GNN-TF for Forecasting Tobacco Use from Brain Imaging and Tabular Data

This paper introduces a novel Graph Neural Network model with Transformer Fusion (GNN-TF) to predict future tobacco use by integrating brain connectivity data (non-Euclidean) and clinical/demographic data (Euclidean). The key contribution is the time-aware fusion of these data modalities, leveraging temporal dynamics for improved predictive accuracy compared to existing methods. This is significant because it addresses a challenging problem in medical imaging analysis, particularly in longitudinal studies.

• •Proposes a novel GNN-TF model for integrating brain imaging and tabular data.
• •Addresses the challenge of forecasting outcomes in longitudinal imaging studies.
• •Demonstrates superior predictive accuracy for future tobacco use compared to existing methods.
• •Leverages temporal dynamics and fuses multiple data modalities.