Multiscale Dynamic Graph Representation for Biometric Recognition With Occlusions.

Occlusion is a common problem with biometric recognition in the wild. The generalization ability of CNNs greatly decreases due to the adverse effects of various occlusions. To this end, we propose a novel unified framework integrating the merits of both CNNs and graph models to overcome occlusion problems in biometric recognition, called multiscale dynamic graph representation (MS-DGR). More specifically, a group of deep features reflected on certain subregions is recrafted into a feature graph (FG). Each node inside the FG is deemed to characterize a specific local region of the input sample, and the edges imply the co-occurrence of non-occluded regions. By analyzing the similarities of the node representations and measuring the topological structures stored in the adjacent matrix, the proposed framework leverages dynamic graph matching to judiciously discard the nodes corresponding to the occluded parts. The multiscale strategy is further incorporated to attain more diverse nodes representing regions of various sizes. Furthermore, the proposed framework exhibits a more illustrative and reasonable inference by showing the paired nodes. Extensive experiments demonstrate the superiority of the proposed framework, which boosts the accuracy in both natural and occlusion-simulated cases by a large margin compared with that of baseline methods.

Pose-Appearance Relational Modeling for Video Action Recognition.

Recent studies of video action recognition can be classified into two categories: the appearance-based methods and the pose-based methods. The appearance-based methods generally cannot model temporal dynamics of large motion well by virtue of optical flow estimation, while the pose-based methods ignore the visual context information such as typical scenes and objects, which are also important cues for action understanding. In this paper, we tackle these problems by proposing a Pose-Appearance Relational Network (PARNet), which models the correlation between human pose and image appearance, and combines the benefits of these two modalities to improve the robustness towards unconstrained real-world videos. There are three network streams in our model, namely pose stream, appearance stream and relation stream. For the pose stream, a Temporal Multi-Pose RNN module is constructed to obtain the dynamic representations through temporal modeling of 2D poses. For the appearance stream, a Spatial Appearance CNN module is employed to extract the global appearance representation of the video sequence. For the relation stream, a Pose-Aware RNN module is built to connect pose and appearance streams by modelling action-sensitive visual context information. Through jointly optimizing the three modules, PARNet achieves superior performances compared with the state-of-the-arts on both the pose-complete datasets (KTH, Penn-Action, UCF11) and the challenging pose-incomplete datasets (UCF101, HMDB51, JHMDB), demonstrating its robustness towards complex environments and noisy skeletons. Its effectiveness on NTU-RGBD dataset is also validated even compared with 3D skeleton-based methods. Furthermore, an appearance-enhanced PARNet equipped with a RGB-based I3D stream is proposed, which outperforms the Kinetics pre-trained competitors on UCF101 and HMDB51. The better experimental results verify the potentials of our framework by integrating various modules.

LFNet: A Novel Bidirectional Recurrent Convolutional Neural Network for Light-Field Image Super-Resolution.

The low spatial resolution of light-field image poses significant difficulties in exploiting its advantage. To mitigate the dependency of accurate depth or disparity information as priors for light-field image super-resolution, we propose an implicitly multi-scale fusion scheme to accumulate contextual information from multiple scales for super-resolution reconstruction. The implicitly multi-scale fusion scheme is then incorporated into bidirectional recurrent convolutional neural network, which aims to iteratively model spatial relations between horizontally or vertically adjacent sub-aperture images of light-field data. Within the network, the recurrent convolutions are modified to be more effective and flexible in modeling the spatial correlations between neighboring views. A horizontal sub-network and a vertical sub-network of the same network structure are ensembled for final outputs via stacked generalization. Experimental results on synthetic and real-world data sets demonstrate that the proposed method outperforms other state-of-the-art methods by a large margin in peak signal-to-noise ratio and gray-scale structural similarity indexes, which also achieves superior quality for human visual systems. Furthermore, the proposed method can enhance the performance of light field applications such as depth estimation.

High WT1 expression is associated with very poor survival of patients with osteogenic sarcoma metastasis.

PURPOSE: Although metastasis is the primary determinant of poor survival of patients with osteogenic sarcoma, some patients live much longer than others, indicating metastatic heterogeneity underlying survival outcome. The purpose of the investigation was to identify genes underlying survival outcome of patients with osteogenic sarcoma metastasis. EXPERIMENTAL DESIGN: We have used microarray to first compare mRNA expression between normal bone and osteogenic sarcoma specimens, identified genes overexpressed in osteogenic sarcoma, and compared expression of the selected gene between a poorly metastatic (SAOS) and two highly metastatic cell lines (LM8 and 143B). Finally, expression of the selected gene was assessed by immunostaining of osteogenic sarcoma samples with known survival outcome. RESULTS: Microarray analysis revealed 5.3-fold more expression of WT1 mRNA in osteogenic sarcoma compared with normal bone and >2-fold overexpression in 143B and LM8 cells compared with SAOS. Furthermore, WT1 mRNA was absent in normal bone (10 of 10) by reverse transcription-PCR but present in osteogenic sarcoma-derived cell lines (5 of 8). One hundred percent (42 of 42) of low-grade osteogenic sarcoma specimens expressed no WT1 as determined by immunostaining; however, 24% (12 of 49) of the high-grade specimens showed intense staining. Mean survival of patients with high-grade metastatic osteogenic sarcoma but low WT1 staining (27 of 37) was 96.5 +/- 129.3 months, whereas mean survival of patients with high-grade metastatic osteogenic sarcoma having intense staining (10 of 37) was 18.3 +/- 12.3 months (P > 0.0143). All splice variants of WT1 mRNA, including a hitherto unknown variant (lacking exons 4 and 5), were found to be expressed in osteogenic sarcoma. CONCLUSION: WT1 seems to be associated with very poor survival of patients with osteogenic sarcoma metastasis.

