Polycystic Ovary Syndrome (PCOS) is a prevalent condition affecting female reproductive health, where early and accurate detection through image analysis can significantly aid diagnosis. This study proposes a hybrid approach for automated binary classification of PCOS that integrates advanced feature extraction techniques with stacking ensemble learning models. Two strategies are investigated. The first approach employs a stacking ensemble of four classifiers, while the second approach introduces Gradient Boosting (GB) as an additional base learner, increasing the ensemble to five classifiers. The PCOSFusion algorithm is utilized during feature extraction to identify distinctive patterns in ovarian medical images. Extracted features are then input to the classifiers for training and evaluation. Both strategies effectively distinguish between PCOS (abnormal) and non-PCOS (normal) cases. Results demonstrate that the stacking ensemble method harnesses the complementary strengths of individual classifiers, with the second approach, which incorporates Gradient Boosting, achieving a slight performance improvement. The best-performing model achieved 98.44% accuracy, 99.35% precision, and 98.49% recall, highlighting the potential of stacking-based ensemble techniques combined with effective feature extraction to improve diagnostic accuracy in medical imaging tasks. These findings support the viability of the proposed method as a valuable tool for assisting medical professionals in the early detection of PCOS.
Luminescent radicals can achieve unit internal quantum efficiency, but solid-state performance remains limited by intricate radical/host interactions. Herein, a cluster model combined with a multi-level conformation search method was adopted to systematically investigate the modulation mechanism of stacking modes on stability and charge injection dynamics in radical/host aggregation. It is found that a quasi-parallel stacking mode with a donor moiety aligned with the host is the energetically favored configuration, which fosters compact π-π stacking, ensures favorable energy alignment and balanced charge injection. Furthermore, quasi-parallel stacking can minimize the energy difference between the charge transfer (CT) and local excited (LE) states. Such state hybridization permits the emissive CT state to borrow oscillator strength from the LE manifold, thereby resulting in a marked enhancement of the radiative transition rate. This work reveals the critical role of the radical/host interface orientation, offering theoretical insights for the rational design of high-performance radical-based optoelectronic devices.
Identifying the energetic and structural properties of amino acid monomers that drive dimer formation can provide key insights into the non-covalent interactions responsible for their association. In this work, density functional theory, energy decomposition analysis (EDA), and conformer sampling are employed to analyze the interactions responsible for the dimerization of the zwitterionic amino acid tryptophan in implicit solvent. Although EDA suggests that charge transfer is the dominant stabilizing interaction, other distinct structure-energy relationships emerge. For instance, end-on and T-shaped conformations are primarily stabilized by charge transfer, contributing more than 1 kcal mol-1 compared to dispersion in some cases. Alternatively, some π-stacked conformations are preferentially stabilized through dispersion by as much as 4 kcal mol-1. To further analyze stabilization effects from hydrogen bonding (e.g., the effects of donor-acceptor orbitals during charge transfer), a complementary occupied-virtual pair analysis reveals that stacked conformers have relatively weaker donor-acceptor contributions to charge transfer compared to their end-on counterparts. Additionally, implicit solvation is found to destabilize binding for stacked conformations more strongly than for end-on structures. The analysis presented here provides insights into the role of the non-covalent interactions that enable amino acid dimerization elucidating the fundamental interactions responsible for self-aggregation.
