Sensor-Token Self-Attention for Matrix-Free Photoacoustic Imaging

The Sensor Attention Network (SAN) uses a Transformer architecture to treat sensor time series as tokens, mapping raw measurements directly to reconstructed images. By removing the dependency on system matrices at inference, it enables real-time clinical reconstruction with a 0.919 mean Pearson correlation.