Understanding belowground changes during stand degradation is essential for the restoration of degraded shelterbelts in sandy areas. This study investigated Populus simonii shelterbelts across four degradation levels (Healthy, Lightly degraded, Moderately degraded, and Severely degraded) in the Kubuqi Desert. We analyzed the relationships between fine-root traits and soil nutrient variation along the 0-100 cm soil profile to characterize their associations during degradation. The results showed that, as degradation intensified, fine-root activity and biomass decreased significantly, whereas lignin content and the lignin-to-cellulose ratio peaked at the Moderately degraded stage, indicating a shift in fine-root resource-use strategy from an acquisitive strategy to a more conservative one. Soil organic matter, total nitrogen, and alkali-hydrolyzable nitrogen generally declined along the soil profile, with the surface layer showing the strongest response. Total nitrogen, total phosphorus, and alkali-hydrolyzable nitrogen were the key soil factors driving fine-root trait variation. Path analysis further showed that changes in total soil nutrients may contribute to reduced fine-root activity and trait differentiation by affecting the supply of available nutrients and fine-root chemical composition. Overall, fine-root trait changes were significantly associated with soil nutrient profile differentiation, and the Moderately degraded stage exhibited a more pronounced transition in belowground functioning. This study deepens our understanding of the degradation of Populus simonii shelterbelts in sandy areas from a belowground perspective and provides a basis for the ecological diagnosis and restoration of degraded shelterbelts.
Influenza-like illness (ILI) remains a persistent global health challenge, necessitating accurate forecasting tools for timely public health response. This study systematically benchmarks fine-tuned large language models (LLMs), e.g., Llama2 and GPT2, for influenza surveillance forecasting in data-limited time-series settings. We develop a lightweight fine-tuning framework that adapts pre-trained LLMs using compact embedding and prediction layers and evaluate it on seven weekly aggregated real-world surveillance datasets. Despite sample sizes of only ∼523 time points per region and the absence of cloud-based data transfer, fine-tuned LLMs consistently outperform SARIMA, LSTM, PatchTST, CoVTransformer, FEDformer, Time-LLM, and GPT4TS in both accuracy and stability, especially for long-term forecasts across diverse geographic settings. Even in zero-shot settings, pre-trained LLMs capture broad epidemic trends with performance comparable to SARIMA. These findings establish fine-tuned LLMs as efficient and robust forecasting tools suitable for privacy-sensitive, data-scarce public health applications.
Within the family of protein kinases there is a subgroup of 'evolutionarily related members of the total human kinome' that are devoid (or have very limited) enzymatic activity. These proteins, so-called pseudokinases, play important functions thanks to their capacity to establish regulatory protein-protein interactions. Specifically, this opinion article focuses on a group of pseudokinases called Tribbles. Tribbles (Trbl) was originally discovered in Drosophila, followed by the subsequent identification of their orthologs in mammals (TRIB1, TRIB2, TRIB3, and STK40). Work over the last decades has shown how these proteins contribute to fine-tuning key signaling pathways involved in the regulation of proliferation, differentiation, inflammation and adaptation to nutritional changes. Accordingly, dysregulation of Tribbles proteins (TRIBs) contributes to the establishment and progression of insulin resistance, obesity, type II diabetes, atherosclerosis and cancer. Here, we will discuss some of the mechanisms by which Tribbles pseudokinases carry out their functions and the crucial importance of cell context in defining the precise role played by each of the TRIBs, with emphasis on TRIB1 and TRIB3, under different physiopathological situations.
Manual acupuncture depends on precise thumb-index finger coordination, yet the biomechanical principles governing this dexterity remain unresolved. Here, we establish a non-contact, video-based platform integrating marker-labeled motion capture with nonlinear dynamic and multi-modal coupling analyses to decode fine finger coordination across standardized training, animal models, and clinical observation. Tracking eight anatomical joints reveals an asymmetric motor architecture: the distal index finger acts as a temporal pacemaker with high rhythmic stability, whereas the distal thumb functions as an amplitude modulator. Multi-modal coupling resolves two biomechanical modules-an index finger chain with hierarchical phase propagation and a thumb chain with strong intra-module synchronization-linked by millisecond-scale temporal offsets. These patterns are encoded as operator-specific "Acupuncture Kinematic Fingerprints," stratifying practitioners into index-dominant and thumb-dominant strategies. This framework transforms subjective manipulation styles into quantitative kinematic signatures for objective skill assessment, standardized training, and mechanistic studies.
