Search Results (797)

Polygonal patterned ground (PPG) is an important remote-sensing indicator of permafrost dynamics, yet its high-resolution distribution in alpine permafrost regions remains unknown. In this study, we developed an ensemble modeling framework to map PPG in the Three-River Source Region (TRSR), northeastern Qinghai–Tibet Plateau. A total of 4200 PPG occurrence samples and multi-source environmental variables were used to train four meta models: Random Forest (RF), Support Vector Machine (SVM), Maximum Entropy (MaxEnt), and the BIOCLIM package. Model performance was evaluated using the area under the curve (AUC) and true skill statistics (TSS). An AUC-weighted ensemble model was constructed from the best-performing models. RF showed the highest accuracy, with an AUC of 0.97 and TSS of 0.85, followed by SVM and MaxEnt. The final ensemble map achieved an overall accuracy of 93.7% and a kappa coefficient of 0.91 based on validation with high-resolution satellite imagery. The mapped PPG area was 59,082 km², accounting for 25.76% of the permafrost area and 16.01% of the TRSR. PPG was mainly distributed in the Yangtze River Source Region, with limited distribution in the Yellow River Source Region and no occurrence in the Lancang River Source Region. Compared with a previous Northern Hemisphere-scale PPG distribution product, our map reduced the estimated PPG area by 21.79% and improved local spatial detail. Variable importance analysis indicated that solar radiation, elevation, freeze–thaw indices, active-layer thickness, topography, soil moisture, and ground ice jointly controlled PPG distribution. PPG mainly occurred at elevations of 4400–5000 m, on gentle slopes of 0–3°, and in areas with 30–40% ground ice content. We also found that PPG can serve as a reliable geomorphic proxy for ice-rich and thermally sensitive permafrost in alpine regions. These findings highlight the utility of PPG for permafrost dynamic assessment and associated eco-hydrological impacts in alpine permafrost regions. Full article

Land-cover (LC) and land-use (LU) mapping is essential for environmental monitoring, yet supervised multi-decadal classification remains constrained by heterogeneous agroforestry mosaics, uneven observation quality, and limited consistent reference data. This study presents a Landsat-based workflow for robust multi-temporal LC/LU classification, combining quality-focused preprocessing, radiometric harmonization, predictor-set evaluation, and Random Forest (RF) hyperparameter-stability assessment. Implemented in south-central Chile, the workflow combines masks with a locally calibrated cloud–snow index and applies per-band histogram matching for color balancing. It also tests progressive predictor designs (seasonal spectral bands, spectral indices, and territorial variables) under sample scarcity. Classification used Google Earth Engine and an RF grid search over hyperparameter ranges. Performance was evaluated through validation-kappa (κ) distributions and a hyperparameter-dispersion metric assessing accuracy and stability. Color balancing improved cross-period consistency and can partly offset the absence of spectral indices, but high performance was achieved only with the full predictor set, even under limited observations. High-performing treatments showed lower hyperparameter dispersion, supporting model selection that jointly considers accuracy and stability rather than one tuned configuration. Under limited reference data, classification depends more on predictor design and robustness-based selection than on increasing sample size alone. Full article

Lipedema and obesity are often misdiagnosed or clinically confused yet arise via distinct mechanisms, complicating diagnosis and treatment. This review synthesizes evidence differentiating these conditions across genetic, hormonal, inflammatory and mechanical pathways to identify therapeutic targets. Lipedema may involve genetic predisposition (forkhead box C2 [FOXC2], prospero homeobox 1 [PROX1]), hormonal dysregulation with aberrant aromatase activity, and altered adipogenesis (peroxisome proliferator-activated receptor gamma [PPARγ], CCAAT/enhancer-binding protein [C/EBP]). A proinflammatory microenvironment with macrophage M1/M2 imbalance, elevated interleukin-6 (IL-6) and tumor necrosis factor-alpha (TNF-α), and extracellular matrix remodeling is hypothesized to drive fibrosis. Emerging evidence implicates gut-derived endotoxemia (lipopolysaccharide [LPS]-toll-like receptor 4 [TLR4]-nuclear factor kappa-B [NF-κB]) and mechanotransduction (Yes-associated protein [YAP]/transcriptional coactivator with PDZ-binding motif [TAZ]) in adipocyte hypertrophy and treatment resistance. Obesity involves systemic metabolic dysfunction with visceral adiposity and cardiometabolic comorbidities. Lipedema patients maintain metabolic health, exhibit gluteofemoral fat distribution and experience neuropathic pain via nociceptor sensitization (transient receptor potential vanilloid 1 [TRPV1] and ankyrin 1 [TRPA1]) with central amplification. Weight-loss interventions are ineffective, necessitating targeted strategies. Promising targets include TLR4 antagonism, vascular endothelial growth factor C/vascular endothelial growth factor receptor-3 (VEGF-C/VEGFR3) modulation for lymphatic enhancement, YAP/TAZ inhibition and neuromodulators for pain. Physical therapy functions as a biological modifier targeting inflammation, lymphatic drainage and mechanotransduction. This review highlights promising but largely hypothesis-generating molecular insights and calls for validated biomarkers, rigorous clinical trials, and mechanism-based therapies. Many of the pathways discussed require further confirmation in human studies. Full article

