A Brain Age Residual Biomarker (BARB): Leveraging MRI-Based Models to Detect Latent Health Conditions in U.S. Veterans

TL;DR

The paper addresses detecting latent health states via brain age residual biomarkers (BARBs) from MRI data. It trains four CNNs on 2D slices from T2-weighted FSE and T2-weighted FLAIR images at two brain locations and aggregates their predictions with a degree-3 polynomial regression to estimate brain age. On a veteran cohort of 1,220 individuals (1,104 with all four images), the approach achieves $R^2=0.816$ and MAE=$5.450$, with residual patterns that relate to the number of ICD-coded conditions, particularly in those aged $>49$. The work demonstrates BARB feasibility in a small, region-specific dataset and highlights the need for broader validation, richer health metrics, and multi-modal data to improve generalizability and clinical utility.

Abstract

Age prediction using brain imaging, such as MRIs, has achieved promising results, with several studies identifying the model's residual as a potential biomarker for chronic disease states. In this study, we developed a brain age predictive model using a dataset of 1,220 U.S. veterans (18--80 years) and convolutional neural networks (CNNs) trained on two-dimensional slices of axial T2-weighted fast spin-echo and T2-weighted fluid attenuated inversion recovery MRI images. The model, incorporating a degree-3 polynomial ensemble, achieved an $R^{2}$ of 0.816 on the testing set. Images were acquired at the level of the anterior commissure and the frontal horns of the lateral ventricles. Residual analysis was performed to assess its potential as a biomarker for five ICD-coded conditions: hypertension (HTN), diabetes mellitus (DM), mild traumatic brain injury (mTBI), illicit substance abuse/dependence (SAD), and alcohol abuse/dependence (AAD). Residuals grouped by the number of ICD-coded conditions demonstrated different trends that were statistically significant ($p = 0.002$), suggesting a relationship between disease states and predicted brain age. This association was particularly pronounced in patients over 49 years, where negative residuals (indicating advanced brain aging) correlated with the presence of multiple ICD codes. These findings support the potential of residuals as biomarkers for detecting latent health conditions.

Figures (8)

• Figure 1: The model presented here is constructed from T2 weighted fast spin-echo (FSE) and T2 weighted fluid attenuated inversion recovery (FLAIR) images of two locations of the brain: the anterior commissure and the frontal horns of the lateral ventricles.
• Figure 2: Distribution of the ICD codes, with a total of 1220 patients.
• Figure 3: Histogram illustrating the distribution of continuous ages in the reduced dataset of 1,104 subjects. Continuous age reflects fractional years, such as 26.25 for an individual aged 26 years and 90 days, providing a precise representation of age variability across the population.
• Figure 4: Architecture of the convolutional neural network (CNN) model used for each image-type. The layers are represented by horizontal bars, color-coded by type along with a table summarizing each layer. The batch size used as 20 and the pooling size was $2 \times 2$.
• Figure 5: Four types of ensemble methods were evaluated for this study, as outlined in Table \ref{['tab:models']}. Ensemble approaches can reduce total prediction error, as described in Equation \ref{['eq:bvd']}, by leveraging the potential for error cancellation across the individual models.