Genetic homogenization due to chronic clonal propagation poses a notable challenge to the sustainable utilization of Coptis chinensis Franch. We developed a novel elite cultivar, 'Chulian No. 1', with superior yield (+34%), disease resistance, and pharmacopoeia-standard alkaloid profiles through comprehensive agronomic evaluation. To understand the genetic basis of these traits, we constructed a high-quality chromosome-level genome assembly, revealing long terminal repeat retrotransposon (comprising 41.92% of the genome) shows significant enrichment near expanded stress/alkaloid biosynthesis genes. Whole-genome resequencing of 235 accessions indicated extreme genetic uniformity (π < 0.0018), but our novel geospatial-autoencoder model successfully identified three eco-geographic groups that could not be discriminated using traditional approaches. Genome-wide association analysis identified the leaf glossiness-associated gene CcHAB1 on chromosome 1. Comparative transcriptomics revealed correlations between CcHAB1 expression and cuticular wax biosynthesis gene regulation via ABA and phenylpropanoid pathways, potentially explaining enhanced disease resistance in glossy-leaf accessions. This work enables molecular breeding for medicinal plants with limited genetic diversity and provides insights into trait maintenance despite genetic homogenization.
To evaluate open-source large language models (LLMs) for extracting cancer-specific phenotypic data, benchmark their performance against GPT4 models, and assess the impact of fine-tuning with training data sizes. Open-source LLMs (Mistral, LLaMa, MAMBA, BioMistral) were evaluated in zero-/one-shot and fine-tuned setups against GPT4-turbo/GPT4o to extract the cancer presence, progression, response, and metastatic sites from radiology impressions of patients with solid tumors treated at Dana-Farber Cancer Institute. Performance metrics (accuracy, precision, recall, F1-score) were computed. McNemar's odds ratio (OR), measuring which model is more likely to be correct when they disagree, was computed with 95% CI. Statistical significance was assessed using the alpha of .000139. This study included 2,623 patients (25,273 radiology impressions). In zero-/one-shot settings, GPT4-turbo/GPT4o outperformed open-source LLMs. However, fine-tuned open-source LLMs achieved higher F1-scores than GPT4 models. Compared with the best-performing GPT4 model, fine-tuned Mistral0.2-7.3B (OR, 0.27 [95% CI, 0.20 to 0.36]; P < .00001), Mistral0.3-7.3B (OR, 0.26 [95% CI, 0.19 to 0.36]; P < .00001), LLaMa2-6.7B (OR, 0.30 [95% CI, 0.22 to 0.40]; P < .00001), LLaMa3.1-8B (OR, 0.37 [95% CI, 0.28 to 0.48]; P < .00001), and MAMBA-2.8B (OR, 0.32 [95% CI, 0.24 to 0.42]; P < .00001) showed significantly better performance in ascertaining disease progression. Performance was consistently better for inferring overall response, any evidence of cancer, and sites of metastases, with no significant differences among fine-tuned open-source LLMs. Fine-tuning gains plateaued at 25% of training data (5,718 impressions) and remained comparable at 5% (1,144 impressions). Open-source LLMs, when fine-tuned using labeled data, can effectively automate the ascertainment of key radiophenotypic variables using only the impression section of radiology reports, without the full report text. Their consistent performance in small training sets suggests that these models may provide a scalable approach for phenotypic characterization of patients with cancer in real-world clinical settings.
