Search Results (6,781)

Background: Baseline lymphopenia is common among advanced solid tumors and may influence the efficacy/safety of immune checkpoint inhibitors (ICIs), but large real-world evidence is limited. We evaluated the association between baseline absolute lymphocyte count (ALC) and clinical outcomes in adults with solid tumors treated with ICIs in routine practice. Methods: We performed a retrospective cohort study using TriNetX. Adults with solid tumors who received pembrolizumab, nivolumab, or atezolizumab (ICI-Naïve) between January 2015 and June 2026 were included. Baseline ALC was measured within 30 days before first treatment and was classified as lymphopenic (ALC < 1.5 × 10⁹/L) or non-lymphopenic (ALC ≥ 1.5 × 10⁹/L). Propensity score matching (1:1) yielded 5249 patients per group. The index date was the first immunotherapy date, and outcomes were assessed at 6, 12, 24, 36 months, and 5 years. The primary outcome was 24-month overall survival (OS); secondary outcomes were OS at 6 and 12 months and 6-month risks of healthcare utilization, immune-related adverse events (irAEs), and serious infections; and exploratory outcomes included OS at 36 months and 5 years. All outcomes were analyzed using Kaplan–Meier analysis, Cox proportional hazards models, and risk ratios. Subgroup analysis included OS stratified by solid tumor subtypes and prior lines of therapy. Results: After matching, patients with baseline lymphopenia had consistently worse OS. Compared with patients without lymphopenia, the lymphopenia cohort had lower OS at 6 months (HR 1.29, 95% CI 1.22–1.37), 12 months (HR 1.28, 95% CI 1.21–1.35), 24 months (HR 1.26, 95% CI 1.2–1.33), and, in exploratory analyses with substantial right censoring and limited observed follow-up, 36 months (HR 1.26, 95% CI 1.2–1.33) and 5 years (HR 1.26, 95% CI 1.2–1.33), though these estimates should be considered hypothesis-generating only. At 6 months, baseline lymphopenia was associated with a greater healthcare utilization (RR 1.05, 95% CI 1.02–1.09), a higher infection risk (RR 1.08, 95% CI 1.01–1.15), and similar rates of clinically coded irAEs (RR 1.0, 95% CI 0.93–1.09), an observation subject to competing risk from early mortality in the lymphopenic cohort. Subgroup analysis, stratified by tumor subtypes and prior lines of therapy, showed consistently lower OS in the lymphopenia group, consistent with the primary outcome results. Conclusions: In this large propensity-matched real-world analysis of 10,498 patients with diverse solid tumors, baseline lymphopenia at ICI initiation was associated with persistently inferior OS at 6, 12, and 24 months (primary and secondary endpoints), greater early healthcare utilization, and a higher serious infection risk. Critically, lymphopenic patients developed irAEs at an identical rate to non-lymphopenic patients despite worse survival, a dissociation suggesting that baseline ALC stratifies patients along mortality risk and immune activation capacity as partially independent axes. These findings could support the use of baseline ALC as a simple, universally available biomarker that informs not only survival prognosis but also the anticipated toxicity profile of ICI therapy and highlight the need for competing-risk analyses and prospective immune phenotyping to characterize this relationship fully. Full article

Accurate landslide identification is crucial for enhancing emergency response capabilities during destructive geological hazards. Although deep-learning-based semantic segmentation has demonstrated effectiveness, substantial variations in landslide scales and environmental similarities continue to challenge existing methods. This paper systematically constructs a new co-seismic landslide dataset for the Yarlung Zangbo River basin based on the 2017 Nyingchi earthquake, effectively filling a critical regional data gap. This paper proposes CETransUNet (coordinate attention and edge-guided attention transformer UNet), a novel landslide detection model that integrates ResNet and Transformer architectures. Specifically, a coordinate attention (CA) module is introduced within the skip connections between the encoder and decoder. This module encodes positional information along both horizontal and vertical spatial directions and dynamically re-weights the feature maps, thereby effectively suppressing background noise caused by semantic gaps and enhancing the model’s ability to localize landslide regions. Additionally, an edge-guided attention (EGA) module is incorporated into the decoder. This module extracts explicit edge priors from the input image using a Laplacian operator and imposes geometric constraints on the predictions via a boundary reverse attention mechanism, thereby significantly alleviating boundary ambiguity and morphological distortion of landslides. Evaluations across datasets from the Yarlung Zangbo River, Iburi-Tobu, and Bijie regions demonstrate that CETransUNet significantly outperforms state-of-the-art models—including TransUNet, SegFormer, and SwinUNet—in terms of IoU, MIoU, and F1-score. Overall, through the synergistic optimization of the coordinate attention and edge-guided attention modules, the CETransUNet model achieves synchronous enhancement of boundary integrity and geometric precision in complex scenarios, providing a reliable technical solution for large-scale intelligent landslide identification. Full article