Hypertension or high blood pressure is a life threatening common cardiovascular disease (CVD) all over the world. In the era of information technology, communication, and artificial intelligence, early prediction of hypertension using various techniques can be advantageous to alert patients. The aim of this research is to automatically detect hypertension using an ensemble of different machine learning models. It carefully studies the use of both clinical and physiological data and compares the performance and explainability of the proposed model with other existing models and published works. This research study proposed a stacked ensemble learning-based model to detect hypertension. The popular classification models K-Nearest Neighbor (KNN), Random Forest (RF), and Light Gradient Boosting Machine (LGBM) are used as a stacked classifier, and at last, a Support Vector Machine (SVM) classifier is used as a meta-classifier. A publicly accessible large dataset of 21,613 patients containing the clinical and physiological data related to hypertension is used in this study. Additionally, data diversity is considered to test the generalization capability of the proposed learning model. Three datasets having only clinical data of both male and female subjects and their combination are used to train and evaluate the proposed model with an emphasis on enhancing the generalization capabilities of the classifier. To resolve data distribution imbalances, the proposed framework employs the Synthetic Minority Oversampling Technique Tomek Link (SMOTE-Tomek), and various feature selection techniques are utilized to compare the impact of features on this model. Various performance evaluation matrices are used to assess and analyze the performance of the classifier under different dataset cases. Moreover, the explainability of the proposed model is inspected using SHapley Additive exPlanations (SHAP) values, and it is perceived that the feature importance given by the model is sensible. The obtained results show that the proposed model achieves the superior accuracy when compared to alternative models and past research investigations. The proposed stacked ensemble model can detect hypertension from clinical and physiological data with the accuracy of 85.90%, 86.72%, and 85.91% for feature sets having feature numbers 21, 10, and 8, respectively. Second, for only clinical data, our model achieves 89.58%, 58.54%, 84.31%, 79.84%, and 77.78% for datasets I, II, III, IV, and I + II, respectively. Comparisons among different combinations of feature sets and with other single and ensemble models are analyzed to achieve the highest accuracy of the proposed model. The outcomes of this research can be useful in the realm of healthcare and predictive analytics of hypertension. By emphasizing timely detection, the research underscores the model's potential in reducing individual health risks and enabling proactive intervention, thus highlighting the significant role of AI technology-driven solutions revolutionizing healthcare practices.
The Kβ/Kα intensity ratios are critical parameters that quantitatively characterize atomic shell transition dynamics and radiative branching probabilities. In this study, we systematically evaluated the capability of machine learning (ML) algorithms to predict these ratios, as well as their advantages over traditional theoretical models (such as Scofield and semi-empirical calculations). A large dataset comprising 2124 experimental measurements compiled from the literature, covering elements with atomic numbers (Z) from 11 to 96, was structured to include more than ten variables, such as atomic number, sample form, excitation source, detector type, and energy resolution. Missing observations were imputed using the multivariate imputation by chained equations (MICE) method in the R programming language. Categorical variables were one-hot encoded, and the data were split into an 80% training set and a 20% test set. Seven heterogeneous individual models (RF, XGBoost, Cubist, SVR, GPR, BRNN, and GLMNET) were constructed, along with seven different stacking combinations derived from them. Following 10×10-fold cross-validation, the highest accuracy was achieved by the stacked model using a BRNN meta-learner (RMSE = 0.009; R2 = 0.973). This model reduced the test error of the Scofield theory by nearly 48% and performed significantly better according to the Diebold-Mariano test (p < 0.001). SHAP analysis revealed that atomic number is the primary determinant, while sample purity and excitation source have secondary yet physically consistent effects. Furthermore, an online R/Shiny-based calculator enhances the practical applicability of the method by enabling users to input their experimental parameters and receive instantaneous Kβ/Kα predictions. These results demonstrate that at the current stage of theoretical and experimental development, data-driven approaches provide significant advantages in both accuracy and interpretability over classical theories for complex atomic parameters such as the Kβ/Κα intensity ratio. Overall, this work constitutes a significant step toward reducing deviations in high-Z elements, improving detector calibration, and establishing new atomic databases.
To evaluate the diagnostic performance of a longitudinal ultrasound (US)-based stack-model for early prediction of pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer, as well as its practicality in assisting radiologists with diagnostic ability. A total of 974 patients who underwent NAC were retrospectively included between January 2017 and March 2022 from three different institutions. For all patients, US imaging was performed before NAC and after two cycles of NAC. The patients from the first hospital were used as the training dataset (n=653) and patients from hospital 2 and 3 were used as the test dataset (n=321)to test five deep learning (DL) models based on different feature sets. The optimal model was selected according to the area under the receiver operating characteristic curve (AUC). A model combining US imaging features with clinical factors was also investigated. Furthermore, the applicability of this model to provide clinical assistance was examined by radiologists with varying degrees of seniority. The Swin Transformer model based on the stacked-feature set achieved the highest AUC. Upon incorporating clinical factors, the combined model demonstrated superior performance in predicting pCR, achieving AUC of 0.935. Diagnostic performance in the early prediction of pCR improved for radiologists across all experience levels when assisted by the combined model. The longitudinal US-based model enables early prediction of pCR. Additionally, the model provided positive diagnostic assistance to radiologists with different experience levels. The longitudinal US-based model enable non-invasive early prediction of response to neoadjuvant chemotherapy in breast cancer while enhancing diagnostic performance across radiologists with varying experience levels.
