High Labeling Research Articles

A novel semi-supervised fault diagnosis method for chillers based on neighbor-optimized graph convolutional network

The fault diagnosis of chillers is of great significance in reducing building energy consumption and extending the operational lifespan of refrigeration equipment. Several popular machine learning-based fault diagnosis methods rely heavily on many labeled samples. However, such samples are difficult to obtain in practice due to sparse fault data and high labeling costs. This limits the application of ML-based fault diagnosis methods based on supervised learning. To reduce the dependence on labeled samples, this paper proposes a novel chiller fault diagnosis method based on neighbor-optimized graph convolutional network. The method improves the utilization efficiency of unlabeled samples by mining the spatio-temporal relationship between a large number of unlabeled samples and a limited number of labeled samples. And it dynamically adjusts the graph's structure by optimizing the number of adjacent samples in the correlation graph to obtain better diagnostic results. Its effectiveness is validated on the authoritative dataset ASHRAE RP-1043 and a more challenging dataset of real-world chillers in a building. Experimental results show that the proposed method can achieve better diagnostic performance than the state-of-the-art methods.

Semi-supervised ensemble fault diagnosis method based on adversarial decoupled auto-encoder with extremely limited labels

Intelligent fault diagnosis can enhance the reliability of mechanical equipment. However, only a few labels are available in a large amount of fault data due to high labeling costs in practical engineering. The fault recognition capability of existing semi-supervised diagnosis methods is severely insufficient with limited labels, especially with extremely limited labels that only a single labeled sample available per fault type. To address this issue, a novel semi-supervised ensemble fault diagnosis framework termed ADAE-LFDM is proposed based on adversarial decoupled auto-encoder (ADAE) and low-dimensional feature distance metric (LFDM). Firstly, the locally selective combination in parallel outlier ensembles (LSCP) method is introduced to efficiently separate normal and fault samples. Subsequently, an ADAE with branching structure and latent space feature regularization strategy is proposed to decouple and capture the fault feature. Finally, a LFDM strategy that contains feature dimensionality reduction, and centroid-based metric is performed to achieve high-accuracy fault diagnosis. Experimental results based on two rotating machinery datasets have demonstrated that the proposed method achieves a diagnostic accuracy of over 97 % when there is only a single labeled sample available per fault type, and an average diagnostic accuracy of 85 % under cross-operating condition, showing the superiority compared to other methods.

Human Embryo Live Imaging Reveals Nuclear DNA Shedding During Blastocyst Expansion and Biopsy

ABSTRACT The proportion of babies born from in vitro fertilization (IVF) is rising at an exponential rate, highlighting the importance that we have a comprehensive understanding of human preimplantation development and determinants of embryo quality. The early divisions of the embryo after fertilization but before implantation and resulting mitotic errors have been studied primarily in the mouse embryo; however, thus far we have lacked the technology to characterize these critical steps in humans. Establishing an approach that can bypass genetic manipulation and microinjections of DNA or mRNA into human embryos would allow us to better uncover the processes patterning preimplantation human development. This study aimed to evaluate human blastocyst preimplantation development with noninvasive imaging using membrane permeable fluorescent dyes. First, the dyes SPY650-DNA, which labels genomic DNA, and SPY555-actin, which labels F-actin, were validated in the mouse embryo where they produced high contrast labelling with similar results to microinjection of fluorescent mRNAs and produced offspring at a similar rate to nondyed control embryos. Live imaging data using these dyes enabled 3-dimensional scans of the embryo at 5- to 10-minute intervals of the following central events during preimplantation development: the main phases of mitosis, visualizing the major changes in cell shape that characterize embryo compaction at the 8-cell stage, detecting cell polarization at the 8-cell stage, establishing apical F-actin rings at the 16-cell stage, tracking the expansion and zippering of F-actin rings, detecting the first internalized cells within the 16-cell embryo, and visualizing blastocyst expansion and hatching from the zona pellucida. Cleavage-stage human IVF embryos were then studied using the fluorescent dyes to track early cellular and morphogenetic processes. Early cell-cycle dynamics and mitotic stages revealed that the duration of human mitosis is similar to that of mice, but interphase is 27% ± 4% longer in humans (16.1 ± 0.9 vs 12.7 ± 0.4 hours; P < 0.01). Using live imaging, compaction dynamics such as increased cell-cell contact and angle between apical membranes with a decrease in cell sphericity were observed beginning at the 12-cell stage and differed significantly from development in mice. In comparison to mice, human compaction was found to be more asynchronous and did not have clear links to apical polarization or inner-outer segregation. Next, chromosomal segregation was analyzed to determine if this approach allowed detection of segregation errors in the human embryo leading to aneuploidy. Lagging chromosomes detected in human embryos using SPY-DNA appeared morphologically similar to those found in mouse embryos and had similar segregation dynamics. During blastocyst expansion, a subset of trophectoderm cell nuclei forms protruding bud-like that is shed into cytoplasm producing cytoplasmic DNA structures (cytDNA), suggesting an additional process of DNA loss different from chromosome segregation errors during mitosis. A perinuclear keratin network developing at the early cavitation stage appears to protect from formation of cytDNA structures, and the mechanical stress of blastocyst expansion appeared to disrupt the keratin network in certain cells, increasing DNA shedding. Next, trophectoderm biopsy was conducted in mouse, and human embryos to determine if the mechanical stress would increase DNA shedding. Biopsy of human blastocysts resulted in a significant increase in nuclear budding (6.0% vs 0.70%; P = 0.0022), suggesting a link between mechanical stress and DNA shedding. This study demonstrates the use of fluorescent dyes and live imaging in characterization of key cellular and morphogenetic processes patterning the preimplantation human embryo and describes a process of DNA shedding in response to mechanical stressors that warrants future research.

