Browse Articles

Detecting infrared UAVs on edge devices through lightweight instance segmentation

Motivation Infrared unmanned aerial vehicle (UAV) detection for surveillance applications faces three conflicting requirements: accurate detection of pixel-level thermal signatures, real-time processing capabilities, and deployment feasibility on resource-constrained edge devices. Current deep learning approaches typically optimize for one or two of these objectives while compromising the third. Method This paper presents YOLO11-AU-IR, a lightweight instance segmentation framework that addresses these challenges through three architectural innovations. First, Efficient Adaptive Downsampling (EADown) employs dual-branch processing with grouped convolutions to preserve small-target spatial features during multi-scale fusion. Second, HeteroScale Attention Network (HSAN) implements grouped multi-scale convolutions with joint channel-spatial attention mechanisms for enhanced cross-scale feature representation. These architectural optimizations collectively reduce computational requirements while maintaining detection accuracy. Third, Adaptive Threshold Focal Loss (ATFL) introduces epoch-adaptive parameter tuning to address the extreme foreground-background imbalance inherent in infrared UAV imagery. Results YOLO11-AU-IR is evaluated on the AUVD-Seg300 dataset, achieving 97.7% mAP@0.50 and 75.2% mAP@0.50:0.95, surpassing the YOLO11n-seg baseline by 1.7% and 4.4%, respectively. The model reduces parameters by 24.5% and GFLOPs by 11.8% compared to YOLO11n-seg, while maintaining real-time inference at 59.8 FPS on an NVIDIA RTX 3090 with low variance. On the NVIDIA Jetson TX2, under INT8 CPU-only deployment, YOLO11-AU-IR retains 95% mAP@0.50 with minimal memory footprint and stable performance, demonstrating its practical edge compatibility. Ablation studies further confirm the complementary contributions of EADown, HSAN, and ATFL in enhancing accuracy, robustness, and efficiency. Code and dataset are publicly available at https://github.com/chen-yuzhi/YOLO11-AU-IR.

Real-time prediction of HFNC treatment failure in acute hypoxemic respiratory failure using machine learning

Abstract Accurate and timely prediction of high-flow nasal cannula (HFNC) treatment failure in patients with acute hypoxemic respiratory failure (AHRF) can lower patient mortality. Previous studies have highlighted inconsistencies in the predictive performance of existing indices, such as ROX and mROX, which are limited by their reliance on oxygenation parameters alone. To address this, we developed a machine learning-based predictive model using temporal data from AHRF patients, aimed at facilitating quicker development of individualized treatment plans and intervention strategies for healthcare professionals. We extracted 15 non-invasive and 15 laboratory features, including patient demographic characteristics, Glasgow Coma Scale, blood gas analysis, chemical assay, and complete blood cell count features. In addition to five machine learning models and an ensemble classifier, an long short-term memory (LSTM) network was included to assess deep learning performance on time-series data. Our study enrolled 427 patients with 498 treatment records. The soft-voting ensemble algorithm achieved an optimal predictive performance with an AUC of 0.839 (95% CI 0.786–0.889) for the all-features model, while logistic regression using common features achieved an AUC of 0.767 (95% CI 0.704–0.825), outperforming ROX and mROX indices. Incorporating blood gas analysis features improved the non-invasive model’s performance by 0.104. This study introduces a machine learning model integrated with a dynamic real-time alert system for predicting HFNC treatment failure in AHRF patients, demonstrating improved performance over traditional indices in internal validation and showing potential for decision support in select healthcare settings.

Identifying microbial functional guilds performing cryptic organotrophic and lithotrophic redox cycles in anaerobic granular biofilms

