Dual Distillation Discriminator Networks for Domain Adaptive Few-Shot Learning

Domain Adaptive Few-Shot Learning (DA-FSL) aims at accomplishing few-shot classification tasks on a novel domain with the aid of a large number of source-style samples and several target-style samples. It is essential for DA-FSL to transfer task knowledge from the source domain to the target domain and overcome the asymmetry amount of labeled data in both domains. To this end, we propose Dual Distillation Discriminator Networks (D3Net) from the perspective of the lack of labeled target domain style samples in DA-FSL. Specifically, we employ the idea of distillation discrimination to avoid the over-fitting caused by the unequal number of samples in the target and source domains, which trains the student discriminator by the soft labels from the teacher discriminator. Meanwhile, we design the task propagation stage and the mixed domain stage respectively from the level of feature space and instances to generate more target-style samples, which apply the task distributions and the sample diversity of the source domain to enhance the target domain. Our D3Net realizes the distribution alignment between the source domain and the target domain and constraints the FSL task distribution by prototype distributions on the mixed domain. Extensive experiments on three DA-FSL benchmark datasets, i.e., mini-ImageNet, tiered-ImageNet, and DomainNet, demonstrate that our D3Net achieves competitive performance.

Parents of preschoolers use multiple strategies to feed their children: Findings from an observational video pilot study

The current study leveraged observational data collection methods to fill gaps in our understanding of parent approach to feeding as well as child responses to various parental approaches. Specifically, the study aimed to: 1) characterize the broad range of food parenting practices used by parents of preschoolers during shared mealtimes at home, including differences by child gender, and 2) describe child responses to specific parent feeding practices. Forty parent-child dyads participated by recording two in-home shared meals. Meals were coded using a behavioral coding scheme that coded the occurrence of 11 distinct food parenting practices (e.g. indirect and direct commands, praise, bribes) and eight child responses (e.g., eat, refuse, cry/whine) to food parenting practices. Results revealed that parents engaged in a broad range of food parenting practices at meals. On average, parents in our sample used 10.51 (SD 7.83; Range 0-30) total food parenting practices per mealtime with a mean use of 3.38 (SD 1.67; Range 0-8) unique food parenting practices per mealtime. Use of indirect and direct commands to eat were most common; direct and indirect commands were used by 97.5% (n = 39) and 87.5% (n = 35) of parents at meals, respectively. No statistically significant differences were observed by child gender. No one specific feeding practice consistently yielded compliance or refusal to eat from the child, instead child responses were often mixed (e.g., compliance followed by refusal and/or refusal followed by compliance). However, use of praise to prompt eating was the practice that most often resulted in child compliance; 80.8% of children complied following parent's use of praise as a prompt to eat. Findings deepen our understanding of the types and frequency of food parenting practices used by parents of preschoolers during meals eaten at home and illuminate child responses to specific food parenting practices.

Measurement of Direct-Photon Cross Section and Double-Helicity Asymmetry at sqrt[s]=510 GeV in p[over →]+p[over →] Collisions

We present measurements of the cross section and double-helicity asymmetry A_{LL} of direct-photon production in p[over →]+p[over →] collisions at sqrt[s]=510  GeV. The measurements have been performed at midrapidity (|η|<0.25) with the PHENIX detector at the Relativistic Heavy Ion Collider. At relativistic energies, direct photons are dominantly produced from the initial quark-gluon hard scattering and do not interact via the strong force at leading order. Therefore, at sqrt[s]=510  GeV, where leading-order-effects dominate, these measurements provide clean and direct access to the gluon helicity in the polarized proton in the gluon-momentum-fraction range 0.02<x<0.08, with direct sensitivity to the sign of the gluon contribution.