State-level health resource systems need precise and timely prediction models, but they are plagued by the ongoing problem of deteriorating model performance because of constantly shifting data distributions (data drift). Predicting uncommon but important events like sickle cell crisis is a classification task where this problem is most noticeable. This paper presents the Evolving AI-Driven Ensemble Learning Framework, which blends novelty detection using the F1-score with sophisticated ensemble approaches (stacking XGBoost, Deep Neural Network, and Random Forest with a meta-learner). Complex, high-dimensional health data are handled using sophisticated feature engineering techniques, such as automated feature selection via evolutionary algorithms and meta-learning (MAML). We empirically assessed a reactive retraining technique that was improved by ensemble stacking and simulated real-time data drift. After five retraining cycles, the improved ensemble and feature engineering showed significant performance improvements over the initial model, achieving substantial improvements: F1-score improved from 0.1250 to 0.9734 (an absolute increase of 0.8484, representing a 678.7% relative improvement), recall from 0.0714 to 0.9767 (an absolute increase of 0.9053), and precision from 0.5000 to 0.9702 (an absolute increase of 0.4702). The framework maintained high specificity (0.9700) and demonstrated outstanding discriminative performance with an AUC-ROC of 0.9909 (an 8.5% improvement). The model’s strong predictive capacity was confirmed by improvements in the Matthews Correlation Coefficient from 0.1000 to 0.9467 (846.7% improvement) and Cohen’s Kappa from 0.0800 to 0.9467 (1083.3% improvement). Model transparency in pipeline development is now made possible by a fixed runtime issued in the SHAP explainability layer. The efficiency of the framework is empirically validated by this study, showing that automated feature engineering and optimized ensemble learning greatly increase model stability and preserve remarkable accuracy for minority classes in complicated data contexts. Full article

Functional near-infrared spectroscopy (fNIRS) has become an important signal source in motor imagery (MI) brain–computer interface research due to its non-invasive nature and high application flexibility. However, fNIRS signals exhibit significant inter-subject variability, complementary information from oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR), and complex spatiotemporal dynamics, making their efficient and robust classification challenging. To address these issues, this paper proposes a Dual-Stream Gated Fusion Network (DSGF-Net). This model employs a dual-branch architecture to perform complementary feature modeling of fNIRS signals: one branch focuses on extracting multi-scale temporal dynamic features, while the other learns the spatial distribution of hemodynamic features across channels, thereby effectively characterizing the signals from different perspectives. Upon this foundation, a gated fusion mechanism was designed to adaptively adjust the importance of different feature dimensions after the fusion of the two feature streams, thereby enhancing the discriminative power of the fused representation. On two public datasets, MI and UFFT, experimental results based on leave-one-subject-out (LOSO) cross-validation show that the proposed method achieves competitive performance across metrics such as classification accuracy, F1-score, and Kappa coefficient. Furthermore, a comparative analysis of performance under different network component configurations validates the contributions of the dual-branch structure and the gated fusion mechanism to performance improvements. Furthermore, complexity analysis results show that DSGF-Net achieves superior classification performance while maintaining a relatively small parameter size, striking a good balance between performance and computational complexity. DSGF-Net provides an effective, lightweight deep learning framework for offline fNIRS-based motor task classification, with potential applications in cross-subject BCI systems and brain signal decoding. Full article