There is limited research examining the association between wildfire smoke, an increasingly frequent exposure, and headache-related emergency department (ED) visits, despite headaches being a leading cause of years lived with disability globally. To examine the association between wildfire-sourced fine particulate matter with an aerodynamic diameter of 2.5 μm or less (PM2.5) and ED visit for migraine and other primary headache syndromes (MOPHS) compared with nonwildfire-sourced PM2.5, and to assess the variation across sociodemographic factors. This case-crossover study used conditional logistic regression to examine ED visits during wildfire seasons (May 1 to October 31) from 2010 to 2023 in Alberta and Ontario provinces in Canada. Case days were matched to referent days by day of week, month, year, and forward sortation area (first 3 postal code characters). Seven-day cumulative lags were evaluated. All ED visits recorded in the National Ambulatory Care Reporting System with a primary diagnosis of MOPHS were included. Statistical analyses were conducted from April to December 2025. Total PM2.5 concentrations and wildfire day (WFD). Daily mean wildfire-sourced PM2.5 was estimated using the Canadian Optimized Statistical Smoke Exposure Model. WFDs were defined by smoke plume presence and total PM2.5 concentration exceeding a predefined threshold based on the mean plus 1.5 SD of PM2.5 on days without a smoke plume. Wildfire-sourced PM2.5 was defined as the difference between total PM2.5 concentration on a WFD and expected background PM2.5 concentration on non-WFDs. ED visits for migraine (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Canada code G43) and other primary headache syndromes (code G44). Among the 997 701 ED visits for 622 753 patients (mean [SD] age, 24.6 [16.8] years; 760 795 females [76.3%]) for MOPHS identified during the study period, 86.2% of visits were for migraines and 13.8% were for other primary headache syndromes. Wildfire-sourced PM2.5 was associated with a 6.07% (95% CI, 5.75%-6.39%) increase in ED visits. Associations appeared attenuated in the least materially and socially deprived quantile (for both provinces combined, the percentage increase in MOPHS was smallest for the Material and Social Deprivation Index quintile 1: 3.61% [95% CI, 1.93%-5.63%] and largest for quintile 5: 10.42% [95% CI, 7.41%-13.49%]). In contrast, total PM2.5 concentration on non-WFDs was not significantly associated with ED visits (0.21%; 95% CI, -0.20% to 0.62%). In this case-crossover study of ED visits in Alberta and Ontario, acute exposure to wildfire-sourced PM2.5 was associated with increased ED visit for MOPHS; associations were greater in magnitude for wildfire-sourced PM2.5 than nonwildfire-sourced PM2.5. These findings support the need for further research into wildfire smoke and severe headache-related outcomes.
The gastrointestinal tract of grazing ruminants contains segment-specific microbial niches that contribute to fermentation metabolism and intestinal function. However, the intestinal bacterial communities and their relationships with fermentation products and mucosal morphology remain insufficiently characterized in alpine sheep breeds. In this study, Tianhua mutton sheep and Gansu Alpine fine-wool sheep maintained under the same grazing conditions were compared using the techniques of hematoxylin and eosin (H&E) staining, gas chromatography, and full-length 16S rRNA sequencing. Histological analysis showed that Tianhua mutton sheep had lower crypt depth and higher villus-height-to-crypt-depth ratios in the duodenum and ileum, while muscular layer thickness differed between breeds in a segment-dependent manner. Short-chain fatty acids (SCFAs) analysis revealed breed-related differences in specific intestinal segments, with Tianhua mutton sheep showing higher levels of several major SCFAs components and total SCFA in the duodenum, jejunum, cecum, and colon. Full-length 16S rRNA sequencing demonstrated distinct bacterial community structures between the two breeds and clear spatial variation along the intestinal tract. Tianhua mutton sheep exhibited higher bacterial richness in several segments and enrichment of Firmicutes and Christensenellaceae_R_7_group, whereas Gansu Alpine fine-wool sheep showed higher abundances of Bacteroidota, Verrucomicrobiota, and Akkermansia in specific segments. Functional prediction indicated that KEGG level 3 differences were mainly concentrated in the cecum and colon, involving pathways related to microbial metabolism, carbon metabolism, ABC transporters, and two-component systems. Spearman correlation analysis further linked selected genera with SCFA traits, whereas fewer associations were observed with small-intestinal histomorphological indices. These findings suggest that Tianhua mutton sheep and Gansu Alpine fine-wool sheep differ in intestinal morphology, fermentation characteristics, and microbial structure and function, providing new information for understanding breed-specific host-microbiota interactions in alpine grazing sheep.
