搜索结果：End-to-end

Background/Objectives: Spontaneous intracranial hypotension (SIH) is caused by spinal cerebrospinal fluid (CSF) leakage and is typically diagnosed by clinical presentation and characteristic MRI signs; however, objective tools for monitoring physiological changes and treatment response remain limited. Cine phase-contrast MRI (PC-MRI) enables noninvasive quantification of aqueductal CSF dynamics, yet reliable analysis is challenging since the cerebral aqueduct is extremely small and susceptible to low contrast, partial volume effects, and ROI-dependent measurement variability-particularly in SIH where CSF pulsatility is often reduced. Methods: We propose an end-to-end automated framework that integrates (1) a cascade localization-segmentation strategy, consisting of Tiny YOLOv4 detection followed by MultiResUNet segmentation on a YOLOv4-derived cropped ROI; (2) physiology-informed pulsatility-based segmentation (PUBS) to refine anatomical masks into functional flow ROIs; and (3) one-dimensional convolutional neural networks (1D-CNNs) to extract exploratory waveform morphology features from 32-phase cardiac-cycle velocity waveforms. The study includes 39 participants, yielding 59 cine PC-MRI examinations: 11 controls, 28 Pre-treatment SIH scans and 20 Post-treatment Recovery scans. Results: The cascade model significantly improves segmentation robustness compared with a full-image baseline, achieving higher Dice scores and markedly lower boundary errors across cohorts (e.g., Pre-treatment SIH HD95: 1.66 ± 0.74 px vs. 15.37 ± 44.98 px). PUBS refinement reduces quantification deviation from expert manual references in SIH (mean relative error: 7.4% to 5.6%) and improves diagnostic performance for multiple hemodynamic parameters (e.g., downward mean flow AUC: 0.747 to 0.792). For waveform morphology analysis, the end-to-end 1D-CNN classifier was evaluated using repeated-seed participant-level grouped LOOCV. The repeated-seed ensemble prediction showed modest out-of-sample discrimination between Normal controls and Pre-treatment SIH scans, with an AUC of 0.646, a bootstrap 95% confidence interval of 0.455-0.826, and a permutation-test p-value of 0.072. Separately, exploratory analysis of the final baseline-trained 1D-CNN latent space showed marked, apparent Normal-versus-SIH separability and an intermediate recovery distribution in PCA space, suggesting that aqueductal waveform morphology may encode SIH-related physiological information. Conclusions: These findings suggest that SIH-related information may be reflected not only in flow magnitude but also in aqueductal CSF waveform morphology. However, the modest and statistically non-significant out-of-sample performance of the end-to-end 1D-CNN classifier indicates that morphology-based AI features should currently be regarded as exploratory biomarker candidates rather than validated stand-alone diagnostic tools. Larger independent cohorts are required to confirm their reproducibility, physiological meaning, and clinical utility.

Accurate detection and fine-scale annotation of dolphin whistles are crucial for understanding marine mammal communication and population dynamics. Dolphin-whistle annotation is challenging due to highly variable signals, overlapping calls including echolocation clicks, attenuation, and noisy backgrounds. Most existing methods treat the task as a spectrogram peak-tracking problem, linking neighboring detected peaks using heuristic or statistical methods, and rely on manual feature engineering. While effective for long, clear whistles, they lack generalizability for short, weak whistles across species. We reformulated dolphin-whistle detection as an instance-segmentation task, introducing an end-to-end transformer model that predicted complete whistle contours directly from spectrograms, eliminating peak-detection and trajectory-reconstruction stages. To overcome manual labeling limitations, we integrated a human-in-the-loop training paradigm that iteratively refined annotations, improving both data quality and model performance. We demonstrated the effectiveness of this architecture on a subset of the detection, classification, localization, and density estimation 2011 corpus where we partitioned training and test data such that test data were from species, locations, and hydrophones that were excluded from the training data. Experiments showed that our end-to-end system generalized effectively in these conditions, achieving 89.99% precision and 80.65% recall for all whistles, and 85.81% precision with 88.44% recall for whistles longer than 150 ms.

