New AttentionMixer framework combines CT scans with patient clinical data to improve brain edema detection using advanced deep learning

AttentionMixer uses a Vision Transformer Autoencoder (ViT-AE++) to process head CT volumes without requiring large labeled datasets through self-supervised learning

Clinical metadata including age, lab values, and scan timing are fused with imaging data through a cross-attention mechanism for dynamic feature modulation

The framework maps heterogeneous data sources into a unified feature space, enabling interpretable multimodal integration for brain edema classification

Cross-attention fusion allows the network to weight imaging features based on patient-specific clinical context rather than naive concatenation