Accurate identification of visually similar targets in Unmanned Aerial Vehicle (UAV) imagery is hindered by significant inter-class ambiguity and viewpoint variability. While hierarchical deep learning mitigates these challenges, existing architectures relieve manual design, introducing subjectivity and limiting cross-domain scalability. In this work, we propose an objective, data-driven method for the automated synthesis of hierarchical classification structures. Our approach uses a hybrid inter-class proximity metric that integrates geometric distances between latent-feature-space centroids with empirical misclassification probabilities. Using a hierarchical agglomerative clustering algorithm optimized via an inconsistency coefficient, we synthesize a coarse-to-fine cascade that deploys YOLOv11 for feature extraction and FT-Transformers for specialized identification. Experimental validation on the VisDrone2019 and UAV123 datasets demonstrates that the automatically generated hierarchy achieves a peak F1-score of 94.9%, outperforming the monolithic YOLOv11 model by 0.8% and matching human-designed cascades. Sensitivity analysis indicates an optimal hybrid weight range of 0.4–0.6. The findings confirm that our automated synthesis provides high adaptability and reliability for real-time edge AI deployments, ensuring robust performance in dynamic monitoring environments without requiring manual redesign. Full article

Background/Objectives: Spinal ependymoma is the most common intramedullary spinal cord tumor in adults, and maximal safe resection is the cornerstone of treatment. Patients aged 75 years and older are underrepresented in surgical neuro-oncology cohorts. We sought to characterize treatment patterns and identify predictors of overall survival in very elderly patients with spinal ependymoma. Methods: We performed a retrospective cohort study of patients aged 65 years or older with spinal ependymoma using the National Cancer Database. The primary cohort was patients aged 75 years or older (very elderly); patients aged 65–74 years served as a comparison cohort. Multivariable Cox proportional-hazards models were fit within each cohort, and a surgery-by-age-cohort interaction was tested. Results: Of 1497 eligible patients aged 65 years or older with spinal ependymoma, 422 patients (28.2%) met criteria for the final analytic cohort. Intramedullary versus extramedullary tumor status was not available in the NCDB PUF and therefore could not be characterized. Very elderly patients were less likely to undergo surgery than the comparison cohort (70% vs. 85%; p < 0.001) despite similar tumor characteristics. Among very elderly patients, median overall survival was 59.7 months without surgery and 106.0 months with surgery, an approximately 46-month difference favoring surgery. Surgery was independently associated with lower mortality (HR 0.46; 95% CI, 0.24–0.89; p = 0.021). Increasing age (HR 1.15 per year; 95% CI, 1.07–1.22; p < 0.001), Charlson–Deyo score ≥ 2 (HR 4.41; 95% CI, 1.65–11.79; p = 0.003), and increasing tumor size (HR 1.02 per mm; 95% CI, 1.01–1.04; p < 0.001) were also independently associated with worse survival. In the 65–74 cohort, no significant association between surgery and overall survival was detected (HR 1.23; 95% CI, 0.54–2.81; p = 0.623), though statistical power was limited by only 7 deaths in the no-surgery arm. The surgery-by-age-cohort interaction was significant (HR 0.37; p = 0.043). Conclusions: Surgical resection was independently associated with improved overall survival in very elderly patients with spinal ependymoma despite lower utilization. Chronological age alone may be an imperfect basis for excluding older adults from surgical consideration. Full article