Extracting key information from vast amounts of documents and data plays a crucial role in knowledge graph construction, intelligence analysis, decision support, and multimodal information retrieval (such as speech sentiment analysis and invoice error detection). While end-to-end OCR-free methods avoid the error propagation issues of traditional two-stage models, they often struggle to balance the extraction of fine-grained character details with the modeling of complex global layouts. To address this, this paper proposes a novel hybrid encoder architecture that synergizes the inductive bias of Convolutional Neural Networks (CNNs) with the global context modeling of Swin Transformers. Unlike standard symmetric architectures, we introduce a geometry-aware asymmetric downsampling strategy: a ConvNext (CN) module first compresses the height to retain horizontal resolution for character distinction, followed by a Swin-T module that reduces width to capture long-range row-column dependencies. Experimental results on the CORD and IIT-CDIP datasets demonstrate that the proposed method outperforms other OCR-free end-to-end information extraction methods in terms of information extraction accuracy and shows potential in advancing intelligent operations and maintenance.

Crack detection in mobile inspection scenarios is constrained by both the extremely slender geometry of crack targets and the real-time inference requirements on edge devices, which expose systematic limitations of general-purpose object detectors. This paper proposes YOLO-Crack, a closed-loop solution that couples geometry-statistics-driven module design with end-to-end edge deployment validation. On the algorithmic side, we first quantify crack geometric properties and then introduce (i) a crack-aware cross-dimensional fusion attention (CFCA) module to strengthen feature representations, (ii) a dual-path feature enhancement module (DFEM) to preserve fine details during upsampling, and (iii) an empirical smooth quality window adjustment with shape consistency regularization to stabilize bounding-box regression for slender cracks. Experiments on the Crack500 dataset show that YOLO-Crack achieves 78.8% precision, 51.4% recall, and 65.7% mAP@0.5, improving over the YOLOv11n baseline by 4.2, 1.7, and 2.9 percentage points, respectively. On the engineering side, we deploy YOLO-Crack on a Jetson Orin NX mobile robot platform and evaluate it in a real ROS pipeline; the measured end-to-end throughput reaches 25.5 FPS, meeting real-time video processing requirements. The proposed framework provides a practical reference workflow for edge vision tasks, from geometry analysis to engineering verification.

Named Data Networking (NDN) represents a paradigm shift toward content-centric architectures but remains critically vulnerable to Interest Flooding Attacks (IFAs), where malicious actors overwhelm router Pending Interest Tables with spurious requests, causing service degradation and denial-of-service. To address the limitations of existing approaches, including high false positives in threshold-based methods and substantial overhead in centralized learning, we propose FL-IFAshield, a novel federated learning framework for adaptive IFA mitigation. Our solution integrates dynamic Poisson-EMA thresholding for accurate flood detection, entropy-aware federated aggregation to handle non-IID traffic distributions across edge routers, and Byzantine-robust mechanisms with differential privacy guarantees. Comprehensive evaluation on the FIT/IoT-LAB testbed with 100 routers demonstrates exceptional performance: 93.1% F1-score in attack detection, only 5% false positives, 28 ms average end-to-end latency ([Formula: see text]), and over 90% legitimate Interest Satisfaction Ratio under sophisticated collusive attacks, while maintaining minimal computational overhead (<9% CPU utilization on ARMv8 routers). FL-IFAshield significantly improves security performance, offering 35% higher accuracy than static thresholding and 60% lower communication overhead than centralized approaches. While simpler heuristic baselines naturally incur marginally lower computational footprints, our solution delivers the optimal overall operational balance among high precision, low end-to-end latency ([Formula: see text]), and resource efficiency in constrained edge computing environments.