Kneeosteoarthritis (KOA) is a common degenerative joint disease that causes pain, stiffness, and impaired mobility. Automated Kellgren–Lawrence (KL) grading from knee X-ray images facilitates efficient screening and follow-up assessment. However, models trained on a single-source dataset frequently suffer performance degradation when applied to external cohorts, due to heterogeneities in image quality, acquisition protocols, class distributions, and annotation patterns. Furthermore, conventional domain adaptation approaches typically treat all source samples uniformly, making them vulnerable to negative transfer induced by ambiguous or distributionally divergent instances. To overcome these limitations, the present study develops a supervised contrastive domain adaptation framework designed for robust KOA severity grading under domain shift. The framework incorporates two task-specific modules: (1) a source-domain sample screening module that dynamically allocates class-wise quotas based on transferability and identifies high-value source samples by evaluating target intra-class affinity, inter-class separability, and source-class compactness; and (2) a target-balanced ordinal contrastive learning module that aligns the screened source samples with target features and imposes stronger constraints on negative pairs with larger KL-grade distances. The framework was evaluated bidirectionally on KneeKL (8260 images) and MedicalExpert-I (1650 images), two public knee radiograph datasets for KOA grading. With ResNet-18, it achieved a Quadratic Weighted Kappa (QWK) of 0.8557 for KneeKL-to-MedicalExpert-I transfer, exceeding source-only training and direct source–target merging by 0.2652 and 0.0468, respectively. Comparisons with representative existing methods and multiple experimental analyses further validate the competitiveness of the proposed framework. Full article

The novel risK-Yield Discriminant OptimizatioN—KYDON can optimize portfolio selection in a volatile market, eliminating the exposure to extreme events: bubbles and busts. Implementing novel methods such as: (i) chaos dynamics, (ii) behavioral welfare, and (iii) optimal AI classifiers, KYDON supports the optimal investment policy. It diagnoses the possibility of bubbles, prepares for their burst under expectations of long growth and reduces the impact of systemic risk. KYDON performs as a dragonslayer, eliminating the loss of investors. Full article

Fine-scale identification of Spartina alterniflora (S. alterniflora) is essential for coastal wetland conservation. However, in tidal-flat environments, lodged S. alterniflora often occurs together with upright S. alterniflora and native vegetation. The two-dimensional spectral features of S. alterniflora are easily affected by senescence, canopy posture, tidal stage and mixed pixels, leading to unstable classification. The integration of UAV multispectral imagery and LiDAR data can effectively address this problem. The study focused on Shangsha Island within Jiuduansha Wetland in the Yangtze Estuary and constructed multidimensional spectral–structural features by integrating the two data sources. The separability of upright S. alterniflora, lodged S. alterniflora, Phragmites australis (P. australis) and Scirpus mariqueter (S. mariqueter) was characterized using point-cloud elevation distributions, vertical organization and canopy density. The results showed that P. australis had a multilayered point-cloud structure with broad vertical extent, S. mariqueter showed a compact and sparse structure, and S. alterniflora was characterized by a continuous and dense single-layer point-cloud structure. Lodged S. alterniflora further showed a more concentrated, single-layered point-cloud structure and stronger grass-layer continuity. Multisource classification achieved an overall accuracy of 98.15% and a Kappa coefficient of 0.97. In the lodging-area comparison experiment, point-cloud fusion increased overall accuracy from 95.18% to 97.99% and Kappa from 0.89 to 0.95, improving boundary continuity and discrimination stability. Experimental results demonstrate that the fusion of UAV multispectral imagery and LiDAR data can improve the identification of lodged S. alterniflora in complex tidal-flat environments. Accurate identification and spatial delineation of S. alterniflora can help reduce field-survey effort and associated costs while supporting more targeted and efficient removal operations. Full article

