Search Results (5,520)

Ovarian cancer (OC) is the second most common gynecological malignancy and remains one of the leading causes of gynecological cancer-related mortality worldwide. A major clinical challenge is the lack of an accurate and widely applicable strategy for identifying patients at high risk of malignancy at an early stage. In this context, artificial intelligence (AI) has emerged as a promising tool to improve diagnostic performance. Among AI technologies, large language models (LLMs) have recently shown considerable potential in healthcare applications. In this study, we evaluated the diagnostic performance of ChatGPT (GPT-5) in classifying 300 adnexal masses as benign or malignant and compared its performance with that of the IOTA Simple Rules, the ADNEX model, and expert subjective assessment. We also assessed ChatGPT’s ability to predict the most likely histological diagnosis for each lesion. All adnexal masses were described using the International Ovarian Tumor Analysis (IOTA) terminology, and histopathological examination served as the reference standard. Our findings showed that expert subjective assessment achieved the highest overall diagnostic performance for both benign/malignant classification (accuracy 87.3%; 95% CI, 83.0–90.9%) and prediction of the presumed histological diagnosis. ChatGPT A and ChatGPT B reached a sensitivity of 72.3% and 73.5%, a specificity of 74.5% and 75.9%, a positive predictive value of 75.2% and 76.5%, and a negative predictive value of 71.5% and 72.8%, respectively (inconclusive responses counted as misclassifications), with an overall accuracy of 73.3% and 74.7%. After adequate validation, large language models might complement existing decision-support tools for less experienced examiners, without replacing expert evaluation. Their ease of use and reliance on standardized ultrasound descriptors make them accessible to ultrasonographers with varying levels of expertise. Full article

Modern network infrastructures are under attack from increasingly sophisticated attacks that static, rule-based defenses cannot adequately mitigate. We introduce a multilayer anomaly detection and mitigation framework for Software-Defined Networking (SDN) environments consisting of four integrated subsystems: (i) a Micro-segmentation Integrated Management and Defense System (MIMDS), (ii) an Adaptive CNN-LSTM-Attention Deep Packet Inspection (MCLA-DPI) module, (iii) a Hybrid Adaptive Cyberattack Prediction (KBGM) framework, and (iv) an AI-driven log analysis pipeline. Detection, prediction and containment operate in parallel (as opposed to conventional approaches) and correlate results through a common risk-scoring mechanism to coordinate policy enforcement via OpenFlow and P4-compatible data planes. The experimental evaluation shows promising performance. MIMDS achieves 94.3%. detection accuracy with full traffic isolation in 10.1 s. MCLA-DPI achieves 98.9% classification accuracy with 14 ms inference latency, outperforming baseline models SVM and LSTM on encrypted traffic. KBGM achieves 98.4% detection accuracy with 7.2 ms mean response time on the CICIDS2017 dataset. All modules are trained in federated learning to preserve data locality with continuous improvements of the global model. The results collectively demonstrate quantifiable improvements over single-paradigm approaches in detection fidelity, response latency, resource efficiency, and privacy compliance. An ablation study isolates the contribution of integration itself. Removing the coordination layer while retaining all four detectors reduces accuracy from 0.892 to 0.634 and raises the false-positive rate from 0.031 to 0.436, while peak rule installation rises from 22.3 to 98.4 rules per second and oscillation events increase by two orders of magnitude; the integrated framework also exceeds its strongest individual subsystem, which reaches 0.831 accuracy at a false-positive rate of 0.117. These figures are obtained from the released reference implementation over a synthetic campaign and are reported separately from the component measurements. Full article