Intrinsically disordered proteins (IDPs) play critical roles in cellular signaling and regulation, yet their dynamic conformational landscapes make them difficult to characterize experimentally and computationally. Phosphorylation, one of the most common post-translational modifications, frequently occurs within intrinsically disordered regions and can modulate protein structure and function. Enhanced sampling molecular dynamics methods offer a potential route to more efficiently explore the diverse conformations accessible to IDPs, but systematic comparisons of their performance sampling the conformational landscape of such systems remain limited. Here, we evaluate the conformational sampling of the intrinsically disordered &#x3b2;-catenin17-48 peptide in both its nonphosphorylated and phosphorylated states using three enhanced sampling approaches: Gaussian-accelerated molecular dynamics (GaMD), metadynamics (METAD), and weighted ensemble simulations (WESTPA), in comparison with conventional molecular dynamics simulations. Two collective variables (CVs) were explored to guide sampling: the &#x3d5; dihedral angles of the phosphorylation sites Ser33 and Ser37 and the end-to-end distance of the peptide. We found that different enhanced sampling methods explored distinct regions of conformational space rather than converging to a single ensemble, with GaMD largely overlapping with unbiased simulations. Notably, METAD and WESTPA more readily accessed conformational regions not observed in unbiased simulations. Analysis of the combined conformational ensembles identified intermediate conformations connecting the nonphosphorylated and phosphorylated states, which are preferentially sampled in simulations employing adaptive strategies. Additionally, the Ser33/Ser37 &#x3d5; angle CV more effectively captures phosphorylation-dependent conformational shifts than the end-to-end distance metric. Together, these results highlight how both the choice of enhanced sampling strategy and the selection of collective variables influence the exploration of IDP conformational landscapes.

Microsurgery is a specialized field within plastic surgery constrained by the high cost and limited availability of surgical microscopes, particularly in resource-limited settings. This lack of access hinders trainees from developing essential microsurgical skills. Previous efforts have explored smartphone-based alternatives, but there remains a need for high-fidelity, accessible training microscopes. This study presents a novel, low-cost, travel-friendly surgical microscope designed for microsurgery training. Modified binocular objective lenses enable near-field stereoscopic viewing using a dual-mirror array. A 3D-printed chassis provides the correct top-down orientation, and an integrated light source operates on standard or battery power, ensuring usability in low-resource environments. Initial testing demonstrated a fixed 6.5&#xd7; magnification, allowing trainees to perform end-to-end anastomoses on 2-mm vessels. Prototypes were deployed for microsurgery training in the United States, Rwanda, Ethiopia, and Vietnam through the SHARE (Surgeons in Humanitarian Alliance for Reconstruction, Research and Education) plastic surgery organization, Nuoy Reconstructive International, and the senior author's home institution. The authors' investigation continues to assess the effectiveness of the microscope compared with state-of-the-art surgical microscopes. By offering a low-cost, portable solution without compromising image quality, this innovation has begun to transform microsurgical education and operating room-based microscopy worldwide, increasing accessibility for trainees and patients in diverse settings.

Point-cloud analysis of sedimentary outcrops using Unmanned Aerial Vehicle (UAV) oblique photogrammetry is a crucial approach to sedimentary system characterization, stratigraphic correlation, and petroleum exploration analog studies. In large-scale field settings, however, outcrops are often scattered and fragmented, vegetation and soil cover is extensive, and class imbalance is pronounced. Manual interpretation is labor-intensive, while existing clustering algorithms, conventional machine learning methods, and general-purpose point-cloud segmentation networks struggle to simultaneously ensure geometric fidelity, rare-class recognition, and multi-scale feature integration. To address these challenges, we propose a method for extracting sedimentary outcrop point clouds from field surface point clouds using a UAV oblique photogrammetry acquisition strategy. The core segmentation module of the method, sedimentary cross-scale self-attention network (SedCSA-Net), is an enhanced version of PointNet++ that integrates collaborative improvements across four dimensions: data augmentation, sampling strategy, feature encoding, and loss optimization. Taking the Cretaceous Qingshuihe Formation in the Louzhuangzi area of the southern Junggar Basin as a case study, our experimental results indicate that SedCSA-Net overcomes the natural variability of UAV oblique photogrammetry point clouds-such as shadows, voids, and uneven density-achieving a mean Intersection over Union(mIoU) of 89.51% and an Overall Accuracy(OA) of 96.08%, with an outcrop-class Intersection over Union(IoU) of 86.90%. Attitude measurements derived from segmentation results deviate by less than 3&#xb0; from manually annotated references, demonstrating that the proposed framework provides an end-to-end, generalizable approach for intelligent segmentation, geometric reconstruction, and attitude extraction of large-scale sedimentary outcrop point clouds.