Fine-scale mapping of forest types is a prerequisite for accurately assessing forest biomass, biodiversity, ecosystem service values, and carbon budgets. Shihe County, located within an ecotone transitioning from subtropical to temperate monsoon climates in China, was selected as the case study area. By integrating Sentinel-1 SAR and Sentinel-2 multispectral imagery, along with derived vegetation indices, texture features, and backscattering coefficients, as well as statistical features extracted from the 2022 NDVI time-series, we employed a hierarchical classification framework and a random forest algorithm to generate the forest type map. The results indicated the following: (1) With a single-date multi-feature dataset, late winter (3 March) was determined to be the optimal period for forest type classification in Shihe County. (2) Shortwave infrared bands, red-edge bands, the modified vegetation index, the normalized difference red-edge index, mean texture features, and VH-polarization data were the most influential variables. (3) Incorporating yearly NDVI time-series statistical features into the single-date winter subset significantly improved classification performance, yielding an overall accuracy of 86.39% and a Kappa of 0.781, which represents a 10.83% improvement over the baseline. Persistent misclassification between bamboo forests and tea plantations remained a primary constraint on further accuracy enhancement, while the classification accuracy for evergreen broadleaf forest and deciduous coniferous forest exhibited considerable uncertainty, likely attributable to limited reference sample sizes. (4) Deciduous broadleaf forests constituted the dominant land cover type in Shihe County, whereas evergreen coniferous forests, bamboo, and tea plantations were also widely distributed, collectively reflecting the region’s transitional ecological characteristics. Full article

Large classes in Chinese College English programmes make frequent analytic assessment of student writing difficult. Large language models (LLMs) may support more frequent formative assessment, but their scores may vary across queries and be systematically harsher or more lenient than local teacher ratings. Using a corpus-based, five-fold cross-validated comparative rater-evaluation design, this study examined whether statistical calibration could make LLM-assisted scores more interpretable for College English writing assessment and where their use should remain limited. Data comprised 414 timed argumentative essays written by Chinese non-English majors at one applied undergraduate institution. Two trained College English teachers independently rated the essays on a seven-dimension analytic rubric informed by China’s Standards of English Language Ability, providing the local reference standard. Three LLMs rated each essay–dimension pair on five occasions. Under five-fold cross-validation, uncalibrated scores were compared with location–scale correction, isotonic calibration, and equipercentile linking, using quadratic weighted kappa, Spearman correlation, mean absolute error, signed bias, and half-point tolerance accuracy. Agreement between models did not imply agreement with teachers: two models showed inter-model kappa values of 0.70–0.78 but an average kappa of only 0.15 with teacher ratings while rating the essays about one band more severely. Calibration removed most of this severity difference and raised pooled kappa to 0.61–0.70 depending on the method (0.63–0.64 under equipercentile linking), compared with a teacher–teacher agreement benchmark of 0.747. The three methods differed little, and the improvement mainly reflected closer alignment of score distributions rather than better judgement of writing quality. Agreement was higher for vocabulary, syntax, and grammar but remained low for cohesion and conventions. The findings suggest that LLM-assisted scoring may support low-stakes formative feedback when calibrated to local teacher standards and used under teacher supervision, while teachers retain responsibility for judging content, coherence, argumentation, and communicative quality. Full article

Background and Objectives: Residual filling material may persist on the walls of oval root canals after mechanical retreatment, and the influence of sodium hypochlorite (NaOCl) concentration and supplementary irrigation procedures on its removal remains unclear. This laboratory study used scanning electron microscopy (SEM) to evaluate the effects of three NaOCl concentrations and five final irrigation protocols on residual filling material in retreated oval canals of mandibular premolars. The protocols tested were conventional needle irrigation (CNI), EDDY, passive ultrasonic irrigation (PUI), XP-endo Finisher R (XPFR), and Shock Wave Enhanced Emission Photoacoustic Streaming (SWEEPS). Materials and Methods: One hundred twenty extracted mandibular premolars with a single root and an oval canal were instrumented with Reciproc R25 and filled with CeraSeal and a matching gutta-percha cone using the single-cone technique. Root canal retreatment was performed using ProTaper Universal Retreatment instruments followed by Reciproc R40. The teeth were then distributed among 15 experimental subgroups according to the final irrigation protocol and NaOCl concentration used: 2.5%, 5.25%, or 8.25% (n = 8). After irrigation, each root was divided longitudinally, and one undamaged canal half was randomly chosen for SEM evaluation. Residual filling material was scored in the coronal, middle, and apical regions using an ordinal scale ranging from 1 to 5. Examiner agreement was determined with quadratically weighted Cohen’s kappa, while treatment effects were examined using a cumulative link mixed model (CLMM). Results: Residual filling material scores differed significantly according to the final irrigation protocol (p < 0.001) and root canal region (p = 0.007), but not according to NaOCl concentration (p = 0.236). The interaction between protocol and canal region was also significant (p = 0.036). In the overall comparisons, CNI produced the highest scores, whereas SWEEPS produced the lowest, and EDDY, PUI, and XPFR did not differ significantly from one another. Apical scores were higher than coronal and middle scores in the EDDY, PUI, and SWEEPS protocols, while XPFR showed no significant variation among canal regions. Conclusions: Residual filling material scores after retreatment were influenced by the final irrigation protocol and root canal region, whereas NaOCl concentration had no significant main effect. SWEEPS resulted in the lowest overall residual filling material scores and CNI in the highest, while EDDY, PUI, and XPFR showed comparable overall performance, although protocol performance varied across root canal regions. Full article

