusting the model for handedness in the training data, raising intriguing questions about potential structural aging differences between the two hemispheres.

When comparing the different groups, the researchers discovered that patients with Alzheimer's disease and mild cognitive impairment exhibited older local brain ages in specific regions known to be susceptible to early neurodegeneration. In Alzheimer's patients, subcortical structures like the pallidum and putamen, which are involved in motor control and certain forms of learning, showed brain age gaps approximately 3.5 to 3.6 years larger than those of cognitively healthy adults. The hippocampus, a seahorse-shaped structure vital for forming new memories, appeared more than three years older on average in the Alzheimer's group compared to the healthy adults. Irimia underscored that the brain does not age uniformly, with different regions exhibiting distinct aging patterns within the same individual, and these regional differences becoming more pronounced in cognitive impairment. This provides a methodology to move beyond a single overall brain age estimate and pinpoint where age-related structural differences occur. The team also investigated the correlation between these localized aging gaps and cognitive test performance. In individuals with Alzheimer's disease, an older local brain age in areas such as the right pallidum showed a strong correlation with poorer scores on general cognitive screening tests and measures of daily functioning. These associations were considerably weaker in individuals with mild cognitive impairment and largely absent in healthy adults. While promising, Irimia cautioned that the tool is currently a research instrument, not yet ready for routine clinical application, and further longitudinal validation is required to determine its predictive power for future disease trajectories.