This paper presents Resilient Edge-IVA (Intelligent Video Analytics), an integrated framework designed to ensure real-time inference stability and high-speed embedding-based similarity search in resource-constrained edge computing environments. Conventional systems often face Quality of Experience (QoE) degradation caused by computational overhead and hardware-level bottlenecks. To address these challenges, this study proposes a “Whole-cycle” methodology employing a perception-driven, three-tier adaptive control algorithm. This algorithm dynamically modulates encoding parameters, such as resolution and bitrate, by utilizing real-time inference latency and CPU utilization as feedback signals. Furthermore, the framework incorporates an event-density-based Data Diet mechanism. This mechanism selectively adjusts video quality based on object detection results, preserving high-fidelity imagery for critical events while significantly reducing data volume during static intervals. The backend implements a hybrid storage architecture combining the Milvus vector database for CLIP-based high-dimensional visual embeddings with a PostgreSQL relational database for structured metadata. These systems are linked via a deterministic hash key to ensure data atomicity and facilitate high-speed, multi-dimensional embedding-based retrieval. Experimental evaluations conducted on a Raspberry Pi 5 and Hailo-8 NPU demonstrate that the proposed framework maintains a frame drop rate below 0.3% even under extreme workloads, providing a 13-fold improvement in operational stability over static configurations. The results also confirm a 54.2% reduction in total storage occupancy and a Hash Mapping Consistency (HMC) score of 0.89. These findings validate the framework’s effectiveness in reconciling real-time processing stability with storage efficiency. Building upon this baseline, future research will extend the framework to multi-class environments, targeting applications such as Intelligent Transport Systems (ITS). Full article

Background/Objectives: Transcranial direct current stimulation (tDCS) has been explored as a strategy for pain management, but no study has investigated its use in Greater Trochanteric Pain Syndrome (GTPS). This study evaluated the effects of the combination of resistance exercises (REs) with tDCS on pain, functionality, and quality of life in patients with GTPS. Methods: In this randomized, double-blind trial, adults with GTPS were allocated to receive tDCS with RE (intervention group, IG) or sham tDCS with RE (control group, CG). Supervised 20 min sessions occurred on four consecutive days. Anodal tDCS (2 mA) was applied over the primary motor cortex. The primary outcome was the VISA-G.BR score at day thirty. Secondary outcomes included pain, functionality, and quality of life at multiple time points, assessed by HAGOS, PQAS, McGill Pain Questionnaire, and SF-36. Results: Thirty patients were included. Both groups improved, but between-group differences were nonsignificant for the primary outcome (VISA-G.BR effect size, −0.16; 95% CI, −0.54 to 0.27; p = 0.460). Secondary outcomes followed a similar pattern. Conclusions: These findings reinforce the value of RE in GTPS while suggesting a limited role for short-term tDCS protocols. Future studies should investigate whether protocols involving a greater number of stimulation sessions may produce superior clinical effects. Full article

The selective detection of small, highly hydrophilic metabolites differing only in stereochemistry represents a major challenge in biosensor development. Here, we present a computational investigation to elucidate the molecular basis of the experimentally observed selectivity of a protein-based electrochemical biosensor toward myo-inositol over D-chiro-inositol. Although the two stereoisomers differ only in the orientation of a single hydroxyl group, they induce distinct dynamic effects on the protein recognition element. Molecular docking revealed comparable binding regions and similar affinity scores, indicating that selectivity does not arise from differences in binding site or docking energy. To investigate dynamic contributions, all-atom molecular dynamics simulations were performed in triplicate (3 × 100 ns) using the AMBER99SB force field and explicit TIP3P water. Trajectory analyses showed that myo-inositol forms a more persistent hydrogen bond network, resulting in reduced residue-level flexibility, more stable ligand–protein interactions, and enhanced local structural stabilization. Overall, these findings support a dynamic model of stereoselective recognition in which ligand-induced modulation of protein conformational ensembles, rather than static affinity, governs biosensor performance. This work highlights the value of molecular dynamics simulations in the rational design of biosensors targeting structurally similar analytes. Full article

Accurate automatic classification of seed varieties is important for seed sorting, quality assurance, and plant breeding, yet reliable discrimination remains difficult when cultivars exhibit highly similar visual characteristics. This study presents a reproducible and interpretable framework for the binary classification of two Turkish pumpkin seed varieties using tabular morphological descriptors extracted from segmented seed images. Unlike many previous machine learning studies in this domain, which offer limited interpretability and leave model decisions largely as a black box, the proposed approach places Explainable Artificial Intelligence (XAI) at the center of the analysis. The framework combines biologically meaningful feature engineering, Optuna-based hyperparameter optimization, repeated stratified cross-validation, and a comparative evaluation of XGBoost, LightGBM, and CatBoost. Model explainability was investigated using SHapley Additive exPlanations (SHAP) to identify the morphological traits driving both global and instance-level predictions, while corrected repeated k-fold t-tests were used to assess the statistical significance of performance differences, which confirmed comparable accuracy among the three boosting models and a significant advantage over the baseline classifiers. All three boosting ensembles consistently outperformed the baseline classifiers (SVM, Logistic Regression, and Random Forest) on the hold-out test set. CatBoost achieved the best overall results, with an accuracy of 0.888, an F1-score of 0.879, and an MCC of 0.777. SHAP analysis consistently highlighted compactness, roundness, eccentricity, and engineered interaction descriptors as the most influential predictors. Overall, the proposed XAI-driven framework provides an accurate and transparent solution for pumpkin seed classification. Full article