Laser powder bed fusion (L-PBF) has emerged as a core metal additive manufacturing technology for high-end sectors, including aerospace and medical device manufacturing. However, melting anomalies that occur during fabrication accumulate layer by layer, leading to degraded surface quality and impaired mechanical performance of as-built components-a critical bottleneck limiting their large-scale industrial adoption. Accurate and robust layer-wise melting quality recognition remains a challenge due to the complex surface morphologies induced by such melting anomalies. This study presents a machine learning-enabled in situ monitoring approach for layer-wise melting quality identification in L-PBF. By systematically varying laser power and scanning speed, 24 parameter combinations were designed to fabricate specimens with three distinct melting states: over-melting (OM), lack of fusion (LOF), and normal melting. A high-resolution complementary meta-oxide-semiconductor (CMOS) camera was used to capture layer-wise surface images of the specimens, and following abnormal layer filtering and manual validation, a high-quality dataset comprising 5110 layer-wise images was constructed. Two mainstream machine learning approaches were systematically evaluated and optimized for melting quality classification: a support vector machine (SVM) model leveraging handcrafted gray-level co-occurrence matrix (GLCM) texture features achieved a classification accuracy of 96.77%, while a convolutional neural network (CNN) model with end-to-end feature learning directly from raw images attained a superior accuracy of 98.14%. In terms of computational efficiency, the CNN model exhibited a faster inference speed with a per-layer inference time of just 0.036 s, nearly half that of the SVM model (0.068 s per layer). Most critically, the CNN model completely eliminated fatal cross-class misclassification between OM and LOF-an error mode common in the SVM model that would trigger erroneous process corrective actions in practical industrial applications. The findings demonstrate that image-based machine learning provides a reliable technical foundation for intelligent in situ monitoring of the L-PBF process. With its high accuracy, strong robustness, and superior computational efficiency, the CNN model can effectively support on-site operational decision-making, reduce material and time losses, and enhance process stability in industrial settings, thus exhibiting significant potential for practical engineering deployment.

As core equipment in high-end manufacturing, computer numerical control machine tools depend critically on the health of their feed systems, which directly affects machining quality and efficiency. To address fault diagnosis challenges under variable-speed and strong-noise conditions, this paper proposes a deep learning model named DARTS-CNN-BiLSTM. The key novelty lies in the first systematic integration of differentiable architecture search (DARTS) with a hybrid CNN-BiLSTM framework. DARTS automatically optimizes the convolutional neural network structure for spatial feature extraction, while the bidirectional long short-term memory (BiLSTM) captures bidirectional temporal dependencies. Global average pooling is used for feature reduction, and a softmax classifier enables end-to-end fault classification. This automated design eliminates the need for manual network tuning and feature engineering. Experimental results on two public datasets and a self-built dataset demonstrate that the proposed method outperforms advanced models such as Inception-BiLSTM and DenseNet. Specifically, our method maintains over 90% diagnostic accuracy under strong noise (signal-to-noise ratio &#x2265; $ \ge $ -6&#xa0;dB) and achieves 98.15% average accuracy on a variable-speed dataset. Ablation studies confirm the advantage of automated architecture design over manually tuned counterparts. These results validate the effectiveness and superiority of the proposed method for complex feed system fault diagnosis.

