Batch ReNormalization Accumulated Residual U-Network for Artifacts Removal in Photoacoustic Imaging

During this ongoing and recent decades, photoacoustic tomography (PAT) is among the modalities that shows extensively growing interest in clinical pre-eminence due to its high resolution and contrast. Hence, along with the other PA related problems, PAT reconstruction is worth to study yet in detail. More precisely, PAT inverse problem is still seeking to optimize the solution concerning fast and accurate recovery under practical impediments. These constraints are analogous to under-sampled artifacts presence due to sparse sampling and degrade the performance of the reconstruction task. Therefore, to overcome the aforesaid problem and constraints, a feasible solution in terms of batch renormalization aggregated residual network (BRn_ResNet) is proposed which further improves the training stability and image quality by eliminating these sub-sampling artifacts. The main motivation behind the inclusion of batch renormalization is, due to its support in optimizing the solution especially for residual networks when use with minibatches. Moreover, in terms of outcomes, an extensive qualitative analysis as well as quantitatively evaluated whereas SSIM almost 0.97 state that the proposed solution provides the robust performance in terms of high-resolution image at the output even training with a low-quality dataset.