Federated Semi-Supervised Learning for Medical Image Segmentation via Pseudo-Label Denoising.

Distributed big data and digital healthcare technologies have great potential to promote medical services, but challenges arise when it comes to learning predictive model from diverse and complex e-health datasets. Federated Learning (FL), as a collaborative machine learning technique, aims to address the challenges by learning a joint predictive model across multi-site clients, especially for distributed medical institutions or hospitals. However, most existing FL methods assume that clients possess fully labeled data for training, which is often not the case in e-health datasets due to high labeling costs or expertise requirement. Therefore, this work proposes a novel and feasible approach to learn a Federated Semi-Supervised Learning (FSSL) model from distributed medical image domains, where a federated pseudo-labeling strategy for unlabeled clients is developed based on the embedded knowledge learned from labeled clients. This greatly mitigates the annotation deficiency at unlabeled clients and leads to a cost-effective and efficient medical image analysis tool. We demonstrated the effectiveness of our method by achieving significant improvements compared to the state-of-the-art in both fundus image and prostate MRI segmentation tasks, resulting in the highest Dice scores of 89.23% and 91.95% respectively even with only a few labeled clients participating in model training. This reveals the superiority of our method for practical deployment, ultimately facilitating the wider use of FL in healthcare and leading to better patient outcomes.

Estimating the Uncertainty in Emotion Class Labels With Utterance-Specific Dirichlet Priors

Emotion recognition is a key attribute for artificial intelligence systems that need to naturally interact with humans. However, the task definition is still an open problem due to the inherent ambiguity of emotions. In this paper, a novel Bayesian training loss based on per-utterance Dirichlet prior distributions is proposed for verbal emotion recognition, which models the uncertainty in one-hot labels created when human annotators assign the same utterance to different emotion classes. An additional metric is used to evaluate the performance by detection test utterances with high labelling uncertainty. This removes a major limitation that emotion classification systems only consider utterances with labels where the majority of annotators agree on the emotion class. Furthermore, a frequentist approach is studied to leverage the continuous-valued "soft" labels obtained by averaging the one-hot labels. We propose a two-branch model structure for emotion classification on a per-utterance basis, which achieves state-of-the-art classification results on the widely used IEMOCAP dataset. Based on this, uncertainty estimation experiments were performed. The best performance in terms of the area under the precision-recall curve when detecting utterances with high uncertainty was achieved by interpolating the Bayesian training loss with the Kullback-Leibler divergence training loss for the soft labels. The generality of the proposed approach was verified using the MSP-Podcast dataset which yielded the same pattern of results.

