What Makes for Hierarchical Vision Transformer?

Recent studies indicate that hierarchical Vision Transformer (ViT) with a macro architecture of interleaved non-overlapped window-based self-attention & shifted-window operation can achieve state-of-the-art performance in various visual recognition tasks, and challenges the ubiquitous convolutional neural networks (CNNs) using densely slid kernels. In most recently proposed hierarchical ViTs, self-attention is the de-facto standard for spatial information aggregation. In this paper, we question whether self-attention is the only choice for hierarchical ViT to attain strong performance, and study the effects of different kinds of cross-window communication methods. To this end, we replace self-attention layers with embarrassingly simple linear mapping layers, and the resulting proof-of-concept architecture termed TransLinear can achieve very strong performance in ImageNet-[Formula: see text] image recognition. Moreover, we find that TransLinear is able to leverage the ImageNet pre-trained weights and demonstrates competitive transfer learning properties on downstream dense prediction tasks such as object detection and instance segmentation. We also experiment with other alternatives to self-attention for content aggregation inside each non-overlapped window under different cross-window communication approaches. Our results reveal that the macro architecture, other than specific aggregation layers or cross-window communication mechanisms, is more responsible for hierarchical ViT's strong performance and is the real challenger to the ubiquitous CNN's dense sliding window paradigm.

Focalizing regions of biomarker relevance facilitates biomarker prediction on histopathological images.

Image-based AI has thrived as a potentially revolutionary tool for predicting molecular biomarker statuses, which aids in categorizing patients for appropriate medical treatments. However, many methods using hematoxylin and eosin-stained (H&E) whole-slide images (WSIs) have been found to be inefficient because of the presence of numerous uninformative or irrelevant image patches. In this study, we introduced the region of biomarker relevance (ROB) concept to identify the morphological areas most closely associated with biomarkers for accurate status prediction. We actualized this concept within a framework called saliency ROB search (SRS) to enable efficient and effective predictions. By evaluating various lung adenocarcinoma (LUAD) biomarkers, we showcased the superior performance of SRS compared to current state-of-the-art AI approaches. These findings suggest that AI tools, built on the ROB concept, can achieve enhanced molecular biomarker prediction accuracy from pathological images.

Condition-Adaptive Graph Convolution Learning for Skeleton-Based Gait Recognition.

Graph convolutional networks have been widely applied in skeleton-based gait recognition. A key challenge in this task is to distinguish the individual walking styles of different subjects across various views. Existing state-of-the-art methods employ uniform convolutions to extract features from diverse sequences and ignore the effects of viewpoint changes. To overcome these limitations, we propose a condition-adaptive graph (CAG) convolution network that can dynamically adapt to the specific attributes of each skeleton sequence and the corresponding view angle. In contrast to using fixed weights for all joints and sequences, we introduce a joint-specific filter learning (JSFL) module in the CAG method, which produces sequence-adaptive filters at the joint level. The adaptive filters capture fine-grained patterns that are unique to each joint, enabling the extraction of diverse spatial-temporal information about body parts. Additionally, we design a view-adaptive topology learning (VATL) module that generates adaptive graph topologies. These graph topologies are used to correlate the joints adaptively according to the specific view conditions. Thus, CAG can simultaneously adjust to various walking styles and viewpoints. Experiments on the two most widely used datasets (i.e., CASIA-B and OU-MVLP) show that CAG surpasses all previous skeleton-based methods. Moreover, the recognition performance can be enhanced by simply combining CAG with appearance-based methods, demonstrating the ability of CAG to provide useful complementary information.

ADSeg: A flap-attention-based deep learning approach for aortic dissection segmentation.

Accurate and rapid segmentation of the lumen in an aortic dissection (AD) is an important prerequisite for risk evaluation and medical planning for patients with this serious condition. Although some recent studies have pioneered technical advances for the challenging AD segmentation task, they generally neglect the intimal flap structure that separates the true and false lumens. Identification and segmentation of the intimal flap may simplify AD segmentation, and the incorporation of long-distance z axis information interaction along the curved aorta may improve segmentation accuracy. This study proposes a flap attention module that focuses on key flap voxels and performs operations with long-distance attention. In addition, a pragmatic cascaded network structure with feature reuse and a two-step training strategy are presented to fully exploit network representation power. The proposed ADSeg method was evaluated on a multicenter dataset of 108 cases, with or without thrombus; ADSeg outperformed previous state-of-the-art methods by a significant margin and was robust against center variation.