To support refined water quality evaluation and short-term early warning in multi-section river systems, this study developed three percentile-based composite indices: the Water Quality Index (WQI), the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), and the Entropy Weight Method (EWM). Monthly multi-parameter monitoring data from 2019 to 2025 were used, covering ten river sections (P1–P5, M1–M5). The three indices were compared in terms of statistical distribution, methodological consistency, and anomaly response. An integrated assessment–prediction framework was further established. Within this framework, a one-step prediction scheme was applied to evaluate four models: Long Short-Term Memory networks (LSTM), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost). The results show that WQI scores were generally high and fluctuated within a narrow range. A clear “ceiling effect” was observed in the moderate-to-high grade intervals. WQI also showed weak consistency with TOPSIS and EWM (r ≈ 0.29–0.32). In contrast, TOPSIS and EWM were more sensitive to water quality fluctuations and extreme risks, and were moderately correlated with each other (r ≈ 0.53). Using TOPSIS < 50 as the threshold, 49 severe anomalous events were identified. These events were mainly clustered in February–April 2020, April–July 2023, and June–September 2025, with sections P4, M1, and M2 acting as high-incidence sites. In several typical events, WQI values remained high, indicating that reliance on WQI alone may delay early warning. Prediction results further reveal that the choice of index strongly affects sequence predictability. Taking XGBoost as the reference, the median validation R² followed a stable gradient: WQI (0.807) > TOPSIS (0.723) > EWM (0.594). XGBoost yielded positive R² values across all indices and sections. It also achieved the most robust overall performance and the strongest cross-site, cross-index generalization capability. Full article

Synthetic Aperture Radar (SAR) is widely used for marine oil spill surveillance due to its all-weather capabilities and sensitivity to sea surface roughness. However, oil slicks often appear as dark formations that can be confounded with visually similar “look-alikes”, making automated detection and boundary delineation challenging. This study proposes a two-stage deep learning framework for oil spill mapping in Sentinel-1 SAR imagery. First, a ConvNeXt-T classifier screens image patches for likely slick presence, reducing the search space for dense prediction. Second, spill boundaries are extracted with a domain-adapted Segment Anything Model (SAM) configured for prompt-free, single-shot segmentation. The input representation is enhanced by combining preprocessed Sentinel-1 VV backscatter with Gray-Level Co-occurrence Matrix (GLCM) texture measures (homogeneity and variance) to better separate oil from heterogeneous background sea at the segmentation level. Quantitative evaluation against established segmentation baselines demonstrates that our adapted SAM achieves the highest overall accuracy, reaching an F1-score of 0.86. This outperforms traditional models such as UNet and CBDNet (0.83), as well as DeepLabV3, SegNeXt, and OFCNet (all at 0.82). Furthermore, an analysis of the wind speed on the test set shows that wind speed affects detectability but does not by itself determine segmentation quality. The results indicate that combining transformer-based screening with efficient foundation-model adaptation can provide accurate and scalable oil spill mapping for operational SAR monitoring. Full article

Conventional recommender systems often rely on shallow collaborative signals, which limits their performance under sparse and popularity-skewed conditions. To address this, we propose a knowledge-aware framework that combines an item hypergraph induced by user interaction histories, a top-k user similarity graph, and one-hop, relation-aware knowledge-graph aggregation. The hypergraph branch learns high-order item co-occurrence representations, which are aggregated into initial user vectors and then refined through user similarity propagation. On the item side, user-conditioned relation attention aggregates one-hop KG neighbors to produce semantic item representations. User and item representations are fused by an MLP scorer, and a lightweight popularity-aware post-scoring adjustment can optionally be applied to moderate head-item dominance. Experiments on MovieLens-1M, Last.FM and Book-Crossing show strong performance among the compared baselines in AUC, ACC, and Recall@K. Full article