Monolithic three-dimensional (M3D) integration is a promising solution for next-generation integrated circuits, offering enhanced signal propagation, high integration density, and lower fabrication costs than planar architectures. Amorphous oxide semiconductors, with room-temperature deposition capability and large-scale uniformity, are well-suited for 3D applications, yet developing multitier high-performance oxide transistors compatible with traditional technologies remains challenging. Here, we present a threshold voltage modulation strategy for indium gallium zinc oxide (IGZO) transistors via channel thickness control and atomic-layer-deposited surface modification. A four-tier vertically stacked IGZO transistor array has been manufactured with sequential layer-by-layer integration; optimized transistors across the tiers exhibited a low subthreshold swing of 150 mV/dec and an on/off ratio exceeding 108. Combined with via-hole interconnects, we demonstrated functional computing-in-memory 3D circuits featuring inverter modules (tier 1-2) and dynamic random access memory (DRAM) components (tier 3-4). The work advances oxide semiconductors' applications in future advanced 3D circuits.
MXene composite thin films, PbI2-Ti3C2T x , were fabricated by sequential dynamic spin-coating using a water/ethanol precursor system that enables in situ MXene incorporation while avoiding strongly coordinating solvents. This study addresses a central gap in halide precursor engineering: PbI2 is commonly treated as a transient phase before perovskite conversion, although its local structure and electronic environment can influence subsequent material formation. Here, we show that Ti3C2T x incorporation modifies PbI2 at the precursor level without disrupting the 2H-PbI2 framework. Profilometry shows that the average film thickness remains nearly constant across the composition series, whereas roughness and surface morphology evolve with MXene loading. Scanning electron microscopy with energy-dispersive X-ray spectroscopy (SEM-EDS) identifies Ti-rich clustered regions associated with Pb-I-containing material, supporting a growth-mediated incorporation pathway. X-ray diffraction confirms preservation of the 2H structure, while the PbI2 (001) basal-envelope line shape evolves with composition, consistent with changes in stacking-related environments. Raman spectroscopy shows preservation of the PbI2 vibrational fingerprint together with mode-selective perturbations. X-ray photoelectron spectroscopy (XPS) reveals statistically significant changes in the Pb 4f-I 3d core-level separation, indicating modification of the local Pb-I electrostatic and chemical environment rather than uniform charging or oxidation-state transformation. Optical measurements show a preserved absorption edge, while photoluminescence quenching and photoelectrical response indicate MXene-associated interfacial deactivation and carrier redistribution. Together, these results show that PbI2 can be treated as an engineerable precursor whose local morphology, stacking-related order, and electronic environment can be tuned before conversion into perovskite.
[This retracts the article DOI: 10.1007/s11042-022-14216-w.].
暂无摘要（点击查看详情）
Fault identification in Nuclear Power Plants (NPPs) is critical for ensuring operational safety, reliability, and efficiency. Traditional diagnostic methods often rely on physical models and expert systems, which may struggle to capture the complex dynamics of transient events. To overcome these limitations, this paper proposes an optimized stacked Graph Attention Network (GAT) for fault detection in NPPs by modeling the complex interdependencies among system components as graphs. Transient operational data are transformed into graph representations, where nodes correspond to system variables, and edges capture physical relationships. The architecture of the proposed model is optimized using a Heteroscedastic and Evolutionary Bayesian Optimization (HEPO), ensuring the use of the best configuration. The proposed GAT-based model, hypertuned by HEPO, is trained to recognize patterns associated with both normal and faulty transient conditions, including sensor anomalies and actuator failures. Based on synthetic data generated from the Personal Computer Transient Analyzer (PCTRAN), the proposed model achieved results above 0.96 for accuracy, precision, recall, and F1-score in a statistical analysis.
