The reconstruction of neuronal populations, a key step in understanding neural circuits, remains a challenge in the presence of densely packed neurites. Here we achieved automatic reconstruction of neuronal populations by partially mimicking human strategies to separate individual neurons. For populations not resolvable by other methods, we obtained recall and precision rates of approximately 80%. We also demonstrate the reconstruction of 960 neurons within 3 h.

Localization-based super resolution microscopy holds superior performances in live cell imaging, but its widespread use is thus far mainly hindered by the slow image analysis speed. Here we show a powerful image analysis method based on the combination of the maximum likelihood algorithm and a Graphics Processing Unit (GPU). Results indicate that our method is fast enough for real-time processing of experimental images even from fast EMCCD cameras working at full frame rate without compromising localization precision or field of view. This newly developed method is also capable of revealing movements from the images immediately after data acquisition, which is of great benefit to live cell imaging.

Drawing the map of neuronal circuits at microscopic resolution is important to explain how brain works. Recent progresses in fluorescence labeling and imaging techniques have enabled measuring the whole brain of a rodent like a mouse at submicron-resolution. Considering the huge volume of such datasets, automatic tracing and reconstruct the neuronal connections from the image stacks is essential to form the large scale circuits. However, the first step among which, automated location the soma across different brain areas remains a challenge. Here, we addressed this problem by introducing L1 minimization model. We developed a fully automated system, NeuronGlobalPositionSystem (NeuroGPS) that is robust to the broad diversity of shape, size and density of the neurons in a mouse brain. This method allows locating the neurons across different brain areas without human intervention. We believe this method would facilitate the analysis of the neuronal circuits for brain function and disease studies.

Abstract. Localization of a single fluorescent molecule is required in a number of superresolution imaging techniques for visualizing biological structures at cellular and subcellular levels. The localization capability and limitation of low-light detectors are critical for such a purpose. We present an updated evaluation on the performance of three typical low-light detectors, including a popular electron-multiplying CCD (EMCCD), a newly developed scientific CMOS (sCMOS), and a representative cooled CCD, for superresolution imaging. We find that under some experimental accessible conditions, the sCMOS camera shows a competitive and even better performance than the EMCCD camera, which has long been considered the detector of choice in the field of superresolution imaging. C 2010 Society of Photo-Optical Instrumentation Engineers.

Developing methods for high-density localization of multiple emitters is a promising approach for enhancing the temporal resolution of localization microscopy while maintaining a desired spatial resolution, but the widespread use of this approach is thus far mainly obstructed by the slow image analysis speed. Here we present a high-density localization method based on the combination of Graphics Processing Unit (GPU) parallel computation, multiple-emitter fitting, and model recommendation via Bayesian Information Criterion (BIC). This method, called PALMER, exhibits satisfactory localization accuracy comparable with the previous reported SSM_BIC method, while executes more than two orders of magnitudes faster. Meanwhile, compared to the conventional localization microscopy which is based on sparse emitter localization, high-density localization microscopy based the PALMER method allows a speed gain of up to ~14-fold in obtaining a super-resolution image with the same Nyquist resolution.

Localization-based super-resolution microscopy (or called localization microscopy) rely on repeated imaging and localization of active molecules, and the spatial resolution enhancement of localization microscopy is built upon the sacrifice of its temporal resolution. Developing algorithms for high-density localization of active molecules is a promising approach to increase the speed of localization microscopy. Here we present a new algorithm called SSM_BIC for such purpose. The SSM_BIC combines the advantages of the Structured Sparse Model (SSM) and the Bayesian Information Criterion (BIC). Through simulation and experimental studies, we evaluate systematically the performance between the SSM_BIC and the conventional Sparse algorithm in high-density localization of active molecules. We show that the SSM_BIC is superior in processing single molecule images with weak signal embedded in strong background.

Computer-assisted diagnosis is key for scaling up cervical cancer screening. However, current recognition algorithms perform poorly on whole slide image (WSI) analysis, fail to generalize for diverse staining and imaging, and show sub-optimal clinical-level verification. Here, we develop a progressive lesion cell recognition method combining low- and high-resolution WSIs to recommend lesion cells and a recurrent neural network-based WSI classification model to evaluate the lesion degree of WSIs. We train and validate our WSI analysis system on 3,545 patient-wise WSIs with 79,911 annotations from multiple hospitals and several imaging instruments. On multi-center independent test sets of 1,170 patient-wise WSIs, we achieve 93.5% Specificity and 95.1% Sensitivity for classifying slides, comparing favourably to the average performance of three independent cytopathologists, and obtain 88.5% true positive rate for highlighting the top 10 lesion cells on 447 positive slides. After deployment, our system recognizes a one giga-pixel WSI in about 1.5 min.

Digital reconstruction of a single neuron occupies an important position in computational neuroscience. Although many novel methods have been proposed, recent advances in molecular labeling and imaging systems allow for the production of large and complicated neuronal datasets, which pose many challenges for neuron reconstruction, especially when discontinuous neuronal morphology appears in a strong noise environment. Here, we develop a new pipeline to address this challenge. Our pipeline is based on two methods, one is the region-to-region connection (RRC) method for detecting the initial part of a neurite, which can effectively gather local cues, i.e., avoid the whole image analysis, and thus boosts the efficacy of computation; the other is constrained principal curves method for completing the neurite reconstruction, which uses the past reconstruction information of a neurite for current reconstruction and thus can be suitable for tracing discontinuous neurites. We investigate the reconstruction performances of our pipeline and some of the best state-of-the-art algorithms on the experimental datasets, indicating the superiority of our method in reconstructing sparsely distributed neurons with discontinuous neuronal morphologies in noisy environment. We show the strong ability of our pipeline in dealing with the large-scale image dataset. We validate the effectiveness in dealing with various kinds of image stacks including those from the DIADEM challenge and BigNeuron project.