Background: Radiation maculopathy (RM) is a common complication after radiotherapy for intraocular tumors, causing permanent visual loss. Intravitreal dexamethasone (IV DEX) is effective for macular edema (ME) resolution, but functional outcomes remain highly heterogeneous. This study aimed to assess the prognostic utility of CEA classification on visual outcomes following IV DEX based on three parameters: largest cyst diameter (C), ellipsoid zone (EZ) disruption (E) and retinal pigment epithelium (RPE) atrophy (A). Methods: A retrospective analysis of 50 patients with RM secondary to Iodine-125 brachytherapy treated with IV DEX was performed. Best-corrected visual acuity (BCVA) and CEA parameters were analyzed with OCT before and after treatment. Results: IV DEX induced significant anatomical improvement (mean Δcyst: −203.3 μm, p < 0.001), regardless of functional response. Visual outcomes were related to baseline CEA stratification: eyes with intact outer retina gained +6.4 ETDRS letters (p = 0.002); eyes with EZ disruption demonstrated stability (−2.8 letters; p = 0.183); eyes with RPE atrophy significantly worsened (−11.7 letters; p = 0.014). Baseline RPE atrophy was the strongest negative predictor (p = 0.007). Subgroup analysis of patients with intact outer retina identified a baseline BCVA cutpoint of ≤70 letters to predict clinically significant visual improvement (≥5 letters), demonstrating a functional ceiling effect. Conclusions: The CEA classification effectively stratifies functional response in RM. EZ disruption and RPE atrophy identify eyes unlikely to achieve visual improvement despite a reduction in ME, aiding clinical decision-making and establishing functional expectations while promoting a fundamental shift from an edema-driven management to a biomarker-centered approach. Full article

Algorithmic management increasingly shapes how gig work is organised, evaluated, and experienced, yet its implications for workers’ affective responses remain fragmented across the literature. This study integrates a systematic review with computational analysis to examine how algorithmically mediated work is represented in research and expressed in online gig-worker discourse. The systematic review followed PRISMA guidelines and synthesised evidence on autonomy, fairness, transparency, evaluation, resource insecurity, and worker experience. The computational component analysed a corpus of 10,000 online texts collected over 12 consecutive months in 2025 from gig-worker-related digital communities and platforms. Sentiment and emotion were examined using NLP-based classification and term extraction, with temporal and contextual patterns assessed across the corpus. The computational analysis found that negative sentiment was more prevalent than neutral and positive sentiment, with frustration, anxiety, and anger among the most prominent expressed emotions. These patterns broadly corresponded with themes identified in the systematic review, particularly concerns regarding fairness, uncertainty, autonomy, transparency, and algorithmic control. However, the temporal findings represent changes in aggregate online discourse rather than within-person emotional trajectories or causal effects. The study concludes that computational analysis provides complementary evidence of how algorithmically mediated work is discussed and affectively expressed online, while individual lived experiences, psychological outcomes, and causal mechanisms require further investigation through longitudinal and mixed-method research. Full article

Reliable classification of brain sequence cases is a challenging problem due to signal ambiguity and noise. Personal concentration is primarily determined by the base frequency of electroencephalogram (EEG) waves, i.e., the task rests on appropriate modelling and recognition of patterns in the corresponding human (in)activity (e.g., reading, relaxation, solving maths problems, etc.). Five underlying types of frequency (alpha, beta, gamma, delta, and theta) were considered as secondary input wave parameters in complex-valued node extensions to the prime amplitude in processing signals. Self-optimisable Artificial Intelligence (AI) methods can process, statistically analyse, and model the series-specific character and time behaviour to recognise untrained session assigned labels. This procedure involves signal pre-processing (transformation) and feature extraction to enhance the representation in time variability, eliminate uncertain cases, and reduce the unacceptable large raw format of data in detailed frequency band recording. This study focuses on improving AI modelling through brain-inspired doubled amplitude/frequency signal processing. This extended concept is based on an analogy with neural activity that generates dynamic frequency pulses as the main information holder in response to time excitations. The model is obtained in partial differential equation (PDE) solutions of evolutionary tree structure nodes—self-computational terms, using the optimal sine/cosine or rational expression. It enables the representation of periodic patterns in their intrinsic form related to primary wave characteristics. Two different machine learning methodologies were compared: the first evolutionary PDE transform and deep learning-based recurrent processing applied to all session records in bloc, assessing only one final class assessment, achieving predictive accuracy above 90% on untrained 1/3 data. The second group of regular modelling techniques evaluates each data row separately to compute its bound-label output in a time-lagged frame, reaching accuracy above 70%. An executable parametric software with a link to the public EEG data repository is available. Full article