[Clinical efficacy and safety analysis of 125I seed implantation in the treatment of mediastinal lymph node metastasis of lung cancer]

Objective: To investigate the clinical efficacy and safety of ¹²⁵I seed implantation in the treatment of mediastinal lymph node metastasis of lung cancer. Methods: Clinical data of 36 patients who underwent CT-guided ¹²⁵I seed implantation for mediastinal lymph node metastasis of lung cancer from August 2013 to April 2020 in three hospitals of the Northern radioactive particle implantation treatment collaboration group were retrospectively collected, including 24 males and 12 females, aged 46 to 84 years. Cox regression model was used to analyze the relationship between local control rate, survival rate and tumor stage, pathological type, postoperative D90, postoperative D100 and other variables, and to analyze the occurrence of complications. Results: The objective response rate of CT-guided ¹²⁵I seed implantation in the treatment of mediastinal lymph node metastasis of lung cancer was 75% (27/36), the median control time was 12 months, the 1-year local control rate was 47.2% (17/36), and the median survival time was 17 months. The 1-year and 2-year survival rates were 61.1% (22/36) and 22.2% (8/36) respectively. Univariate analysis showed that in the treatment of mediastinal lymph node metastasis with CT-guided ¹²⁵I implantation, factors related to local control included tumor stage (HR=5.246, 95%CI: 2.243-12.268, P<0.001), postoperative D90 (HR=0.191, 95%CI: 0.085-0.431, P<0.001), postoperative D100 (HR=0.240, 95%CI: 0.108-0.533, P<0.001); The factors affecting survival were tumor stage (HR=2.712, 95%CI: 1.356-5.425, P=0.005), postoperative D90 (HR=0.110, 95%CI: 0.041-0.294, P<0.001), postoperative D100 (HR=0.212, 95%CI: 0.092-0.489, P<0.001). Multivariate analysis showed that tumor stage (HR=5.305, 95%CI: 2.187-12.872, P<0.001) and postoperative D100 (HR=0.237, 95%CI: 0.099-0.568, P<0.001) were correlated with local control rate. Tumor stage (HR=2.347, 95%CI: 1.095-5.032, P=0.028) and postoperative D90 (HR=0.144, 95%CI: 0.051-0.410, P<0.001) were correlated with survival. In terms of complications, 9 of the 36 patients had pneumothorax, and 1 of them was cured by closed thoracic drainage for severe pneumothorax; 5 cases developed pulmonary hemorrhage and 5 cases developed hemoptysis, which recovered after hemostasis treatment. One case developed pulmonary infection and recovered after anti-inflammatory treatment. No radiation esophagitis and radiation pneumonia occurred; No grade 3 or higher complications occurred. Conclusion: ¹²⁵I seed implantation in the treatment of lung cancer mediastinal lymph node metastasis has a high local control rate and controllable adverse effects.

Knowledge Distillation Classifier Generation Network for Zero-Shot Learning

In this article, we present a conceptually simple but effective framework called knowledge distillation classifier generation network (KDCGN) for zero-shot learning (ZSL), where the learning agent requires recognizing unseen classes that have no visual data for training. Different from the existing generative approaches that synthesize visual features for unseen classifiers' learning, the proposed framework directly generates classifiers for unseen classes conditioned on the corresponding class-level semantics. To ensure the generated classifiers to be discriminative to the visual features, we borrow the knowledge distillation idea to both supervise the classifier generation and distill the knowledge with, respectively, the visual classifiers and soft targets trained from a traditional classification network. Under this framework, we develop two, respectively, strategies, i.e., class augmentation and semantics guidance, to facilitate the supervision process from the perspectives of improving visual classifiers. Specifically, the class augmentation strategy incorporates some additional categories to train the visual classifiers, which regularizes the visual classifier weights to be compact, under supervision of which the generated classifiers will be more discriminative. The semantics-guidance strategy encodes the class semantics into the visual classifiers, which would facilitate the supervision process by minimizing the differences between the generated and the real-visual classifiers. To evaluate the effectiveness of the proposed framework, we have conducted extensive experiments on five datasets in image classification, i.e., AwA1, AwA2, CUB, FLO, and APY. Experimental results show that the proposed approach performs best in the traditional ZSL task and achieves a significant performance improvement on four out of the five datasets in the generalized ZSL task.