Expression of a novel alternatively spliced UCP-2 transcript in osteogenic sarcoma.

BACKGROUND: Development of chemoresistance is common in patients with osteogenic sarcoma (OGS); however, the underlying mechanism is largely unknown. Many anticancer drugs exert their therapeutic action by generating reactive oxygen radicals, which might be countered by the cancer cell through induction of uncoupling protein 2 (UCP-2). UCP-2 has been shown to be able to protect tumor cells from the cytotoxic actions of chemotherapeutic drugs. Because OGS is seldom completely cured by current chemotherapy regimens, we hypothesized that increased expression of UCP-2 underlies this phenomenon. The primary initial interest of our research was to evaluate the level of UCP-2 mRNA in OGS. METHODS: The level of UCP-2 mRNA was determined by reverse transcriptase polymerase chain reaction (RT-PCR) comparing expression in normal-bone-derived specimens and OGS-derived specimens. Semiquantification of mRNA expression was achieved by radioactive RT-PCR. Nucleotide sequencing was performed using automated instruments. RESULTS: Interestingly, we failed to observe induction of UCP-2 mRNA in OGS tumor specimens and OGS-derived primary cell lines compared to the expression level in normal bone. However, we found expression of a hitherto unknown UCP-2 transcript in eight of eight OGS-derived and one EWS-derived cell lines and in nine of ten OGS biopsy specimens but in only one of six normal bone-derived specimens. Thus, tumor samples express both types (normal and the novel one) of UCP-2 mRNAs, whereas normal bone expresses only the wild-type form. Further experiments identified the novel mRNA species as an alternatively spliced UCP-2 transcript (termed UCP-2as). UCP-2as has a 22-nucleotide insertion from the 3' end of intron 3 that introduces an early stop codon in exon 4, which theoretically can produce a protein 79 amino acids long. CONCLUSIONS: We have identified a hitherto unknown UCP-2 transcript. Expression of the novel transcript appears to be OGS-specific, implying a function advantageous to the tumor.

Intragenic codon bias in a set of mouse and human genes.

To better conceptualize the mechanism underlying the evolution of synonymous codons, we have analysed intragenic codon usage in chosen "regions" of some mouse and human genes. We divided a given gene into two regions: one consisting of a trinucleotide repeat (TNR) and the other consisting of the "rest of the coding region" (RCR). Usually, a TNR is composed of a repetitive single codon, which may reflect its frequency in a gene. In contrast, a non-random frequency of a codon in the RCR versus TNR (or vice versa) of a gene should indicate a bias for that codon within the TNR. We examined this scenario by comparing codon frequency between the RCR and the cognate TNR(s) for a set of human and mouse genes. A TNR length of six amino acids or more was used to identify genes from the Genbank database. Twenty nine human and twenty one mouse genes containing TNRs coding for nine different amino acid runs were identified. The ratio of codon frequency in a TNR versus the corresponding RCR was expressed as "fold change" which was also regarded as a measure of codon bias (defined as preferential use either in TNR or in RCR). Chi-square values were then determined from the distribution of codon frequency in a TNR vs. the cognate RCR. At p<0.001, 22% and 27%, respectively, of human and mouse TNRs showed codon bias. Greater than 40% of the TNRs (29 out of 69 in human, and 18 of 42 in mouse) showed codon bias at p<0.05. In addition, we identify eight single-codon TNRs in mouse and ten in human genes. Thus, our results show intragenic codon bias in both mouse and human genes expressed in diverse tissue types. Since our results are independent of the Codon Adaptation Index (CAI) and starvation CAI, and since the tRNA repertoire in a cell or in a tissue is constant, our data suggest that other constraints besides tRNA abundance played a role in creating intragenic codon bias in these genes.

The human homolog of yeast SEP1 is a novel candidate tumor suppressor gene in osteogenic sarcoma.

The hSEP1 gene is the human homolog of yeast SEP1. Yeast SEP1 is a multifunctional gene that regulates a variety of nuclear and cytoplasmic functions including homologous recombination, meiosis, telomere maintenance, RNA metabolism and microtubule assembly. The function of hSEP1 is not known. We show loss or reduced expression of hSEP1 messenger RNA (mRNA) in three of four primary osteogenic sarcoma (OGS)-derived cell lines and in eight of nine OGS biopsy specimen. In addition, we find a heterozygous missense mutation (Valine(1484)>Alanine) at a conserved amino acid in the primary OGS-derived cell line U2OS. Importantly, we identified a homozygous missense mutation involving a CG-dinucleotide leading to a change in a conserved amino acid, aspartic acid(1137) >asparagine, in the primary OGS-derived cell line, TE85. hSEP1 mRNA expression was nearly undetectable in TE85 and low in U2OS cell lines. None of these mutations were identified in 20 normal samples consisting of bone, cartilage and fibroblast. The hSEP1 gene is located in chromosome 3 at 3q25-26.1 between markers D3S1309 and D3S1569. An adjacent locus defined by the polymorphic markers D3S1212 and D3S1245 has previously been reported to undergo loss of heterozygosity (LOH) at a >70% frequency in OGS and claimed to harbor an important tumor suppressor gene in osteosarcoma. The homozygous mutation in the hSEP1 mRNA in TE85 cell line suggest that this gene itself is subject to LOH. Taken together, these results suggest that hSEP1 acts as a tumor suppressor gene in OGS.