Search

Magnetic resonance (MR) imaging often lacks standardized acquisition protocols across various sites, leading to contrast variations that reduce image quality and hinder automated analysis. MR harmonization improves consistency by integrating data from multiple sources, ensuring reproducible analysis. Recent advances leverage image-to-image translation and disentangled representation learning to decompose anatomical and contrast representations, achieving consistent cross-site harmonization. However, these methods face two significant drawbacks: imbalanced contrast availability during training affects adaptation performance, and insufficient utilization of spatial variability in local anatomical structures limits model adaptability to different sites. To address these challenges, we propose Domain-aware Adaptive Weighting with Fine-Grain Attention (DAW-FA) for Unsupervised MRI Harmonization. DAW-FA incorporates an adaptive weighting mechanism and enhanced self-attention to mitigate MR contrast imbalance during training and account for spatial variability in local anatomical structures. This facilitates robust cross-site harmonization without requiring paired inter-site images. We evaluated DAW-FA on MR datasets with varying scanners and acquisition protocols. Experimental results show DAW-FA outperforms existing methods, with an average increase of 1.92 ± 0.56 in Peak Signal-to-Noise Ratio (PSNR) and 0.023 ± 0.011 in Structural Similarity Index Measure (SSIM). Additionally, we demonstrate DAW-FA’s impact on downstream tasks: Alzheimer’s disease classification and whole-brain segmentation, highlighting its potential clinical relevance.

Abstract Point cloud completion is prevalent due to the insufficient results from current point cloud acquisition equipments, where a large number of point data failed to represent a relatively complete shape. Existing point cloud completion algorithms, mostly encoder‐decoder structures with grids transform (also presented as folding operation), can hardly obtain a persuasive representation of input clouds due to the issue that their bottleneck‐shape result cannot tell a precise relationship between the global and local structures. For this reason, this article proposes a novel point cloud completion model based on a Stack‐Style Folding Network (SSFN). Firstly, to enhance the deep latent feature extraction, SSFN enhances the exploitation of shape feature extractor by integrating both low‐level point feature and high‐level graphical feature. Next, a precise presentation is obtained from a high dimensional semantic space to improve the reconstruction ability. Finally, a refining module is designed to make a more evenly distributed result. Experimental results shows that our SSFN produces the most promising results of multiple representative metrics with a smaller scale parameters than current models.

In this paper, two types (i.e., type-A and type-B) of hybrid microstrip/defected ground structure (DGS) cells are proposed for passive circuit implementation with ultra-wide stopband. Both cells consist of the stepped-impedance DGS and embedded folded slotline on the ground, which could obtain the dual-resonances. In type-A cell, a microstrip T-stub on the top side is introduced, which can not only allocate a strong coupling to the DGS with slotline on the bottom side, but also act as the input/output port. To finely adjust the dual-resonances of the type-B cell, a grounded microstrip patch is used. Meanwhile, such compact cells could feature an ultra-wide upper stopband, due to their own slow-wave effect. Based on the aforementioned hybrid microstrip/DGS cells, two dual-band bandpass filters (BPFs) and a dual-band filtering power divider (FPD) are proposed and fabricated. Measured and simulated results are in a fairly-close agreement. Both dual-band BPFs exhibit the ultra-wide upper stopband, which extends up to 40 GHz with a high rejection level about 30 dB. Besides, the dual-band FPD has merits of more than 18.7 dB of in-band isolation and 28 dB stopband rejection levels up to 40 GHz.

Accurate tooth segmentation from CBCT images is a crucial step for specialist to perform quantitative analysis, clinical diagnosis and operation. In this paper, we present a symmetric full convolutional network with residual block and Dense Conditional Random Field (DCRF), which can achieve accurate segmentation automatically for tooth images. The proposed method can not only strengthen feature propagation, but also boost feature reuse, which can be credited to the contracting path and the expanding path that extract and recover pixel cues sufficiently. To this end, we apply special deep bottleneck architectures (DBAs) and summation-based skip connection into our network to ensure accurate segmentation for much deeper neural network. Compared with previous methods which are based on conditional random field for original image intensity, our approach applies DCRF to the posterior probability generated by the proposed network. To avoid the interferences of noises around the tooth, we combine the pixel-level prediction capability of DCRF, which further enhance the segmentation performance. In the experiments, we verify the capabilities of our methods based on four evaluation indicators, which demonstrates the superiority of our method.

