Search Results (1,022)

Real-time roadway surveillance now leans hard on automated detection. How a model trained in one geographic context actually behaves on another, though, is still underexplored for Southeast Asian deployments. We answer that question for Thai roadway closed-circuit television with a cross-domain validation framework. A YOLOv11n (Ultralytics v8.2.0; Ultralytics, Los Angeles, CA, USA) detector trained with focal loss feeds a confidence-aggregation step that turns per-detection scores into a per-frame accident score, and we put four aggregation operators head-to-head. Reliability comes from DeLong variance estimation paired with non-parametric bootstrap on 1245 Thai frames that carry 23 positive accident events. Under maximum-class aggregation the proposed configuration reaches a frame-level AUROC of 0.959 ± 0.020 across three random seeds. Under top-K aggregation it reaches 0.965 ± 0.018. Per-seed DeLong 95 percent intervals exclude chance performance throughout. We also evaluate three baseline configurations: YOLOv5su comes in at 0.738, YOLOv8n at 0.868, and a Chiang Mai-tuned YOLOv11n variant at 0.918. The architectural progression seen on standard benchmarks therefore carries cleanly into the cross-domain setting. The same Chiang Mai-tuned variant reached an in-domain mAP50 of 0.952 yet only 0.918 cross-region AUROC on a separate Thai region, which is a quiet but clear signal that geographic proximity within a country does not on its own remove distributional shift. Bounding-box localisation appears as a secondary diagnostic because the operational target here is frame-level alerting rather than pixel-precise annotation. Edge deployment optimisation falls outside the present scope. What the work leaves behind is a reproducible baseline and a statistical protocol that follow-up Southeast Asian roadway-safety research can build on. Full article

Background: Normothermic machine perfusion (NMP) is increasingly being used to improve organ utilization in liver transplantation (LT). However, its non-physiological perfusion setting may cause focal hepatic hypoperfusion (FHH), which remains insufficiently characterized in terms of its incidence, risk factors, and clinical impact. Methods: Data on liver grafts that underwent NMP prior to LT at the Department of General, Visceral, and Transplant Surgery, University Hospital Münster, between October 2019 and August 2024 were retrospectively analyzed. Recipients who underwent contrast-enhanced computed tomography within 30 days post-LT were included. The primary outcomes were the Comprehensive Complication Index (CCI) and overall graft survival rate. Ninety-one patients met the inclusion criteria and were stratified according to the presence of FHH in the FHH+ (n = 27) and FHH- (n = 64) groups. Results: FHH was detected in 29.7% of the grafts. Higher graft weight was the only independent predictor of FHH. In addition, graft weight correlated with the extent of FHH (τ = 0.40, p < 0.001). FHH did not affect graft or patient survival but was associated with higher CCI scores (p = 0.001) and prolonged intensive care unit length of stay (p = 0.028). Conclusions: FHH is a common radiological finding after NMP. Although it does not affect graft loss, its association with a higher complication burden warrants further attention. Whether avoiding NMP in very heavy grafts could reduce the incidence of FHH remains to be determined. Full article

Computed tomography angiography (CTA)-based segmentation of the superficial temporal arteries (STAs) and facial vessels (FVs) is important for neurosurgical and reconstructive planning. Nevertheless, segmentation of STAs and FVs remains challenging because of their small caliber, tortuous courses, and proximity to high-intensity bony structures. This study aims to develop a deep learning framework for accurate automated segmentation of these craniofacial vessels. A single-input 3D nnU-Net v2 model was trained using raw CTA volumes, while a Fusion-based Vesselness Map (FVM) was constructed from multiscale vessel-enhancement filters to emphasize small vascular structures and suppress irrelevant regions such as the skull and skin. Instead of being used as an additional input channel, the FVM was incorporated into the loss function as a spatial prior to guide the network toward vessel boundaries and distal branches. In 72 clinical cases, the FVM-guided model improved segmentation accuracy compared with a baseline model trained with Dice Focal Loss, particularly in boundary delineation. For the STAs, the Average Symmetric Surface Distance decreased from 6.543 mm to 2.941 mm. Qualitative evaluation further showed reduced segmentation noise and fewer false positives near bone and distal branches. These findings suggest that integrating classical vessel enhancement into deep learning supervision can improve morphologically consistent craniofacial vessel segmentation and support preoperative surgical planning. Full article