Chronic coastal pollution can drive progressive mangrove degradation, yet its spatiotemporal trajectories and post-disturbance recovery remain poorly quantified from satellite observations. In this study, Landsat time-series imagery and the LandTrendr algorithm implemented on Google Earth Engine (GEE) were used to characterize mangrove responses to a kaolin pollution event in Tieshan Port, Guangxi, China. NDVI, NDMI, and NBR trajectories were first compared to identify the most sensitive indicator of contamination-induced stress, and LandTrendr was then applied to extract the timing, magnitude, duration, and spatial extent of mangrove disturbance and recovery. Results showed that kaolin contamination imposed persistent chronic stress on mangroves from 2017 to 2021, with degradation first occurring near Langen Village and then expanding northward across the port. Moderate and severe degradation were mainly distributed along tidal creeks and patch edges, indicating strong spatial control by local hydrodynamics and geomorphology. Among the tested indices, NDVI provided the earliest and clearest response to contamination, whereas NDMI and NBR showed delayed or less consistent responses. The disturbance mapping achieved an overall accuracy of 86.5% with a Kappa coefficient of 0.73. Recovery remained limited after pollution discharge ceased, suggesting persistent environmental constraints on mangrove regeneration. These findings demonstrate that Landsat–LandTrendr trajectories provide an effective framework for monitoring chronic pollution-driven mangrove degradation and recovery in coastal wetlands. Full article

Hyperspectral imagery (HSI) provides rich spectral information for land-cover classification, while Light Detection and Ranging (LiDAR) data provide complementary elevation and structural cues. Existing HSI–LiDAR fusion methods can achieve strong performance, but many rely on complex cross-modal interaction modules with substantial computational cost. This paper proposes the HSI-Primary Residual Adaptation with Consistency Regularization Network (SRAC-Net) for lightweight HSI–LiDAR classification. The method treats HSI as the primary spectral–spatial modality and introduces LiDAR features as an adapted residual correction. A learnable channel-wise residual scaling vector controls the contribution of the LiDAR residual in each feature channel. In addition, the HSI-primary branch is explicitly supervised and provides a stop-gradient reference distribution for consistency regularization of the fused prediction. Experiments on Houston2013, MUUFL, and Trento show that SRAC-Net achieves the highest mean OA, AA, and Kappa values among the evaluated internal baselines and selected representative fusion methods under the adopted protocol. The ablation results show that the complete configuration obtains the best mean performance among the evaluated variants. LiDAR perturbation experiments on Houston2013 further show smaller mean OA reductions than direct residual fusion under the tested Gaussian-noise, random-dropout, and block-occlusion settings. The method also maintains a compact parameter scale and low measured inference latency relative to several heavier multimodal architectures. These results suggest that HSI-primary residual adaptation with consistency regularization is an effective lightweight fusion alternative for the evaluated HSI–LiDAR classification settings. Full article

Conventional landslide susceptibility assessment (LSA) generally relies on static land-use/land-cover (LULC) data, limiting its ability to capture the influence of future land-use evolution on landslide susceptibility. To address this limitation, this study proposes a dynamic LSA framework by integrating the Patch-generating Land Use Simulation (PLUS) model with a slope unit-based Generalized Additive Model (GAM). Using Wushan County in the Three Gorges Reservoir Area (TGRA), China, as a case study, historical LULC data were first analyzed, and future LULC scenarios for 2026 and 2030 were simulated using the PLUS model. Landslide susceptibility under different LULC scenarios was then evaluated using the interpretable GAM, while the statistical association between LULC categories and the spatial distribution of LSI was quantified using the GeoDetector model. The results show that the PLUS model accurately reproduced LULC evolution with a Kappa coefficient of 0.963. The GAM exhibited robust predictive performance, with 100 repeated spatial cross-validations (SCVs) yielding a mean AUC value exceeding 0.75. Although dynamic LULC had only a limited influence on the overall spatial pattern of landslide susceptibility, the proportion of very high-susceptibility areas gradually increased from 4.54% in 2022 to 4.62% in 2030. The interpretable model further revealed distinct nonlinear responses of environmental variables and demonstrated that different LULC categories contributed differently to the spatial distribution of landslide susceptibility. The proposed framework provides a practical approach for incorporating future land-use dynamics into landslide susceptibility assessment and offers valuable support for long-term landslide risk management and sustainable land-use planning. Full article