Preparation and evaluation of 111In-labeled liposomes containing phosphatidylglycerol for detection of macrophages in atherosclerotic plaques

Macrophage infiltration is a characteristic feature of atherosclerotic plaque progression. Since liposomes containing 1,2-distearoyl-sn-glycero-3-phosphoglycerol (DSPG) are efficiently phagocytosed by macrophages, we deduced that radiolabeled liposomes containing DSPG could potentially be used for nuclear imaging of vulnerable atherosclerotic plaques. Indium-111 (111In)-labeled liposomes containing different ratios of DSPG were developed with a high labeling efficiency. 111In-labeled liposomes with higher DSPG content showed higher uptake by macrophage-like RAW264 cells. A biodistribution study demonstrated rapid blood clearance and selective accumulation in the liver and spleen, especially in normal mice injected with 111In-labeled liposomes with higher DSPG content. Accumulation in atherosclerotic plaques was evaluated using 111In-labeled DSPG liposomes, which had the highest DSPG content among the studied liposomes. 111In-labeled DSPG liposomes accumulated in the plaques and the radioactive regions were mostly consistent with the distribution of macrophages. The target-to-non-target ratio of 111In-labeled DSPG liposomes was higher than that of 111In-labeled control liposomes without DSPG. These results suggest that 111In-labeled liposomes containing DSPG are useful for nuclear medical diagnosis of atherosclerotic plaques.

Boosting SAR Aircraft Detection Performance with Multi-Stage Domain Adaptation Training

Deep learning has achieved significant success in various synthetic aperture radar (SAR) imagery interpretation tasks. However, automatic aircraft detection is still challenging due to the high labeling cost and limited data quantity. To address this issue, we propose a multi-stage domain adaptation training framework to efficiently transfer the knowledge from optical imagery and boost SAR aircraft detection performance. To overcome the significant domain discrepancy between optical and SAR images, the training process can be divided into three stages: image translation, domain adaptive pretraining, and domain adaptive finetuning. First, CycleGAN is used to translate optical images into SAR-style images and reduce global-level image divergence. Next, we propose multilayer feature alignment to further reduce the local-level feature distribution distance. By applying domain adversarial learning in both the pretrain and finetune stages, the detector can learn to extract domain-invariant features that are beneficial to the learning of generic aircraft characteristics. To evaluate the proposed method, extensive experiments were conducted on a self-built SAR aircraft detection dataset. The results indicate that by using the proposed training framework, the average precision of Faster RCNN gained an increase of 2.4, and that of YOLOv3 was improved by 2.6, which outperformed other domain adaptation methods. By reducing the domain discrepancy between optical and SAR in three progressive stages, the proposed method can effectively mitigate the domain shift, thereby enhancing the efficiency of knowledge transfer. It greatly improves the detection performance of aircraft and offers an effective approach to address the limited training data problem of SAR aircraft detection.

Radiolabeled Human Serum Albumin Nanoparticles Co-Loaded with Methotrexate and Decorated with Trastuzumab for Breast Cancer Diagnosis.

Breast cancer is a leading cause of cancer-related mortality among women worldwide, with millions of new cases diagnosed yearly. Addressing the burden of breast cancer mortality requires a comprehensive approach involving early detection, accurate diagnosis, effective treatment, and equitable access to healthcare services. In this direction, nano-radiopharmaceuticals have shown potential for enhancing breast cancer diagnosis by combining the benefits of nanoparticles and radiopharmaceutical agents. These nanoscale formulations can provide improved imaging capabilities, increased targeting specificity, and enhanced sensitivity for detecting breast cancer lesions. In this study, we developed and evaluated a novel nano-radio radiopharmaceutical, technetium-99m ([99mTc]Tc)-labeled trastuzumab (TRZ)-decorated methotrexate (MTX)-loaded human serum albumin (HSA) nanoparticles ([99mTc]-TRZ-MTX-HSA), for the diagnosis of breast cancer. In this context, HSA and MTX-HSA nanoparticles were prepared. Conjugation of MTX-HSA nanoparticles with TRZ was performed using adsorption and covalent bonding methods. The prepared formulations were evaluated for particle size, PDI value, zeta (ζ) potential, scanning electron microscopy analysis, encapsulation efficiency, and loading capacity and cytotoxicity on MCF-7, 4T1, and MCF-10A cells. Finally, the nanoparticles were radiolabeled with [99mTc]Tc using the direct radiolabeling method, and cellular uptake was performed with the nano-radiopharmaceutical. The results showed the formation of spherical nanoparticles, with a particle size of 224.1 ± 2.46 nm, a PDI value of 0.09 ± 0.07, and a ζ potential value of -16.4 ± 0.53 mV. The encapsulation efficiency of MTX was found to be 32.46 ± 1.12%, and the amount of TRZ was 80.26 ± 1.96%. The labeling with [99mTc]Tc showed a high labeling efficiency (>99%). The cytotoxicity studies showed no effect, and the cellular uptake studies showed 97.54 ± 2.16% uptake in MCF-7 cells at the 120th min and were found to have a 3-fold higher uptake in cancer cells than in healthy cells. In conclusion, [99mTc]Tc-TRZ-MTX-HSA nanoparticles are promising for diagnosing breast cancer and evaluating the response to treatment in breast cancer patients.

