Multimodal brain age prediction using machine learning: combining structural MRI and 5-HT2AR PET-derived features.

Area of Science:

• Neuroscience
• Radiology
• Biomarker Development

Background:

• Accurate biomarkers are crucial for assessing neurodegenerative disorders and neuroprotective interventions.
• Brain serotonin 2A receptors (5-HT2AR) exhibit significant age-related decline and are implicated in neurodegenerative conditions like Alzheimer's disease.

Purpose of the Study:

• To evaluate 5-HT2AR binding, measured via PET, as a potential biomarker for brain aging.
• To compare the predictive accuracy of 5-HT2AR binding and gray matter (GM) volume for brain age.
• To determine if combining 5-HT2AR and GM volume data improves brain age prediction.

Main Methods:

• Utilized PET and MRI data from 209 healthy individuals (18-85 years).
• Estimated 5-HT2AR binding and GM volume in 14 brain regions.
• Applied machine learning algorithms to predict chronological age using 5-HT2AR binding, GM volume, and combined measures, evaluated by Mean Absolute Error (MAE) and cross-validation.

Main Results:

• Cerebral 5-HT2AR binding accurately predicted chronological age (mean MAE = 6.63 years).
• Gray matter volume also accurately predicted chronological age (mean MAE = 6.95 years).
• Combining 5-HT2AR binding and GM volume significantly improved age prediction accuracy (mean MAE = 5.54 years).

Conclusions:

• 5-HT2AR binding measured by PET shows promise as a biomarker for brain aging.
• Combining 5-HT2AR PET imaging with MRI-derived GM volume offers a more precise method for quantifying brain age.