Search

Intravoxel incoherent motion imaging (IVIM) is a new magnetic resonance imaging (MRI) technique that can detect both diffusion and pseudo perfusion microscopic tissue information. However, it requires to acquire the diffusion-weighted (DW) images with multiple b-values, which leads to long acquisition time. To reduce the time, the spatial resolution of IVIM is usually not high enough to reflect the image details. To solve this problem, in this study, a multi-similarity block-based super-resolution reconstruction (SRR) method for IVIM was proposed. The fact that the image block similarity of DW images occurs not only across or within the images at different scales, but also the adjacent slices with the same b-values. The intra-slice and inter-slice similarity at multi-scales is defined as multi-similarity in this work. To fully explore the multi-similarity patches to assist SRR, a chain of Gaussian pyramids of a given image slice and the corresponding adjacent slices with the same b-value are established first, and then a random search strategy is used to find and identify the most similar blocks in low resolution (LR) image pyramids. Next, the matched blocks are mapped to high resolution (HR) image pyramids and then unwrapped by inverse affine transformation to replace original LR blocks. Finally, the super-resolution images are obtained through an iterative back-projection algorithm. To evaluate the performance of proposed method, we compared the brain and liver IVIM reconstruction results with several state-of-art methods. The results indicate that the multi-similarity-based method can achieve the best reconstruction results in both DW images and the corresponding IVIM parameter maps. [ABSTRACT FROM AUTHOR]

The task to recognize named entities is often modeled as a sequence labeling process, which selects a label path whose probability is maximum for an input sentence. Because it makes the assumption that the input sentence has a flattened structure, it often fails to recognize nested named entities. In our previous work, a boundary assembling (BA) model was proposed. It is a cascading framework, which identifies named entity boundaries first, and then assembles them into entity candidates for further assessment. This model is effective to recognize nested named entities, but still suffers from poor performance caused by the sparse feature problem. In this article, the BA model is remodeled with the advancement of neural networks, which enables the model to capture semantic information of a sentence by using word embeddings pretrained in external resources. In our experiments, it shows an impressive improvement on the final performance, outperforming the state of the art more than 17% in F-score. [ABSTRACT FROM AUTHOR]