Cyanine lipids promote the shedding of extracellular vesicles from cell membranes

<img src=” https://s3.amazonaws.com/production.scholastica/article/88009/medium/prnano_1052023ga.jpg?1694809016”> Extracellular vesicles (EV) have garnered significant attention in the fields of drug delivery, imaging, and immunotherapy. There is a need in methods to enhance release of EVs from cells. We found that at high labeling concentrations (100µM), indocarbocyanine lipids DiD and DiR that are commonly used for labeling cells, nanoparticles and EVs, promoted shedding of cell membrane tetraspanins with concomitant release of EVs in the medium. To further investigate this phenomenon, we screened a library of lipids and liposomal formulations for the release of membrane marker CD63 from THP-1 cells, and membrane red nanolantern (RNL) from 4T1 cells. We found a strong dependency of the EV release on lipid structure. In general, lipids that had a cyanine headgroup were more efficient than PEGylated phospholipids, neutral and cationic liposomes, with some lipids enhancing the release of CD63 up to 4-fold, and of RNL up to 8-fold, over vehicle treated control. A side-by-side comparison of cyanine lipid derivatives and corresponding precursor lipids confirmed that the cyanine headgroup significantly promoted shedding of RNL. Mutation of an exosome biogenesis regulator UNC13D did not hinder the release. Lipid-released EV could be modified with anti-interleukin 13 receptor alpha 2 antibody and targeted to glioma cells, suggesting potential utility in drug delivery. Furthermore, the impact of extraneously added lipids on cell membrane integrity should be carefully considered in cell labeling and drug delivery applications.

Synthesis, Preclinical Evaluation, and First-in-Human PET Study of [68Ga]-Labeled Biphenyl-Containing PSMA Tracers.

Radioisotope-labeled prostate-specific membrane antigen (PSMA) PET tracers have gained popularity in diagnosing prostate cancer (PCa). This study aimed to improve the affinity and tumor-targeting capabilities of new PSMA tracers by increasing the number of pharmacophores that specifically bind to PSMA. Using biphenyl as a core scaffold, we investigated the relationship among spacer segments, affinity, and pharmacokinetic properties. In preclinical PET studies on mice with 22Rv1 tumors, compared with [68Ga]Ga-PSMA-11 (SUVmax = 3.37), [68Ga]Ga-PSMA-D5 (Ki = 0.15) showed higher tumor uptake (SUVmax = 3.51) and lower renal uptake (T/K = 1.84). In the first-in-human study, [68Ga]Ga-PSMA-D5 effectively detected small PCa-associated lesions and distant metastases. The advantages of [68Ga]Ga-PSMA-D5 include high tumor uptake, straightforward synthesis, and labeling, making it a promising PSMA PET tracer. Furthermore, [68Ga]Ga-PSMA-D5 contains a DOTA chelator, allowing convenient labeling with therapeutic radionuclides such as 177Lu and 225Ac, providing the potential for targeted radioligand therapy in PCa.

Dual-Targeting PET Tracers Enable Enzyme-Mediated Self-Assembly for the PET Imaging of Legumain Activity.

Legumain, a lysosomal cysteine protease overexpressed in a variety of tumors, has been considered a promising biomarker for various cancers. Precise detection of legumain activity in the lysosome represents an important strategy for early diagnosis and prognosis of tumors. Small-molecule probes with the property of target-enabled self-assembly hold great potential for molecular imaging. In this study, we reported two dual-targeting radiotracers ([18F]SF-AAN-M and [18F]SF-AAN-HEM) with a property of legumain-mediated self-assembly for positron emission tomography (PET) imaging. Both the radiotracers were synthesized with high labeling yield (>50%) and the radiochemical purity was over 99% via one-step straightforward 18F-labeling. Both tracers were efficiently activated by the reducing agent and legumain to self-assemble into aggregates and showed enhanced retention in legumain-overexpressed MDA-MB-468 cells and tumors, indicating that the introduction of lysosome-targeting morpholine increased the tumor uptake and extended the retention of radiotracers in legumain-overexpressed tumors. In addition, [18F]SF-AAN-HEM with a hydrophilic (histidine-glutamate)3 tag displayed significantly reduced liver uptake with no conspicuous reduction in tumor uptake, affording high signal-to-noise ratios (tumor/liver and tumor/muscle). All of these results suggest that dual-targeting tracer [18F]SF-AAN-HEM could provide a promising tool for in vivo monitoring legumain activity in tumors.

