The contrast and resolution of images obtained with optical microscopes can be improved by deconvolution and computational fusion of multiple views of the same sample, but these methods are computationally expensive for large datasets. Here we describe theoretical and practical advances in algorithm and software design that result in image processing times that are tenfold to several thousand fold faster than with previous methods. First, we show that an ‘unmatched back projector’ accelerates deconvolution relative to the classic Richardson–Lucy algorithm by at least tenfold. Second, three-dimensional image-based registration with a graphics processing unit enhances processing speed 10- to 100-fold over CPU processing. Third, deep learning can provide further acceleration, particularly for deconvolution with spatially varying point spread functions. We illustrate our methods from the subcellular to millimeter spatial scale on diverse samples, including single cells, embryos and cleared tissue. Finally, we show performance enhancement on recently developed microscopes that have improved spatial resolution, including dual-view cleared-tissue light-sheet microscopes and reflective lattice light-sheet microscopes.

We present Richardson–Lucy network (RLN), a fast and lightweight deep learning method for three-dimensional fluorescence microscopy deconvolution. RLN combines the traditional Richardson–Lucy iteration with a fully convolutional network structure, establishing a connection to the image formation process and thereby improving network performance. Containing only roughly 16,000 parameters, RLN enables four- to 50-fold faster processing than purely data-driven networks with many more parameters. By visual and quantitative analysis, we show that RLN provides better deconvolution, better generalizability and fewer artifacts than other networks, especially along the axial dimension. RLN outperforms classic Richardson–Lucy deconvolution on volumes contaminated with severe out of focus fluorescence or noise and provides four- to sixfold faster reconstructions of large, cleared-tissue datasets than classic multi-view pipelines. We demonstrate RLN’s performance on cells, tissues and embryos imaged with widefield-, light-sheet-, confocal- and super-resolution microscopy.

Accurate and robust reconstruction of the radioactivity concentration is of great importance in positron emission tomography (PET) imaging. Given the Poisson nature of photo-counting measurements, we present a reconstruction framework that integrates sparsity penalty on a dictionary into a maximum likelihood estimator. Patch-sparsity on a dictionary provides the regularization for our effort, and iterative procedures are used to solve the maximum likelihood function formulated on Poisson statistics. Specifically, in our formulation, a dictionary could be trained on CT images, to provide intrinsic anatomical structures for the reconstructed images, or adaptively learned from the noisy measurements of PET. Accuracy of the strategy with very promising application results from Monte-Carlo simulations, and real data are demonstrated.

Executive function (EF) refers as to a set of high-level cognitive abilities that are critical to many aspects of daily life. Despite its importance in human daily life, the neural networks responsible for the development of EF in childhood are not well understood. The present study thus aimed to examine the development of task-dependent brain network organization and its relationship to age-related improvements in EF. To address this issue, we recruited eighty-eight Chinese children ranging in age from 7 to 12 years old, and collected their functional magnetic resonance imaging (fMRI) data when they performed an EF task. By utilizing graph theory, we found that the task-dependent brain network modules became increasingly segregated with age. Specifically, the intra-module connections within the default-mode network (DMN), frontal-parietal network (FPN) and sensorimotor network (SMN) increased significantly with age. In contrast, the inter-module connections of the visual network to both the FPN/SMN decreased significantly with age. Most importantly, modular segregation of the FPN significantly mediated the relationship between age and EF performance. These findings add to our growing understanding of how development changes in task-dependent brain network organization support vast behavioral improvements in EF observed during childhood.

Accurate and robust tomographic reconstruction from dynamic positron emission tomography (PET) acquired data is a difficult problem. Conventional methods, such as the maximum likelihood expectation maximization (MLEM) algorithm for reconstructing the activity distribution-based on individual frames, may lead to inaccurate results due to the checkerboard effect and limitation of photon counts. In this paper, we propose a stacked sparse auto-encoder based reconstruction framework for dynamic PET imaging. The dynamic reconstruction problem is formulated in a deep learning representation, where the encoding layers extract the prototype features, such as edges, so that, in the decoding layers, the reconstructed results are obtained through a combination of those features. The qualitative and quantitative results of the procedure, including the data based on a Monte Carlo simulation and real patient data demonstrates the effectiveness of our method.