CAMBRIDGE, UK – A groundbreaking development in artificial intelligence (AI) is set to revolutionize the early detection of neurodegenerative diseases. In March, researchers announced they had developed sophisticated AI-powered technology that can estimate a person’s biological brain age from MRI scans, providing a powerful method for detecting conditions like dementia, Alzheimer’s, Parkinson’s, and multiple sclerosis, and identifying individuals at risk before clinical symptoms appear.
The research team, whose findings were published in the journal NeuroImage, performed an in-depth analysis of current trends in artificial intelligence (AI) applied to brain age estimation. By leveraging advanced neuroimaging data from the UK Biobank (UKBB) and using modern machine learning (ML) and deep learning (DL) techniques, they have created robust brain clocks that show significant potential in predicting brain age and detecting early signs of neurodegenerative diseases.
An Advancement in Brain Age Prediction
The team’s approach involved processing T1-weighted MRI scans from the UKBB using FastSurfer, a leading neuroimaging tool. This allowed them to derive standardized images and extract detailed image-derived phenotypes, which were then fed into their models. Their analysis encompassed a wide array of ML models, ranging from traditional methods like LASSO regression to innovative deep learning architectures such as the Simple Fully Convolutional Network.
One of the more challenging hurdles in brain age prediction has been the systematic bias observed across different age groups, where younger people’s brain age tends to be overestimated, and older individuals’ ages are often underestimated. To counteract this, the researchers used advanced correction methods, including Cole’s, Lange’s, and Zhang’s age-bias correction. These methods generally led to improved model stability and performance across all age brackets.
While the models were exclusively trained on UKBB data, their generalizability was rigorously tested using two external datasets: the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the National Alzheimer’s Coordinating Center (NACC).
High Accuracy and Generalisability
The results of the study are highly promising. The models achieved a Mean Absolute Error (MAE) of less than 1 year, demonstrating remarkable accuracy in brain age prediction. This accuracy was maintained across all 5-year age bins for individuals between 55 and 85 years of age across all analyzed databases.
Furthermore, these models demonstrated strong potential as biomarkers for neurodegenerative conditions, such as dementia and multiple sclerosis. In distinguishing healthy individuals from those with dementia or multiple sclerosis, brain age prediction achieved an AUROC (Area Under the Receiver Operating Characteristic curve) of up to 0.90.
The best-performing models for both age and disease prediction were K-fold cross-validated penalized linear models, typically LASSO, enhanced with Zhang’s age-bias correction. These ML-based models demonstrated excellent generalizability to the external datasets, underscoring the robustness of the approach. For deep learning models, however, generalization to external datasets proved more challenging, highlighting the need for further fine-tuning to enhance their robustness. Notably, certain architectures, particularly ResNet18, showed significant improvements in age prediction when corrected using the Lange method. Despite these gains in age prediction, their performance in disease classification remained limited.
Zhang’s correction emerged as the most effective method for age prediction. However, it may introduce numerical issues, especially in non-linear models prone to overfitting. For disease prediction, Cole’s correction and uncorrected models generally performed better. Still, there were instances where Zhang- and Lange-corrected models also yielded strong results. This highlights the ongoing need for improved correction techniques that can effectively reduce age bias while preserving the strong disease prediction capabilities of uncorrected models.
Clinical Implications and Future Directions
The ability of these brain age models to detect deviations from healthy aging patterns suggests their profound potential as non-invasive biomarkers for early-stage diagnosis of neurodegenerative conditions. Looking ahead, the research team aims to integrate multimodal data sources and explore new trends in transfer learning to further enhance the accuracy and clinical utility of these models. Their future work will also focus on incorporating additional MRI modalities to build an even more comprehensive predictive framework.
“At Oxcitas, we are excited to be at the forefront of this innovative intersection of AI and neuroimaging,” the research team stated. “Our research highlights the promise of AI-driven brain age estimation as a transformative tool in the early detection and management of neurological diseases.”
