The explosion of image data on the Internet has the potential to foster more sophisticated and robust models and algorithms to index, retrieve, organize and interact with images and multimedia data. But exactly how such data can be harnessed and organized remains a critical problem. We introduce here a new database called "ImageNet", a largescale ontology of images built upon the backbone of the WordNet structure. ImageNet aims to populate the majority of the 80,000 synsets of WordNet with an average of 500-1000 clean and full resolution images. This will result in tens of millions of annotated images organized by the semantic hierarchy of WordNet. This paper offers a detailed analysis of ImageNet in its current state: 12 subtrees with 5247 synsets and 3.2 million images in total. We show that ImageNet is much larger in scale and diversity and much more accurate than the current image datasets. Constructing such a large-scale database is a challenging task. We describe the data collection scheme with Amazon Mechanical Turk. Lastly, we illustrate the usefulness of ImageNet through three simple applications in object recognition, image classification and automatic object clustering. We hope that the scale, accuracy, diversity and hierarchical structure of ImageNet can offer unparalleled opportunities to researchers in the computer vision community and beyond.

Convolutional neural network (CNN) depth is of crucial importance for image super-resolution (SR). However, we observe that deeper networks for image SR are more difficult to train. The lowresolution inputs and features contain abundant low-frequency information, which is treated equally across channels, hence hindering the representational ability of CNNs. To solve these problems, we propose the very deep residual channel attention networks (RCAN). Specifically, we propose a residual in residual (RIR) structure to form very deep network, which consists of several residual groups with long skip connections. Each residual group contains some residual blocks with short skip connections. Meanwhile, RIR allows abundant low-frequency information to be bypassed through multiple skip connections, making the main network focus on learning high-frequency information. Furthermore, we propose a channel attention mechanism to adaptively rescale channel-wise features by considering interdependencies among channels. Extensive experiments show that our RCAN achieves better accuracy and visual improvements against state-of-the-art methods.

Fluorescent bioprobes are powerful tools for analytical sensing and optical imaging, which allow direct visualization of biological analytes at the molecular level and offer useful insights into complex biological structures and processes. The sensing and imaging sensitivity of a bioprobe is determined by the brightness and contrast of its fluorescence before and after analyte binding. Emission from a fluorophore is often quenched at high concentration or in aggregate state, which is notoriously known as concentration quenching or aggregation-caused quenching (ACQ). The ACQ effect limits the label-to-analyte ratio and forces researchers to use very dilute solutions of fluorophores. It compels many probes to operate in a fluorescence "turn-off" mode with a narrow scope of practical applications. The unique aggregation-induced emission (AIE) process offers a straightforward solution to the ACQ problem. Typical AIE fluorogens are characterized by their propeller-shaped rotorlike structures, which undergo low-frequency torsional motions as isolated molecules and emit very weakly in solutions. Their aggregates show strong fluorescence mainly due to the restriction of their intramolecular rotations in the aggregate state. This fascinating attribute of AIE fluorogens provides a new platform for the development of fluorescence light-up molecules and photostable nanoaggregates for specific analyte detection and imaging. In this Account, we review our recent AIE work to highlight the utility of AIE effect in the development of new fluorescent bioprobes, which allows the use of highly concentrated fluorogens for biosensing and imaging. The simple design and fluorescence turn-on feature of the molecular AIE bioprobes offer direct visualization of specific analytes and biological processes in aqueous media with higher sensitivity and better accuracy than traditional fluorescence turn-off probes. The AIE dot-based bioprobes with different formulations and surface functionalities show advanced features over quantum dots and small molecule dyes, such as large absorptivity, high luminosity, excellent biocompatibility, free of random blinking, and strong photobleaching resistance. These features enable cancer cell detection, long term cell tracing, and tumor imaging in a noninvasive and high contrast manner. Recent research has significantly expanded the scope of biological applications of AIE fluorogens and offers new strategies to fluorescent bioprobe design. We anticipate that future development on AIE bioprobes will combine one- or multiphoton fluorescence with other modalities (e.g., magnetic resonance imaging) or functionalities (e.g. therapy) to fully demonstrate their potential as a new generation of theranostic reagent. In parallel, the advances in molecular biology will provide more specific bioreceptors, which will enable the development of next generation AIE bioprobes with high selectivity and sensitivity for molecular sensing and imaging.