Background: Chlorine dioxide (ClO₂), formulated as an aqueous solution (CDS), has been proposed as a therapeutic agent, yet its classification among oxidizing compounds is frequently confounded with chemically distinct species—sodium chlorite (NaClO₂) and sodium hypochlorite (NaOCl). Objective: This study aimed to synthesize the physicochemical, mechanistic and cellular basis of ClO₂ action and to delineate the distinctions between ClO₂ and related oxidants, with explicit separation of established data from hypothesized data. Methods: The methodology includes a narrative synthesis of electrochemical data, primary toxicological and clinical literature (PubMed) and the redox-biology literature on low-dose oxidant signaling. Results: ClO₂ is a neutral free radical that acts via one-electron transfer (E°′ = +0.95 V vs. SHE for the half-reaction ClO₂ + e⁻ → ClO₂⁻, essentially pH-independent over pH 4–8, pH-independent) without chlorination. Its antimicrobial action is attributable to oxidation of thiol-dependent enzymes and membrane constituents and to the inhibition of bacterial respiratory metabolism; host-cell sparing is consistent with differential glutathione-dependent buffering. Controlled human ingestion studies at low concentrations report a favorable short-term safety profile; conversely, the principal metabolite (chlorite) is associated with documented developmental toxicity in animals and regulatory caution by EPA/WHO. Low-dose ClO₂ exposure is hypothesized—but not yet directly demonstrated—to elicit dose-dependent adaptive redox responses and redox-regulatory modulation. Conclusions: The one-electron, non-chlorinating chemistry of ClO₂ provides a coherent mechanistic framework. The translation of this framework into clinical application requires prospective trials and direct mechanistic measurement, ideally conducted by independent investigators. The following mechanisms are hypothesized and have not yet been directly demonstrated in vivo: modulation of oxidative phosphorylation, mitochondrial oxygenation, and associated redox-regulatory signaling. Reported antimicrobial efficacy typically occurs in the low-ppm range (0.1–5 ppm), whereas adverse effects on mammalian cells in the reviewed literature are reported at concentrations at least one order of magnitude higher; direct head-to-head cytotoxicity comparisons remain scarce and are identified as a priority for future work. This differential thiol buffering—in particular, the higher glutathione content of many mammalian cells relative to many pathogens—has been proposed as a contributing basis for host-cell sparing; direct quantitative confirmation in vivo is not yet available. Bacterial species possess their own thiol-based defense systems (e.g., mycothiol in actinomycetes, bacillithiol in Firmicutes), and the interplay between ClO₂ and these systems requires further investigation. Full article

Yellow rust (YR) is one of the most damaging diseases of wheat, causing substantial economic losses and threatening food security worldwide. The potential of Synthetic Aperture Radar (SAR) for YR monitoring remains largely unexplored compared with optical observations. This study investigates the ability of Sentinel-1 SAR time series to discriminate YR-infected and healthy wheat fields and compares their temporal responses with Sentinel-2 observations. Sentinel-1 intensity-based and polarimetric features and Sentinel-2 vegetation indices were computed and assessed using correlation analysis, the Mann–Whitney U test, Youden’s J statistic, and Random Forest classification. Sentinel-2 indices showed the strongest discrimination in May, with Mann–Whitney effect sizes of up to r = 0.60 and Youden’s J = 0.70 for TVI. The best-performing Random Forest models were based on NDMI and DSWI-1 in May and achieved a macro F1-score of 0.84. In contrast, Sentinel-1 features generally showed limited sensitivity, but some features revealed significant differences as early as March, several months before the fields were identified as severely YR-infected. ${SE}_I$ reached r = 0.60 and J = 0.60, while the best Sentinel-1 Random Forest model, based on LPR, achieved a macro F1-score of 0.70 in March. These findings indicate that Sentinel-2 vegetation indices provide stronger discrimination of YR-infected and healthy fields during the period of greatest disease spread while suggesting that Sentinel-1 may capture early differences associated with moisture conditions that favor subsequent YR development. Full article