Background/Objectives: To explore potential associations between pediatric obesity and retinal and anterior segment ocular structures using OCT and ocular biometry. This study was designed as an exploratory, hypothesis-generating analysis without a pre-specified primary endpoint; all findings should be interpreted accordingly. Methods: This retrospective cross-sectional study included 52 children (104 eyes): 27 obese children (body mass index (BMI) percentile ≥95%) and 25 healthy controls (BMI percentile 5–85%). Optical coherence tomography (OCT) and ocular biometry were used to assess retinal nerve fiber layer (RNFL), ganglion cell complex (GCC), focal loss volume (FLV), global loss volume (GLV), Early Treatment Macular Map 5 (EMM5), corneal parameters, axial length (AL), anterior chamber depth (ACD), and white-to-white corneal diameter (WTOW). Group comparisons and cluster-robust bootstrap regression adjusted for inter-eye dependency, age, and sex; Bonferroni correction was applied. Results: Obese children showed nominally higher GCC average thickness, RNFL, and EMM5 values and shallower ACD; however, no parameter survived Bonferroni correction. ACD showed the most internally consistent exploratory pattern (unadjusted p = 0.006; adjusted p = 0.018; Bonferroni p = 0.249); however, this finding did not survive Bonferroni correction and should not be interpreted as a confirmed association. Other corneal and biometric parameters were not significantly different. Conclusions: Pediatric obesity may be associated with subtle ocular structural variations, but all findings are exploratory and hypothesis-generating. Larger prospective, pre-registered studies are needed to determine whether pediatric obesity is associated with structural ocular changes. Full article

Maritime search and rescue is an important component of emergency response frameworks and primarily relies on Unmanned Aerial Vehicles (UAVs) for maritime object detection. However, maritime accidents frequently occur in low-visibility environments, such as foggy or low-light conditions, which lead to low contrast, blurred object boundaries, and degraded texture representations. Most existing maritime object detection algorithms are developed for natural light scenes, and their performance deteriorates markedly when deployed directly in low-visibility environments, primarily due to reduced image quality that hinders feature extraction and semantic information aggregation. Although several studies incorporate image enhancement techniques prior to detection to improve image quality, these approaches often introduce significant additional computational overhead, limiting their practical deployment on UAV platforms. To tackle these challenges, this paper proposes a lightweight model built upon a recent YOLO framework, termed Multi-Scale Adaptive YOLO (MSA-YOLO), for maritime detection using UAVs in low-visibility environments. The proposed model systematically optimizes the backbone, neck, and detection head networks. Specifically, an improved StarNet backbone is designed by integrating Efficient Channel Attention (ECA) mechanisms and multi-scale convolutional kernels, which strengthen feature extraction capability while maintaining low computational overhead. In the neck network, a high-frequency enhanced residual block branch is inserted into the C3k2 module to capture richer detailed information, while depthwise separable convolution is utilized to further reduce computational cost. Moreover, a non-parametric attention module is incorporated into the detection head to adaptively optimize features in the classification and regression branches. Finally, a joint loss function that combines bounding box regression, classification, and distribution focal losses is utilized to improve detection accuracy and training stability. Experimental results on the constructed AFO, Zhoushan Island, and Shandong Province datasets demonstrate that, relative to YOLOv11-s, MSA-YOLO reduces model parameters and FLOPs by 52.07% and 41.36%, respectively, while achieving improvements of 1.11% and 1.33% in mAP@0.5:0.95 and mAP@0.5. These results indicate that the proposed method effectively balances computational efficiency and detection accuracy, rendering it suitable for practical maritime search and rescue applications in low-visibility environments. Full article

Dendritic cells (DCs) are key antigen-presenting cells essential for the initiation of immune responses. Their migration is tightly regulated by adhesive structures, including podosomes and focal adhesions (FAs), allowing for interactions with the extracellular matrix (ECM) for coordinated cell movement. The organization and dynamics of these structures are controlled by actin and microtubule cytoskeletons; however, the mechanisms governing their balance in distinct DC subsets are not completely understood. In this study, we investigated cytoskeletal regulation of the interplay between podosomes and FAs in GM-CSF-derived inflammatory-like DCs (GM-BMDCs) and Flt3L-derived conventional DCs (FL-BMDCs). GM-BMDCs showed a higher capacity to form podosomes compared with FL-BMDCs, which exhibited fewer and less prominent structures. Actin depolymerization resulted in the complete loss of podosomes, whereas disruption of microtubules induced podosome reorganization and altered the structure of FAs. Importantly, cytoskeletal perturbation in both DC subsets led to podosome dissolution, highlighting the requirement of cytoskeletal integrity for their maintenance. Furthermore, actin integrity was essential for podosome-mediated ECM degradation and efficient migration of GM-BMDCs, while microtubules fine-tuned the balance between podosome and focal adhesion dynamics, thereby regulating DC motility. Full article

