Stiffness matrix-based physics-constrained neural network for transcranial photoacoustic imaging (Conference Presentation)

The major obstacle in transcranial photoacoustic imaging is the severe acoustic distortion caused by the skull. Herein, we propose a Stiffness Matrix-based Physics-Constrained Neural Network (SM-PCNN) for reconstructing intracranial vessels. Our approach leverages the five-layer structure of tPAI—cerebral blood vessels, cortical bone, cancellous bone, cortical bone, and coupling fluid. By integrating the elastic solid wave equation and Biot's theory, coupled with joint angular-frequency spectrum decomposition, we precisely calculate the transcranial transmission coefficient of intracranial vessels. This coefficient serves as a physics-based constraint within our neural network architecture, guiding the training process on transcranial vascular imaging data. SM-PCNN demonstrates high-resolution transcranial reconstructed images with sharp vascular edges and clean backgrounds. By integrating the stiffness matrix with the neural network, our method reduces the reliance of neural networks on large amounts of data and effectively corrects skull-induced distortions in transcranial photoacoustic imaging.