Fibroblast growth factor 5 expression predicts the progression of oral squamous cell carcinoma

Fibroblast growth factor (FGF) 5 is a member of the FGF family that functions as a regulator of tissue growth and regeneration. Aberrant FGF5 expression has been previously associated with the progression of a number of different malignancies. However, its potential role in oral cancer remains unclear. In this study, we explored the relationship between the expression of FGF5 protein in oral squamous cell carcinomas (OSCCs) and the clinicopathological parameters of OSCCs and whether the expression of FGF5 protein in OSCCs could be a prognostic factor for OSCC patients. The FGF5 protein expression was examined in 64 OSCC and 34 normal oral mucosal specimens by immunohistochemical staining. Stress induced upregulation and intracellular redistribution of FGF5 were verified using xenograft animal model and OSCC cell lines. The mean FGF5 protein labelling index was significantly higher in OSCC than in normal oral mucosal samples, with high FGF5 protein labelling index (>58%) being correlated with advanced stage and poor survival of OSCC patients. Apart from the peri-cytoplasmic staining pattern characteristic of paracrine growth factors, FGF5 protein was localized as distinct punctate structures in the cytoplasm of advanced stage or stressed-induced cells. This redistribution and upregulation of FGF5 protein could be sustained after termination of the stress induction in cell line and xenograft animal models. FGF5 can be induced by cellular stress and risk factors of OSCC, where high expression levels of FGF5 is potentially a useful parameter for predicting OSCC progression and patient survival.

BlockDetective: A GCN‐based student–teacher framework for blockchain anomaly detection

AbstractThe anonymous and tamper‐proof nature of the blockchain poses significant challenges in auditing and regulating the behaviour and data on the chain. Criminal activities and anomalies are frequently changing, and fraudsters are devising new ways to evade detection. Moreover, the high volume and complexity of transactions and asymmetric errors make data classification more challenging. Also, class imbalances and high labelling costs are hindering the development of effective algorithms. In response to these issues, the authors present BlockDetective, a novel framework based on GCN that utilizes student–teacher architecture to detect fraudulent cryptocurrency transactions that are related to money laundering. The authors’ method leverages pre‐training and fine‐tuning, allowing the pre‐trained model (teacher) to adapt better to the new data distribution and enhance the prediction performance while teaching a new, light‐weight model (student) that provides abstract and top‐level information. The authors’ experimental results show that BlockDetective outperforms state‐of‐the‐art research methods by achieving top‐notch performance in detecting fraudulent transactions on the blockchain. This framework can assist regulators and auditors in detecting and preventing fraudulent activities on the blockchain, thereby promoting a more secure and transparent financial system.

Clinicopathological significance of TUBB3 in upper tract urothelial carcinoma and possible application in urine cytology.

βIII-Tubulin, encoded by the TUBB3 gene, is a microtubule protein. We previously reported that TUBB3 is overexpressed in renal cell carcinoma. We investigated the clinicopathological significance of TUBB3 in upper tract urothelial carcinoma (UTUC) by immunohistochemistry. In normal tissue, TUBB3 expression was weak or absent. In contrast, TUBB3 overexpression was observed in urothelial carcinoma (UC) tissues in 51 (49%) of 103 UTUC cases. TUBB3 overexpression was associated with nodular/flat morphology, high-grade disease, high T stage, and a poor prognosis. Similar results were obtained in The Cancer Genome Atlas bladder cancer cohort. TUBB3 expression was also associated with high Ki-67 labeling index, CD44v9, HER2, EGFR, and p53 expression in UTUC. Among representative cancer-related molecules, TUBB3 was an independent predictor of progression-free survival and high-grade UC. Finally, using urine cytology samples, we analyzed TUBB3 expression by immunocytochemistry. TUBB3 expression was more frequently found in UC cells than in nonneoplastic cells. The diagnostic accuracy of urine cytology was improved when combined with TUBB3 immunostaining. The findings suggest the importance of TUBB3 in tumor progression and its potential application as a biomarker for high-grade disease and the prognosis of UC. Moreover, combination with TUBB3 immunostaining might improve the diagnostic accuracy of urine cytology.