This study examines whether digital infrastructure contributes to sustainable development by improving green total factor productivity (GTFP)—a comprehensive measure that jointly evaluates economic output and environmental performance—in the Yangtze River Economic Belt. We exploit the staggered implementation of the “Broadband China” pilot policy as a quasi-natural experiment and estimate its effects using panel data for 107 prefecture-level cities from 2010 to 2022. The empirical strategy combines a staggered difference-in-differences design with an event study framework. The baseline results show that the average treatment effect for the full sample is positive but not statistically significant at conventional levels under standard TWFE estimation; however, the Sun–Abraham interaction-weighted estimator confirms a significant positive effect (ATT = 0.080, p < 0.05), and the Goodman-Bacon decomposition shows that the TWFE estimate is driven primarily by clean comparisons (91% weight, 0% negative weights). Further analysis reveals substantial regional heterogeneity. The estimated effect is significantly positive in the central region (0.171, p < 0.05), positive but not significant in the eastern region (0.097), and negligible in the western region (−0.042). A similar pattern emerges across income groups: digital infrastructure generates significant gains in GTFP in high- and middle-income cities, whereas the effect is not identifiable in low-income cities. These results remain robust to propensity score matching, placebo tests, alternative specifications, and alternative measures. Exploratory mechanism analysis provides limited evidence that technological innovation and industrial upgrading mediate the effect of digital infrastructure on GTFP within the sample period, though the causal interpretation of mediation is constrained by the sequential ignorability assumption. The findings suggest that the environmental returns to digital infrastructure depend on local complementary conditions, especially human capital, institutional capacity, and industrial foundations. These results imply that digital infrastructure policy should be differentiated across regions rather than implemented uniformly. By demonstrating that the environmental returns to digital infrastructure are conditional on local complementary conditions, this study contributes to the sustainability literature by providing a framework for quantifying and monitoring the sustainability impacts of digital infrastructure policies, with implications for sustainable development strategies in developing economies. Full article

Background: Robot-assisted total knee arthroplasty (raTKA) improves the precision of component positioning and coronal alignment restoration, but it remains uncertain whether that technical accuracy modifies the clinical effect of alignment strategy in different varus phenotypes. The present report evaluates alignment strategy and correction magnitude, explicitly as a post hoc exploratory deformity-subgroup analysis within a randomized raTKA cohort. Methods: This single-center, open-label, randomized study enrolled 296 patients with varus knee osteoarthritis who underwent raTKA between 2023 and 2025 using either mechanical alignment (MA; n = 149) or limited/restricted kinematic alignment (lim.-KA; n = 147). The parent randomized comparison was conducted at the whole-cohort level; the deformity-based subgroups reported here were defined after the whole-cohort analysis and are therefore post hoc and exploratory. Patients were stratified according to preoperative varus severity into a mild-deformity subgroup (≤10°; lim.-KA-I n = 99, MA-I n = 102) and a moderate-deformity subgroup (11–20°; lim.-KA-II n = 48, MA-II n = 47). Outcomes included hip–knee–ankle angle (HKA), correction angle, range of motion (ROM), visual analog scale (VAS; 0–10 points), Knee Society Score (KSS; knee and function), Oxford Knee Score (OKS), and Forgotten Joint Score-12 (FJS-12) over 12 months. Estimates are presented with 95% confidence intervals where applicable. Because multiple post hoc subgroup comparisons were performed without formal multiplicity adjustment, p-values are interpreted descriptively and in conjunction with effect sizes and 95% confidence intervals. Results: The primary whole-cohort randomized comparison did not demonstrate an overall between-group advantage of either alignment strategy. The post hoc moderate-varus subgroup showed favorable unadjusted 12-month differences for lim.-KA versus MA in KSS-knee (+6.8 points; 95% CI 5.3 to 8.3; nominal p < 0.001), KSS-function (+4.0 points; 95% CI 2.7 to 5.2; nominal p < 0.001), OKS (+6.4 points; 95% CI 4.5 to 8.3; nominal p < 0.001), and FJS-12 (+11.3 points; 95% CI 9.4 to 13.1; nominal p < 0.001). In contrast, ROM favored MA rather than lim.-KA in the moderate-varus subgroup (−11.8°; 95% CI −16.6 to −7.0; nominal p < 0.001), indicating greater 12-month ROM after MA, and VAS pain, reported on a 0–10 scale, did not support a lim.-KA pain advantage (+0.26 points; 95% CI 0.05 to 0.48; higher scores indicate worse pain; nominal p = 0.018). Exploratory, unadjusted, post hoc 12-month alignment-by-deformity interaction terms were significant for ROM, KSS-knee, KSS-function, OKS, and FJS-12, but not for VAS. Because multiple post hoc comparisons were performed without formal multiplicity adjustment, the results are interpreted descriptively, along with effect sizes and confidence intervals. Conclusions: The primary randomized comparison did not demonstrate a clinical advantage of lim.-KA over MA in the whole cohort. In post hoc exploratory analyses, mild varus deformity was associated with outcomes broadly similar to those after both alignment strategies. In the moderate-varus subgroup, patient-level analyses suggested a possible phenotype-dependent signal for KSS-knee, KSS-function, OKS, and FJS-12 after lim.-KA, whereas ROM favored MA, and VAS pain did not support a lim.-KA pain advantage. These subgroup findings should be interpreted separately from the primary randomized result, considered hypothesis-generating only, and not used in isolation to change clinical practice without prospective confirmation. Full article