Granular biofilms used in anaerobic digester systems contain diverse microbial populations that interact to hydrolyze organic matter and produce methane within controlled environments. Prior research investigated the feasibility of utilizing granular biofilms obtained from an anaerobic digester to remove nitrate without the addition of exogenous electron donors. These granules possessed a unique structure of alternating light and dark iron sulfide and pyrite rich layers that potentially served as both an electron source and sink, linking carbon, nitrogen, sulfur, and iron cycles. To characterize the functional roles of diverse microbial populations enriched within these layered biofilms, we analyzed metagenomes obtained from three different granules. Comparisons between the functional gene content of forty metagenome assembled genomes (MAGs) identified phylogenetically cohesive functional guilds. Each of these functional MAG clusters was assigned to specific steps in anaerobic digestion (hydrolysis, acidogenesis, acetogenesis, and methanogenesis) and anaerobic respiration (denitrification and sulfate reduction). Comparisons with metagenomes derived from a variety of natural and engineered ecosystems confirmed that the enriched denitrifying bacteria were similar to populations typically found in wetlands and biological nitrogen removal systems. Analysis of read alignments to individual genes within the forty MAGs identified conserved genomic features that were representative of the functions that distinguished functional guilds. Overall, this research illustrates the utility of functional based classification of microorganisms for characterizing ecosystem functions and highlights the potential application of engineered ecosystems to serve as experimental models for complex natural ecosystems.

Patterns and influencing factors of smokeless tobacco use among pregnant and lactating mothers in urban slums of bhubaneswar, Odisha

Associations between Kynurenine pathway metabolites and cognitive dysfunction in major depressive disorder

This research sought to investigate the relationship between cognitive impairment and kynurenine pathway metabolites in individuals diagnosed with major depressive disorder (MDD). A total of 67 patients diagnosed with MDD and 61 healthy controls (HC) were enrolled in this study. Cognitive function was assessed utilizing the MATRICS Consensus Cognitive Battery. Plasma levels of tryptophan (TRP), kynurenine (KYN), kynurenic acid (KYNA), and quinolinic acid (QUIN) were quantified by liquid chromatography-tandem mass spectrometry. Subsequently, we examined the potential associations between metabolites of the KYN pathway and cognitive dysfunction. MDD patients exhibited significantly poorer performance across all cognitive domains, including processing speed, attention/vigilance, working memory, verbal learning, visual learning, reasoning and problem-solving, and social cognition. Inter-group comparisons indicated that levels of KYN, QUIN, and the KYN/TRP ratio in MDD patients were significantly lower than those in HC, whereas KYNA and the KYNA/QUIN ratio were significantly higher. In MDD patients, a negative correlation was observed between KYN levels and working memory (r = −0.302, p = 0.020), and the KYN/TRP ratio was also negatively correlated with working memory (r = −0.307, p = 0.018). Our findings indicate that impaired working memory in MDD is correlated with increased KYN levels and KYN/TRP ratio. This suggests that the KYN pathway may play a role in the pathological mechanisms underlying neurocognitive dysfunction, particularly working memory deficits, in MDD.

Risk factors for the progression of distal adding-on phenomenon after surgery in patients with Lenke type 1 and 2 adolescent idiopathic scoliosis

Scalable architecture for autonomous malware detection and defense in software-defined networks using federated learning approaches

Abstract This paper proposes a scalable and autonomous malware detection and defence architecture in software-defined networks (SDNs) that employs federated learning (FL). This architecture combines SDN’s centralized management of potentially significant data streams with FL’s decentralized, privacy-preserving learning capabilities in a distributed manner adaptable to varying time and space constraints. This enables a flexible, adaptive design and prevention approach in large-scale, heterogeneous networks. Using balanced datasets, we observed detection rates of up to 96% for controlled DDoS and Botnet attacks. However, in more realistic simulations that utilized diverse, real-world imbalanced datasets (such as CICIDS 2017 and UNSW-NB15) and complex scenarios like data exfiltration, the performance dropped to an overall accuracy of 59.50%. This reflects the challenges encountered in real-world deployments. We analyzed performance metrics such as detection accuracy, latency (less than 1 s), throughput recovery (from 300 to 500 Mbps), and communication overhead comparatively. Our architecture minimizes privacy risks by ensuring that raw data never leaves the device; only model updates are shared for aggregation at the global level. While it effectively detects high-impact incursions, there is room for improvement in identifying more subtle threats, which can be addressed with enriched datasets and improved feature engineering. This work offers a robust, privacy-preserving framework for deploying scalable and intelligent malware detection in contemporary network infrastructures.

