Abstract: Brain Age Prediction was developed to address the need for early identification and monitoring of neurodegenerative changes before clinical symptoms appear. Traditional methods rely on cognitive testing and expert MRI interpretation, often leading to late diagnoses. This project introduces a web portal that converts routine MRI scans into predictions of healthy-brain lifespan using machine learning and large neuroimaging datasets. Using the OpenBHB dataset (3,985 individuals aged 5.9–88), four volumetric biomarkers—total intracranial volume, cerebrospinal fluid volume, gray matter volume, and white matter volume—were extracted via K-means clustering and standardized. Chronological age served as the regression target. Three neural architectures were developed: Ultra_ResDNN (residual connections), Ultra_WideDeep (wide-linear + deep-MLP), and Ultra_Attention (multi-head self-attention). Outputs were ensembled to yield robust predictions. On a held-out test set, the ensemble achieved a mean absolute error of 0.58 years and explained 96.38% of variance (R² = 0.9638), demonstrating clinical-grade accuracy.

Downloads: | DOI: 10.17148/IJIREEICE.2025.131215