This paper deals with a data-driven framework for transient-based leak detection in water distribution networks (WDNs) under limited field data. Leak detection are characterized as three supervised tasks: (i) leaking-pipe classification, (ii) leak-location regression, and (iii) leak-size regression. Large partially calibrated simulation datasets are generated to produce transient pressure responses, but models trained solely on these simulations perform poorly when applied to measurement data, reflecting a pronounced simulation-to-reality gap. To address this problem, we train several advanced artificial neural network (ANN) architectures, including Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), Gated Recurrent Units (GRU), and Transformer models, on simulated transients contaminated with measurement-based noise. Subsequently, we apply transfer learning to fine-tune the pre-trained models using experimental records and evaluate the adaptation through leave-one-out cross-validation. The transfer-learning tests include both the original 17-scenario experimental dataset and an expanded 40-scenario dataset collected under a different PRV set pressure. Across architectures, transfer learning consistently improves leaking-pipe classification and leak-location regression relative to direct simulation-to-experiment testing; in the expanded 40-scenario evaluation, the best after-transfer classification accuracy reaches 62.50%, compared with a best direct-test accuracy of 30.00%, while the best leak-location MAE decreases from 18.44 m in direct testing to 6.57 m after fine-tuning. To interpret the learned models, we use Integrated Gradients technique to identify the sensor-time regions that drive the predictions and to compare their behavior before and after transfer learning. The attribution maps show that the networks rely primarily on later-time, multiply scattered transients rather than solely on the first-half characteristic wave period, i.e., the time required for the wave to travel from the transient source to the furthest boundaries and back, and that fine-tuning (via transfer learning) suppresses spurious time features while preserving physically meaningful features. Overall, the proposed framework provides a robust leak-detection strategy that leverages simulations, limited experimental data, and interpretability tools to improve the reliability of transient-based defect detection in practical WDN applications.
Patients with de novo metastatic cancer develop local organ-related complications leading to hospitalizations, procedures, and reduced quality of life. Prophylactic local therapy (e.g., radiotherapy) has been proposed to reduce morbidity, but the incidence and timing of severe primary organ-related events (SPOREs) remain poorly defined. We quantified the cumulative incidence and spectrum of SPOREs to inform trials of prophylactic local therapy. We conducted a retrospective cohort study of 358 patients with de novo metastatic prostate, bladder, and pancreatic cancers treated between 2010 and 2023. SPOREs were defined using clinically meaningful, organ-specific criteria, excluding events within 30 days of diagnosis. Death was treated as a competing event, and cumulative incidence functions were estimated with the Fine-Gray method. A total of 104 prostate, 59 bladder, and 195 pancreatic cancer patients (adenocarcinoma n = 178; neuroendocrine n = 17; including head and body tumors) met inclusion criteria. At 24 months, cumulative incidence of SPOREs was 3.3%, 27.2%, 40.0%, and 45.5%, respectively (p < 0.001). Overall survival differed across disease sites but did not parallel SPORE risk. SPORE incidence plateaued after 24 months for bladder and pancreatic adenocarcinoma, while prostate cancer and pancreatic neuroendocrine tumors continued to increase. Urinary obstruction predominated in prostate and bladder cancer, whereas biliary obstruction or cholangitis was most common in pancreatic cancer. Within pancreatic tumors, there was a trend toward higher SPORE incidence for head versus body tumors (48.5% vs 32.7% at 24 months, p = 0.054). SPOREs represent a substantial source of morbidity in de novo metastatic cancer. Across disease sites, risk and timing appear driven significantly by anatomic vulnerability of the primary tumor rather than metastatic virulence or prognosis. Despite differing biology, pancreatic adenocarcinoma and neuroendocrine tumors demonstrated high local morbidity. These findings suggest anatomic context may better identify patients most likely to benefit from prophylactic local therapy aimed at reducing morbidity.