Three-dimensional multi-color optical nanoscopy at sub-10-nm resolution based on small-molecule organic probes

Achieving nanometer-scale resolution remains challenging in expansion microscopy due to photon loss. To address this concern, here we develop a multi-color expansion stimulated emission depletion technique based on small-molecule probes to realize high labeling density and intensity. Our method substantially lowers the barrier to visualizing diverse intracellular proteins and their interactions in three dimensions. It enables us to achieve sub-10-nm resolution in structures such as microfilaments, lysosomes, and mitochondria, providing new insights into cell biology.

Observing Extracellular Vesicles Originating from Endothelial Cells in Vivo Demonstrates Improved Astrocyte Function Following Ischemic Stroke via Aggregation-Induced Emission Luminogens.

Extracellular vesicles (EVs) obtained from endothelial cells (ECs) have significant therapeutic potential in the clinical management of individuals with ischemic stroke (IS) because they effectively treat ischemic stroke in animal models. However, because molecular probes with both high labeling efficiency and tracer stability are lacking, monitoring the actions of EC-EVs in the brain remains difficult. The specific intracellular targets in the brain that EC-EVs act on to produce their protective effects are still unknown, greatly impeding their use in clinical settings. For this research, we created a probe that possessed aggregation-induced emission (AIE) traits (namely, TTCP), enabling the effective labeling of EC-EVs while preserving their physiological properties. In vitro, TTCP simultaneously had a higher EC-EV labeling efficiency and better tracer stability than the commercial EV tags PKH-67 and DiI. In vivo, TTCP precisely tracked the actions of EC-EVs in a mouse IS model without influencing their protective effects. Furthermore, through the utilization of TTCP, it was determined that astrocytes were the specific cells affected by EC-EVs and that EC-EVs exhibited a safeguarding impact on astrocytes following cerebral ischemia-reperfusion (I/R) injury. These protective effects encompassed the reduction of the inflammatory reaction and apoptosis as well as the enhancement of cell proliferation. Further analysis showed that miRNA-155-5p carried by EC-EVs is responsible for these protective effects via regulation of the c-Fos/AP-1 pathway; this information provided a strategy for IS therapy. In conclusion, TTCP has a high EC-EV labeling efficiency and favorable in vivo tracer stability during IS therapy. Moreover, EC-EVs are absorbed by astrocytes during cerebral I/R injury and promote the restoration of neurological function through the regulation of the c-Fos/AP-1 signaling pathway.

Absolute Quantification of Dynamic Cellular Uptake of Small Extracellular Vesicles via Lanthanide Element Labeling and ICP-MS.

Small extracellular vesicles (sEVs) are increasingly reported to play important roles in numerous physiological and pathological processes. Cellular uptake of sEVs is of great significance for functional regulation in recipient cells. Although various sEV quantification, labeling, and tracking methods have been reported, it is still highly challenging to quantify the absolute amount of cellular uptake of sEVs and correlate this information with phenotypic variations in the recipient cell. Therefore, we developed a novel strategy using lanthanide element labeling and inductively coupled plasma-mass spectrometry (ICP-MS) for the absolute and sensitive quantification of sEVs. This strategy utilizes the chelation interaction between Eu3+ and the phosphate groups on the sEV membrane for specific labeling. sEVs internalized by cells can then be quantified by ICP-MS using a previously established linear relationship between the europium content and the particle numbers. High Eu labeling efficiency and stability were demonstrated by various evaluations, and no structural or functional alterations in the sEVs were discovered after Eu labeling. Application of this method revealed that 4020 ± 171 sEV particles/cell were internalized by HeLa cells at 37 °C and 61% uptake inhibition at 4 °C. Further investigation led to the quantitative differential analysis of sEV cellular uptake under the treatment of several chemical endocytosis inhibitors. A 23% strong inhibition indicated that HeLa cells uptake sEVs mainly through the macropinocytosis pathway. This facile labeling and absolute quantification strategy of sEVs with ppb-level high sensitivity is expected to become a potential tool for studying the functions of sEVs in intracellular communication and cargo transportation.

