Limited Availability Of Samples Research Articles

Enhancing Semi-Supervised Few-Shot Hyperspectral Image Classification via Progressive Sample Selection

Hyperspectral images (HSIs) provide valuable spatial–spectral information for ground analysis. However, in few-shot (FS) scenarios, the limited availability of training samples poses significant challenges in capturing the sample distribution under diverse environmental conditions. Semi-supervised learning has shown promise in exploring the distribution of unlabeled samples through pseudo-labels. Nonetheless, FS HSI classification encounters the issue of high intra-class spectral variability and inter-class spectral similarity, which often lead to the diffusion of unreliable pseudo-labels during the iterative process. In this paper, we propose a simple yet effective progressive pseudo-label selection strategy that leverages the spatial–spectral consistency of HSI pixel samples. By leveraging spatially aligned ground materials as connected regions with the same semantic and similar spectrum, pseudo-labeled samples were selected based on round-wise confidence scores. Samples within both spatially and semantically connected regions of FS samples were assigned pseudo-labels and joined subsequent training rounds. Moreover, considering the spatial positions of FS samples that may appear in diverse patterns, to fully utilize unlabeled samples that fall outside the neighborhood of FS samples but still belong to certain connected regions, we designed a matching active learning approach for expert annotation based on the temporal confidence difference. We identified samples with the highest training value in specific regions, utilizing the consistency between predictive labels and expert labels to decide whether to include the region or the sample itself in the subsequent semi-supervised iteration. Experiments on both classic and more recent HSI datasets demonstrated that the proposed base model achieved SOTA performance even with extremely rare labeled samples. Moreover, the extended version with active learning further enhances performance by involving limited additional annotation.

Generating Synthetic Health Sensor Data for Privacy-Preserving Wearable Stress Detection.

Smartwatch health sensor data are increasingly utilized in smart health applications and patient monitoring, including stress detection. However, such medical data often comprise sensitive personal information and are resource-intensive to acquire for research purposes. In response to this challenge, we introduce the privacy-aware synthetization of multi-sensor smartwatch health readings related to moments of stress, employing Generative Adversarial Networks (GANs) and Differential Privacy (DP) safeguards. Our method not only protects patient information but also enhances data availability for research. To ensure its usefulness, we test synthetic data from multiple GANs and employ different data enhancement strategies on an actual stress detection task. Our GAN-based augmentation methods demonstrate significant improvements in model performance, with private DP training scenarios observing an 11.90-15.48% increase in F1-score, while non-private training scenarios still see a 0.45% boost. These results underline the potential of differentially private synthetic data in optimizing utility-privacy trade-offs, especially with the limited availability of real training samples. Through rigorous quality assessments, we confirm the integrity and plausibility of our synthetic data, which, however, are significantly impacted when increasing privacy requirements.

Abstract PO1-14-04: A multivariate biomarker to guide antibody-drug conjugate selection and provide insight on response differences across breast cancer subtypes