Background/Objectives: Herpes zoster (HZ) vaccination rates remain low globally despite the availability of effective recombinant vaccines. This study aimed to assess knowledge of HZ and HZ vaccination, willingness to vaccinate, and factors associated with vaccination willingness among at-risk populations in both outpatient clinic and online survey settings in Poland. Methods: A questionnaire-based cross-sectional study was conducted between May and September 2025 in outpatient primary care clinics operated by Medical Center Medyk (Podkarpackie Voivodeship, southeastern Poland) and via online social media platforms. Participants included individuals aged ≥50 years with a history of HZ and individuals aged ≥18 years with at least one recognised HZ risk factor. Multivariable logistic regression and classification and regression tree (CART) analysis were employed to identify predictors of vaccination willingness. Results: Valid responses were obtained from 212 onsite and 241 online participants. A prior history of HZ was significantly associated with correct responses to several knowledge items concerning HZ and HZ vaccination in both settings. Over 40% of initially hesitant or unwilling onsite participants reported a positive change in vaccination intention following medical consultation. CART models demonstrated good discriminative performance (AUC: online 0.841, onsite 0.861). The primary difference between settings was the source of information: online participants relied more heavily on the internet, whereas onsite participants more frequently cited healthcare professionals. Conclusions: Personal experience with HZ and engagement with healthcare professionals significantly influence vaccination willingness. Targeted educational interventions delivered through appropriate communication channels, with emphasis on physician engagement, may help improve vaccination willingness among at-risk populations. Full article

Advanced Metering Infrastructure (AMI) underpins bidirectional communication in smart grids, a foundational layer of smart city energy systems, facilitating the flow of data, real-time monitoring, and demand-responsive control. But its connectivity exposes smart meters to data tampering, denial-of-service attack, and false-data-injection attacks. Most intrusion detection systems (IDSs) for AMI only report that an intrusion has occurred but cannot reconstruct how it propagated or where it originated, leaving the operators without the forensic evidence to perform containment. This paper proposes a multi-stage approach coupling detection with forensic analysis. A recurrent neural network (RNN) extracts temporal features, a support vector classifier (SVC) performs binary classification, and ant colony optimization (ACO) serves two purposes: feature selection before classification and a backward path reconstruction after an intrusion is confirmed. The proposed framework is evaluated on a simulated AMI network with forensic ground truth and further validated on the public UNSW-NB15 benchmark. The detection accuracy exceeds $96\%$, while ACO reduces the feature set from 40 to 14. A McNemar’s test ($p=0.265$) indicates that this feature reduction does not significantly alter the per-sample error pattern. With the use of the improved tracer, the Path Overlap Score increases from $0.29$ to $0.40$, while the False-Positive Path Rate decreases from $0.39$ to $0.19$, relative to the centroid baseline tracer used for forensic tracking. This improvement in the Path Overlap Score is statistically significant ($p=4.39\times {10}^{-8}$). However, the Source Localization Rate remains relatively low ($9\%$–$11\%$) for both methods, owing to the intrinsic difficulty of identifying the true source meter from incomplete alert data. Therefore, the proposed framework not only reliably detects intrusions but also significantly outperforms the baseline tracer in path overlap and false-positive rate, although precise source localization remains an open challenge. Full article

Post-quench natural aging can alter the response of Al–Mg–Si alloys to further artificial aging, while post-artificial-aging hardness, H_post-AA, may obscure changes in the retained artificial-aging response. Five high-purity alloys containing 0.95–0.99 mol% Mg + Si, with Mg/Si ratios of 0.52, 1.1, 1.9, 3.0, and 4.0, were solution-treated at 848 K for 3.6 ks, quenched, and artificially aged at 473 K. Under direct artificial aging, after a minimum practical delay of approximately 0.1 ks, the Mg/Si = 1.1 alloy showed the highest H_peak of approximately 72–73 HV0.1, whereas Mg/Si = 4.0 reached approximately 53–55 HV0.1. The precipitate areal density was highest near Mg/Si = 1 and decreased markedly in Mg-rich alloys. Operational HRTEM classification indicated that β″-related precipitates predominated in the Si-excess alloys, whereas β′-like and parallelogram-type precipitates were more prominent in the Mg-rich alloy. For Mg/Si ratios of 0.52, 1.9, and 4.0, quench-to-aging delays of up to 6000 ks produced non-monotonic changes in H_post-AA. However, the additional hardening increment, ΔH_AA, decreased from 38.5 to 26.1 HV0.1 at Mg/Si = 0.52 and from 33.5 to 26.9 HV0.1 at Mg/Si = 1.9. These results show that both H_post-AA and ΔH_AA are required to evaluate quench-to-aging delays. Full article

