Deep learning-based reconstruction of ultrasound images from raw channel data

• DOI: 10.1007/s11548-020-02197-w
• Publication Date: 2020-06-03T15:29:09Z
• Authors: Hannah Strohm; Sven Rothlübbers; Klaus Eickel; Matthias Günther
• Keywords: Ground truth; Hyperparameter; Data set

We investigate the feasibility of reconstructing ultrasound images directly from raw channel data using a deep learning network. Starting data, we present network full measurement information, allowing for more generic reconstruction to form, as compared common reconstructions constrained by physical models fixed speed sound assumptions.We propose U-Net-like architecture given task. Additional layers with strided convolutions downsample data. Hyperparameter optimization was used find suitable rate. train and test our approach on plane wave single insonification angle. The dataset includes phantom well in vivo data.The produced method are visually comparable ones reconstructed conventional delay sum algorithm. Deviations between prediction ground truth likely be related speckle noise. For set, mean absolute error is [Formula: see text] result shows opens up new research directions regarding information retrieval As networks performance limited quality images, other technique or image types target would interest.