Abstract Background: Antibody-drug conjugates (ADCs), including trastuzumab deruxtecan (T-DXd; targeting HER2) and sacituzumab govitecan (SG; targeting TROP2), have transformed the management of metastatic breast cancer, with additional ADCs approved in other solid tumors or in late-stage development. To date, expression of ADC target alone has been poorly predictive of objective response rates (ORRs) in both breast cancer and other tumors. Hence, there is an urgent need to develop better predictive biomarkers to guide ADC vs. other treatments, T-DXd vs. SG treatment, and optimized predictive medicine opportunities involving other ADCs. Additionally, whether ADCs are similarly effective in less common breast cancer subtypes—such as invasive lobular carcinoma (ILC) vs. invasive ductal carcinoma (IDC), is unclear. Given the limited availability of trial tissue samples and cohorts with long-term follow-up, herein we sought to determine whether a tissue based, pan-solid tumor, multivariate biomarker algorithm could predict observed ORRs across ADCs and tumor types, as has been used previously to associate tumor mutation burden with immunotherapy ORRs. Methods: From 15,032 FFPE tumor tissue samples (from 21 tumor types) tested by clinical comprehensive genomic profiling plus RNA based quantitative transcriptional profiling (qTP) as part of the observational Strata Trial (NCT03061305), the ADC treatment response score (TRS) was discovered and validated to predict published ORRs across tumor types and approved/late-stage ADCs (n=16 observed ORRs [from 7 tumor types and 8 ADCs; SG not included]). The best performing 3-factor algorithm (by Pearson correlation coefficient of observed ORRs vs. tumor type and ADC specific predicted biomarker positivity rates) included only qTP components and combined ADC target expression, cell proliferation, and extracellular matrix adhesion (the latter being negatively associated with ORRs). Importantly, predictive biomarker positivity rates of TRS was more correlated vs. observed ORRs (n=16, r=0.81, p=0.0001) than target expression alone (n=16, r=0.54, p=0.03). TRS was then validated using held out SG ORRs from nine tumor types in the IMMU-132-01 basket trial and two ADC ORRs from ASCO 2023 abstracts (enfortumab vedotin [EV; targeting NECTIN-4] in head and neck cancer and patritumab vedotin [targeting HER3] in breast cancer), with TRS predictive biomarker positivity rates again being more significantly correlated vs. observed ORRs than target expression alone (n=11, TRS r=0.91, p=0.0001; target expression alone r=44, p=0.17). Lastly, the locked TRS model was then applied to the DESTINY-PanTumor02 T-DXd dataset presented at ASCO 2023 (n=21 tumor type/HER2 expression groups matched to the trial groups), with TRS predictive biomarker positivity rates again being highly correlated vs. observed ORRs (n=21, TRS r=0.80, p< 0.0001). Across the 21 tumor types in the Strata Trial dataset, breast cancer had the highest percentage (76%) of patients predicted positive for at least one ADC, with 23% and 55% positive for the approved ADCs mirvetuximab soravtansine (targeting FOLR1) and EV, respectively. Lastly, patients with ILC vs. IDC had significantly greater TRS positivity rates for T-DXd (61% vs. 47%) and SG (57% vs. 48%) due to significantly decreased extracellular matrix adhesion in ILC vs. IDC. Clinical outcomes data for patients treated with ADCs and available TRS enrolled in the Strata Trial are maturing and will be presented at the meeting. Conclusion: We have developed and validated TRS, a multivariate RNA based tumor tissue algorithm that predicts observed ORRs across tumor types and approved/late-stage ADCs. More than 75% of all patients with metastatic breast cancer are predicted to be responsive to one or more ADCs, including those approved in other tumor types. Citation Format: Laura Lamb, Jonathan Mowers, Azadeh Nasrazadani, Mark Burkard, Nickolay Khazanov, Daniel Hovelson, Kat Kwiatkowski, Scott Tomlins, D. Bryan Johnson, Daniel Rhodes. A multivariate biomarker to guide antibody-drug conjugate selection and provide insight on response differences across breast cancer subtypes [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO1-14-04.

Per- and polyfluoroalkyl substances (PFAS) exposure in plasma and their blood–brain barrier transmission efficiency–A pilot study

Per- and polyfluoroalkyl substances (PFAS) have been shown to penetrate the blood–brain barrier (BBB) and accumulate in human brain. The BBB transmission and accumulation efficiency of PFAS, as well as the potential health risks from human co-exposure to legacy and emerging PFAS due to differences in transport efficiency, need to be further elucidated. In the present pilot study, 23 plasma samples from glioma patients were analyzed for 17 PFAS. The concentrations of PFAS in six paired brain tissue and plasma samples were used to calculate the BBB transmission efficiency of PFAS (RPFAS). This RPFAS analysis was conducted with utmost care and consideration amid the limited availability of valuable paired samples. The results indicated that low molecular weight PFAS, including short-chain and emerging PFAS, may have a greater potential for accumulation in brain tissue than long-chain PFAS. As an alternative to perfluorooctane sulfonic acid (PFOS), 6:2 chlorinated polyfluorinated ether sulfonate (6:2 Cl-PFESA) exhibited brain accumulation potential similar to that of PFOS, suggesting it may not be a suitable substitute concerning health risk in brain. The BBB transmission efficiencies of perfluorooctanoic acid, PFOS, and 6:2 Cl-PFESA showed similar trends with age, which may be an important factor influencing the entry of exogenous compounds into the brain. A favorable link between perfluorooctane sulfonamide (FOSA) and the development and/or progression of glioma may be implicated by a strong positive correlation (r2 = 0.94; p < 0.01) between RFOSA and Ki-67 (a molecular marker of glioma). However, a causal relationship between RFOSA and glioma incidence were not established in the present study. The present pilot study conducted the first examination of BBB transmission efficiency of PFAS from plasma to brain tissue and highlighted the importance of reducing and/or controlling exposure to PFAS.