Rockburst intensity classification supports risk assessment in deep underground engineering, but its performance is often constrained by limited sample size and class imbalance. This study evaluates the response of five fixed classifier configurations to jointly controlled changes in sample size and class imbalance. A cleaned 331-case four-class database was used to construct 16 conditions defined by $N=\left\{50, 100, 150, 192\right\}$ and $I{R}_{\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{e}\mathrm{t}}=\left\{1, 2, 4, 6\right\}$. RF, SVM, CW-SVM, SMOTE-SVM, and TabPFN were evaluated over 10 controlled repeats using shared four-fold stratified partitions. Macro-F1 was the primary metric, with balanced accuracy and MCC as complementary metrics. After study-wide Holm correction, the Friedman tests remained significant in 15 of 16 conditions for Macro-F1 and balanced accuracy and in 14 of 16 for MCC. TabPFN achieved the highest mean Macro-F1 in 10 conditions, RF in five, and SMOTE-SVM in one. None of TabPFN’s 10 mean leads over the second-ranked configuration were significant after condition-specific Holm correction. The only single-member Macro-F1 competitive set occurred at $N=100$ and $I{R}_{\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{e}\mathrm{t}}=6$ for RF, while none of the 480 pairwise Wilcoxon tests remained significant after study-wide Holm correction. In the 12-case independent engineering case validation, CW-SVM achieved the highest accuracy and MCC, whereas TabPFN achieved the highest Macro-F1 and balanced accuracy. These results indicate condition-dependent mean performance and statistical competition without evidence of universal superiority among the evaluated configurations. Full article

Background/Objectives: Advances in cancer treatment and an aging population have increased ICU admissions among patients with cancer, while mortality remains high and ICU resources are limited. We aimed to identify mortality predictors available at admission and factors arising during the ICU stay and to evaluate retrospective concordance with ICU admission-prioritization guidelines. Methods: This retrospective single-center cohort included 287 patients with an active malignancy admitted to an oncology ICU in Lublin, Poland, from 1 January 2024 through 31 December 2025. Two independent reviewers, blinded to outcomes, retrospectively assigned ICU admission priorities according to Polish Society of Anaesthesiology and Intensive Therapy guidelines. Univariable and multivariable logistic regression analyses were used to identify factors associated with ICU mortality. Results: Overall ICU mortality was 41.1% (118/287). In the adjusted admission-time model, APACHE II score (odds ratio (OR), 1.15 per point), ECOG 4 (OR, 6.37), and postoperative admission (OR, 0.29) were associated with mortality (area under the receiver operating characteristic curve (AUC), 0.84). In the exploratory ICU-course model, mechanical ventilation >48 h or death within 48 h while mechanically ventilated (OR, 6.79) and vasopressor therapy >48 h or death within 48 h while receiving vasopressors (OR, 4.33) were associated with mortality. The proportion of priority-4 admissions decreased from 19.2% (24/125) in 2024 to 4.9% (8/162) in 2025 (p < 0.001). Priority-4 classification remained associated with ICU mortality even after adjustment for APACHE II score. Conclusions: Cancer, even at an advanced stage, should not be regarded as an absolute contraindication to intensive care. Clinical and oncological information available at admission may support evidence-based ICU triage. Structured prioritization may help identify patients with a very limited likelihood of benefiting from intensive care and support responsible allocation of critical care resources. Full article

Non-specific lipid transfer proteins (nsLTPs) are small cysteine-rich proteins belonging to the cysteine-rich peptide (CRP) antimicrobial peptide superfamily, widely distributed across land plants and involved in key physiological processes including fertilization, cell wall organization, wax deposition, and responses to biotic and abiotic stresses. Recent genomic and evolutionary analyses have revealed an extensive diversification of nsLTPs across plant lineages, which has led to multiple classification systems based on molecular mass, cysteine spacing, intron structure, and GPI (glycosylphosphatidylinositol-anchored proteins)-anchor motifs. This diversity reflects their adaptive expansion during plant terrestrialization. This review proposes that nsLTPs act as central nodes connecting lipid metabolism, developmental signals, and stress response networks, functioning as integrative signaling hubs rather than simple lipid carriers. Building on this central thesis, we synthesize current knowledge on nsLTP structure, classification, and evolution, and examine their multifaceted roles in plant development and defense. We also highlight emerging biotechnological applications and identify key gaps that warrant further investigation. Full article