With the increasing exploitation of unconventional oil and gas resources in China, the tight sandstones of the Upper Triassic Xujiahe Formation in the northeastern (NE) Sichuan Basin have emerged as a critical exploration frontier. However, the complex controlling factors of reservoir development under multiprovenance convergence remain poorly understood. This study focuses on the second member of the Upper Triassic Xujiahe Formation (the T3x2) in the northeastern region of the Sichuan Basin as its research subject, through core observations, thin-section petrography, scanning electron microscopy (SEM), and light/heavy mineral analyses to clarify the differential controls exerted by various provenance systems. Results indicate that the T3x2 reservoir in the study area was jointly supplied by three distinct systems: the Longmen mountain, Micang mountain, and Daba mountain. Specifically, the Longmen mountain region is dominated by feldspathic lithic sandstones, featuring pore-type reservoir spaces with medium-to-fine pore throats. The Micang mountain region primarily comprises lithic sandstones and lithic quartz sandstones, characterized by fracture-pore type reservoirs with fine pore throats. The Daba mountain region develops feldspathic lithic sandstones dominated by micropore-throat systems. Porosity evolution analysis reveals that initial porosity was predetermined by provenance-related lithofacies. In the Longmen mountain and Daba mountain regions, porosity enhancement was driven by chlorite coatings and dissolution; however, porosity reduction in the Longmen mountain area was mainly caused by compaction, whereas the Daba mountain area was restricted by low-calcite cementation. Conversely, porosity in the Micang mountain region was preserved through dissolution and tectonic fracturing, with reduction primarily resulting from the combined effects of compaction and cementation. Furthermore, the timing and mechanisms of reservoir densification exhibit significant spatial heterogeneity. The Micang mountain reservoirs densified earliest during the Middle to Late Jurassic, primarily controlled by quartz frameworks and late-stage tectonic fracturing. The Longmen mountain and Daba mountain reservoirs reached the tight threshold later, from the latest Late Jurassic to the Early Cretaceous, with the former controlled by feldspar grains, chlorite coatings, and dissolution, and the latter governed by the synergy of chlorite coatings and dissolution. This study reveals the systematic control of provenance on lithology, pore-throat structure, evolutionary pathways, and densification history, providing critical geological insights for hydrocarbon exploration in multiprovenance convergent basins.
Molecular Tumor Boards (MTBs) generate highly technical recommendations. The language used in their protocols is rarely accessible to patients. Lay-language patient protocols could support patient-clinician communication, yet manual production is difficult to sustain in high-volume oncology settings. Large language models (LLMs) may offer scalable drafting assistance, yet clinical usability remains largely uninvestigated under real-world deployment constraints. Existing evaluations rely predominantly on synthetic data or closed-source models that are incompatible with strict data protection requirements. This study evaluated whether open-weight LLMs can provide clinically usable drafting support for German MTB patient protocols under real-world deployment constraints and developed a transferable evaluation framework for patient-facing text generation. Eight open-weight LLMs were evaluated under zero-shot (A1) and one-shot (A2) prompting with constrained decoding, which ensures section-schema compliance. Automatic evaluation used ROUGE-1 (Recall-Oriented Understudy for Gisting Evaluation), BERTScore-F1 (Bidirectional Encoder Representations From Transformers Score), Wiener Sachtextformel version 4, and DistilBERT (Distilled Version of Bidirectional Encoder Representations From Transformers)-based complexity using a corpus of 316 MTB protocols and 47 expert-written patient protocols. For expert evaluation, 7 medical oncologists evaluated 50 protocols from the best-performing model across 3 International Organization for Standardization 9241-11 usability dimensions using fine-grained error annotation, perceived postediting effort (PPEE), and net promoter score. Critical errors were defined as bearing the risk of patient harm. Llama-3.3-70B-Instruct achieved the strongest automatic performance. Across models, A2 significantly improved most automatic metrics compared to A1. However, expert usability evaluation of Llama-3.3-70B-Instruct showed the opposite picture: the proportion of protocols containing at least 1 critical error doubled under A2 (10/25, 40% vs 5/25, 20%) compared with A1, and the dominant error type shifted from language (40/108, 37%) errors to factual errors (69/145, 48%). Overall, 16% (230/1420) of the annotated paragraphs contained errors. Median PPEE was 2 (IQR 2.0-3.0; low), and median net promoter score was 7 (IQR 5.0-9.0). Detractors (46/100, 46%) outweighed promoters (29/100, 29%), which suggests hesitation toward routine adoption. These differences in expert evaluation between A2 and A1 were directionally consistent but did not reach individual statistical significance for the paired samples (n=25). Prompting strategies that improve automatic metrics can simultaneously increase the number of critical errors. Surface-level metric gains were, therefore, insufficient proxies for clinical safety. This was observed as a consistent directional pattern for a single model, but generalization to other models remains to be investigated. Nonetheless, the low paragraph-level error rate and favorable PPEE suggest that structured open-weight LLM generation may be a useful drafting support in a clinician-supervised setting. The proposed evaluation framework provides a text-quality-focused basis for future assessment of patient-facing LLM applications in real-world clinical settings.