FeatWalk: Enhancing Few-Shot Classification through Local View Leveraging

Few-shot learning is a challenging task due to the limited availability of training samples. Recent few-shot learning studies with meta-learning and simple transfer learning methods have achieved promising performance. However, the feature extractor pre-trained with the upstream dataset may neglect the extraction of certain features which could be crucial for downstream tasks. In this study, inspired by the process of human learning in few-shot tasks, where humans not only observe the whole image (`global view') but also attend to various local image regions (`local view') for comprehensive understanding of detailed features, we propose a simple yet effective few-shot learning method called FeatWalk which can utilize the complementary nature of global and local views, therefore providing an intuitive and effective solution to the problem of insufficient local information extraction from the pre-trained feature extractor. Our method can be easily and flexibly combined with various existing methods, further enhancing few-shot learning performance. Extensive experiments on multiple benchmark datasets consistently demonstrate the effectiveness and versatility of our method.The source code is available at https://github.com/exceefind/FeatWalk.

Determining the state of guidance on pediatric biobanking for researchers, HRECS, and families: Regulatory mapping of international guidance

Biobanking—the storage of human biological samples, including tissue, blood, urine, and genetic data—raises many ethical, legal, and social issues, including confidentiality and privacy. Pediatric biobanking is more complicated, with difficulties arising because children lack capacity to consent and acquire this capacity upon maturity when the research is still ongoing. Yet given the limited availability of pediatric samples, the translational nature of biobanking presents a unique opportunity to share samples and produce clinically necessary information about pediatric development and diseases. Guidance on navigating these legal and ethical difficulties is needed for those involved in pediatric biobanking—including researchers, participants, and families, and those involved in biobank governance. This paper seeks to map the current regulatory framework governing pediatric biobanking to determine what guidance is currently offered. Regulatory mapping of current international and national guidelines on pediatric biobanking addressing the ethical, legal, and social nuances of pediatric biobanking was undertaken. This paper finds that international guidelines around biobanking are mostly for adults, and even when pediatric-specific, documents are non-binding, inconsistent, or only limited guidance is offered on a range of important issues specific to pediatric biobanks. Conclusion: This paper shows a need for consistent, comprehensive, and clear regulation on pediatric biobanking so that research can more quickly, efficiently, and ethically be translated to useful information and treatment in pediatric care.What is Known:• Pediatric biobanking presents new opportunities to conduct valuable translational research to benefit pediatric populations. However, the storage of pediatric biological samples raises many ethical, legal and social issues—in part because child participants may be considered to lack capacity to consent but can acquire this capacity upon maturity when the research is still ongoing. Pediatric biobanks must grapple with issues of consent, confidentiality and privacy, and long-term participation regarding child participants.What is New:• Regulatory guidance on these ethical, legal, and social issues is needed for researchers, participants, and families and those involved in biobank governance. This paper identifies nationally specific and international guidance on biobanking and summarizes the guidance provided in relation to these pediatric specific issues. It finds that most guidance is non-binding and inconsistent between guidance documents and may offer only limited guidance to stakeholders. A need for consistent, comprehensive, and clear regulation on pediatric biobanking is needed at an international level to enable research.

A lightweight image-level segmentation method for steel surface defects based on cross-layer feature fusion

Steel is widely used in the aerospace, machinery and automotive industries. Surface defects not only have a negative impact on the appearance of steel but also significantly reduce its wear resistance, high temperature resistance, corrosion resistance and fatigue strength. Therefore, the detection of steel surface defects is very important to improve the quality of steel production. The limited availability of surface defect samples in the industrial sector poses significant challenges for the accurate detection of defects in high-quality materials. In addition, the existing defect detection model is highly complex and not easy to deploy. To solve this problem, a lightweight defect detection network suitable for steel defects is proposed. The cross-layer feature fusion (CFF) in the design enables effective utilisation of multi-layer semantic features, facilitating the detection of small defects in steel. Secondly, a new loss function is designed to make up for the problems of small data volume and uneven data distribution. The experimental results demonstrate that the steel surface defect detection method proposed in this paper achieves the highest detection performance on widely used public datasets such as RSDDS, NEUS and NRSD-CR(test), while maintaining the lowest model complexity.

