Search

Sepsis is a life-threatening systemic syndrome characterized by various biological, biochemical, and physiological abnormalities. Due to its high mortality, identifying sepsis patients with high risk of in-hospital death early and accurately will help doctors make optimal clinical decisions and reduce the mortality of sepsis patients. In this paper, we propose a length insensitive TCN-based model to predict sepsis patient's death risk in the future k hours, which is the first work for sepsis death risk early warning model only based on vital signs time series to our best knowledge. Furthermore, we design residual connections between temporal residual blocks to improve the prediction performance and stability especially on short input sequences. We validate and evaluate our model on two freely-available datasets, i.e., MIMIC-IV and eICU, from which 16,520 and 29,620 patients are selected respectively. The experiment results show that our model outperforms LSTM and other machine learning methods, as it has the highest sensitivity and Youden index in almost all cases. Meanwhile, the Youden index of the TCN-based model only slightly decreases by 0.0233 and 0.0307 when the time range of the input sequence changes from 24 to 4 h for k equal to 6 and 12, respectively.

Speckle noise reduction is an important issue in synthetic aperture radar (SAR) imaging. Because SAR images are distinct in being complex valued and susceptible to corruption owing to multiplicative fluctuations, specialized methods for speckle reduction are needed. Techniques based on nonlocal means perform denoising by exploiting the natural redundancy of patterns within an image. They calculate a weighted average of pixels whose neighborhoods are close to one another, where this significantly reduces noise while preserving most image content. While this method performs well on flat areas and textures, its results are excessively smooth in low-contrast areas, and leave residual noise around edges and singular structures. Another variational denoising method uses total variation (TV) minimization to restore regular images but is prone to excessively smooth textures, the staircasing effect, and contrast losses. Our proposed model is intended for the logarithmic domain of SAR data, and combines the above two methods by minimizing an adaptive TV using a nonlocal data fidelity term. In the variational functionals developed here, weighted parameters of nonlocal regularization are adaptively tuned based on local heterogeneity information and noise in the images. A fast iterative shrinkage/thresholding algorithm (FISTA) is then used to solve the optimization problem. The results of experiments on real SAR images verify the effectiveness of the proposed method in terms of speckle reduction.

Inspired by structures of antenna-reaction centers in photosynthesis, the complex micelle was prepared from zinc tetra-phenyl porphyrin (ZnTPP), fullerene derivative (PyC60) and poly(ethylene glycol)-block-poly(e-caprolactone) (PEG-b-PCL). The core-shell structure made the hydrophobic donor-acceptor system work in aqueous. In micellar core, coordination interaction occurred between ZnTPP and PyC60 molecules which ensured the enhanced energy migration from the donor to the acceptor. The enhanced interaction between porphyrin and fullerene was confirmed by absorption, steady-state fluorescence and transient fluorescence. The generation of singlet oxygen and superoxide radical was detected by iodide method and reduction of nitro blue tetrazolium, respectively, which confirmed that electron transfer reaction in the complex micellar core occurred. Moreover, the complex micelle exhibited effective electron transfer performance in photodebromination of 2,3-dibromo-3-phenylpropionic acid. The complex micellar structure endowed the donor-acceptor system with improved stability under irradiation. This strategy could be helpful for designing new electron transfer platform and artificial photosynthetic system.

Block-matching and 3D-filtering (BM3D) is a state of the art denoising algorithm for image/video, which takes full advantages of the spatial correlation and the temporal correlation of the video. The algorithm performance comes at the price of more similar blocks finding and filtering which bring high computation and memory access. Area, memory bandwidth and computation are the major bottlenecks to design a feasible architecture because of large frame size and search range. In this paper, we introduce a novel structure to increase data reuse rate and reduce the internal static-random-access-memory (SRAM) memory. Our target is to design a phase alternating line (PAL) or real-time processing chip of BM3D. We propose an application specific integrated circuit (ASIC) architecture of BM3D for a 720×576 BT656 PAL format. The feature of the chip is with 100 MHz system frequency and a 166-MHz 32-bit double data rate (DDR). When noise is σ = 25, we successfully realize real-time denoising and achieve about 10 dB peak signal to noise ratio (PSNR) advance just by one iteration of the BM3D algorithm.

This paper proposes a fusion-based edge-sensitive interpolation method (FEID) for intra-field deinterlacing. The proposed FEID is composed of three steps: (1) region classification by a gradient-based region selection approach, (2) pre-interpolation by a 6-tap fixed coefficient Wiener filter, (3) data fusion by the linear minimum mean square-error estimation (LMMSE) technique. Specifically, three directional neighboring pixel sets are defined in three directions (45°, 90°, and 135°) for every missing pixel. And each set produces an estimate of the pixel to be interpolated with a Wiener filter. With the information that gathered from the three directional neighboring pixel sets, a more robust estimate is obtained by fusing these directional estimates with the LMMSE technique. For fast implementation, we propose a gradient-based region selection approach that classifies a local region into two different classes, Region 1 and Region 2. The LMMSE-based data fusion method is used in Region 1; a fast deinterlacing algorithm is used in Region 2 to reduce the computational complexity. Compared with existing deinterlacing methods, the proposed method FEID improves the visual quality of the interpolated edges while maintaining a higher peak signal-to-noise---ratio (PSNR) level.

With the characteristic size reducing as well as the power densities exponentially increasing, elevated chip temperatures are true limiters to the performance and reliability of integrated circuits. To address these thermal issues, it is essential to use a set of on-chip thermal sensors to monitor temperatures during operation. These temperature sampling results are then used by thermal management techniques to appropriately manage chip performance. In this paper, we propose a surface spline interpolation method to reconstruct the full thermal characterization of integrated circuits with non-uniform thermal sensor placements. We construct the thermal surface function using the mathematical tool of surface spline with the matrix calculation of the non-uniform sample data. Then, we take the coordinates of the points at grid locations into the surface function to get its temperature value so that we can reconstruct the full thermal signals. To evaluate the effectiveness of our method, we develop an experiment for reconstructing full thermal status of a 16-core processor. Experimental results show that our method outperforms the inverse distance weighting method based on dynamic Voronoi diagram and spectral analysis techniques both in the average absolute error metric and the hot spot absolute error metric with short enough runtime to meet the real-time process demand. Besides, our method still has the advantages such as its mathematical simplicity with no need of pre-process.

This paper provides a new way to calculate the temperature of steel components in localized fires. A newly developed concept named adiabatic surface temperature (AST) on fire-structural interface research has been used in the calculation to represent the fire exposed boundary conditions. Based on AST, a simple heat transfer model with the capability of considering heat conduction among different parts of the steel sections has been proposed. The popular CFD code FDS (version 5.3.0), which can output ASTs, has been adopted to simulate the heating of a steel beam and a steel column in a localized fire. Steel temperatures calculated by the proposed model agree very well with the results from CFD simulation. Studies also obviously show that using the surrounding gas temperatures for thermal calculation of steel members exposed to fire might yield unreasonable and unsafe results.