Sb2Se3/CdS/ZnO photodetectors based on physical vapor deposition for color imaging applications

The reported antimony selenide (Sb₂Se₃) photodetectors (PDs) are still far away from color camera applications mainly due to the high operation temperature required in chemical vapor deposition (CVD) and the lack of high-density PD arrays. In this work, we propose a Sb₂Se₃/CdS/ZnO PD created by physical vapor deposition (PVD) operated at room temperature. Using PVD, a uniform film can be obtained, so the optimized PD has excellent photoelectric performance with high responsivity (250 mA/W), high detectivity (5.6 × 10¹² Jones), low dark current (∼10^-9 A), and short response time (rise: < 200 μs; decay: < 200 μs). With the help of advanced computational imaging technology, we successfully demonstrate color imaging applications by the single Sb₂Se₃ PD; thus, we expect this work can bring Sb₂Se₃ PDs in color camera sensors closer.

Automatic Classification Method of Arrhythmias Based on 12-Lead Electrocardiogram

Cardiovascular disease is one of the main causes of death worldwide. Arrhythmias are an important group of cardiovascular diseases. The standard 12-lead electrocardiogram signals are an important tool for diagnosing arrhythmias. Although 12-lead electrocardiogram signals provide more comprehensive arrhythmia information than single-lead electrocardiogram signals, it is difficult to effectively fuse information between different leads. In addition, most of the current researches working on automatic diagnosis of cardiac arrhythmias are based on modeling and analysis of single-mode features extracted from one-dimensional electrocardiogram sequences, ignoring the frequency domain features of electrocardiogram signals. Therefore, developing an automatic arrhythmia detection algorithm based on 12-lead electrocardiogram with high accuracy and strong generalization ability is still challenging. In this paper, a multimodal feature fusion model based on the mechanism is developed. This model utilizes a dual channel deep neural network to extract different dimensional features from one-dimensional and two-dimensional electrocardiogram time-frequency maps, and combines attention mechanism to effectively fuse the important features of 12-lead, thereby obtaining richer arrhythmia information and ultimately achieving accurate classification of nine types of arrhythmia signals. This study used electrocardiogram signals from a mixed dataset to train, validate, and evaluate the model, with an average of F1 score and average accuracy reached 0.85 and 0.97, respectively. Experimental results show that our algorithm has stable and reliable performance, so it is expected to have good practical application potential.

High-Performance Solar-Blind UV Phototransistors Based on ZnO/Ga2O3 Heterojunction Channels

High-performance phototransistor-based solar-blind (200-280 nm) ultraviolet (UV) photodetectors (PDs) are constructed with a low-cost thin-film ZnO/Ga₂O₃ heterojunction. The optimized PD shows high spectral selectivity (R254/R365 > 1 × 10³) with a photo-to-dark current ratio of ∼10⁴, a responsivity of 113 mA/W, a detectivity of 1.25 × 10¹² Jones, and a response speed of 41 ms under 254 nm UV light irradiation. It is found that the gate electrode of a three-terminal phototransistor can amplify the responsivity and increase the photo-to-dark current ratio because of the different densities of field-induced electrons at different gate biases. In addition, the built-in electric field at the ZnO/Ga₂O₃ heterojunction interface can control the distribution of the photoinduced electrons and the total conductivity of the heterojunction, which can further enhance device performance. Together with the simple fabrication process, the achieved results suggest that the three-terminal ZnO/Ga₂O₃ heterojunction phototransistor is a promising candidate for highly sensitive solar-blind PDs.

Semantic-Aware Dynamic Generation Networks for Few-Shot Human-Object Interaction Recognition

Recognizing human-object interaction (HOI) aims at inferring various relationships between actions and objects. Although great progress in HOI has been made, the long-tail problem and combinatorial explosion problem are still practical challenges. To this end, we formulate HOI as a few-shot task to tackle both challenges and design a novel dynamic generation method to address this task. The proposed approach is called semantic-aware dynamic generation networks (SADG-Nets). Specifically, SADG-Net first assigns semantic-aware task representations for different batches of data, which further generates dynamic parameters. It obtains the features that highlight intercategory discriminability and intracategory commonality adaptively. In addition, we also design a dual semantic-aware encoder module (DSAE-Module), that is, verb-aware and noun-aware branches, to yield both action and object prototypes of HOI for each task space, which generalizes to novel combinations by transferring similarities among interactions. Extensive experimental results on two benchmark datasets, that is, humans interacting with common objects (HICO)-FS and trento universal HOI (TUHOI)-FS, illustrate that our SADG-Net achieves superior performance over state-of-the-art approaches, which proves its impressive effectiveness on few-shot HOI recognition.