CCNet: Criss-Cross Attention for Semantic Segmentation.

Contextual information is vital in visual understanding problems, such as semantic segmentation and object detection. We propose a criss-cross network (CCNet) for obtaining full-image contextual information in a very effective and efficient way. Concretely, for each pixel, a novel criss-cross attention module harvests the contextual information of all the pixels on its criss-cross path. By taking a further recurrent operation, each pixel can finally capture the full-image dependencies. Besides, a category consistent loss is proposed to enforce the criss-cross attention module to produce more discriminative features. Overall, CCNet is with the following merits: 1) GPU memory friendly. Compared with the non-local block, the proposed recurrent criss-cross attention module requires 11× less GPU memory usage. 2) High computational efficiency. The recurrent criss-cross attention significantly reduces FLOPs by about 85 percent of the non-local block. 3) The state-of-the-art performance. We conduct extensive experiments on semantic segmentation benchmarks including Cityscapes, ADE20K, human parsing benchmark LIP, instance segmentation benchmark COCO, video segmentation benchmark CamVid. In particular, our CCNet achieves the mIoU scores of 81.9, 45.76 and 55.47 percent on the Cityscapes test set, the ADE20K validation set and the LIP validation set respectively, which are the new state-of-the-art results. The source codes are available at https://github.com/speedinghzl/CCNethttps://github.com/speedinghzl/CCNet.

VoxelTrack: Multi-Person 3D Human Pose Estimation and Tracking in the Wild.

We present VoxelTrack for multi-person 3D pose estimation and tracking from a few cameras which are separated by wide baselines. It employs a multi-branch network to jointly estimate 3D poses and re-identification (Re-ID) features for all people in the environment. In contrast to previous efforts which require to establish cross-view correspondence based on noisy 2D pose estimates, it directly estimates and tracks 3D poses from a 3D voxel-based representation constructed from multi-view images. We first discretize the 3D space by regular voxels and compute a feature vector for each voxel by averaging the body joint heatmaps that are inversely projected from all views. We estimate 3D poses from the voxel representation by predicting whether each voxel contains a particular body joint. Similarly, a Re-ID feature is computed for each voxel which is used to track the estimated 3D poses over time. The main advantage of the approach is that it avoids making any hard decisions based on individual images. The approach can robustly estimate and track 3D poses even when people are severely occluded in some cameras. It outperforms the state-of-the-art methods by a large margin on four public datasets including Shelf, Campus, Human3.6 M and CMU Panoptic.

Evaluation of Thermophysical and Mechanical Properties of Sandstone Due to High-Temperature.

In this study, thermophysical and mechanical tests were conducted on sandstone samples from room temperature to 1000 °C. Based on the test results, the thermophysical properties (such as specific heat capacity, thermal conductivity, and thermal expansion coefficient) of sandstone after high-temperature treatment and the variations of mechanical properties (including peak strength, peak strain, elastic modulus, and whole stress-strain curve) with temperature were analyzed. Indeed, the deterioration law of sandstone after high-temperature treatment was also explored with the aid of a scanning electron microscope (SEM). The results show that with the increase in temperature, the specific heat capacity and thermal expansion coefficient of sandstone samples after high-temperature treatment increase first and then decrease, while the thermal conductivity gradually decreases. The range from room temperature to 1000 °C witnesses the following changes: As temperature rises, the peak strength of sandstone rises initially and falls subsequently; the elastic modulus drops; the peak strain increases at an accelerated rate. Temperature change has a significant effect on the deterioration rules of sandstone, and the increase in temperature contributes to the transition in the failure mode of sandstone from brittle failure to ductile failure. The experimental study on the thermophysical and mechanical properties of sandstone under the action of high temperature and overburden pressure has a guiding significance for the site selection and safety evaluation of UCG projects.

