Search Results (1,063)

Accurate bee counting is important for colony monitoring, pollination assessment, and precision beekeeping, but manual counting and dense point annotation are labor-intensive. This study proposes M3DANet, a lightweight semi-supervised density regression network with a handheld edge deployment system for bee colony counting. A dataset containing 586 valid high-resolution images and 34,869 point annotations was constructed for training and evaluation. M3DANet uses the first seven stages of MobileNetV3-Large as the lightweight backbone and combines multi-scale context encoding, attention-guided low-level feature fusion, and teacher–student consistency learning with confidence masking and warm-up training. The 10%, 30%, and 50% labeled data settings refer to the proportions of labeled images in the training set, and the remaining training images are used as unlabeled data. Mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation metrics. On the main dataset, M3DANet achieved MAE values of 9.937, 7.003, and 5.570 and RMSE values of 13.093, 9.387, and 7.620 under the 10%, 30%, and 50% settings, respectively, outperforming representative semi-supervised baselines. Under the fully supervised setting, it achieved an MAE of 5.201 and an RMSE of 6.989 with only 2.095 M parameters and 416.64 FPS, using 87.1% fewer parameters and running 17.7 times faster than CSRNet. Cross-species experiments confirmed its low-label generalization ability. Jetson Orin NX deployment achieved 65.75 ms/image inference latency and 10.44 FPS complete-pipeline throughput. These results show that M3DANet balances counting accuracy, annotation efficiency, generalization, and edge deployment practicality. Full article

Background/Objectives: This meta-analysis aimed to quantify the combined ergogenic effects of caffeine (CAF) and carbohydrate (CHO) on high-intensity interval exercise (HIIE) performance and to identify potential moderating factors. Methods: Four databases were systematically searched to identify randomized crossover trials assessing CAF combined with CHO (CAF + CHO) on HIIE performance (i.e., exercise time, distance, or total work). A three-level random-effects meta-analysis was performed to calculate pooled Hedge’s g (g) values. Moderator effects were explored through subgroup analyses, including control group (placebo-, CHO-, and CAF-controlled), CHO administration (mouth rinse and ingestion), and training status (recreationally active and trained). Results: Eleven studies met the inclusion criteria (n = 105; 8 female). CAF + CHO significantly enhanced HIIE performance (g = 0.44, 95% CI: 0.23–0.66). Subgroup analyses indicated that CAF + CHO mouth rinsing (g = 0.91, CI: 0.49–1.33) yielded superior effects compared to CAF + CHO ingestion (g = 0.33, CI: 0.14–0.52) (p for subgroup < 0.05). Performance improvements with CAF + CHO were observed for CHO- and placebo-controlled trials, but not in CAF-controlled trials, without significant subgroup effects (p for subgroup > 0.05). Importantly, evidence of publication bias was identified, and the overall certainty of evidence was graded as low according to the GRADE framework. Conclusions: CAF + CHO appears to be effective for enhancing HIIE performance, with greater benefits observed when CHO is administered via mouth rinsing rather than ingestion. Preliminary evidence suggests that CAF may play a key role in CAF + CHO strategies. However, given the limited number of female participants, the generalizability of these findings to both sexes is limited. Additional high-quality trials are needed to establish more definitive recommendations. Full article

The understanding of oxidative stress is being refined leading to the use of the terms “oxidative distress” and “eustress”. This reflects the dual role of reactive oxygen species (ROS) and reactive nitrogen species (RNS) in both pathology and physiology, emphasizing the complexity of the mechanisms influencing the redox status. This review discusses how these redox mechanisms interact with key signaling pathways, specifically the mammalian/mechanistic target of rapamycin (mTOR) and peroxisome proliferator-activated receptor-gamma coactivator (PGC-1α), which are crucial for mitochondrial health and muscle recovery. During exercise, the contraction of skeletal muscles increases ROS production which, through redox signaling, triggers mitochondrial biogenesis, enhances the antioxidant defenses and stimulates glucose metabolism, contributing to cardiovascular function and health. There is a large consensus about the importance of physical exercise in maintaining the redox homeostasis. However, the redox status could be disturbed after an intense and/or long physical effort, and signs such as markers of oxidative distress were identified. In that context, antioxidant strategies are warranted to prevent oxidative damage and help recovery. Given the many factors influencing the redox status of the body, including the training status, the duration and type of exercises and effort, diet, lifestyle, genetic polymorphisms, and circulating cytokines, a personalized approach is necessary. Targeted therapeutic interventions become important for preventing oxidative damage and helping recovery. In this review, we discuss the potential benefits of micro-immunotherapy (MI), as a multi-target approach utilizing signaling molecules, including cytokines at low doses (LD, typically 3–5 centesimal Hahnemannian CH dilutions) and ultra-low doses (ULD, from 6 CH upwards). We focused specifically on the investigational MI medicine 2LMIREG, and propose its application in preventing oxidative distress and restoring redox balance. Additionally, this review explores how the redox status interplays with the immune system, presenting preclinical data on 2LMIREG as a proof-of-concept for a tailored immunoregulatory strategy to enhance both immune and oxidative adaptations. Full article