The accurate alignment of serial histological sections is essential for preserving anatomical continuity in 3D reconstruction. Although automated registration tools can efficiently correct global alignment errors, they often fail to resolve local misalignments caused by sectioning artifacts, tissue deformation, staining variability, or missing slices. Thus, we propose a practical two-step registration workflow that uses MultiStackReg (an ImageJ/Fiji plugin) for automatic alignment and AlignRef (a standalone application for interactive adjustment) for manual refinement. In the first step, MultiStackReg performs global registration by using rigid body transformations. In the second step, AlignRef corrects residual misalignments through semi-transparent overlay visualization, keyboard-based translation and rotation, and batch propagation of recorded transformations across selected slice ranges. We applied our workflow to 135 serial sections of a Carnegie Stage 15 embryo from the Virtual Human Embryo dataset that were stained with hematoxylin and eosin. MultiStackReg resolved most global inconsistencies, whereas AlignRef enabled the precise adjustment of subtle local deviations, particularly in curved structures such as neural tubes and limb buds. After automatic registration with MultiStackReg, the subsequent manual refinement step using AlignRef was completed in approximately 30 min and produced a suitable stack for 3D reconstruction. This two-step workflow balances automation with expert-guided correction and provides an accessible, reproducible, and anatomically precise method for the serial section alignment of morphological and developmental anatomy.
In homo- or hetero-interface of two dimensional materials, beyond the rigid moiré patterns, the marginal twisting ensures local commensurate stacking registry via reconstruction and thus introduces versatile structural subtleties into the many-body interplay, enabling the coexistence and tessellation of multiple domain-specific emergent states. In this work, we demonstrate that monolayer 1T-TiSe2 epitaxially grown on 2H-NbSe2 undergoes spontaneous moiré reconstruction, therein developing distinct charge density wave configurations dictated by stacking orientation. In parallel-stacked regions, the native 2×2 charge order of TiSe2 persists but experiences domain-selective modulation. Antiparallel-stacked systems host distinctive 3×3 and 2×1 charge orders which exhibit intertwined phenomena unveiled by scanning tunneling microscopy-including bias-toggled negative differential conductance, competitive inter-order penetration, and defect-mediated local ordering reconstitution. These observations coalesce into a unified paradigm, where the 3×3 charge order is quantum confined and isolated in single domain, while the 2×1 charge order permeates across domains, forming a percolative network. Our results demonstrate that the marginal-twist moiré reconstruction is a designer platform to generate rich emergent charge density wave landscapes, which also serves as a nanoscale testbed to decipher their disparate microscopic nature.
Hospitalized patients taking direct factor Xa inhibitors (DFXaIs), such as apixaban and rivaroxaban, often need to be transitioned to unfractionated heparin (UFH). This transition is complicated by residual DFXaI activity, which can lead to anticoagulant stacking and possible increased bleeding risk. The guidance on this transition from the manufacturers does not account for interpatient variability in drug clearance and may not be appropriate for all hospitalized patients. Residual DFXaI anticoagulation at the time of UFH initiation can result in a period of dual anticoagulation or anticoagulant stacking. This issue is identified most often when heparin is monitored using the anti-factor Xa assay as this assay is highly sensitive to the DFXaI. Several retrospective descriptive studies have reported supratherapeutic heparin levels (above 0.7 units/mL), which can last for days depending on the presence of patient-specific factors that reduce DFXaI clearance. The period of anticoagulant stacking has been associated with bleeding events. With the UFH anti-factor Xa assay detecting both anticoagulants, alternative methods are needed for heparin monitoring when transitioning from a DFXaI to UFH. Limited data describe alternative approaches that can be categorized as approaches that ignore residual DFXaI anticoagulation and ones that account for residual DFXaI anticoagulation. Observational data suggest that approaches that measure baseline residual DFXaI anticoagulation allow for a delay in UFH initiation and can reduce bleeding risk. The authors provide recommendations for hospitals or health systems to consider for optimizing management of the transition from DFXaI to UFH therapy. Each hospital or health system should develop a structured approach for managing DFXaI to UFH transitions. Ideally, the transition should be managed by pharmacists and involve measurement of residual DFXaI anticoagulation to inform the timing of UFH initiation. Prospective studies are needed to better define best practices.
