Deep model-based optoacoustic image reconstruction (DeepMB)

Christoph Dehner, Guillaume Zahnd, Vasilis Ntziachristos, Dominik Jüstel

Keywords: Multispectral Optoacoustic Tomography (MSOT), Model-based reconstruction, Inverse problems, Real-time imaging, Synthesized training data

Abstract: Multispectral optoacoustic tomography requires image feedback in real-time to locate and identify relevant tissue structures during clinical interventions. Backprojection methods are commonly used for optoacoustic image reconstruction in real-time but only afford imprecise images due to oversimplified modelling assumptions. Herein, we present a deep learning framework, termed DeepMB, that infers optoacoustic images with state-of-the-art quality in 31 ms per image.