Methodological optimisation of thymocyte isolation and cryopreservation of human thymus samples

Premature lymphocytes develop into non-autoreactive, mature naïve CD4+ or CD8+ T cells in the thymus before entering the circulation. However, in-depth characterization of human thymocyte development remains challenging due to limited availability of human thymus samples and the fragile nature of thymocyte populations. Thymocytes often do not survive cryopreservation and thawing procedures, especially the fragile CD4+CD8+ double positive population. It is generally recommended to use fresh human thymus tissue on the day of excision to avoid any biases in thymocyte composition. This hampers the possibility to perform multiple experiments on the same thymus sample. To establish how the thymocyte viability and composition can be maintained, we compared two thymocyte isolation methods used for human and/or mice thymi, three cryopreservation methods in combination with our most gentle thawing technique. Based on our findings we established that fresh human thymi remain viable in cold storage for up to two days post-surgery without compromising thymocyte composition. Thymocytes can be cryopreserved if required, although the CD4+CD8+ double positive populations may be reduced. Our study provides thoroughly optimized methods to study human thymocyte development over a considerable time-frame post-surgery.

Understanding Spondyloarthritis Pathogenesis: The Promise of Single-Cell Profiling.

Single-cell profiling, either in suspension or within the tissue context, is a rapidly evolving field. The purpose of this review is to outline recent advancements and emerging trends with a specific focus on studies in spondyloarthritis. The introduction of sequencing-based approaches for the quantification of RNA, protein, or epigenetic modifications at single-cell resolution has provided a major boost to discovery-driven research. Fluorescent flow cytometry, mass cytometry, and image-based cytometry continue to evolve. Spatial transcriptomics and imaging mass cytometry have extended high-dimensional analysis to cells in tissues. Applications in spondyloarthritis include the indexing and functional characterization of cells, discovery of disease-associated cell states, and identification of signatures associated with therapeutic responses. Single-cell TCR-seq has provided evidence for clonal expansion of CD8+ T cells in spondyloarthritis. The use of single-cell profiling approaches in spondyloarthritis research is still in its early stages. Challenges include high cost and limited availability of diseased tissue samples. To harness the full potential of the rapidly expanding technical capabilities, large-scale collaborative efforts are imperative.

Advanced Cellular Models for Rare Disease Study: Exploring Neural, Muscle and Skeletal Organoids.

Organoids are self-organized, three-dimensional structures derived from stem cells that can mimic the structure and physiology of human organs. Patient-specific induced pluripotent stem cells (iPSCs) and 3D organoid model systems allow cells to be analyzed in a controlled environment to simulate the characteristics of a given disease by modeling the underlying pathophysiology. The recent development of 3D cell models has offered the scientific community an exceptionally valuable tool in the study of rare diseases, overcoming the limited availability of biological samples and the limitations of animal models. This review provides an overview of iPSC models and genetic engineering techniques used to develop organoids. In particular, some of the models applied to the study of rare neuronal, muscular and skeletal diseases are described. Furthermore, the limitations and potential of developing new therapeutic approaches are discussed.

Challenges and Innovations in Forensic Identification: Case Study of Decomposed Remains in a Crocodile's Body for Missing Child Identification

Identifying a missing child from decomposed remains in a crocodile's body poses significant challenges and complexity. Forensic investigations involving human remains within animal’s body can present unique difficulties apart from conventional investigative scenarios. Decomposed remains added an additional layer of difficulty as the digestive process of a crocodile can significantly degrade the body and its features. Forensic anthropology examinations were impeded due to the limited availability of tiny bone samples. In this scenario decomposed bone pieces were the only remaining material for DNA analysis, the choice of an efficient DNA recovery procedure becomes crucial for successful outcome. In the study, prior to the DNA extraction process, a decontamination technique was employed, followed by decalcification method on the bone sample. The resulting, clean and decalcified bone sample was then subjected to extraction using a silica-based method, and the extracted DNA was subsequently typed through STR amplification. The derived DNA profile information from the bone sample pertaining to the child is compared with the profile details of parents for identification purposes.

A hierarchical transformer-based adaptive metric and joint-learning network for few-shot rolling bearing fault diagnosis

Recently, deep learning techniques have significantly bolstered the advancement of intelligent fault diagnosis. However, in engineering practice, the limited availability of fault samples poses considerable challenges to the existing methods. To address this problem, a hierarchical Transformer-based adaptive metric and joint-learning network (HTAMJN) is suggested in this paper. Firstly, a hierarchical progressive fusion encoder based on orthogonal self-attention is devised, which effectively enhances the model’s attention to local features and establishes comprehensive interdependencies. To reduce the demand for labeled data, a metric method with a coupling mechanism is proposed to adaptively enhance the metric benchmark, making the category representation more discriminative. Additionally, a joint optimization scheme based on the episodic paradigm is formulated to improve the recognition capability of the network for different categories under few-shot scenarios. The outcomes demonstrate that the HTAMJN outperforms several comparison models in both few-shot diagnosis and cross-domain recognition tasks in experiments conducted on three bearing datasets, exhibiting superior performance.