This study evaluates the accuracy of various preprocessing methods applied to scanned archival aerial photographs for reconstructing historical terrain in the Czech Republic. Seven workflows were tested on identical imagery and control points, varying parameters such as resolution unification, brightness normalization, focal length calibration, and AI-based denoising. Accuracy was assessed using GNSS checkpoints and high-resolution LiDAR data. Results show that basic brightness correction reduced the vertical RMSE by 59% (to 5.69 m). In contrast, standalone AI preprocessing was associated with increased geometric instability (RMSE 16.48 m) due to over-smoothing and the loss of essential micro-texture. However, the evaluated “Fusion AI” workflow—combining AI enhancement with strict focal length constraints—successfully mitigated this degradation. By restricting the internal orientation, it stabilized the vertical accuracy at 6.48 m, closely matching the best traditional approaches. Statistical analysis revealed strong spatial autocorrelation and non-normal error distributions, highlighting the need for robust validation. Ultimately, this study confirms that AI can be effectively utilized to enhance visual clarity in data-scarce historical reconstruction without sacrificing spatial reliability, provided it is strictly geometrically constrained. This offers an optimal compromise and a tested, reproducible workflow that supports heritage preservation and long-term environmental analysis. Full article

Drainage pipe defect detection is essential for maintaining the normal operation of urban infrastructure. In recent years, deep learning-based object detection methods have provided an effective technical solution for drainage pipe defect recognition. Among them, YOLO-series models have demonstrated strong potential in visual detection tasks due to their end-to-end architecture and high inference efficiency. However, directly applying baseline YOLO models may still face challenges such as limited detection accuracy, relatively high model complexity, and insufficient adaptability for lightweight deployment scenarios. To address these issues, this paper proposes a lightweight drainage pipe defect detection method based on an improved YOLO11 network. Rather than treating detection enhancement and model compression as two separate procedures, the proposed method integrates feature enhancement, adaptive pruning, and distillation-based recovery into a unified lightweight detection framework. Specifically, an improved SimAM attention mechanism is introduced into the backbone and integrated with the C3k2 module to construct the C3K2_SWS module, aiming to enhance the representation capability of critical defect features. In the neck network, a focused diffusion pyramid network with a dimension-aware selective fusion structure, termed FDPN-DASI, is designed to strengthen multi-scale feature interactions. In addition, an adaptive-threshold focal loss (ATFL) is introduced to improve the learning capability for hard samples. For efficient deployment, the LAMP pruning algorithm is further improved, and an entropy-guided entropy-adaptive magnitude-based pruning method (EA-LAMP) is proposed to enable adaptive allocation of pruning ratios across different network layers. Moreover, BCKD knowledge distillation is applied after pruning to mitigate the accuracy degradation caused by model compression. Experimental results indicate that the proposed lightweight YOLO11-SFA+EA+BCKD framework achieves a precision of 92.4%, a recall of 88.5%, and an mAP50 of 93.3%, while maintaining a compact model size of 1.6 M parameters and 4.5 G FLOPs. Compared with the baseline model, the proposed method improves precision, recall, and mAP50 by 5.9%, 5.0%, and 4.7%, respectively, while reducing the number of parameters, FLOPs, and model size by 1.0 M, 1.8 G, and 2.1 M, respectively. These results suggest that the proposed framework can improve detection performance while reducing model complexity under the current experimental setting, indicating its potential for lightweight drainage pipe defect detection tasks. Full article

Background/Objectives: Deep learning models for digital mammography sensor data are increasingly deployed across hospitals using different X-ray detector technologies and patient populations. Whether models trained on one sensor platform and population maintain accuracy when transferred to another has not been tested for the latest generation of mammography-specific foundation models under one controlled protocol. Methods: We fine-tuned five backbone architectures (ResNet-50, DINOv2-B14, Rad-DINO, Mammo-CLIP B5, and Mammo-FM) on CBIS-DDSM (film-digitized, USA, n = 714 validation) with three seeds, ablated a density-aware focal loss across three auxiliary weights, and evaluated transfer to three external sensor cohorts: CMMD (full-field digital, China, n = 1032), DMID (mixed digital, India, n = 509), and MIAS (film-digitized, UK, n = 322). Significance used paired DeLong z-tests with Benjamini–Hochberg FDR correction; temperature scaling tested post hoc recalibration at all transfer targets. Results: Within this single-source three-seed evaluation, ResNet-50 outperformed all four foundation models on CBIS-DDSM (AUC 0.867 vs. 0.847, 0.846, 0.813, and 0.703; all gaps p_adj < 0.05). The density-aware focal loss degraded both AUC and calibration at every weight tested. At transfer, every model lost 0.165 to 0.320 AUC points relative to in-distribution performance, with sensitivity at 95% specificity collapsing from 0.31 to 0.47 in-distribution to 0.11 to 0.22 across the three external targets. A per-seed Stouffer meta-analysis confirms that Mammo-CLIP B5 and Mammo-FM significantly outperformed ResNet-50 on DMID and Mammo-CLIP on CMMD, after BH-FDR; MIAS comparisons remained directional only. In the extremely dense subgroup (BI-RADS D4), Mammo-FM reached AUC 0.870 versus ResNet-50 at 0.842, a directional observation whose 95% CIs overlap heavily at the n = 140 sample size and which we do not interpret as a statistically supported advantage. Conclusions: In this single training-source, three-seed protocol, mammography-specific pretraining did not deliver the in-distribution AUC premium reported in the originating papers, and no architecture reached a level at which transfer deployment without local validation would be defensible. We frame these as observations specific to the present protocol rather than as broader conclusions about foundation models for mammography classification. The findings argue for sensor-stratified and population-stratified external validation and for local recalibration as practical prerequisites before clinical use. Code and weights are released under MIT license. Full article