Perovskite-based color camera inspired by human visual cells

There are two primary types of photoreceptor cells in the human eye: cone cells and rod cells that enable color vision and night vision, respectively. Herein, inspired by the function of human visual cells, we develop a high-resolution perovskite-based color camera using a set of narrowband red, green, blue, and broadband white perovskite photodetectors as imaging sensors. The narrowband red, green, and blue perovskite photodetectors with color perceptions mimic long-, medium-, and short-wavelength cones cells to achieve color imaging ability. Also, the broadband white perovskite photodetector with better detectivity mimics rod cells to improve weak-light imaging ability. Our perovskite-based camera, combined with predesigned pattern illumination and image reconstruction technology, is demonstrated with high-resolution color images (up to 256 × 256 pixels) in diffuse mode. This is far beyond previously reported advanced perovskite array image sensors that only work in monochrome transmission mode. This work shows a new approach to bio-inspired cameras and their great potential to strongly mimic the ability of the natural eye.

[Validation of calculation method for dose distribution around radioactive iodine-125 particles based on AAPM TG43 report]

Objective: According to the formula provided by the TG43 report [AAPM TG43 (2004)] proposed by the American Association of Physicists in Medicine (AAPM) in 2004, we calculated the dose distribution around the radioactive iodine-125 particles, and verified the calculation accuracy of the radioactive iodine-125 particles treatment planning system. Methods: AAPM TG43 (2004) report provides two calculation methods when calculating the dose around a single radioactive source. The calculation method that does not consider the geometric structure of the radioactive source is called point source calculation method, and the calculation method that considers the geometric structure of the radioactive source is called line source calculation method. Assuming a single Amersham 6711 radioactive iodine-125 particle with an activity of 100 U, the following point doses were calculated according to the two calculation methods provided by AAPM TG43 (2004) report, at 0°, 90° directions, distances 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5 and 6 cm; In the direction of 45°, the doses at 0.71, 1.41, 2.12, 2.83, 3.54, 4.24, 4.95, 5.66, 6.36, 7.07, 7.78 and 8.49 cm. On the clinically used brachytherapy planning system variseeds 8.0, the above two calculation methods are used to calculate the corresponding activity and the dose around the corresponding type of radioactive iodine-125 particles, and the function of capturing points to templates built in the planning system is used to accurately find the above corresponding point position, using a single measurement of the above corresponding point dose; and comparation of the results were performed to see if there is a statistical difference. Results: The AAPM TG43 report uses point source calculation method to calculate the dose of single Amersham 6711 radioactive iodine-125 particles with activity of 100 U at 0° and 90° directions. The points with the same distance and the same dose are 8 082.18, 1 870.08, 756.58, 381.47, 217.11, 131.91, 86.55, 58.32, 39.97, 27.42, 19.74, 14.13 Gy, respectively, at 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5 and 6 cm away from them. In the 45° direction, the doses at the distances of 0.71, 1.41, 2.12, 2.83, 3.54, 4.24, 4.95, 5.66, 6.36, 7.07, 7.78 and 8.49 cm are 3 957.37, 865.83, 329.99, 155.69, 84.10, 48.50, 28.49, 17.80, 11.37, 7.38, 4.98 and 3.39 Gy, respectively; For line source calculation method, radioactive particles are at the same distance as above. The doses at each point in the direction of 0° are 3 128.71, 755.44, 330.30, 180.53, 107.74, 68.56, 46.40, 32.22, 22.70, 16.00, 11.51, 8.24 Gy, respectively. The doses at each point in the direction of 90° are 8 306.46, 1 981.01, 802.74, 405.38, 230.60, 140.03, 91.83, 61.84, 42.36, 29.05, 20.91, 14.97 Gy; In the 45° direction, the dose at the corresponding distance as above is 4 020.78, 877.43, 333.49, 156.93, 84.69, 48.81, 28.65, 17.89, 11.42, 7.41, 4.99 and 3.40 Gy, respectively. The maximum dose difference (0.3%) between the two methods is 7.78 cm in the 45° direction, the maximum difference (-0.3%) between the two methods is 8.49 cm in the 45° direction, and the value of other sampling points is less than 0.3%. The closer the Amersham 6711 iodine-125 particles are to the source in the directions of 0°, 45°, and 90°, the faster the dose will drop, and the dose will drop gradually as the distance increases. Conclusion: The brachytherapy planning system variseeds 8.0 and the AAPM TG43 report calculate a maximum dose difference of 0.3%, which can accurately calculate the dose distribution around radioactive iodine-125 seeds, and provide a reliable tool for the clinical implementation of radioactive iodine-125 particles implantation for tumor treatment.