Rift Valley fever (RVF) is a vector-borne zoonotic disease whose occurrence is strongly associated with climatic and environmental conditions, making data-driven approaches potentially valuable for epidemiological surveillance and risk assessment. Using a publicly available historical dataset comprising 180,288 monthly observations from geographically defined administrative units across Kenya between 1981 and 2010, this study investigates machine learning (ML) for the retrospective classification of reported RVF occurrence from contemporaneous climatic, environmental, topographic, and seasonal predictors under an extremely imbalanced classification setting. The dataset provides broad geographic coverage across Kenya over a 30-year historical period; however, because the outcome reflects reported events in historical surveillance records, it is not assumed to constitute a formally population-representative national sample or to capture all underlying RVF transmission. Each observation represents a geographic unit and observation month, and the response indicates whether an RVF event was reported during that corresponding period. Therefore, the present analysis should be interpreted as contemporaneous outbreak classification rather than as a fixed-horizon prospective forecast. Thirteen classifiers representing distinct learning paradigms were systematically evaluated: Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Classification and Regression Tree, Naive Bayes, Support Vector Machine, Weighted Logistic Regression, XGBoost, LightGBM, CatBoost, Balanced Random Forest, EasyEnsemble, and RUSBoost. Model performance was assessed before and after SMOTENC-based rebalancing using overall and class-specific metrics, including accuracy, precision, sensitivity, specificity, F1-score, ROC–AUC, and precision–recall-based measures. Under the retrospective stratified hold-out benchmark, XGBoost, CatBoost, Balanced Random Forest, and LightGBM achieved ROC–AUC values of 0.9176, 0.9175, 0.9114, and 0.9062, respectively. Balanced Random Forest attained the highest outbreak sensitivity (0.8851), although at the cost of very low precision, illustrating that high rare-event detection can generate a substantial false-alert burden in surveillance settings. SMOTENC produced strongly model-dependent effects: it increased outbreak sensitivity for XGBoost, LightGBM, CatBoost, KNN, CART, and RUSBoost, but substantially reduced sensitivity for Balanced Random Forest and EasyEnsemble. SHAP-based interpretability analysis indicated that month, rainfall, and slope were among the most influential predictors and further showed that class rebalancing can alter the distribution of feature contributions. Overall, the findings demonstrate that modeling reported RVF occurrence under severe class imbalance requires joint evaluation of minority-class detection, false-positive behavior, discrimination, and model interpretability rather than overall accuracy alone. The present results establish a retrospective classification benchmark for climate-informed RVF risk assessment, but they should not be interpreted as an autonomous outbreak-warning system. Translation into prospective early-warning prediction will require an explicit forecasting horizon, predictors constructed exclusively from information available before the target period, temporally and geographically independent validation, and decision thresholds evaluated against an operationally acceptable false-alert burden. Full article

Precision Approach Path Indicator (PAPI) systems are critical visual aids for supporting flight crews during the final approach phase by providing visual information on the aircraft’s vertical position relative to the desired glide path. Their correct operation and angular setting are therefore essential for maintaining operational safety at aerodromes. This study develops and field-evaluates a deterministic UAS-based workflow for automatic PAPI light localization, RED/WHITE/UNDEFINED state classification, red–white transition detection, and angular reconstruction from geotagged UAS observations. Data were acquired at two Colombian aerodromes, Perales Airport (SKIB) and Flaminio Suárez Camacho Aerodrome (SKGY), using a DJI Matrice 400 equipped with Zenmuse P1 and H30T optical payloads under manual vertical flight profiles and two illumination conditions. The proposed algorithm integrates luminance-based saliency extraction, four-light geometric validation, localized ROI-based chromatic classification, and RED → WHITE transition-event detection to estimate the transition angle of each PAPI unit from geotagged UAS observations. Results differed between the two evaluated site–payload configurations. In the SKGY–H30T configuration, the use of zoom and background-attenuation settings facilitated target isolation, and all 255 processed images were correctly classified during manual verification. In the SKIB–P1 configuration, 96.57% image-level classification accuracy was obtained, with the observed misclassifications mainly associated with city lights and luminous halos between adjacent PAPI units. Angular agreement with theodolite measurements also differed between the two datasets. Because each optical payload was evaluated at a different aerodrome, these differences cannot be attributed independently to payload characteristics; they represent the combined response of the payload, acquisition settings, background complexity, illumination, and site-specific conditions. Full article