Kombucha is traditionally produced by fermenting Camellia sinensis tea and sugar in a consortium of microorganisms called SCOBY (Symbiotic Culture Of Bacteria and Yeasts). Short- and medium-chain fatty acids and other organic acids in K are mainly produced by acetic acid bacteria, which contribute to the typical K taste. Coffee is one of the most widely consumed beverages in the world and one of the most traded commodities globally. Harvesting during coffee production generates tons of byproducts generally considered of low value, including cascara (CC), composed of dried pulp and skin, and leaves (CL). To date, few studies have investigated the production of short- and medium-chain fatty acids and monosaccharide’s profile during traditional kombucha fermentation, and their composition in kombuchas prepared from substrates other than C. sinensis is even scarcer. This study followed the changes in sugars and the production of short- and medium-chain fatty acids during K fermentation of black tea (BT), CC, and CL and associated their concentrations with physicochemical parameters (total soluble solids (TSS), pH, and titratable acidity (TA)) and the perceived acidity of the beverages evaluated by a trained panel and untrained consumers. BT K, a SCOBY, and 10% sucrose were added to infusions of arabica CC, CL, or BT. The mixture was fermented for 0, 3, 6, and 9 days. Organic acids were analyzed by GC-MS; sucrose and monosaccharides were analyzed by HPLC-RID. The Rate All That Apply (RATA) test was used for sensory analysis. Results were treated by ANOVA–Fisher and Pearson correlation tests with significance at p < 0.05. Glucose, fructose, arabinose, xylose, cellobiose and glycerol were identified in the infusions. On average, sucrose concentration decreased by 28% up to day 9, considering all K samples, accompanied by TSS decrease. Eight organic acids were semi-quantified, with acetic being the major acid in all beverages (8.4 to 1971 mg L⁻¹) and isovaleric being the lead minor acid (0.7 to 17.7 mg L⁻¹). Additional acids identified were: butanoic, 2-methylpropanoic, pentanoic, 3-methylpentanoic, hexanoic, and octanoic acids. TA values and sourness perceived by consumer assessors increased generally, even though in CC Ks, the acid concentration decreased by day 9. TA, sourness, and sparkling and fizzy mouthfeel correlated positively in all Ks. In general, although the total acid concentration was mainly higher on days 3 or 6, CO₂ formation, among other organic acids, probably increased TA and sourness on day 9. Although it is generally accepted that pH and organic acid concentrations are directly associated with sour taste, it is not possible to accurately predict and modify sour taste intensity in kombucha based only on these parameters, given that other factors, such as the production of CO_2, the existence of buffer systems, and the presence of sugars and other soluble solids, will probably affect the perceived acidity and sourness. Full article

Nitrogen control in basic oxygen furnace (BOF) steelmaking is critical, as dissolved nitrogen concentrations exceeding 30–40 ppm detrimentally affect the mechanical properties and formability of low-carbon steel products; however, no prior study has applied a physics-informed machine learning model to nitrogen prediction at this process stage. A NitroPINN model was developed incorporating a multiplicative prediction structure that embeds Sievert’s law equilibrium, Wagner interaction coefficients, and Byrne–Belton surface blockage theory directly into the model. The model was trained and evaluated on 66 matched industrial heats from a top-blown 170-ton BOF converter, characterized by 16 physics-informed features, and benchmarked against ridge regression and a pure multilayer perceptron (MLP) under five-fold cross-validation. The NitroPINN achieved the lowest mean absolute error (MAE = 5.60 ppm) and mean absolute percentage error (MAPE = 27.2%) among the three models, whilst the learned equilibrium attainment factor η averaged 0.456 ± 0.028, consistent with sub-equilibrium nitrogen conditions imposed by intense CO flushing during oxygen blowing. All three models exhibited comparable overall accuracy, confirming that dataset size constitutes the principal performance bottleneck. The primary advantage of the NitroPINN lies in its physical interpretability, constraining predictions to metallurgically plausible ranges and providing a transparent decomposition into thermodynamic and kinetic contributions. Full article