Dynamic model-assisted transferable network for liquid rocket engine fault diagnosis using limited fault samples

The accurate detection and diagnosis of faults in Liquid Rocket Engines (LREs) are critical for ensuring space mission safety. However, the limited availability of actual fault samples and the diversification of potential faults present significant challenges in achieving precise diagnosis. To overcome these obstacles, we propose a dynamic model-assisted transfer learning approach. In this study, we first modularize the LRE and establish a dynamic model based on the mathematical principles of each module. Subsequently, we artificially established a fault module model, injected faults into the normal model, and simulated various fault modes to expand the fault sample library. Leveraging this augmented dataset in combination with a limited number of actual fault samples, we employ transfer learning to fine-tune a Convolutional Neural Network (CNN). Compared with other classic methods, the migrated CNN effectively adapts to the distribution of real data and significantly improves the accuracy of LRE fault diagnosis.

Segregation of formulated powders in direct compression process and evaluations by small bench-scale testers

Powder segregation can cause severe issues in processes of pharmaceutical drugs for control of content uniformity if the powder is likely to be free or easy flowing. Assessing segregation intensity of formulated powders in a process is challenging at the formulation stage because of the limited availability of samples. An advanced segregation evaluation using small bench-scale testers can be useful for formulation decisions and suggestions of operation conditions in the process, which has not been practically investigated before. In this study, eight formulations (two co-processed excipients blended with one active pharmaceutical ingredient at different ratios) were used for the segregation study on two types of bench-scale testers (air-induced and surface rolling segregation tester), and a pilot simulation process rig as a comparative study. The results show that segregation measured on the bench-scale testers can give a good indication of the segregation intensity of a blend if the segregation intensity is not more than 20%. The comparison also shows that both the bench-scale testers have a good correlation to the process rig, respectively, which means either segregation tester can be used independently for the evaluation. A linear regression model was explored for prediction of segregation in the process.

Deep reference autoencoder convolutional neural network for damage identification in parallel steel wire cables

This paper addresses the challenge of overcoming the limited availability of training samples for the damage state of parallel steel wire cables in practical engineering and the difficulty of detecting minor damages. To tackle this issue, we propose an unsupervised deep learning damage identification technique called Deep Reference Autoencoder Convolutional Neural Network (DRACNN) for analyzing the damage state of parallel steel wire cables in bridge engineering. The DRACNN method utilizes multi-dimensional cross-correlation function (CCF) derived from acceleration signals at various health stages as input to train the network structure and obtain optimal parameters. Subsequently, we analyze the layer decomposition to identify neurons in the lowest hidden layer indicating damage. The neuronal change information is then extracted using an Exponentially Weighted Moving Average (EWMA) Control Chart to determine the damage state of the structure. Finally, we present a comprehensive numerical analysis describing the method's flow and network architecture and demonstrate the feasibility of this approach through experiments.

Fabrication of Single-Crystalline CoCrFeNi Thin Films by DC Magnetron Sputtering: A Route to Surface Studies of High-Entropy Alloys.

High-entropy alloys (HEAs) with their almost limitless number of possible compositions have raised widespread attention in material science. Next to wear and corrosion resistive coatings, their application as tunable electrocatalysts has recently moved into the focus. On the other hand, fundamental properties of HEA surfaces like atomic and electronic structure, surface segregation and diffusion as well as adsorption on HEA surfaces are barely explored. The lack of research is caused by the limited availability of single-crystalline samples. In the present work, the epitaxial growth of face centered cubic (fcc) CoCrFeNi films on MgO(100) is reported. Their characterization by X-ray diffraction (XRD), energy dispersive X-ray spectroscopy (EDX), and transmission electron microscopy (TEM) demonstrates that the layers with a homogeneous and close to equimolar elemental composition are oriented in [100] direction and aligned with the substrate to which they form an abrupt interface. X-ray photoelectron spectroscopy (XPS), low-energy electron diffraction (LEED), and angle-resolved photoelectron spectroscopy are employed to study chemical composition and atomic and electronic structure of CoCrFeNi(100). It is demonstrated that epitaxially grown HEA films have the potential to fill the sample gap, allowing for fundamental studies of properties of and processes on well-defined HEA surfaces over the full compositionalspace.