Early detection of pulmonary embolism (PE) is critical for clinical outcomes, yet existing deep-learning models often fail to generalize across institutions. The recently released Google CT Foundation model, pre-trained on a large, diverse CT corpus, produces compact volumetric embeddings that may transfer to downstream tasks without fine-tuning. We evaluate a classification pipeline that pairs these frozen embeddings with three lightweight heads - Multi-Layer Perceptron (MLP), Random Forest (RF), and a stacking ensemble - for central PE detection, training on the public RSNA CTPA dataset and externally validating on the Stanford INSPECT cohort. The pipeline first reproduces the published data-size scaling behavior of the foundation model on other medical pathologies. The highest point-estimate test AUC of 0.79 was obtained by the RF trained on a balanced training set of 408 CT studies and evaluated on the held-out RSNA test set of 144 studies; paired DeLong tests showed that this advantage over the MLP and stacking heads was not statistically significant. Direct comparison to specialized state-of-the-art PE pipelines is task-mismatched - those target any-PE rather than central PE - so the result anchors a non-fine-tuned baseline rather than a deficit. In a low-data regime, hard vascular segmentation did not improve performance. On the external INSPECT cohort, AUC dropped by 0.17 and specificity at the transferred operating point collapsed, so simple global threshold recalibration does not restore deployability-frozen generalist embeddings alone do not guarantee cross-institutional reliability.
This study systematically investigated how whey protein isolate (WPI) to gum arabic (GA) mass ratios govern the structural transition from fine-stranded to particulate hydrogel networks and analyzed the resulting chemical, viscoelastic, and mechanical properties. WPI slab hydrogels with protein concentrations of 10-20% (w/v), and various WPI:GA mass ratios (20:1 to 2.5:1), were characterized via electron microscopy, Fourier-transform infrared spectroscopy, oscillatory rheology, and mechanical testing, including: frequency and strain sweeps, uniaxial compression, creep recovery, stress relaxation, and Burgers viscoelastic modeling. The WPI:GA mass ratio was found to alter the hydrogels' microstructures and scale mechanical properties over multiple orders of magnitude, enabling production of biomaterial-based gels with widely varied characteristics for engineering applications. The addition of GA to WPI gels up to an approximate 5:1 WPI:GA mass ratio yielded enhanced Young's elastic modulus (E) and dynamic moduli (G', G") while developing the gels' permanent crosslink network as quantified by relaxation modulus G(t), frequency dependence, and creep recovery behavior. The 5:1 WPI:GA ratio also represented an approximate phase transition point with gels above this ratio adopting fine-stranded microstructures and gels below adopting particulate microstructures. The fundamental structure-property relationships established here provide a foundation for development of WPI:GA hydrogels as a tunable platform.