Target tracking based on Siamese network has reached the state-of-the-art performance. However is still limited in semantic feature extraction. In this paper, we propose a novel method to distinguish positive and negative samples. Taking deep neural network as the backbone, we fuse the feature maps from different layers and feed it to RPN (Region Proposal Network). In addition, we use a loss term for loss function to achieve self-adjusting and learn more discriminative embedding features of target objects with similar semantics. In the tracking stage, one-shot detection is used as the reference, fix the first frame as the weight of tracking to track the subsequent frames. Our method has achieved outstanding performance on several benchmark data set, such as: OTB2015, VOT2016, VOT2018, and VOT2019 et al.

Currently, object detectors based on CNN, such as RetinaNet, Faster-RCNN, CornerNet series, can achieve good performance, but have some common drawbacks, like large calculation cost, high model complexity and slow detection speed. In this paper, a new lightweight object detector is proposed, which adopted a density-based approach to merge the real boxes. To reduce calculation cost and improve detection speed, the tactic of multi-scale output is adopted to predict objects of different sizes with features of different scales. Furthermore, a new lightweight network model is proposed, which can show better performance in computation, FPS, and model complexity. Meanwhile, the separation of convolution is used to improve the basic convolution layer, which can achieve better results under the same number of filters. In the experiments, we verified the capability of our methods based on ablation experiment and model evaluation, which demonstrates the superiority of our method. Moreover, we have also conducted deep network and multichannel experiments on MS-COCO2014 datasets and achieved 20.9% mAP performance.

Constructing a robust appearance model of the visual object is a crucial task for visual object tracking. Recently, more and more studies combine spatial feature with a temporal feature to improve the tracking performance. These methods successfully apply the features from spatial and temporal to address the problem for tracking. This paper presents a novel method for visual object tracking based on spatiotemporal feature combined with correlation filters. In this paper, the visual features of a target object are extracted from a spatial-temporal residual network (STResNet) appearance model with two sub-networks. The STResNet appearance model learns separately spatial feature and temporal feature, respectively, so that we can effectively utilize spatial context around the surrounding of the target object in each frame and the temporal relationship between successive frames to refine the appearance representation of the target object. Finally, our spatiotemporal fusion feature from STResNet appearance model is incorporated into the correlation filter for robust visual object tracking. The experimental results show that our method achieves similar or better performance compared with the other tracking methods based on convolutional neural networks or correlation filter.

The 3-D reconstruction methods based on structure from motion (SfM) pipeline mainly use the traditional scale invariant feature detecting methods for feature matching. This type of methods suffers from the accuracy with affine matching in the image-based modeling system. In this paper, we propose an affine-invariant feature detection and matching method which is accurate and fast based on the three types of critical points in Morse theory called precise extreme feature region (PEFR). We also exploit the evaluation method for the reconstruction results without the ground truth. We propose the average descending rate to quantitatively evaluate the accuracy of the reconstruction model. Using PEFR in an SfM pipeline to build 3-D point-cloud, our method presents better results and less runtime compared to the state-of-the-art approaches. In addition, our quality assessment method presents a practical comparison of the reconstruction result based on the geometric accuracy of the reconstruction models. The experimental results prove that our PEFR method is faster and more accurate and works well on the reconstruction pipeline. The assessment method is fit for the quality assessment on 3-D reconstruction.

Aerial object detection acts a pivotal role in searching and tracking applications. However, the large model, limited memory, and computing power of embedded devices restrict aerial pedestrian detection algorithms’ deployment on the UAV (unmanned aerial vehicle) platform. In this paper, an innovative method of aerial infrared YOLO (AIR-YOLOv3) is proposed, which combines network pruning and the YOLOv3 method. Firstly, to achieve a more appropriate number and size of the prior boxes, the prior boxes are reclustered. Then, to accelerate the inference speed on the premise of ensuring the detection accuracy, we introduced Smooth-L1 regularization on channel scale factors, and we pruned the channels and layers with less feature information to obtain a pruned YOLOv3 model. Meanwhile, we proposed the self-built aerial infrared dataset and designed ablation experiments to perform model evaluation well. Experimental results show that the AP (average precision) of AIR-YOLOv3 is 91.5% and the model size is 10.7 MB (megabyte). Compared to the original YOLOv3, its model volume compressed by 228.7 MB, nearly 95.5 %, while the model AP decreased by only 1.7%. The calculation amount is reduced by about 2/3, and the inference speed on the airborne TX2 has been increased from 3.7 FPS (frames per second) to 8 FPS.