Despite advances in lifestyle-based therapy, achieving clinically meaningful reductions in blood lipid levels remains a major challenge in obese men with secondary hypercholesterolemia. Hypoxic exposure encompassing both training sessions and nocturnal rest may offer a novel adjunct to conventional interventions; however, no study has evaluated such a protocol in this population. Twenty sedentary men with obesity-related hypercholesterolemia were randomly allocated to a hypoxic group (H) or normoxic control group (C). Both groups completed an identical two-week high-intensity training program under an individualized calorie-restricted diet, residing at the same lowland location (~100 m above sea level). The H group trained and rested under normobaric hypoxia (FiO₂ = 14.4%, simulated altitude ~3000 m, 8 h nightly); C remained under normoxic conditions. The H group demonstrated significantly greater reductions in body mass (−4.1%) and fat mass (−11.0%). Significant reductions in total cholesterol (−20.1%), low-density lipoprotein cholesterol (−21.3%), non-high-density lipoprotein cholesterol (−23.1%), atherogenic index of plasma (−42.4%), and Castelli Risk Index I (−19.4%) occurred exclusively in the H group, accompanied by a strong downward trend in Castelli Risk Index II (p = 0.072). High-density lipoprotein cholesterol did not change; for triglycerides, a clear downward trend was observed in the H group, approaching statistical significance within-group (p = 0.052). The magnitude of cholesterol reduction was significantly associated with body mass and fat loss (r = 0.61–0.67). A two-week intervention combining hypoxic training with nocturnal normobaric hypoxic exposure and caloric restriction produces clinically relevant improvements in lipid profile and body composition in men with obesity-related hypercholesterolemia. Full article

Handball requires athletes to sustain intermittent high-intensity effort while maintaining rapid cognitive processing and technical skills. Caffeine is widely used as an ergogenic aid, yet its dose-dependent effects across physical, cognitive, and technical performance outcomes remain unclear in female handball players. This randomized, double-blind, placebo-controlled crossover study examined the acute effects of low-dose caffeine (LCAF; 3 mg/kg), moderate-dose caffeine (MCAF; 6 mg/kg), and placebo (PLA) in trained female handball players. Participants (n = 20) completed three experimental sessions separated by 72 h. Cognitive performance was assessed using the Simplified Eriksen Flanker Test, throwing performance was evaluated through maximal ball velocity, and intermittent running capacity was measured with the Yo-Yo Intermittent Recovery Test Level 1. Both LCAF and MCAF significantly improved Yo-Yo performance compared with PLA (η²p = 0.415, representing improvements of approximately ~23.5% and 29.0%), with no difference between caffeine doses. MCAF significantly reduced overall Flanker response time (η²p = 0.486, ~18.5%) and congruent and incongruent trial response time compared with PLA and LCAF. No significant effects were observed for throwing velocity, Flanker accuracy and interference scores. These findings suggest that acute caffeine intake has performance-specific effects in female handball players: intermittent running performance responded to both doses, whereas cognitive enhancement was limited to response time, with no improvements in inhibitory control or accuracy. Full article

Breast cancer survivors frequently experience adverse changes in body composition, cardiometabolic biomarkers, functional capacity and quality of life that may worsen long-term prognosis, yet the comparative effectiveness of lifestyle interventions across delivery formats and supervision levels remains unclear. Background/Objectives: This systematic review assessed the effects of structured diet and exercise interventions on body composition, metabolic and inflammatory biomarkers, functional capacity, dietary habits and quality of life in breast cancer survivors. Methods: Following PRISMA guidelines, Cochrane, PubMed, Scopus and Web of Science were searched for randomized controlled trials and quasi-experimental studies published in English between 2016 and 2026. Risk of bias was assessed with RoB 2 and ROBINS-I and certainty of evidence with GRADE. Results: Of 1413 records, 15 studies (11 RCTs; mean age 46–60 years; mostly overweight or obese post-treatment women) met the inclusion criteria; twelve interventions were supervised and three home-based or web-based. Within the assessed domains, many studies reported significant improvements in body composition, quality of life and metabolic or inflammatory biomarkers. Effects were larger in multimodal supervised programs combining caloric restriction with moderate-to-vigorous aerobic plus resistance training (5–8% weight loss; 19–29% visceral fat reduction; improved insulin, IGF-1, leptin, adiponectin and EORTC QLQ-C30 scores), whereas digital or low-intensity interventions produced smaller, less uniform objective effects despite improving dietary behaviors. GRADE certainty ranged from very low to moderate–high. Conclusions: Multimodal supervised programs offer the most robust benefits; digital formats require additional supervision. Standardized protocols and longer follow-up are needed. Full article