ObjectiveDigital mental health screening is increasingly explored through the use of AI systems. Yet, most models provide limited insight into how predictions are derived, restricting clinical trust and patient-centered adoption. We investigate whether Large Language Models (LLMs) can generate structured DSM-5-aligned rationales for depression assessment when trained with Group Relative Policy Optimization (GRPO), a reinforcement learning method that encourages outputs aligned with DSM-5 diagnostic criteria.MethodsWe fine-tuned LLMs (1B-27B parameters) on the ReDSM5 dataset, covering 1,484 Reddit posts annotated by a licensed psychologist for DSM-5 depressive symptoms and accompanied by expert rationales. We compared standard Supervised Fine-Tuning (SFT) with GRPO-based optimization using a composite reward integrating symptom classification accuracy and the quality of generated reasoning judged against clinical rationales.ResultsGRPO consistently improved symptom detection over SFT, with relative gains exceeding 10% for mid-sized models and weighted F1 scores above 0.60. Models trained to generate structured DSM-5-aligned rationales exhibited additional performance boosts (0.09-0.39 F1), particularly for complex symptoms requiring complex contextual interpretation. Qualitative analysis shows that GRPO encourages models to reference symptom-relevant evidence rather than relying on superficial cues.ConclusionsGRPO enables LLMs to produce clinically grounded explanations while improving classification accuracy, representing a promising direction for interpretable AI in social media-based mental health screening, with potential to support patient-centered applications pending clinical validation.
Background and Clinical Significance: Odontogenic myxoma (OM) is a rare benign neoplasm of the jawbones characterized by spindle-shaped cells embedded in a myxoid stroma. Despite its benign histological nature, it demonstrates locally aggressive behavior, significant invasiveness, and a high risk of recurrence. OM ranks as the third most common odontogenic tumor after odontoma and ameloblastoma. It affects both sexes and occurs more frequently in the mandible than in the maxilla, typically during the second to fourth decades of life. Macroscopically, OM is non-encapsulated, whitish-gray, and gelatinous. Radiographically, it usually presents as a radiolucent lesion with fine bony trabeculae, producing a characteristic "tennis racket" appearance. Case Presentation: We report a case of a 27-year-old male diagnosed with maxillary odontogenic myxoma measuring 2.3 × 1.7 cm. Clinical, radiographic, and histopathological findings were evaluated, and the lesion was treated conservatively by surgical enucleation and curettage. Results: The surgical procedure was completed without complications. Histopathological analysis confirmed the diagnosis of odontogenic myxoma. The patient showed satisfactory postoperative healing, and no evidence of recurrence was observed during a 10-month follow-up period. Conclusions: Although odontogenic myxoma is benign, its locally aggressive nature and recurrence potential require accurate diagnosis and appropriate management. Conservative treatment by enucleation and curettage may be effective for small, well-defined lesions, provided that careful long-term follow-up is maintained to monitor for recurrence.
Detecting salient objects in complex traffic environments remains a challenging task for ground-based systems. Recently, unmanned aerial vehicles (UAVs) have emerged as an effective solution by offering flexible perspectives and the ability to capture complementary RGB and thermal images. However, direct fusion of these modalities often introduces artifacts caused by spatial misalignment and semantic discrepancies. To overcome these challenges, we propose a Semantic-Spatial Driven Alignment Network (S2ANet) for salient object detection in UAV-based unregistered RGB-T imagery. The proposed network performs progressive semantic and spatial alignment through a Semantic-Spatial Alignment (SSA) module, achieving precise cross-modal registration. After alignment, three specialized components are designed to enhance feature representation: the Deep Positional Awareness (DPA) module extracts accurate positional cues via self-attention; the Cross-Hierarchical Contextual Interaction (CCI) module captures both intra- and inter-feature dependencies; and the Multi-scale Detail Perception (MDP) module refines fine-grained details through multi-receptive-field convolutions and spatial attention. Finally, a Dual-Cascade Feature Fusion (DFF) module integrates positional, contextual, and detailed information to generate high-quality saliency maps. Extensive experiments validate that S2ANet achieves accurate feature alignment and superior detection performance in UAV-based RGB-T scenarios.
