Performing Aperture Domain Model Image REconstruction (ADMIRE) Using a Deep Neural Network Sparse Encoder

Aperture Domain Model Image REconstruction (ADMIRE) is an advanced ultrasound beamforming method that uses a model-based approach to suppress sources of acoustic clutter and improve ultrasound image quality. However, although effective, ADMIRE requires solving an inverse problem that is ill-posed, which means that there are infinitely many solutions that can have different impacts on image quality. Currently, linear regression with elastic-net regularization is used to obtain a solution, but there are potentially better methods for performing model fitting. Therefore, in this work, we propose using a deep neural network sparse encoder for performing the model fits of ADMIRE. In particular, we unfold the iterations of the iterative shrinkage and thresholding algorithm (ISTA) as a feedforward neural network and train it using different training schemes to perform sparse coding. Test results using both simulated and in vivo data demonstrate that ADMIRE using a deep neural network sparse encoder has the potential to outperform conventional ADMIRE in terms of ultrasound image quality while still preserving the model-based intuition of ADMIRE.