The dose-volume histogram (DVH) is a clinically relevant criterion to evaluate the quality of a treatment plan. It is hence desirable to incorporate DVH constraints into treatment plan optimization for intensity modulated radiation therapy. Yet, the direct inclusion of the DVH constraints into a treatment plan optimization model typically leads to great computational difficulties due to the non-convex nature of these constraints. To overcome this critical limitation, we propose a new convex-moment-based optimization approach. Our main idea is to replace the non-convex DVH constraints by a set of convex moment constraints. In turn, the proposed approach is able to generate a Pareto-optimal plan whose DVHs are close to, or if possible even outperform, the desired DVHs. In particular, our experiment on a prostate cancer patient case demonstrates the effectiveness of this approach by employing two and three moment formulations to approximate the desired DVHs.

Deep Residual Networks (ResNets) have recently achieved state-of-the-art results on many challenging computer vision tasks. In this work we analyze the role of Batch Normalization (BatchNorm) layers on ResNets in the hope of improving the current architecture and better incorporating other normalization techniques, such as Normalization Propagation (NormProp), into ResNets. Firstly, we verify that BatchNorm helps distribute representation learning to residual blocks at all layers, as opposed to a plain ResNet without BatchNorm where learning happens mostly in the latter part of the network. We also observe that BatchNorm well regularizes Concatenated ReLU (CReLU) activation scheme on ResNets, whose magnitude of activation grows by preserving both positive and negative responses when going deeper into the network. Secondly, we investigate the use of NormProp as a replacement for BatchNorm in ResNets. Though NormProp theoretically attains the same effect as BatchNorm on generic convolutional neural networks, the identity mapping of ResNets invalidates its theoretical promise and NormProp exhibits a significant performance drop when naively applied. To bridge the gap between BatchNorm and NormProp in ResNets, we propose a simple modification to NormProp and employ the CReLU activation scheme. We experiment on visual object recognition benchmark datasets such as CIFAR-10/100 and ImageNet and demonstrate that 1) the modified NormProp performs better than the original NormProp but is still not comparable to BatchNorm and 2) CReLU improves the performance of ResNets with or without normalizations.

No abstract

During the past decades, with the implementation of pneumococcal polysaccharide vaccine (PPV) and pneumococcal conjugate vaccines (PCVs), a dramatic reduction in vaccine type diseases and transmissions has occurred. However, it is necessary to develop a less expensive, serotype-independent pneumococcal vaccine due to the emergence of nonvaccine-type pneumococcal diseases and the limited effect of vaccines on colonization. As next-generation vaccines, conserved proteins, such as neuraminidase A (NanA), elongation factor Tu (Tuf), and pneumolysin (Ply), are promising targets against pneumococcal infections. Here, we designed and constructed a novel fusion protein, NanAT1-TufT1-PlyD4, using the structural and functional domains of full-length NanA, Tuf and Ply proteins with suitable linkers based on bioinformatics analysis and molecular cloning technology. Then, we tested whether the protein protected against focal and lethal pneumococcal infections and examined its potential protective mechanisms. The fusion protein NanAT1-TufT1-PlyD4 consists of 627 amino acids, which exhibits a relatively high level of thermostability, high stability, solubility and a high antigenic index without allergenicity. The purified fusion protein was used to subcutaneously immunize C57BL/6 mice, and NanAT1-TufT1-PlyD4 induced a strong and significant humoral immune response. The anti-NanAT1-TufT1-PlyD4 specific IgG antibody assays increased after the first immunization and reached the highest value at the 35th day. The results from in vitro experiments showed that anti-NanAT1-TufT1-PlyD4 antisera could inhibit the adhesion of Streptococcus pneumoniae (S. pneumoniae) to A549 cells. In addition, immunization with NanAT1-TufT1-PlyD4 significantly reduced S. pneumoniae colonization in the lung and decreased the damage to the lung tissues induced by S. pneumoniae infection. After challenge with a lethal dose of serotype 3 (NC_WCSUH32403), a better protection effect was observed with NanAT1-TufT1-PlyD4-immunized mice than with the separate full-length proteins and the adjuvant control; the survival rate was 50%, which met the standard of the marketed vaccine. Moreover, we showed that the humoral immune response and the Th1, Th2 and Th17-cellular immune pathways are involved in the immune protection of NanAT1-TufT1-PlyD4 to the host. Collectively, our results support that the novel fusion protein NanAT1-TufT1-PlyD4 exhibits extensive immune stimulation and is effective against pneumococcal challenges, and these properties are partially attributed to humoral and cellular-mediated immune responses.

proposed) 128x128 Stage2 generation VAE Figure 1: Comparison demonstrating our channel-recurrent VAE-GAN's superior ability to model complex bird images. Based on the high-quality generation of Stage1 64×64 images, higher-resolution Stage2 images can be further synthesized unsupervisedly. AbstractDespite recent successes in synthesizing faces and bedrooms, existing generative models struggle to capture more complex image types (Figure 1), potentially due to the oversimplification of their latent space constructions. To tackle this issue, building on Variational Autoencoders (VAEs), we integrate recurrent connections across channels to both inference and generation steps, allowing the high-level features to be captured in global-to-local, coarse-to-fine manners. Combined with adversarial loss, our channelrecurrent VAE-GAN (crVAE-GAN) outperforms VAE-GAN in generating a diverse spectrum of high resolution images while maintaining the same level of computational efficacy. Our model produces interpretable and expressive latent representations to benefit downstream tasks such as image completion. Moreover, we propose two novel regularizations, namely the KL objective weighting scheme over time steps and mutual information maximization between transformed latent variables and the outputs, to enhance the training.