Remote sensing object detection remains challenging because objects often appear with large scale variation, dense spatial layouts, and strong interference from complex geographical backgrounds. To address these coupled difficulties, we propose EDM-Net, an end-to-end multi-scale detector that organizes feature processing into three coordinated stages: adaptive extraction, intra-scale interaction, and cross-scale fusion. First, an efficient sparse mixture-of-experts (ES-MoE) module is embedded in the backbone to allocate scale-specific convolutional experts according to scene-level feature responses, providing a more adaptive feature basis than a single static extraction path. Second, a dynamic mixing intra-scale feature interaction (DMIFI) module is introduced into the Transformer encoder. This module combines global self-attention with dynamic spatial mixing, thereby preserving long-range context while reintroducing local two-dimensional inductive bias for dense and small objects. Third, a multi-scale synergistic attention fusion (MSAF) module aligns adjacent feature levels through parallel local and global attention branches and structural re-parameterization, reducing semantic dilution during feature aggregation. Comprehensive experiments on three large-scale remote sensing benchmark datasets, DIOR, NWPU VHR-10, and RSOD, demonstrate that EDM-Net consistently improves over the re-trained RT-DETR-R18 baseline under the same experimental protocol, attaining mAP50 scores of 83.7%, 95.6%, and 95.8% respectively. Additional ablation and scale-specific analyses indicate that the three modules contribute complementary gains, especially for small and densely distributed objects. These results suggest that coordinated extraction, interaction, and fusion can improve remote sensing object detection under complex scale and background conditions.

Real-time biomechanical feedback during table tennis training demands both low latency and high recognition accuracy, yet existing systems sacrifice one for the other due to cloud-transmission delays and the computational constraints of edge devices. This paper presents EdgeFusionNet, an integrated edge-cloud collaborative architecture that delivers actionable stroke-level feedback within 32&#xa0;ms under realistic network conditions. At its core, a lightweight cross-modal attention fusion network (LCA-FNet) fuses temporally aligned features from high-speed vision, inertial measurement, and surface electromyography streams through shared-projection cross-modal attention and adaptive channel gating, achieving 93.6&#x2009;&#xb1;&#x2009;0.4% recognition accuracy (Macro-F1&#x2009;=&#x2009;0.927&#x2009;&#xb1;&#x2009;0.005, mean&#x2009;&#xb1;&#x2009;SD over five seeds) across seven canonical stroke types with only 1.48 million parameters, and retaining 90.4&#x2009;&#xb1;&#x2009;2.3% accuracy under a strict leave-one-subject-out evaluation. A hardware-triggered synchronization mechanism maintains sub-millisecond cross-modal alignment, while a two-level knowledge distillation strategy recovers 97.8% of the cloud-resident teacher model's accuracy after aggressive structural compression. An adaptive computation offloading agent, trained via Q-learning with explicitly defined state, action and reward spaces, dynamically partitions inference between the edge node and cloud server based on prediction entropy and network quality, sustaining sub-32&#xa0;ms P95 end-to-end latency even at 5 Mbps uplink bandwidth. Field deployment over thirty training sessions with twelve athletes per study arm confirmed ecological validity, yielding 91.8&#x2009;&#xb1;&#x2009;0.7% accuracy under uncontrolled gymnasium conditions, a Cohen's kappa of 0.874 against the consensus of two expert coaches (whose own inter-coach kappa was 0.892), and a 10.3-percentage-point gain in standardized multi-ball hit rate over a matched control group after four weeks (p&#x2009;<&#x2009;0.001). These results demonstrate that principled co-design of multimodal fusion, model compression, and adaptive offloading can bridge the gap between laboratory-grade recognition performance and the stringent latency requirements of live athletic training.

In the resistor images used in this study, many defective regions are weak coating-like marks rather than obvious scratches or pits. Their appearance is close to the epoxy background, and some visible defects were missing from the original annotation files. If these labels are used directly, the detector treats the missed defects as background samples during training. We therefore corrected the supervision before changing the feature constraint. An early YOLO26s model was first used to nominate low-overlap boxes, and these candidates were then checked manually. Only confirmed defects were merged into the labels. After this step, a scale-gated prototype consistency term was added during training to reduce the model's bias toward the dominant tiny-defect group. On the fixed corrected benchmark, mAP50 improved from 28.14% to 63.20%, and Recall increased from 18.42% to 62.20%. In the end-to-end deployment view, where the raw and cleaned validation sets answer different practical questions, mAP50 changed from 43.66% to 63.15%, and Recall changed from 30.01% to 62.24%. For normal-size defects, Recall increased from 26.09% to 56.52%. A prototype-only transfer study on the public MVTec AD benchmark further evaluates whether the feature constraint generalizes when the label-repair stage is not applicable to clean public annotations. Since the prototype term is removed after training, the deployed detector remains the original YOLO26s model without an additional inference branch.