Highway–rail grade crossing (HRGC) safety analysis is often based on raw incident counts or site-level models that do not adequately control for regional crossing-density exposure and frequently ignore spatial dependence. This limits the ability to identify where risk is structurally concentrated across the rail network. The problem is important because misidentifying high-risk environments leads to inefficient allocation of limited safety resources and weakens corridor-level intervention strategies. This study introduces accumulated incidents per crossing (AIPX), a crossing-count-normalized cumulative incident intensity metric that measured cumulative incident burden at the county level over a 51-year period (1975–2025). The study developed an algorithmic framework that integrates data reconciliation with spatial autocorrelation analysis, distributional modeling, and nonparametric machine learning to identify and interpret high-intensity risk environments. Global Moran’s I indicates statistically significant positive spatial autocorrelation (I = 0.359, p = 0.001), suggesting that incident intensity is spatially clustered rather than random. Local indicators identify statistically significant high and low intensity county clusters. Distributional analysis shows that AIPX within high-intensity clusters was best represented by lognormal and Johnson SU distributions. Machine learning models achieved strong classification performance (AUC ≈ 0.85). Explainability methods consistently identified temperature, train direction, crossing warning configuration, train composition, and track class as dominant associated features. These variables function as proxies for broader geographic, operational, exposure, and network-structure differences rather than direct causal drivers. The findings indicate a pattern consistent with regional and network-level exposure regimes concentrated along freight-intensive corridors. The study provides a transparent analytical workflow that supports corridor-level prioritization of safety interventions and more effective allocation of infrastructure investments. Full article

Background: Multimodal rehabilitation represents standard practice in musculoskeletal care, where exercise therapy is routinely combined with manual therapy, electrotherapy, education, and cognitive–behavioral strategies. However, research has largely evaluated these modalities in isolation, and no bibliometric synthesis has characterized multimodal rehabilitation despite its predominance in routine practice. Objective: To characterize global research activity, thematic clusters, and diagnostic patterns underpinning multimodal musculoskeletal rehabilitation and to examine their alignment with contemporary rehabilitation guidelines and practice models. Methods: A bibliometric and altmetric analysis was performed using Web of Science Core Collection (1989–2026). Studies indexed under exercise therapy, manual therapy, electrotherapy, education, and cognitive–behavioral approaches were included. Network analyses (co-occurrence, co-authorship, thematic evolution, and bibliographic coupling) were conducted using Bibliometrix and VOSviewer. Diagnostic subgroups included osteoarthritis, low back pain, chronic musculoskeletal pain, tendinopathy, and shoulder disorders. Results: A total of 409 publications were identified. Five multimodal combinations were recurrent: exercise + education, exercise + cognitive–behavioral therapy, exercise + manual therapy, exercise + electrotherapy, and mixed multimodal programs. Diagnostic subgrouping showed distinct patterns, with osteoarthritis and low back pain clustering around exercise + education, chronic musculoskeletal pain around exercise + CBT/self-management, and tendinopathy/shoulder disorders around exercise + manual therapy. Temporal analyses demonstrated a shift from unimodal electrophysical agents toward guideline-aligned biopsychosocial models. Altmetric signals suggested relevant dissemination and policy attention. Conclusions: Multimodal musculoskeletal rehabilitation is research-intensive, diagnosis-specific, and aligned with guideline recommendations prioritizing exercise, education, self-management, and behavioral strategies. These findings support multimodal rehabilitation as a maturing evidence-based practice model with implications for pragmatic trials, guideline implementation, and clinical service delivery. Beyond research implications, these patterns are relevant for musculoskeletal care pathways, training of rehabilitation professionals and health system planning. Full article