The hemoglobin concentration in blood is vital for diagnosing anemia and monitoring the various health conditions. However, conventional measurement methods need invasive blood sampling so that they might have limited accessibility and uncomfortable for patients. Today, non-invasive alternatives powered by machine learning techniques provide promising solutions for point-of-care facilities and remote healthcare systems. This paper presents a methodology through a comprehensive research and development process to estimate hemoglobin levels from facial videos using multi-modal feature extraction and ensemble learning techniques. A dataset of 260 participants with various blood hemoglobin levels was processed to extract the features from pre-trained convolutional neural-networks (MobileNetV2, ResNet152), remote photoplethysmography (rPPG) signals, and color statistical features. Using these features, hemoglobin concentration was estimated via a number of machine learning models including XGBoost, Random Forest, and Stacking Regressor, respectively. Stacking Regressor provided the best estimation scores with a mean-absolute error of 0.7754 g/dL, Pearson correlation-coefficient of 0.7878, and [Formula: see text] score of 0.5852. ResNet152 model based features were combined with XGBoost, which achieved comparable performance (MAE: 0.6635 g/dL, [Formula: see text]: 0.4977). Experimental results demonstrated that multi-modal feature strategy outperformed single-modality approaches in terms of prediction accuracy and robustness. The proposed video-based estimation of hemoglobin concentration system achieves clinically relevant accuracy levels, outperforms to literature methods, comparable to point-of-care instruments demonstrating strong potential for use in anemia screening and remote patient monitoring.
High-order pancake bonding arising from the overlap of multiple π-type orbitals in π-conjugated molecules is exceedingly rare. Recently reported cofacially stacked hexaazatrinaphthylene trianions ([HAN]3-) stabilized by tetravalent actinides exhibit six-electron triple pancake bonds, but the presence of counterions hides the intrinsic nature of the bonding. Here, we designed a neutral HAN derivative via hydrogen coordination, 1,5,9-trihydro-1,4,5,8,9,12-hexaazatriphenylene (HATH3), which features a quartet ground state with three π-type singly occupied molecular orbitals. The HATH3 monomer dimerizes both in trans- and cis-cofacial arrangements, with ultrashort intermolecular separation of 2.968 and 2.971 Å with substantial interaction energy of -158.6 and -135.1 kJ/mol, respectively. The stability of these dimers occurs primarily through orbital interactions, three electron-sharing π-orbitals between two HATH3 fragments. Electrostatic interactions and dispersion make smaller but significant bonding contributions to the overall stability. These neutral dimers exhibit a genuine triple pancake bond, providing new insight into the nature of high-order π-stacking interactions. These strong intermolecular interactions can be important in aggregate formation and crystal formation.
The development of high-performance covalent organic framework (COF)-based photocatalysts is hindered by their dense layer stacking, lack of intrinsic active sites, and limited light absorption. Herein, we report a "functionalization-assisted exfoliation" strategy that simultaneously addresses these challenges by integrating plasmonic Ni-Co-Mo sulfide nanocavities (NCMS) into a COF matrix. The in situ formation of interfacial CoN and MoN bonds acts as "molecular wedges" to exfoliate the bulk TP-BD covalent organic framework into ultrathin nanosheets (∼1.4 nm) while concurrently establishing covalent "electron highways" for rapid charge transport. The plasmonic NCMS core serves a dual role: it drives the structural transformation and provides dual-pathway plasmonic enhancement via intense photothermal heating (raising the local temperature to ∼99 °C) and hot-carrier generation, which collectively boost light harvesting, optimize reaction kinetics, and elevate the electron reservoir density. The optimized hybrid, denoted MT-3, exhibits exceptional photocatalytic hydrogen evolution rates of 95.6 and 56.4 mmol g-1 h-1 under full-spectrum and visible-light irradiation, respectively, with an apparent quantum yield of 25.6% at 420 nm. This performance represents a 956-fold enhancement over the pristine COF and surpasses that of the benchmark Pt-loaded COF by a factor of 18. Comprehensive spectroscopic and microscopic studies elucidate the synergistic interplay of morphological engineering, plasmonic energy conversion, and interfacial electronic coupling. This work provides a versatile design principle for converting inert organic semiconductors into efficient, stable photocatalysts for solar-to-fuel conversion.