Equipment fault prediction in maglev train systems poses substantial challenges: fault events are inherently rare, class distributions are severely imbalanced, and individual equipment units are subject to complex spatial and functional couplings that single-device statistical approaches fundamentally cannot capture. To address these challenges, this study proposes a Knowledge Graph-driven Graph Neural Network (KG-GNN) framework. A fault knowledge graph encompassing equipment, fault, temporal, and environmental entities is constructed to unify multi-source maintenance data. Graph connectivity is established via three spatial relation types (co-location, co-zone, and co-level), with edge weights derived from Laplacian-smoothed Lift scores quantifying fault co-occurrence strength. A two-layer GATv2Conv-based graph attention network is designed: the first layer employs four-head attention with explicit edge-weight integration to capture heterogeneous neighborhood influences, while the second layer produces compact node embeddings via single-head attention. A Top-20 sparsification strategy suppresses weak-association noise, and training under severe class imbalance is stabilized through Focal Loss and F2-Score-guided early stopping. On the test set, the proposed method achieves an F2-Score of 0.5703, Recall of 0.6825, and AUC-ROC of 0.9329 (single-run evaluation); multi-seed evaluation (5 seeds) yields F2 = 0.5645 ± 0.0035, Recall = 0.6789 ± 0.0095, and AUC-ROC = 0.9298 ± 0.0026, outperforming the MLP baseline by 18.3% in F2-Score and substantially exceeding GCN (F2 = 0.1476 ± 0.0176) and GATConv (F2 = 0.4284 ± 0.0097). Ablation studies confirm the individual contributions of authentic graph topology, precise edge weighting, and graph sparsification to overall performance. Full article

Obtaining high-quality labeled data for supervised learning is costly, motivating the use of crowdsourcing, which distributes the annotation process across multiple workers with varying levels of expertise. A key challenge in crowdsourced data is annotation sparsity, as each worker labels only a limited subset of instances. This sparsity can amplify class imbalance, reduce supervision for minority classes, and bias standard cross-entropy-based models toward the majority classes. To address this problem, we propose a correlated chained Gaussian process framework trained on a focal-loss-based variational objective (CCGPFL). This probabilistic framework jointly models latent ground-truth and instance-dependent annotator reliability while accounting for correlations among annotators. In addition, the focal-weighted objective mitigates the imbalance induced by sparse annotations by assigning greater importance to harder examples during training. Experiments on synthetic, semi-synthetic, and fully real multi-annotator datasets show that CCGPFL achieves competitive and often superior performance relative to state-of-the-art learning-from-crowds baselines in terms of Overall Accuracy (OA) and Area Under the ROC Curve (AUC). Full article

To address the challenges of detecting sugar orange fruits in complex natural orchard environments—where fruits are frequently occluded by leaves and branches and may be mutually occluded due to dense growth, leading to missed detections, false positives, and low detection confidence—we propose an improved algorithm based on YOLOv8n, named YOLO-FSEP. A Spatial-Channel Synergistic Attention (SCSA) module is introduced into the main network to enhance feature extraction capabilities; the IoU loss function is replaced with Focal_SIOU to improve the detection accuracy for difficult samples; and an SE attention mechanism is embedded in the detection head, with the addition of a P6 high-resolution detection layer to optimize multi-scale object performance. Experimental results on a self-built sugar orange dataset show that, compared to the baseline YOLOv8n, the improved model achieves a 0.9% increase in accuracy, a 1.3% increase in recall, and a 3.2% increase in mAP50-95, while maintaining an inference speed of 62.6 FPS. To evaluate the model under dynamic conditions, we performed a 200-frame continuous test of the 3D localization pipeline on a laptop with a RealSense D435i camera. The average YOLO inference time was 49.90 ms, post-processing (depth extraction and 3D coordinate conversion) took 0.24 ms, and the total processing time was 50.15 ms. Given that the typical response time for a robotic arm’s single positioning operation is 100–200 ms, this real-time performance meets the dynamic localization requirements of sugar orange harvesting. Full article