Memorizing Complementation Network for Few-Shot Class-Incremental Learning

Few-shot Class-Incremental Learning (FSCIL) aims at learning new concepts continually with only a few samples, which is prone to suffer the catastrophic forgetting and overfitting problems. The inaccessibility of old classes and the scarcity of the novel samples make it formidable to realize the trade-off between retaining old knowledge and learning novel concepts. Inspired by that different models memorize different knowledge when learning novel concepts, we propose a Memorizing Complementation Network (MCNet) to ensemble multiple models that complements the different memorized knowledge with each other in novel tasks. Additionally, to update the model with few novel samples, we develop a Prototype Smoothing Hard-mining Triplet (PSHT) loss to push the novel samples away from not only each other in current task but also the old distribution. Extensive experiments on three benchmark datasets, e.g., CIFAR100, miniImageNet and CUB200, have demonstrated the superiority of our proposed method.

Complementary Calibration: Boosting General Continual Learning with Collaborative Distillation and Self-Supervision

General Continual Learning (GCL) aims at learning from non independent and identically distributed stream data without catastrophic forgetting of the old tasks that don't rely on task boundaries during both training and testing stages. We reveal that the relation and feature deviations are crucial problems for catastrophic forgetting, in which relation deviation refers to the deficiency of the relationship among all classes in knowledge distillation, and feature deviation refers to indiscriminative feature representations. To this end, we propose a Complementary Calibration (CoCa) framework by mining the complementary model's outputs and features to alleviate the two deviations in the process of GCL. Specifically, we propose a new collaborative distillation approach for addressing the relation deviation. It distills model's outputs by utilizing ensemble dark knowledge of new model's outputs and reserved outputs, which maintains the performance of old tasks as well as balancing the relationship among all classes. Furthermore, we explore a collaborative self-supervision idea to leverage pretext tasks and supervised contrastive learning for addressing the feature deviation problem by learning complete and discriminative features for all classes. Extensive experiments on six popular datasets show that our CoCa framework achieves superior performance against state-of-the-art methods. Code is available at https://github.com/lijincm/CoCa.

Mosaic-free compound eye camera based on multidirectional photodetectors and single-pixel imaging

Compound-eye wide field-of-view (FOV) imaging generally faces the disadvantages of a complex system, low resolution, and complicated image mosaic. Single-pixel imaging has proven to very beneficial in building a high-resolution and simple wide-FOV camera, but its ability to overcome the problem of image mosaics still needs to be demonstrated. In this Letter, we propose a novel, to the best of our knowledge, kind of artificial compound eye based on multidirectional photodetectors (PDs) and demonstrate theoretically and experimentally that mosaics are unnecessary in multidirectional PD-based single-pixel imaging. In addition, we show experimentally that only nine multidirectional PDs are needed to obtain wide-angle images in a hemisphere to realize wide-FOV mosaic-free imaging. This work greatly simplifies the concept of compound-eye cameras and is very enlightening for detector design in wide-FOV single-pixel imaging, plausibly leading to the development of single-pixel endoscopic imaging.

Endovascular treatment of acute ischemic stroke with a fully radiopaque retriever: A randomized controlled trial

Objective

The Neurohawk retriever is a new fully radiopaque retriever. A randomized controlled non-inferiority trial was conducted to compare the Neurohawk and the Solitaire FR in terms of safety and efficacy. In order to evaluate the efficacy and safety of endovascular treatment in acute ischemic stroke (AIS) caused by intracranial atherosclerotic disease (ICAD) larger vessel occlusion (LVO), a sub-analysis was performed.