The role of extracellular vesicles in non-small-cell lung cancer, the unknowns, and how new approach methodologies can support new knowledge generation in the field

Extracellular vesicles (EVs) are nanosized particles released from most human cell types that contain a variety of cargos responsible for mediating cell-to-cell and organ-to-organ communications. Current knowledge demonstrates that EVs also play critical roles in many aspects of the progression of Non-Small-Cell Lung Cancer (NSCLC). Their roles range from increasing proliferative signalling to inhibiting apoptosis, promoting cancer metastasis, and modulating the tumour microenvironment to support cancer development. However, due to the limited availability of patient samples, intrinsic inter-species differences between human and animal EV biology, and the complex nature of EV interactions in vivo, where multiple cell types are present and several events occur simultaneously, the use of conventional preclinical and clinical models has significantly hindered reaching conclusive results. This review discusses the biological roles that EVs are currently known to play in NSCLC and identifies specific challenges in advancing today's knowledge. It also describes the NSCLC models that have been used to define currently-known EV functions, the limitations associated with their use in this field, and how New Approach Methodologies (NAMs), such as microfluidic platforms, organoids, and spheroids, can be used to overcome these limitations, effectively supporting future exciting discoveries in the NSCLC field and the potential clinical exploitation of EVs.

Alveolar epithelial-to-mesenchymal transition in scleroderma interstitial lung disease: Technical challenges, available evidence and therapeutic perspectives

The alveolar epithelial-to-mesenchymal transition is the process of transformation of differentiated epithelial cells into mesenchymal-like cells through functional and morphological changes. A partial epithelial-to-mesenchymal transition process can indirectly contribute to lung fibrosis through a paracrine stimulation of the surrounding cells, while a finalized process could also directly enhance the pool of pulmonary fibroblasts and the extracellular matrix deposition. The direct demonstration of alveolar epithelial-to-mesenchymal transition in scleroderma-related interstitial lung disease is challenging due to technical pitfalls and the limited availability of lung tissue samples. Similarly, any inference on epithelial-to-mesenchymal transition occurrence driven from preclinical models should consider the limitations of cell cultures and animal models. Notwithstanding, while the occurrence or the relevance of this phenomenon in scleroderma-related interstitial lung disease have not been directly and conclusively demonstrated until now, pre-clinical and clinical evidence supports the potential role of epithelial-to-mesenchymal transition in the development and progression of lung fibrosis. Evidence consolidation on scleroderma-related interstitial lung disease epithelial-to-mesenchymal transition would pave the way for new therapeutic opportunities to prevent, slow or even reverse lung fibrosis, drawing lessons from current research lines in neoplastic epithelial-to-mesenchymal transition.

A general approach for automatic segmentation of pneumonia, pulmonary nodule, and tuberculosis in CT images

Proposing a general segmentation approach for lung lesions, including pulmonary nodules, pneumonia, and tuberculosis, in CT images will improve efficiency in radiology. However, the performance of generative adversarial networks ishampered by the limited availability of annotated samples and the catastrophicforgetting of the discriminator, whereas the universality of traditional morphology-based methods is insufficient for segmenting diverse lung lesions. A cascaded dual-attention network with a context-aware pyramid feature extraction module was designed to address these challenges. A self-supervised rotation loss was designed to mitigate discriminator forgetting. The proposed model achieved Dice coefficients of 70.92, 73.55, and 68.52% on multi-center pneumonia, lung nodule, and tuberculosis test datasets, respectively. No significant decrease in accuracy was observed (p>0.10) when a small training sample size was used. The cyclic training of the discriminator was reduced with self-supervised rotation loss (p<0.01). The proposed approach is promising for segmenting multiple lung lesion types in CT images.

A semisupervised autoencoder-based method for anomaly detection in cutting tools

Detecting anomalies such as breakage and excessive wear of cutting tools in the machining process is crucial to prevent damage and improve productivity. Data-driven anomaly detection (AD) methods suffer from limited availability of anomaly samples, which is ineluctable in practice owing to strict reliability restrictions. Therefore, we propose a semisupervised AD approach in which only failure-free samples are required to establish an AD model. The key strategy is to learn the characteristics of failure-free samples using an improved autoencoder (AE) and discern observations by deviations from the characteristics. We rebuild the loss function of AE to impel the model to learn the common characteristics in latent space. We propose a factor that reflects the anomaly degree as the decision-making function to implement AD. The proposed approach is verified on an experimental cutting tool breakage dataset and a public cutting tool wear dataset. The experimental results demonstrate the validity of the proposed approach. The comparisons with conventional methods substantiate that the proposed approach outperforms existing AD methods.