Gender-specific associations of metabolic and circadian syndromes with melanoma risk: insights from NHANES 2007–2018

Impact of a self management mobile application on quality of life and limb circumference in women with breast cancer related lymphedema

Bite-sized self-compassion: a pilot cohort study of a well-being tool for healthcare workers

An effective hybrid method for the optimal control of fraud rumor propagation on online social networks

Precise correction of G6PD Viangchan mutation in iPSCs by prime editing strategy

HBV reactivation and prognosis after systemic therapy in HCC with undetectable HBV DNA: a multicenter retrospective study

Unveiling diurnal metamorphosis in the atmospheric boundary layer during an annular solar eclipse through large eddy simulations

Molecular epidemiology of foot and mouth disease virus in Iran during 2019 and 2023

The roots of belonging: childhood predictors of belonging in 22 countries

A study of the correlation between urinary perchlorate, nitrate, thiocyanate, and serum liver function indices

Prevalence of paediatric diarrhoea in Arba minch government health institutions, Southern Ethiopia and associated factors

Artificial intelligence-assisted optimization of Eichhornia crassipes extracts and evaluation of their biological activities

Abstract In this research, the extraction conditions for Eichhornia crassipes (Mart.) Solms were optimized using Response Surface Methodology (RSM) and Artificial Neural Network–Genetic Algorithm (ANN–GA) techniques to enhance the biological efficacy of the extracts. The optimization focused on three key variables: extraction temperature, duration, and the ethanol-to-water solvent ratio. Through the ANN–GA model, the optimal parameters were identified as 56.85 °C for temperature, 7.62 h for extraction time, and 23.93% for the ethanol/water proportion. The obtained extracts showed statistically significantly higher values ​​compared to RSM in terms of antioxidant capacity (FRAP: 152.89 mg TE/g; DPPH: 121.48 mg TE/g), total phenolic content (TPC: 209.47 mg GAE/g) and flavonoid content (TFC: 263.86 mg QE/g). In addition, ANN–GA extract exhibited high anticholinesterase activity with lower IC₅₀ values ​​against acetylcholinesterase (AChE: 61.69 µg/mL) and butyrylcholinesterase (BChE: 81.40 µg/mL) enzymes. In in vitro tests on A549 cell line, its antiproliferative effect increased significantly in a dose-dependent manner and significant decreases in cell viability were observed especially at high concentrations. LC-MS/MS analyses revealed that pharmacologically important phenolic compounds such as quercetin (10295.26 mg/kg), kaempferol (8656.31 mg/kg) and naringenin (5364.56 mg/kg) were present in high concentrations in the optimized extracts. In conclusion, ANN–GA based extraction approach stands out as an effective method for obtaining phenolic compound rich and biologically effective extracts of E. crassipes. These findings indicate that this aquatic plant should be evaluated for its pharmaceutical, neuroprotective and anticancer potential.

Estimating vastus lateralis muscle volume from a single ultrasound image

Abstract The assessment of skeletal muscle volume is valuable for fundamental research and clinical practice, but remains limited in larger cohorts due to its time-consuming nature. Here, we developed a method to accurately estimate vastus lateralis (VL) muscle volume based on a single measurement of anatomical cross-sectional area (ACSA) or tissue thickness. Sixty-nine healthy participants (20–91 years) volunteered. In a subgroup (n = 34) we measured VL volume and ACSAs at 10% intervals along the muscle length to derive a VL muscle shape factor. We subsequently estimated VL volume by multiplying this muscle shape factor with muscle length and a single measure of ACSA at 50% muscle length (ACSAVL50%) or an estimated ACSAVL50% from a single ultrasound scan of tissue thickness in an independent cohort (n = 35). VL muscle shape factor was determined by integrating a fourth-order polynomial of muscle length and ACSA, and was dependent on muscle size. Estimating muscle volume had a high accuracy (R²=0.976, CCC = 0.987), low bias and error (< 8.5%) in both the main cohort and an independent validation group. Estimating muscle volume from stitching 2D images at 50% muscle length or estimating ACSA with a geometric model explained 91–95% of variance in measured volumes, with high accuracy and concordance correlation coefficients. VL muscle volume can be estimated by multiplying a muscle shape factor with muscle length and ACSAVL50% from a single ultrasound image. We present a novel, cost-effective, rapid, yet accurate assessment of VL muscle mass for (large-scale) studies and clinical practice.