Process Evaluation for Diversified Academic Assessment Mechanism in Higher Education Institutions by Use of Data Mining

A diversified academic assessment mechanism can effectively improve students’ learning motivation, make up for the possible blind spots of a single assessment method, and better guide students’ learning and teachers’ teaching. Using data mining methods to process evaluation data for diversified academic assessment mechanisms in colleges and universities can discover patterns in students’ learning, find key factors affecting academic performance, and provide a basis for teaching reform. Most of the current process evaluation data mining methods focus on hard skills, such as academic performance and classroom participation, but it is difficult to evaluate soft skills such as critical thinking and teamwork. To this end, this paper studies the process evaluation data mining methods for a diversified academic assessment mechanism in colleges and universities. It constructs an indicator system for process evaluation of diversified academic assessment mechanism in colleges and universities, gives a quantitative method for indicators, and performs fuzzy comprehensive evaluation based on AHP-entropy weight method. For the evaluation of text-based indicators, a consistency training method is introduced to train the process evaluation correlation mining model using a large amount of unlabeled process evaluation examples, which effectively solves the problems of lack of labeled data, high labeling cost, and changes in data distribution, and improves the performance and availability of the model. The experimental results verify the effectiveness of the proposed method.

CSSRS: Pore segmentation method of sandstone cast thin section images based on weak supervised learning

Semantic segmentation based on deep learning has been gradually applied to digital rocks, but the amount of training sets and the fineness of data labelling affect the accuracy of the final segmentation, as well as the pixel-level annotation would cost a lot of human effort and be accompanied by certain errors. To address the problem of high data labelling cost of sandstone cast thin section images and segmentation accuracy affected by manual bias, this paper proposes a new method combining Smoothed Score Class Activation Mapping (SSCAM) and Restricted Similarity Dual Contrast Learning (RSimDCL), and we named it CSSRS. Using an image-level annotated dataset, the study applied CSSRS to the images with pore segments (pore images) presented and those without (background images). CSSRS utilized SSCAM to obtain the general location of the pores from the cast thin section map, adopted RSimDCL to obtain the fine boundaries of the pore area, and combined these two methods to generate more accurate pore segmentation of sandstone cast thin section images. Finally, pore segmentation was performed on the same test dataset using CSSRS and RSimDCL, and the segmentation results were compared with various Class Activation Mapping (CAM) algorithms based on weakly supervised learning and various Generative Adversarial Networks (GAN) algorithms. Results showed that CSSRS demonstrated the highest Precision of 79.35% (18.65% higher than the second best method RSimDCL). In addition, the study applied Mean Intersection over Union (MIoU), Dice, Pixel Accuracy (PA), False Positive Rate (FPR) comparison metrics and found that CSSRS consistently demonstrated a better performance of 5%–6% improvement than the second best method in each metric. In conclusion, the CSSRS method based on weakly supervised learning can be applied to pore segmentation of sandstone cast thin section images, which significantly reduces the data annotation workload, avoids the manual bias, and also increases the segmentation effect to a greater extent.

Spatial attention and quantization-based contrastive learning framework for mmWave massive MIMO beam training

Deep learning (DL)-based beam training schemes have been exploited to improve spectral efficiency with fast optimal beam selection for millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems. To achieve high prediction accuracy, these DL models rely on training with a tremendous amount of labeled environmental measurements, such as mmWave channel state information (CSI). However, demanding a large volume of ground truth labels for beam training is inefficient and infeasible due to the high labeling cost and the requirement for expertise in practical mmWave massive MIMO systems. Meanwhile, a complex environment incurs critical performance degradation in the continuous output of beam training. In this paper, we propose a novel contrastive learning framework, named self-enhanced quantized phase-based transformer network (SE-QPTNet), for reliable beam training with only a small fraction of the labeled CSI dataset. We first develop a quantized phase-based transformer network (QPTNet) with a hierarchical structure to explore the essential features from frequency and spatial views and quantize the environmental components with a latent beam codebook to achieve robust representation. Next, we design the SE-QPTNet including self-enhanced pre-training and supervised beam training. SE-QPTNet pre-trains by the contrastive information of the target user and others with the unlabeled CSI, and then, it is utilized as the initialization to fine-tune with a reduced volume of labeled CSI. Finally, the experimental results show that the proposed framework improves beam prediction accuracy and data rates with 5% labeled data compared to existing solutions. Our proposed framework further enhances flexibility and breaks the limitation of the quantity of label information for practical beam training.