Organic electrodes suffer from poor active site accessibility, sluggish charge transport, and structural degradation upon cycling, limiting their practical application for energy storage. To address these challenges, this work elucidates a precise electronic and structural modulation strategy for polyimide (PI) via polyoxometalate (POM) hybridization. The key advancement lies in the multiple regulatory effects imparted by POM, enabling the construction of novel hybrid electrodes for high-performance SIBs. Specifically, the covalently anchored phosphomolybdic acid (PMo12) clusters disrupt π-π stacking to expose abundant active C[double bond, length as m-dash]O sites and serve as electron-withdrawing modulators to lower the LUMO level, thereby enhancing Na+ uptake and transport kinetics. Simultaneously, they function as an electron-buffering reservoir to dissipate charge accumulation during discharge, preventing structural degradation of the PI matrix. This multi-scale synergy endows the PI-PMo12 anode with significantly improved reversible capacity, rate capability, and cycling stability, offering a promising molecular engineering strategy for developing organic-inorganic hybrid electrodes in next-generation energy storage systems.
In-home spatiotemporal data, such as the movement trajectory data and the spatial time series data, contains potential predictive utility for detection of geriatric conditions including Mild Cognitive Impairment (MCI), frailty, and cognitive frailty. However, few have explored spatiotemporal learning models for learning and fusion of such disparate spatiotemporal data, owing to the lack of a generalized machine learning model that can jointly model these different spatiotemporal data types. This work reports a multimodal spatiotemporal machine learning model based on a class of self-organizing neural networks that can integrate different spatiotemporal data types for MCI detection. Specifically, Episodic Memory Adaptive Resonance Theory (EM-ART) and SpatioTemporal Episodic Memory (STEM) were employed to model movement trajectory and spatial time-series data of in-home room trips, respectively. For detecting behavioral changes, a contrastive layer was added to the individual models, generating latent representative features normalized for the different layouts of homes. A three-channel Fusion ART layer was stacked above the contrastive layers, integrating the latent representative features of the two models for the MCI classification task. Using a longitudinal real-world dataset collected from high-frequency passive infrared motion sensors paired with annual neuropsychological assessments over a period of two years, the EM-ART and STEM latent representative features achieved a Receiver Operating Characteristics-Area Under the Curve (ROC-AUC) of 0.6-0.7, demonstrating the efficacy of each model for MCI detection. By integrating the two models, the multimodal contrastive model achieves predictive accuracy and F1 rates of 84.2% and 66.7%, respectively, outperforming state-of-the-art models including Support Vector Machines (SVM) and Long Short-Term Memory (LSTM) based on cyclomatic complexity features in terms of specificity and accuracy, paving the way for clinical possibility to early detect geriatric conditions, particularly the MCI.
GaN p-channel field-effect transistors (p-FETs) are critical for enabling complementary integration of power and logic, while offering high breakdown capability. However, achieving high on-current in GaN p-FETs remains a significant challenge, primarily due to the low hole concentration, poor mobility, and challenges in forming stable, low-resistance ohmic contacts. Here, we present a novel GaN p-FET architecture that exhibits unconventional electron conduction, which helps enhance both thermal response and current modulation. We conduct a comprehensive investigation of heavily Mg-doped p++-GaN layers, focusing on contact optimization for high-temperature operation. Structural and interfacial characterization confirms high crystal quality and a thermally robust Ni/Au contact stack stabilized by an interfacial NixOy layer. This strategic interfacial layer, combined with moderate-temperature annealing, promotes Mg activation and suppresses oxygen-related traps, resulting in a ~ 73% reduction in contact resistance. Temperature-dependent analysis further reveals non-monotonic Schottky barrier modulation driven by interface evolution. Integrated into the device, this contact strategy enables thermally enhanced operation, with a drastically increase in on-state current and threshold voltage shifting positively by ~ 69% with temperature. These findings highlight a model shift in GaN p-FET design, where interface study and transport-mode innovation enable high-performance, thermally resilient devices for next-generation power integration.