$Σ$-net: Ensembled Iterative Deep Neural Networks for Accelerated Parallel MR Image Reconstruction
This work addresses the problem of fast and high-quality image reconstruction in parallel MRI for medical imaging applications, representing an incremental improvement with ensembling techniques.
The paper tackles accelerated parallel MRI reconstruction by proposing an ensembled iterative deep neural network called Σ-net, which combines parallel coil and sensitivity networks with various data consistency methods and achieves robust high SSIM scores through model ensembling.
We explore an ensembled $Σ$-net for fast parallel MR imaging, including parallel coil networks, which perform implicit coil weighting, and sensitivity networks, involving explicit sensitivity maps. The networks in $Σ$-net are trained in a supervised way, including content and GAN losses, and with various ways of data consistency, i.e., proximal mappings, gradient descent and variable splitting. A semi-supervised finetuning scheme allows us to adapt to the k-space data at test time, which, however, decreases the quantitative metrics, although generating the visually most textured and sharp images. For this challenge, we focused on robust and high SSIM scores, which we achieved by ensembling all models to a $Σ$-net.