Accurate segmentation of the ureter on abdominal computed tomography (CT) remains challenging due to its thin tubular structure and limited expert-annotated training data. While recent deep learning approaches have shown promise on non-contrast CT, arterial-phase imaging remains under-researched. We systematically compared nnU-Net-based configurations for ureter segmentation on arterial-phase CT using 25 radiologist-annotated cases from Seoul St. Mary’s Hospital. Seven training strategies were evaluated with five-fold cross-validation: binary ureter-only segmentation, multi-organ training with anatomical context from eight structures, alternative encoder architectures (ResEncM), specialized loss functions (Tversky, clDice), and a multi-phase fusion architecture. Multi-organ training with Tversky-Focal loss (Config 6) achieved the highest mean Dice of 0.743 ± 0.021 with the best clDice connectivity score (0.800 ± 0.046) and lowest fragmentation (6.56 connected components). Multi-phase fusion yielded a mean Dice of 0.713 on the 12-case subset; a controlled arterial-phase single-channel ablation on the identical 12-case subset achieved 0.721, marginally exceeding the two-channel fusion result (0.713). These findings are scoped to a single-institution exploratory cohort and should be interpreted as internally comparative benchmarking results; they may not generalize to other centres, scanners, or patient populations, and do not constitute clinical validation. Full article

Accurate traffic sign detection is important for the safety of autonomous driving systems. However, fully supervised methods require a large amount of manual annotation, which is cost-prohibitive and time-consuming. Semi-supervised methods employ a small amount of labeled data and a large amount of unlabeled data to train the models, hence largely reducing the annotation costs. However, these methods have the following challenges: (1) with an imbalanced long-tail class distribution of traffic signs, they tend to achieve poor performance on tail classes; (2) they often fail to detect small traffic signs. To solve these issues, we propose a Semi-Supervised Traffic Sign Detection method with Dynamic Pseudo-Label Selection and Gated Feature Fusion-based Proposal Refinement. Firstly, we design a Class Distribution-based Dynamic Pseudo-Label Selection module (CD-DPLS) to select pseudo-labels for different classes based on the class distribution information, which reduces the tendency to select more pseudo-labels from head classes instead of tail classes, thereby improving the tail class detection performance. Secondly, we employ a Gated Feature Fusion-based Proposal Refinement strategy (GFF-PR) to refine detection proposals by fusing different-scale features with a gating mechanism, which facilitates the detection of small traffic signs. In addition, we use an Adaptive-Weight Focal Loss (AWFL), with which the weight of each pseudo-label is determined by the ratio between its classification confidence and the corresponding class-specific classification-confidence threshold. Experiments on traffic sign datasets demonstrate that the proposed method outperforms state-of-the-art semi-supervised approaches, with mAP₅₀ scores of 10.8% and 34.9% using only 1% and 10% labeled data, respectively. Full article

The 2018 classification of periodontal and peri-implant diseases introduced a multidimensional diagnostic framework integrating staging, grading, and disease extent, representing a major advance over earlier severity-based systems. By incorporating structural destruction, treatment complexity, spatial distribution, and estimated risk of progression, the classification aimed to support more individualized and biologically informed diagnosis. However, increasing clinical application has revealed interpretive challenges, particularly in cases where different components of the system appear discordant. This perspective examines these challenges through a conceptual and clinical lens, focusing on the distinction between focal severity and overall disease burden in staging, the biological meaning of disease distribution, the interpretation of tooth loss as a historical rather than current indicator of disease status, and the need to differentiate between observed progression and risk-based modifiers in grading. Rather than reflecting deficiencies of the classification itself, these discordances are understood as a consequence of applying categorical systems to a biologically heterogeneous and temporally dynamic disease. A biologically grounded interpretive hierarchy is proposed, prioritizing observed tissue behavior and realized tissue destruction over probabilistic risk indicators while integrating structural parameters, historical outcomes, and susceptibility modifiers within their appropriate conceptual roles. This approach enhances diagnostic coherence and supports a more phenotype-oriented interpretation of periodontal disease. Full article