Background: Infertility is a common reproductive health disorder that affects roughly 10–15% of couples during their reproductive period. Hysterosalpingography (HSG) is a widely utilized imaging modality for assessing uterine cavity morphology and fallopian tube patency and continues to play a central role in infertility investigations. Nevertheless, the interpretation of HSG findings may vary according to the experience and expertise of the evaluator, potentially leading to inconsistencies in clinical decision-making. Although artificial intelligence (AI) has demonstrated considerable potential in medical image analysis across various specialties, evidence regarding its application in the interpretation of HSG examinations remains scarce. Therefore, this study aimed to evaluate the level of agreement among radiologists, gynecologists, and an AI-based system in the assessment of identical HSG images. Methods: In this retrospective study, a total of 1443 HSG images obtained from 414 women who underwent hysterosalpingography as part of an infertility evaluation between January 2021 and January 2025 were reviewed. Cases with incomplete clinical records or suboptimal image quality were excluded from the analysis. All examinations were independently assessed by an experienced radiologist, a gynecologist specializing in infertility management, and a multimodal artificial intelligence system based on ChatGPT-5, with each evaluator blinded to the assessments of the others and to the patients’ clinical information. Image interpretation included the evaluation of contrast distribution, peritoneal spill, uterine cavity findings, tubal patency, and overall HSG impression, which were categorized according to predefined diagnostic criteria. The primary outcome was the degree of interobserver agreement among the evaluators. Agreement analyses were performed using Cohen’s kappa (κ) and Gwet’s AC1 coefficients. Analyses were conducted using IBM SPSS Statistics (version 30.0; IBM Corp., Armonk, NY, USA) and R statistical software (version 4.4.0; R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was set at p < 0.05 (two-sided). Results: A total of 1443 HSG images obtained from 414 women were included in the final analysis. The mean age of the study population was 30.97 ± 5.59 years, and primary infertility accounted for 87.9% of cases. The average number of images acquired per examination was 3.49 ± 1.05. According to Cohen’s kappa analysis, the highest levels of agreement were observed for the assessment of image artifacts and contrast medium distribution. Agreement between the AI system and the radiologist was particularly strong for contrast medium distribution (κ = 0.757). For the overall interpretation of HSG findings, AI demonstrated substantial agreement with the radiologist (κ = 0.637), exceeding the level of agreement observed between the radiologist and the gynecologist (κ = 0.363). In contrast, concordance involving AI was lower for the evaluation of uterine abnormalities, intrauterine filling defects, and tubal patency. When agreement was reassessed using Gwet’s AC1 statistic, concordance coefficients were consistently higher than the corresponding kappa values across all evaluator pairs. Near-perfect agreement between AI and the radiologist was identified for contrast medium distribution (AC1 = 0.954), peritoneal spill (AC1 = 0.893), and patterns of peritoneal contrast passage (AC1 = 0.841). Procedures performed under local anesthesia yielded a significantly greater number of images than those conducted under general anesthesia (3.86 ± 0.86 vs. 3.08 ± 1.10, p < 0.001). No significant associations were detected between abnormal HSG findings and either infertility type or anesthetic technique. In multivariable analysis, the use of general anesthesia was independently associated with a lower image count, whereas the presence of tubal pathology emerged as an independent predictor of acquiring a greater number of images during the examination. Conclusions: Our findings indicate that AI-assisted interpretation of HSG images has the potential to complement expert assessment, showing substantial concordance in several key diagnostic domains. While the technology appears promising as a decision-support tool in infertility evaluation, further research and refinement are warranted, particularly regarding the assessment of tubal and uterine pathologies. Full article