Methods

Acute ischemic stroke patients aged 18-80 years with LVO in the anterior circulation were randomly assigned to undergo thrombectomy with either the Neurohawk or the Solitaire FR. The primary efficacy endpoint was successful reperfusion (mTICI 2b-3) rate by the allocated retriever. A relevant non-inferiority margin was 12.5%. Safety outcomes were symptomatic intracranial hemorrhage (sICH) and all-cause mortality within 90 days. Secondary endpoints included first-pass effect (FPE), modified FPE, and favorable outcomes at 90 days. In subgroup analysis, the patients were divided into the ICAD group and non-ICAD group according to etiology, and baseline characteristics, angiographic, and clinical outcomes were compared.

Results

A total of 232 patients were involved in this analysis (115 patients in the Neurohawk group and 117 in the Solitaire group). The rates of successful reperfusion with the allocated retriever were 88.70% in the Neurohawk group and 90.60% in the Solitaire group (95%CI of the difference, -9.74% to 5.94%; p = 0.867). There were similar results in FPE and mFPE in both groups. The rate of sICH seemed higher in the Solitaire group (13.16% vs. 7.02%, p = 0.124). All-cause mortality and favorable outcome rates were comparable as well. In subgroup analysis, 58 patients were assigned to the ICAD group and the remaining 174 to the non-ICAD group. The final successful reperfusion and favorable outcome rates showed no statistically significant differences in two groups. Mortality within 90 days was relatively lower in the ICAD group (6.90% vs. 17.24%; p = 0.054).

Conclusion

The Neurohawk retriever is non-inferior to the Solitaire FR in the mechanical thrombectomy of large vessel occlusion-acute ischemic stroke (LVO-AIS). The sub-analysis suggested that endovascular treatment including thrombectomy with the retriever and essential rescue angioplasty is effective and safe in AIS patients with intracranial atherosclerotic disease-larger vessel occlusion (ICAD-LVO).

Clinical trial registration

https://clinicaltrials.gov/ct2/show/NCT04995757, number: NCT04995757.

Mechanical thrombectomy combined with stenting for radiation-induced carotid stenosis-related stroke with high-load embolization: A case report

Radiation therapy in patients with nasopharyngeal carcinoma can cause chronic progressive carotid artery injury, but acute ischemic stroke caused by carotid artery high-load thrombosis rarely occurs in patients with tandem lesions. We performed carotid mechanical thrombectomy combined with angioplasty in a 57-year-old man who received radiotherapy for nasopharyngeal carcinoma more than 10 years before presentation. He presented with acute-onset left hemiplegia, confusion, and mixed aphasia. Head CT revealed a hyper-dense sign in the right middle cerebral artery M1 region, and angiography disclosed occlusion in the right internal carotid artery C5 region with extremely severe stenosis in the middle C1 region. Intra-arterial mechanical thrombectomy with carotid stenting was performed, and re-canalization was achieved. Re-examination angiography after 3 months revealed worsening of ulcerative plaques and pseudoaneurysms in the left common carotid artery. Consequently, we performed carotid stenting over the left common carotid artery, and the patient recovered well postoperatively. Our experience suggests that early detection of large blood vessel damage and intervention are necessary to prevent large-vessel ischemic stroke in patients who received radiotherapy for nasopharyngeal carcinoma.

Perovskite Wide-Angle Field-Of-View Camera

Researchers have attempted to create wide-angle field-of-view (FOV) cameras inspired by the structure of the eyes of animals, including fisheye and compound eye cameras. However, realizing wide-angle FOV cameras simultaneously exhibiting low distortion and high spatial resolution remains a significant challenge. In this study, a novel wide-angle FOV camera is developed by combining a single large-area flexible perovskite photodetector (FP-PD) using computational technology. With this camera, the proposed single-photodetector imaging technique can obtain high-spatial-resolution images using only a single detector, and the large-area FP-PD can be bent further to collect light from a wide-angle FOV. The proposed camera demonstrates remarkable features of an extraordinarily tunable wide FOV (greater than 150°), high spatial resolution of 256 × 256 pixels, and low distortion. It is believed that the proposed compatible and extensible camera prototype will promote the development of high-performance versatile FOV cameras.