MicroRNA-101-3p Suppresses mTOR and Causes Mitochondrial Fragmentation and Cell Degeneration in COPD.

Background: Cigarette smoke is assumed to cause the loss of airway wall structure in chronic obstructive pulmonary disease (COPD) by reducing airway smooth muscle cell (ASMC) function. It also modifies mTOR activity, microRNA (miR)-101-3p expression, and mitochondria function. Here, the link between miR-101-3p and mTOR-regulated mitochondria integrity and ASMC deterioration was assessed. Methods: Disease-specific miR-101-3p expression was determined by RT-PCR in primary ASMC (non-COPD smokers: n = 6; COPD: n = 8; healthy: n = 6). The regulatory effect of miR-101-3p modification on mTOR expression, mitochondrial fragmentation, and remodeling properties (α-SMA, fibronectin, MTCO2, and p70S6 kinase) was assessed in ASMC (healthy nonsmokers: n = 3; COPD: n = 3) by Western blotting and immunofluorescence microscopy. MiR-101-3p was modified by specific mimics or inhibitors, in ASMC stimulated with TNF-α (10 ng/ml) or cigarette smoke extract (CSE). Results: MiR-101-3p expression was significantly higher in ASMC of COPD patients, compared to ASMC of healthy or active smokers. MiR-101-3p expression was increased by TNF-α or CSE. TNF-α or miR-101-3p deteriorated ASMC and mitochondria, while decreasing mTOR signaling, α-SMA, fibronectin, and MTCO2. MiR-101-3p inhibition reduced ASMC deterioration and mitochondrial fragmentation. Conclusion: Constitutive high miR-101-3p expression characterizes COPD-ASMC, causing increased mitochondrial fragmentation and ASMC deterioration. Thus, reactivation mTOR or blocking miR-101-3p presents a potential new strategy for COPD therapy.

Self-supervised learning enables 3D digital subtraction angiography reconstruction from ultra-sparse 2D projection views: A multicenter study.

3D digital subtraction angiography (DSA) reconstruction from rotational 2D projection X-ray angiography is an important basis for diagnosis and treatment of intracranial aneurysms (IAs). The gold standard requires approximately 133 different projection views for 3D reconstruction. A method to significantly reduce the radiation dosage while ensuring the reconstruction quality is yet to be developed. We propose a self-supervised learning method to realize 3D-DSA reconstruction using ultra-sparse 2D projections. 202 cases (100 from one hospital for training and testing, 102 from two other hospitals for external validation) suspected to be suffering from IAs were conducted to analyze the reconstructed images. Two radiologists scored the reconstructed images from internal and external datasets using eight projections and identified all 82 lesions with high diagnostic confidence. The radiation dosages are approximately 1/16.7 compared with the gold standard method. Our proposed method can help develop a revolutionary 3D-DSA reconstruction method for use in clinic.

Downregulation of PLK4 expression induces apoptosis and G0/G1-phase cell cycle arrest in keloid fibroblasts.

OBJECTIVES: Keloids are benign fibroproliferative tumors that display many cancer-like characteristics, such as progressive uncontrolled growth, lack of spontaneous regression, and extremely high rates of recurrence. Polo-like kinase 4 (PLK4) was recently identified as a master regulator of centriole replication, and its aberrant expression is closely associated with tumorigenesis. This study aimed to investigate the expression and biological role of PLK4 in the pathogenesis of keloids. MATERIALS AND METHODS: We evaluated the expression of PLK4 in keloids and adjacent normal skin tissue samples. Then, we established PLK4 knockdown and overexpression cell lines in keloid fibroblasts (KFs) and normal skin fibroblasts (NFs), respectively, to investigate the roles of PLK4 in the regulation of proliferation, migration, invasion, apoptosis, and cell cycle in KFs. Centrinone B (Cen-B), a highly selective PLK4 inhibitor, was used to inhibit PLK4 activity in KFs to evaluate the therapeutic effect on KFs. RESULTS: We discovered that PLK4 was overexpressed in keloid dermal samples and KFs compared with adjacent normal skin samples and NFs derived from the same patients. High PLK4 expression was positively associated with the proliferation, migration, and invasion of KFs. Furthermore, knockdown of PLK4 expression or inhibition of PLK4 activity by Cen-B suppressed KF growth, induced KF apoptosis via the caspase-9/3 pathway, and induced cell cycle arrest at the G0/G1 phase in vitro. CONCLUSIONS: These findings demonstrate that PLK4 is a critical regulator of KF proliferation, migration, and invasion, and thus, Cen-B is a promising candidate drug for keloid treatment.