Speech emotion recognition systems are typically trained on audio sampled at conventional bandwidths that exclude frequencies above approximately 8 kHz, yet the contribution of extended high-frequency information to vocal emotion recognition remains unclear. This study examines how spectral bandwidth influences automatic emotion classification using the RAVDESS corpus of acted speech and song. Recordings were low-pass filtered to simulate multiple bandwidth conditions (8, 12, and 16 kHz, along with the original full-bandwidth signal), and classification was performed using a Random Forest model trained on mel-spectral features. In addition to classification accuracy, we analyzed permutation-based spectral feature importance and the geometry of the classifier’s posterior-probability space. Bandwidth restriction had relatively modest effects on classification accuracy overall, with mean accuracy ranging from approximately 55% to 77% across conditions, although its impact was greater for speech than for song. Feature-importance analyses indicated that the model depends primarily on low- and mid-frequency spectral information, whereas higher-frequency and EHF regions show increased importance when available. Geometry analyses showed no reliable evidence that bandwidth altered the global structure of the stimulus-level emotion space, although spectral truncation reduced separability for certain emotion contrasts, particularly in speech at normal emotional intensity. These results indicate that most acoustic information supporting categorical emotion recognition resides in lower spectral regions, while EHF information provides supplementary acoustic information that may refine some emotional distinctions under specific conditions. Full article

Background: Predicting rehabilitation outcomes at admission supports tailored therapy plans and efficient use of resources for patients undergoing intensive inpatient rehabilitation, including those with stroke, orthopedic, and other neurological conditions. Nonetheless, current machine learning (ML) methods face limitations, including the ceiling effect in absolute functional gain measures, the uniform treatment of diverse patient groups, and reliance on black-box models that lack clinical transparency. Methods: This retrospective observational study analyzed a fully anonymized, publicly available dataset of 3419 patients admitted to the intensive rehabilitation unit at IRCCS San Raffaele Hospital, Rome, Italy, from 2015 to 2018. To mitigate the ceiling effect, a normalized Barthel Index gain metric was developed. K-means clustering (K = 2, trained solely on the training set) identified patient admission profiles based on functionality, which were then used as predictive features. Eight machine learning classifiers were tested across three groups (All Patients, Orthopedic, Neurological). SHAP-based explainability was employed at four levels: global, diagnostic group, patient functional profile, and individual. Finally, clinical decision rules and bedside stratification profiles were derived and validated with an internal held-out test set (n = 684). Results: Normalization significantly increased the correlation between admission BI and gain (r = 0.130 to r = 0.520), supporting the presence of a ceiling-related limitation in absolute gain metrics. Two distinct functional admission profiles with statistically significant group differences were identified—High-Burden (38% below-median recovery) and Moderate-Burden (21%)—with cluster membership the third most important predictor (13.9% SHAP importance). The highest AUC-ROC values were 0.831 for all patients (XGBoost), 0.864 for neurological patients (Gradient Boosting), and 0.839 for orthopedic patients (Gradient Boosting). Multilevel SHAP analysis showed age as the primary predictor for neurological patients (mean |SHAP| = 0.360) but the third for orthopedic patients (0.350), highlighting clinical relevance. Validation using SHAP values from the Gradient Boosting model showed a Spearman correlation of ρ = 0.925 (p = 1.13 × 10⁻³⁰), with eight of the top ten features overlapping, indicating that these patterns are not model-specific but reflect the underlying data. Risk zone stratification found 80.7% of patients in high-confidence zones (accuracy > 80%). The clinical decision rules achieved 70.8% accuracy with full transparency, and the elderly (≥75 years) combined with a low BI (<25) profile showed an 89.6% model accuracy with only 10.4% recovery above the median. Conclusions: This explainable, profile-informed ML pipeline addresses key methodological limitations in predicting rehabilitation outcomes. It also provides a foundation for integrating models into clinical practice, pending prospective, external validation of the results. Before clinical implementation, validation across multicenter cohorts is essential. Full article

