SIMBA: Scalable Inversion in Optical Tomography Using Deep Denoising Priors

Abstract

Two features desired in a three-dimensional (3D) optical tomographic image reconstruction algorithm are the ability to reduce imaging artifacts and to do fast processing of large data volumes. Traditional iterative inversion algorithms are impractical in this context due to their heavy computational and memory requirements. We propose and experimentally validate a novel scalable iterative mini-batch algorithm (SIMBA) for fast and high-quality optical tomographic imaging. SIMBA enables highquality imaging by combining two complementary information sources: the physics of the imaging system characterized by its forward model and the imaging prior characterized by a denoising deep neural net. SIMBA easily scales to very large 3D tomographic datasets by processing only a small subset of measurements at each iteration. We establish the theoretical fixedpoint convergence of SIMBA under nonexpansive denoisers for convex data-fidelity terms. We validate SIMBA on both simulated and experimentally collected intensity diffraction tomography (IDT) datasets. Our results show that SIMBA can significantly reduce the computational burden of 3D image formation without sacrificing the imaging quality.

Single‐Shot Ultraviolet Compressed Ultrafast Photography

“…In software, the quality of the reconstruction may be further improved by incorporating state‐of‐the‐art deep‐learning‐based denoisers into the PnP‐ADMM framework. [ 46,47 ] UV‐CUP will open up many new possibilities in single‐shot observation of transient UV phenomena, including laser‐induced UV plasma emission [ 48 ] and UV‐fluorescence in graphene oxides. [ 49 ]…”

Stochastic Deep Unfolding for Imaging Inverse Problems

“…The CNN within SCRED-Net is trained in an end-to-end fashion to remove artifacts due to the imaging system and stochastic processing of the measurements. We demonstrate the practical relevance of SCRED-Net by reconstructing images in intensity diffraction tomography (IDT) [15,16] and sparse-view CT. Our numerical results corroborate the effectiveness of SCRED-Net in achieving comparable imaging quality to batch deep unfolding networks, at a fraction of computational complexity. SCRED-Net thus addresses an important gap in the current literature on deep unfolding by providing a flexible and scalable framework applicable to a wide variety of computational imaging problems.…”

“…The excellent performance of RED under learned CNN denoisers has been reported in superresolution, phase retrieval, and compressed sensing [20,21]. Additionally, prior work has developed a scalable online variant of RED that is well suited for tomographic applications with a large number of projections [16]. However, unlike deep unfolding, the CNN in RED is not jointly trained with the measurement model, limiting its ability to capture the non-iid nature of the artifacts within iterations [11].…”

“…However, unlike deep unfolding, the CNN in RED is not jointly trained with the measurement model, limiting its ability to capture the non-iid nature of the artifacts within iterations [11]. SCRED-Net introduced in the next section is a natural extension of online RED [16] towards CNNs trained for artifactremoval, which leads to a scalable and end-to-end trainable deep network.…”

A novel deep learning conditional generative adversarial network for producing angiography images from retinal fundus photographs

“…classification 20 , object detection 21 , semantic 22 and instance segmentation 23 , and de-noising algorithms 24 , but it also has introduced novel computational frameworks such as super-resolution 25 , image generation 26 , and styletransfer 27,28 . Despite its late arrival to the field of ophthalmology compared to other medical domains 29,30 , deep learning has already started to play a transformative role in ophthalmology ranging from noise removal 31 , to disease classification [32][33][34] , to disease marker segmentation 33,35,36 . These advances resulted in the first automatic AI-enabled Diabetic Retinopathy (DR) system, called IDx-DR 37 , to be approved by the FDA in 2018 38 .…”