AlignSeg: Feature-Aligned Segmentation Networks.

Aggregating features in terms of different convolutional blocks or contextual embeddings has been proven to be an effective way to strengthen feature representations for semantic segmentation. However, most of the current popular network architectures tend to ignore the misalignment issues during the feature aggregation process caused by step-by-step downsampling operations and indiscriminate contextual information fusion. In this paper, we explore the principles in addressing such feature misalignment issues and inventively propose Feature-Aligned Segmentation Networks (AlignSeg). AlignSeg consists of two primary modules, i.e., the Aligned Feature Aggregation (AlignFA) module and the Aligned Context Modeling (AlignCM) module. First, AlignFA adopts a simple learnable interpolation strategy to learn transformation offsets of pixels, which can effectively relieve the feature misalignment issue caused by multi-resolution feature aggregation. Second, with the contextual embeddings in hand, AlignCM enables each pixel to choose private custom contextual information adaptively, making the contextual embeddings be better aligned. We validate the effectiveness of our AlignSeg network with extensive experiments on Cityscapes and ADE20K, achieving new state-of-the-art mIoU scores of 82.6 and 45.95 percent, respectively. Our source code is available at https://github.com/speedinghzl/AlignSeg.

Effect of Dietary Inclusion of Riboflavin on Growth, Nutrient Digestibility and Ruminal Fermentation in Hu Lambs.

The study evaluated the influences of riboflavin (RF) supply on the growth performance, nutrient digestibility and ruminal fermentation in lambs. Forty-eight Hu lambs were randomly assigned into four groups receiving RF of 0, 15, 30 and 45 mg/kg dry mater (DM), respectively. Increasing RF supply did not affect the DM intake, but quadratically increased the average daily gain and linearly decreased feed conversion ratio. Total-tract DM, neutral detergent fibre, acid detergent fibre and crude protein digestibility increased quadratically. Rumen pH and propionate molar percentage decreased linearly, total volatile fatty acids concentration, acetate proportion and the ratio of acetate to propionate increased linearly, but ammonia nitrogen concentration was unchanged with increasing RF supply. Linear increases were observed on the activities of carboxymethyl-cellulase, xylanase, pectinase and protease, and the populations of bacteria, fungi, protozoa, dominant cellulolytic bacteria, Ruminobacter amylophilus and Prevotella ruminicola. Methanogens population was not affected by RF supplementation. The microbial protein amount and urinary total purine derivatives excretion increased quadratically. The results indicated that 30 mg/kg DM RF supply improved growth performance, rumen fermentation and nutrient digestion in lambs.

The Associations of Plasma/Serum Carotenoids with Alzheimer's Disease: A Systematic Review and Meta-Analysis.