Commodity object retrieval is a key issue in the application of self-service shopping and so on. In this paper, we propose a saliency object detection-based unsupervised commodity object retrieval scheme. Since most commodity objects are conspicuous and not complicated in commodity images, saliency detection could predict a saliency box that indicates approximate position information of objects. The proposed scheme utilizes the saliency box to filter the proposals extracted by selective search. The reserved proposals have a big overlapping ratio with saliency box to a large extent. This paper composes both the saliency box and the reserved proposals as saliency proposals. Furthermore, we propose a channel weighting generalized mean pooling feature to represent saliency proposals. On one hand, the reduction of proposals’ number after filtering significantly improves the computational efficiency; on the other hand, the new feature more accurately represents the objects to be retrieved, which results in higher retrieval precision. In addition, we built and manually annotated a commodity data set named PRODUCT to evaluate the proposed method. Extensive experiments are also conducted on the databases INSTRE and Flick32. The results demonstrate the superior performance of our scheme compared with the other state-of-the-art methods.

Accurate segmentation of entity categories is the critical step for 3D scene understanding. This paper presents a fast deep neural network model with Dense Conditional Random Field (DCRF) as a post-processing method, which can perform accurate semantic segmentation for 3D point cloud scene. On this basis, a compact but flexible framework is introduced for performing segmentation to the semantics of point clouds concurrently, contribute to more precise segmentation. Moreover, based on semantics labels, a novel DCRF model is elaborated to refine the result of segmentation. Besides, without any sacrifice to accuracy, we apply optimization to the original data of the point cloud, allowing the network to handle fewer data. In the experiment, our proposed method is conducted comprehensively through four evaluation indicators, proving the superiority of our method.

The development of the Internet has made social communication increasingly important for maintaining relationships between people. However, advertising and fraud are also growing incredibly fast and seriously affect our daily life, e.g., leading to money and time losses, trash information, and privacy problems. Therefore, it is very important to detect anomalies in social networks. However, existing anomaly detection methods cannot guarantee the correct rate. Besides, due to the lack of labeled data, we also cannot use the detection results directly. In other words, we still need human analysts in the loop to provide enough judgment for decision making. To help experts analyze and explore the results of anomaly detection in social networks more objectively and effectively, we propose a novel visualization system, egoDetect, which can detect the anomalies in social communication networks efficiently. Based on the unsupervised anomaly detection method, the system can detect the anomaly without training and get the overview quickly. Then we explore an ego’s topology and the relationship between egos and alters by designing a novel glyph based on the egocentric network. Besides, it also provides rich interactions for experts to quickly navigate to the interested users for further exploration. We use an actual call dataset provided by an operator to evaluate our system. The result proves that our proposed system is effective in the anomaly detection of social networks.

A fetoscope is an optical endoscope, which is often applied in fetoscopic laser photocoagulation to treat twin-to-twin transfusion syndrome. In an operation, the clinician needs to observe the abnormal placental vessels through the endoscope, so as to guide the operation. However, low-quality imaging and narrow field of view of the fetoscope increase the difficulty of the operation. Introducing an accurate placental vessel segmentation of fetoscopic images can assist the fetoscopic laser photocoagulation and help identify the abnormal vessels. This study proposes a method to solve the above problems. A novel encoder-decoder network with a dual-path structure is proposed to segment the placental vessels in fetoscopic images. In particular, we introduce a channel attention mechanism and a continuous convolution structure to obtain multi-scale features with their weights. Moreover, a switching connection is inserted between the corresponding blocks of the two paths to strengthen their relationship. According to the results of a set of blood vessel segmentation experiments conducted on a public fetoscopic image dataset, our method has achieved higher scores than the current mainstream segmentation methods, raising the dice similarity coefficient, intersection over union, and pixel accuracy by 5.80%, 8.39% and 0.62%, respectively. [ABSTRACT FROM AUTHOR]