The coexistence of cancer and severe aortic stenosis (AS) is increasing as a result of population aging and substantial improvements in cancer survival. Transcatheter aortic valve implantation (TAVI) has transformed the management of AS; however, patients with active malignancy or a history of cancer remain markedly under-represented in pivotal randomized trials. This under-representation has resulted in persistent uncertainty regarding patient selection, risk stratification, and the expected benefit of TAVI in this growing and clinically heterogeneous population. This review provides a comprehensive and contemporary synthesis of the evidence on TAVI in patients with cancer, integrating cardiovascular (CV), oncologic, and geriatric perspectives. Available data on epidemiological overlap, cancer-specific procedural challenges, and short- and long-term outcomes following TAVI are critically examined, with particular emphasis on distinctions between active cancer and cancer survivorship. Key modifiers of risk and benefit—including prior thoracic radiotherapy, competing thrombotic and bleeding risk, immunosuppression, frailty, sarcopenia, and nutritional status—are discussed in detail. Limitations of conventional surgical risk scores in oncology populations are highlighted, underscoring the need for individualized assessment beyond traditional CV metrics. Across registries and meta-analyses, TAVI is associated with high procedural success and comparable short-term outcomes in patients with and without cancer. Excess mortality observed during mid- and long-term follow-up is driven predominantly by non-CV causes related to malignancy rather than valve-related complications. Importantly, patients with cancer in remission demonstrate outcomes similar to those of non-cancer populations, whereas prognosis in active cancer is strongly influenced by disease stage, biology, and competing risks. Overall, cancer diagnosis alone should not preclude consideration of TAVI. Optimal management requires multidisciplinary, goal-oriented decision-making that integrates oncologic prognosis, functional status, and patients’ priorities. As cancer survivorship continues to expand, prospective studies, integrated risk stratification tools, and closer alignment between cardio-oncology and structural heart programs are essential to guide evidence-based and equitable care. Full article

Weed detection is an important part of precision agriculture because it allows farmers to manage weeds more efficiently and reduce unnecessary herbicide use. With the use of UAVs, it is now possible to capture high-resolution images of agricultural fields, but identifying weeds from these images is still challenging due to complex backgrounds, lighting variations, and the visual similarity between crops and weeds. In this study, an improved YOLOv7-based approach is developed to address these challenges using UAV imagery collected from rainfed cotton fields in the Texas Panhandle. The original dataset consisted of high-resolution UAV images, which were divided into smaller patches and manually annotated to label weed and cotton classes. After cleaning the dataset and applying simple augmentation techniques, a total of 8396 images were used for training and testing. To improve detection performance, two modifications were introduced: Convolutional Block Attention Module (CBAM) to help the model focus on important features and Bidirectional Feature Pyramid Network (BiFPN) to improve how information is shared across different scales. Three models—YOLOv7-CBAM, YOLOv7-BiFPN, and the combined CB-YOLOv7—were evaluated. The results show that CBAM helps detect more weed instances, BiFPN reduces false detections, and the combined model gives the best overall performance, achieving an mAP@0.5 of 0.89 and an F1-score of 0.84. Overall, the study shows that improving both the dataset and the model can lead to more reliable weed detection under real field conditions. The proposed approach can be useful for identifying weeds in cotton fields using UAV imagery and can support better crop management and more efficient use of herbicides in precision agriculture. Full article