BACKGROUND: Multiple lines of evidence indicate protective effects of carotenoids in Alzheimer's disease (AD). However, previous epidemiological studies reported inconsistent results regarding the associations between carotenoids levels and the risk of AD. OBJECTIVE: Our study aims to evaluate the associations of six major members of carotenoids with the occurrence of AD by conducting a systematic review and meta-analysis. METHODS: Following PRISMA guidelines, a comprehensive literature search of PubMed, Web of Science, Ebsco, and PsycINFO databases was conducted, and the quality of each included studies was evaluated by a validated scoring systems. Standardized mean differences (SMD) with 95% confidence intervals (CI) were determined by using a random effects model. Heterogeneity was evaluated by I2 statistics. Publication bias was detected using funnel plots and Egger's test. RESULTS: Sixteen studies, with 10,633 participants were included. Pooled analysis showed significantly lower plasma/serum levels of lutein (SMD = -0.86, 95% CI: -1.67 to -0.05, p = 0.04) and zeaxanthin (SMD = -0.59; 95% CI: -1.12 to -0.06, p = 0.03) in patients with AD versus cognitively intact controls, while α-carotene (SMD = 0.21, 95% CI: -0.68 to 0.26, p = 0.39), ß-carotene (SMD = 0.04, 95% CI: -0.57 to 0.65, p = 0.9), lycopene (SMD = -0.12, 95% CI: -0.96 to 0.72, p = 0.78), and ß-cryptoxanthin (SMD = -0.09, 95% CI: -0.83 to 0.65, p = 0.81) did not achieve significant differences. CONCLUSION: Of six major members of carotenoids, only lutein and zeaxanthin concentrations in plasma/serum were inversely related to the risk of AD. More high-quality longitudinal studies are needed to verify these findings.

Particle Size Effect and Temperature Effect on the Pore Structure of Low-Rank Coal.

High moisture content and high volatile content are typical characteristics of low-rank coal. To acquire the pore structure characteristics of low-rank coal accurately, the particle sizes and the pretreatment temperatures are two key parameters that should be considered when the low-pressure liquid-nitrogen adsorption is used. In this study, a low-rank coal sample was collected from Ordos Basin, and it was polished into four different particle sizes, 40-80 mesh, 80-120 mesh, 120-160 mesh, and 160-200 mesh, respectively. Besides, the low-rank coal samples are handled under seven various pretreatment temperatures (ranging from 120 to 300 °C); then, the pore structure characteristics of low-rank coal under various particle sizes and pretreatment temperatures are acquired. The dynamic change of pore volume and pore-specific surface area for low-rank coal is coincident. Under the same pretreatment temperatures, the mesopores' volume continuously decreases. When the pretreatment temperature reaches 300 °C, a faint increase in their volume is observed. These results mean the mesopores are damaged during the progressive pulverization and heating procedures. When it comes to the same particle sizes, the mesopores' volume also decreased with the increased pretreatment temperatures. Contrarily, the macropore volume is stable. This is mainly due to the decomposition of volatile matters and collapse of mesopores under the high pretreatment temperatures. However, the enrichment of ash in the mesopores could maintain the coal skeleton. The particle size effect and temperature effect mainly relate to the mesopores in low-rank coal, and the pores with the aperture below 5 nm contribute predominantly, followed by the pores with the aperture ranging from 5 to 10 nm.

FNA++: Fast Network Adaptation via Parameter Remapping and Architecture Search.

Deep neural networks achieve remarkable performance in many computer vision tasks. Most state-of-the-art (SOTA) semantic segmentation and object detection approaches reuse neural network architectures designed for image classification as the backbone, commonly pre-trained on ImageNet. However, performance gains can be achieved by designing network architectures specifically for detection and segmentation, as shown by recent neural architecture search (NAS) research for detection and segmentation. One major challenge though is that ImageNet pre-training of the search space representation (a.k.a. super network) or the searched networks incurs huge computational cost. In this paper, we propose a Fast Network Adaptation (FNA++) method, which can adapt both the architecture and parameters of a seed network (e.g., an ImageNet pre-trained network) to become a network with different depths, widths, or kernel sizes via a parameter remapping technique, making it possible to use NAS for segmentation and detection tasks a lot more efficiently. In our experiments, we apply FNA++ on MobileNetV2 to obtain new networks for semantic segmentation, object detection, and human pose estimation that clearly outperform existing networks designed both manually and by NAS. We also implement FNA++ on ResNets and NAS networks, which demonstrates a great generalization ability. The total computation cost of FNA++ is significantly less than SOTA segmentation and detection NAS approaches: 1737× less than DPC, 6.8× less than Auto-DeepLab, and 8.0× less than DetNAS. A series of ablation studies are performed to demonstrate the effectiveness, and detailed analysis is provided for more insights into the working mechanism. Codes are available at https://github.com/JaminFong/FNA.