Background: Although early mobilization has been shown to improve clinical outcomes after intensive care unit (ICU)-acquired weakness, its implementation remains limited in routine clinical practice. This study aimed to evaluate the feasibility, safety, and preliminary clinical outcomes of a mobile application-based rehabilitation program in non-ambulatory patients during the early ward phase following ICU discharge. Methods: This prospective single-arm pilot trial included adult patients (≥19 years) who had received ICU care and demonstrated limited ambulatory function, defined as Functional Ambulatory Category (FAC) ≤3. Participants received an individualized, application-guided exercise program comprising two daily sessions over two weeks. Primary outcomes were programmatic feasibility, safety, and patient satisfaction. Rehabilitation compliance was quantified using application usage logs and categorized as high (≥50%) or low (<50%). Secondary functional outcomes, such as Medical Research Council Sum Score (MRC-SS), ICU Mobility Scale, FAC, muscle strength measures, health-related quality of life, and pain scores, were assessed at baseline, week 1, and week 2. Results: Of the 25 initially enrolled patients, 5 dropped out due to clinical status changes or transfers, yielding a retention rate of 80.0%. For the 20 analyzed patients (mean age 52.7 ± 13.9 years; 45% male), the overall mean rehabilitation compliance was 40.6%. No serious adverse events related to the intervention were reported, and overall patient satisfaction and application usability were high. Progressive increases in exercise intensity and training levels were observed throughout the intervention period. Significant improvements over time were found in MRC-SS, ICU Mobility Scale, FAC, grip strength, health-related quality of life, and pain scores (all p < 0.05). Although compliance-based recovery trajectories were confounded by small subgroup sizes and baseline clinical imbalance, exploratory analyses nonetheless identified statistically significant time × compliance interaction effects for MRC-SS and straight leg raise performance. Conclusions: This pilot study demonstrates that a mobile application-based rehabilitation program is a feasible and safe approach to implement in deconditioned patients after ICU discharge. These preliminary functional recovery trajectories provide encouraging signals, suggesting that this digital platform may serve as a potential adjunct to conventional care. Rigorous, randomized controlled trials are required to confirm its definitive clinical efficacy and scalability. Full article

Heart failure (HF) remains a major global health challenge, necessitating accurate yet accessible diagnostic tools. While the left ventricular ejection fraction (LVEF) is the primary metric for classifying HF into preserved (HFpEF), mid-range (HFmrEF), and reduced (HFrEF) phenotypes, conventional imaging modalities such as echocardiography are resource intensive. In contrast, the electrocardiogram (ECG) offers a low-cost, non-invasive alternative for continuous cardiac assessment. This paper proposes a multi-algorithm artificial intelligence (AI) framework for automated HF phenotype classification using high-resolution ECG signals from 303 patients with chronic heart failure from the MUSIC cohort. After preprocessing (normalization, bandpass filtering), we employed a hybrid approach combining the Pan–Tompkins algorithm for robust R-peak detection with the NeuroKit2 toolbox for the precise delineation of P, Q, S, and T waves. ECG recordings were then segmented using an adaptive beat-centric windowing strategy. From the segmented beats, we extracted a comprehensive set of temporal, morphological, and energy-based features, including RR, QRS, and QT intervals, along with P-wave, QRS-complex, and T-wave energies. These features were used to train and evaluate several ensemble machine learning models—Random Forest, XGBoost, CatBoost, LightGBM, and a stacking classifier—using a stratified 70–15–15 train–validation–test split with 5-fold cross-validation. The LightGBM model achieved the highest performance with a test accuracy of 98.45%, an AUC of 0.9989, and a macro F1-score of 0.9804, outperforming other ensembles and the stacking classifier. The results demonstrate that an AI-driven analysis of ECG-derived morpho-energy features can serve as a reliable, non-invasive screening tool for the accurate and early discrimination of HF phenotypes, potentially supporting clinical decision making and improving patient management in resource-limited settings. Full article

Sustainable aviation fuel (SAF) is widely recognized as a critical pathway for aviation decarbonization; however, its life-cycle carbon performance is highly sensitive to supply chain configurations. This study proposes a data-driven framework integrating life-cycle assessment (LCA) with a generative adversarial network (GAN) to model and optimize SAF supply chain pathways under structural constraints. A rule-constrained synthetic dataset comprising feasible pathways is constructed, incorporating feedstock sources, refinery locations, airport demand nodes, conversion technologies, transport modes, and distances. Each pathway is encoded into a numerical feature vector, and a GAN model is trained to learn the distribution of feasible configurations. Generated pathways are further validated through LCA-based post-processing to ensure physical feasibility and emission consistency. The results show that pathway-level carbon intensity varies significantly across configurations, with differences exceeding 30% under varying feedstock–transport combinations. The model successfully captures the multimodal distribution of carbon emissions and identifies structurally consistent low-carbon pathways. In particular, localized supply structures and reduced transport distances are found to play a dominant role in minimizing emissions. This study provides a scalable methodological framework for SAF pathway modeling and offers insights into supply chain design and spatial configuration for achieving aviation carbon reduction targets. Full article