Single-lead electrocardiogram (ECG) is widely used in wearable devices for atrial fibrillation (AF) screening. Nevertheless, subtle pathological characteristics like P-waves and f-waves in practical signals are vulnerable to noise contamination. Meanwhile, the scarcity of high-quality annotated abnormal data instances leads to severe class imbalance. To mitigate these issues, we present an end-to-end framework designed for arrhythmia diagnosis using single-lead ECG signals, which integrates quality-aware data augmentation with a Peak-Enhanced attention mechanism. First, to mitigate the problem of data imbalance, a Quality-Aware Generative Adversarial Network (QA-GAN) is designed. This network integrates a signal quality evaluation module based on signal kurtosis, together with a dynamic soft-label training scheme, guiding the generator to prioritize learning high-quality morphological features, thereby synthesizing high-fidelity minority class samples. Second, to accurately capture subtle pathological features in electrocardiograms, a Peak-Enhanced Attention Convolutional Network (PEAC-Net) classification model is proposed. This model incorporates a Peak-Enhanced Attention (PE-Att) module, which employs learnable derivative convolutional kernels to precisely identify the transition points in the ECG signal. Furthermore, by integrating one-dimensional multi-scale dilated convolution (DSGC1D) with bidirectional LSTM, the model achieves effective capturing of both fine-grained local morphological features and long-range global rhythm patterns. Experimental results on the PhysioNet 2017 dataset indicate that the presented model attains an accuracy of 0.902 and a macro-F1 score of 0.880, respectively, outperforming other state-of-the-art models and also exhibiting robust data adaptability on the MIT-BIH dataset.

Rule-based multi-agent system (MAS) architectures for healthcare coordination rely on hardcoded decision trees that cannot generalise to novel clinical scenarios or self-correct reasoning errors. These limitations are acute in surgical continuum care, where patients traverse presurgical risk stratification, intraoperative monitoring, postsurgical ICU, ward care, and remote rehabilitation over days to weeks-a complexity no fixed-policy agent architecture can address without prohibitive rule engineering. We present the first agentic large language model (LLM) framework for autonomous end-to-end surgical continuum monitoring, superseding the prior rule-based MAS Digital Twin. Six ReAct-driven tool-use agents replace fixed-policy agents with dynamic reasoning, multi-hop evidence retrieval, and Reflexion self-correction while maintaining mandatory confidence-gated Human-in-the-Loop (HITL) gating at every care-pathway-modifying decision. The framework is grounded in the ReAct paradigm and Reflexion self-evaluation, embedded within the DETER Digital Twin state engine S(t). Each agent is specified by a ReAct loop signature, a ten-function clinical tool registry, and confidence-gated HITL escalation logic. Inter-agent coordination replaces the rule-based Priority Queue Manager with an LLM-mediated Coordination Supervisor Agent reasoning over competing resource requests. The framework delivers: (i) six formally specified ReAct-loop agents with explicit tool registries and authorisation boundaries; (ii) a confidence-gated HITL architecture that reduces alert fatigue while preserving safety for ambiguous clinical scenarios; (iii) an extended conflict resolution function P(p,t,context) incorporating surgical phase and DETER deterioration trajectory gradient; (iv) Reflexion self-correction with a formal N_max = 2 termination condition and Clinical Factuality Verification Layer; and (v) a multi-phase Digital Twin state engine extending S(t) to the full surgical continuum. The proposed framework represents a fundamental architectural departure from rule-based clinical AI-from hardcoded policies to dynamic reasoning, from static retrieval to multi-hop tool-use chains, and from fixed escalation thresholds to confidence-gated self-evaluation-providing a formally specified, clinically deployable foundation for next-generation autonomous surgical care coordination.