No abstract

Studying monogenic mitochondrial cardiomyopathies may yield insights into mitochondrial roles in cardiac development and disease. Here, we combine patient-derived and genetically engineered iPSCs with tissue engineering to elucidate the pathophysiology underlying the cardiomyopathy of Barth syndrome (BTHS), a mitochondrial disorder caused by mutation of the gene Tafazzin (TAZ). Using BTHS iPSC-derived cardiomyocytes (iPSC-CMs), we defined metabolic, structural, and functional abnormalities associated with TAZ mutation. BTHS iPSC-CMs assembled sparse and irregular sarcomeres, and engineered BTHS “heart on chip” tissues contracted weakly. Gene replacement and genome editing demonstrated that TAZ mutation is necessary and sufficient for these phenotypes. Sarcomere assembly and myocardial contraction abnormalities occurred in the context of normal whole cell ATP levels. Excess levels of reactive oxygen species mechanistically linked TAZ mutation to impaired cardiomyocyte function. Our study provides new insights into the pathogenesis of Barth syndrome, suggests new treatment strategies, and advances iPSC-based in vitro modeling of cardiomyopathy.

Polymer encapsulated organic nanoparticles have recently attracted increasing attention in the biomedical field because of their unique optical properties, easy fabrication and outstanding performance as imaging and therapeutic agents. Of particular importance is the polymer encapsulated nanoparticles containing conjugated polymers (CP) or fluorogens with aggregation induced emission (AIE) characteristics as the core, which have shown significant advantages in terms of tunable brightness, superb photo- and physical stability, good biocompatibility, potential biodegradability and facile surface functionalization. In this review, we summarize the latest advances in the development of polymer encapsulated CP and AIE fluorogen nanoparticles, including preparation methods, material design and matrix selection, nanoparticle fabrication and surface functionalization for fluorescence and photoacoustic imaging. We also discuss their specific applications in cell labeling, targeted in vitro and in vivo imaging, blood vessel imaging, cell tracing, inflammation monitoring and molecular imaging. We specially focus on strategies to fine-tune the nanoparticle property (e.g. size and fluorescence quantum yield) through precise engineering of the organic cores and careful selection of polymer matrices. The review also highlights the merits and limitations of these nanoparticles as well as strategies used to overcome the limitations. The challenges and perspectives for the future development of polymer encapsulated organic nanoparticles are also discussed.

Premature aging syndromes often result from mutations in nuclear proteins involved in the maintenance of genomic integrity. Lamin A is a major component of the nuclear lamina and nuclear skeleton. Truncation in lamin A causes Hutchinson-Gilford progerial syndrome (HGPS), a severe form of early-onset premature aging. Lack of functional Zmpste24, a metalloproteinase responsible for the maturation of prelamin A, also results in progeroid phenotypes in mice and humans. We found that Zmpste24-deficient mouse embryonic fibroblasts (MEFs) show increased DNA damage and chromosome aberrations and are more sensitive to DNA-damaging agents. Bone marrow cells isolated from Zmpste24-/- mice show increased aneuploidy and the mice are more sensitive to DNA-damaging agents. Recruitment of p53 binding protein 1 (53BP1) and Rad51 to sites of DNA lesion is impaired in Zmpste24-/- MEFs and in HGPS fibroblasts, resulting in delayed checkpoint response and defective DNA repair. Wild-type MEFs ectopically expressing unprocessible prelamin A show similar defects in checkpoint response and DNA repair. Our results indicate that unprocessed prelamin A and truncated lamin A act dominant negatively to perturb DNA damage response and repair, resulting in genomic instability which might contribute to laminopathy-based premature aging.

K-Nearest Neighbor Graph (K-NNG) construction is an important operation with many web related applications, including collaborative filtering, similarity search, and many others in data mining and machine learning. Existing methods for K-NNG construction either do not scale, or are specific to certain similarity measures. We present NN-Descent, a simple yet efficient algorithm for approximate K-NNG construction with arbitrary similarity measures. Our method is based on local search, has minimal space overhead and does not rely on any shared global index. Hence, it is especially suitable for large-scale applications where data structures need to be distributed over the network. We have shown with a variety of datasets and similarity measures that the proposed method typically converges to above 90% recall with each point comparing only to several percent of the whole dataset on average.