Accurate classification of brain tumors in MRI scans is critical for effective clinical decision making, yet manual assessment is labor-intensive and subject to variability. Although deep learning models, particularly CNNs and Transformers, have shown potential, each architecture alone has limitations: CNNs excel at local feature extraction but lack global context awareness, while Transformers capture global relationships but may overlook fine-grained details. To develop a hybrid Transfer Learning-Transformer model that integrates CNN-driven local feature modeling with Transformer-based global reasoning to enhance multi-class brain tumor classification. The proposed workflow comprises three stages: (1) benchmarking standalone CNN (ResNet50, VGG19, ConvNeXtBase, EfficientNetV2B0) and Transformer models (ViT, Swin, DeiT, PoolFormer); (2) constructing two intra-family hybrids-HTL (ResNet50 + ConvNeXtBase) and HTF (PoolFormer + ViT); and (3) fusing them into a final hybrid model (HF). Two public MRI datasets (Figshare, Kaggle) were used. Models were trained, validated, and tested on the Kaggle dataset, while the Figshare dataset was used exclusively for external validation. Performance was assessed using accuracy, precision, recall, F1-score, AUC, Friedman's aligned-rank test, Kendall's W, Holm post hoc analysis, TOPSIS-based ranking, calibration analysis (Brier score), and Grad-CAM++ for interpretability. The HF model achieved near-perfect classification (∼99.4% on Kaggle and up to 100% on Figshare), significantly outperforming all baseline and single-architecture models. Statistical and multi-criteria analyses consistently ranked HF as the top-performing model, while calibration results confirmed reliable probability estimation. Grad-CAM++ further indicated tumor-focused decision-making. The proposed hybrid model delivers high accuracy, strong generalizability, and reliable, interpretable predictions across MRI datasets, positioning it as a promising solution for AI-assisted brain tumor diagnosis.
Upper gastrointestinal subepithelial lesions (SELs) are detected during routine screening endoscopy and are usually identified as asymptomatic protrusions covered by normal-appearing mucosa. Although most SELs follow a benign clinical course, some lesions, including gastrointestinal stromal tumors, neuroendocrine tumors, lymphomas, and metastatic tumors, possess malignant potential and require further evaluation. Appropriate risk stratification is therefore essential to guide management. Endoscopy remains the initial diagnostic modality, while endoscopic ultrasonography (EUS) plays a central role in characterizing lesion size, layer of origin, echogenicity, and other high-risk features. Recent advances in tissue acquisition techniques, including EUS-guided fine-needle biopsy and mucosal incision-assisted biopsy, have improved diagnostic yield and histologic accuracy. Accumulating evidence from large retrospective and prospective studies indicates that most small asymptomatic SELs remain stable during long-term follow-up, whereas larger lesions, interval growth, mucosal surface changes, and suspicious EUS findings are associated with an increased risk of progression. Management strategies should be individualized according to lesion characteristics, malignant potential, and patient factors. Surveillance is generally appropriate for small lesions without high-risk features, while tissue diagnosis or resection should be considered for lesions demonstrating growth, symptoms, or imaging findings that are concerning. Advances in therapeutic endoscopy, including endoscopic submucosal dissection, submucosal tunneling endoscopic resection, and endoscopic full-thickness resection, have expanded minimally invasive treatment options. Current international and Korean guidelines emphasize EUS-based risk assessment and selective intervention rather than routine resection of all SELs. This review summarizes the epidemiology, natural history, diagnostic approaches, management strategies, and current guideline recommendations for asymptomatic upper gastrointestinal SELs.
Lenalidomide is an immunomodulatory agent widely used in the treatment of multiple myeloma and is generally associated with a favorable safety profile. Although pigmentary alterations such as cutaneous hyperpigmentation and hair repigmentation have been reported, oral mucosal involvement remains poorly documented. An 80-year-old Black female with multiple myeloma undergoing lenalidomide therapy presented with asymptomatic brown macules on the tongue, hard palate, bilateral buccal mucosa and gingiva. Histopathological examination of an incisional biopsy on the tongue revealed many fine granular, brownish pigmentation of approximately the same size throughout all epithelial layers, as well as collagen fibers, vascular walls, neural tissue, and skeletal striated muscle. These findings represent an unusual pattern of pigment distribution not previously described in patients under lenalidomide therapy. This case emphasizes the need for thorough oral examination in patients undergoing lenalidomide therapy, as early recognition of drug-induced pigmentation is essential for proper diagnosis and management.