Search Results (14,260)

Optimizing fertilizer management is crucial for increasing crop yields while reducing environmental impact. However, traditional methods rely on extensive field trials, which are costly and limit their scalability. To overcome these limitations, this study developed data-driven yield prediction models (YPM) for wheat, rice, and maize by integrating multiple feature selection and machine learning algorithms with explainable ensemble learning, namely stacking regression (SR) and voting mean (VM). The optimal YPM was subsequently combined with the hybrid optimization strategy to construct an intelligent fertilization decision model (IFDM), and the economic–environmental benefits were subsequently evaluated. The best-performing models were SHAP-SR for wheat and rice and GBM-SR for maize, achieving R² values of 0.79, 0.69, and 0.67, and RMSEs of 681.69, 725.35, and 1091.49 kg ha⁻¹, respectively. Based on the IFDM, the recommended application ranges for nitrogen (N), phosphorus (P₂O₅), and potassium (K₂O) were as follows: for wheat, 122.1–256.3, 45.4–98.2, and 30.6–60.7 kg ha⁻¹; for rice, 170.8–261.2, 55.1–91.4, and 40.6–98.5 kg ha⁻¹; and for maize, 157.5–293.4, 84.2–156.4, and 30.1–62.7 kg ha⁻¹. Simulation-based evaluation suggested that adopting these recommendations could potentially increase average yields by 9.2–12.4% and enhance economic–environmental benefits by 32.86–97.73% across the three crops. This study indicates that coupling interpretable ensemble learning with a hybrid optimization strategy can support efficient decision-making for field-scale fertilization and provides a data-driven and cost-effective approach for precision fertilization, with potential applicability to arid agricultural regions under similar agro-ecological conditions. Full article

Recognition of the American Sign Language (ASL) alphabet is of utmost importance in bridging the communication gap between the hearing-impaired and the hearing. However, robust classification remains difficult because some hand gestures are morphologically very similar. To address this problem, this study presents CT-Net, a hybrid deep learning architecture that integrates ConvNeXt-Tiny with a lightweight Transformer encoder. CT-Net combines convolutional feature extraction and self-attention mechanisms, which enable it to capture fine-grained local patterns and long-range spatial dependencies effectively. The proposed model was extensively compared with various architectures including traditional CNNs, Transformer-based models, hybrid machine-learning approaches and recent lightweight hybrid networks. The experimental results show that CT-Net achieved the best overall performance with a peak accuracy of 95.67% on the enhanced ASL dataset. Ablation studies demonstrate the effectiveness of our design choice. CT-Net achieves a strong trade-off between recognition accuracy and computational efficiency with an inference rate of 163.55 Frames Per Second (FPS). These findings highlight the potential of hybrid frameworks as a powerful tool for fine-grained gesture recognition tasks. Full article

In response to the dual pressures of energy consumption and environmental pollution faced by the global shipping industry, this paper proposes an optimization method for the energy storage capacity of a ship’s hybrid energy system based on a two-layer optimization model, aiming to enhance the energy utilization efficiency and operational stability of the system. A DNN-IPSO optimization framework integrating deep neural networks (DNN) and the improved particle swarm optimization algorithm (IPSO) was constructed, and combined with robust control strategies, it optimized the energy storage capacity configuration problem under complex dynamic conditions. The results show that the proposed method exhibits superior performance in terms of energy utilization efficiency, system dynamic response, and stability. The energy utilization efficiency of the system has been increased to 91.3%, the bus voltage fluctuation has been reduced to 3.98%, the load tracking error has been decreased to 17.6 kW, and the average convergence iteration times have been reduced to 71 times. The 17.6 kW load tracking error accounts for only 1.76% of the rated propulsion power of the 1 MW-level experimental platform, which is approximately 38% lower than that of the GA-PSO method. The experimental results on the real ship show that after using the DNN-IPSO optimization, the unit voyage energy consumption has been reduced to 41.7 kWh/km, the propulsion power stability coefficient has been increased to 0.956, the system transient recovery time has been shortened to 3.2 s, and the power reserve margin has been increased to 18.4%. The proposed method can effectively enhance the energy management capability, dynamic response performance, and operational stability of the ship’s hybrid energy system in the actual operating environment, providing reliable technical support for the engineering application of the integrated energy system of ships. Full article

The basic renovation of old urban communities is an important livelihood project for urban renewal, but there are many problems in the decision-making of renovation schemes, such as strong dependence on experience, lack of quantitative basis for multi-objective trade-off, and difficulty in describing residents’ risk attitude. Combining Case-Based Reasoning (CBR) and utility theory, this paper constructs a set of intelligent decision support models driven by data and knowledge. First of all, through literature analysis and expert investigation, a decision-making index system is established, which includes four dimensions and 16 quantitative indicators: policy and financial support, residential conditions and needs, residents’ consensus and social coordination, and implementation management and long-term maintenance. Secondly, the framework representation method is used to describe the reconstruction case, a hybrid retrieval strategy combining inductive retrieval and nearest-neighbor retrieval is designed, and the subjective and objective data combination weights are calculated by using AHP and the entropy method. On this basis, a loss utility function and risk aversion coefficient based on accident and public opinion data (a = 0.02) are introduced to modify the similarity calculation results to describe the risk avoidance behavior of decision-makers. Through 40 real renovation projects, a case base is built, and two types of target cases, “typical inclusive” (F5) and “key renovation” (F35), are selected for empirical verification. The results show that the model can effectively retrieve similar cases, and the similarity ranking changes in line with risk aversion expectations after utility correction. Taking F5 as an example, by reusing and revising the reconstruction scheme of a similar case, targeted suggestions are generated, which give consideration to safety, economy and operability. This model provides a new quantifiable and reusable method for scientific decision-making in basic renovation of old residential areas. Full article

Electroencephalography (EEG) has emerged as a promising non-invasive tool for the diagnosis of neurodegenerative disorders, and artificial intelligence (AI) has shown significant potential in this domain, as demonstrated by recent studies. However, strong inter-subject variability remains a major challenge, limiting the ability of AI-based models to learn disease-specific features that generalize across individuals, thereby hindering the development of clinically deployable subject-independent systems. In this work, we propose a cross-subject, AI-based EEG classification framework to distinguish between Alzheimer’s disease (AD), Creutzfeldt–Jakob disease (CJD), and healthy control subjects using clinical EEG data collected from a local hospital. A lightweight hybrid deep learning model is developed, combining a two-layer one-dimensional convolutional neural network with a two-layer Transformer encoder to capture both local temporal patterns and long-range dependencies in EEG signals. The proposed model achieves an average classification accuracy of $97\%$, representing a $3\%$ improvement over a baseline model evaluated on a cohort of 36 subjects. To assess deployment feasibility in real-time clinical settings, the trained model is implemented and evaluated on an edge-AI platform (NVIDIA Jetson AGX Orin), demonstrating energy efficiency for the inference with a compact model footprint. These results indicate that the proposed approach provides an accurate, efficient, and practically deployable solution for subject-independent EEG-based classification of neurological disorders. Full article

This research aimed to develop a rapid, non-destructive, and accurate method for detecting adulteration in turmeric using Visible–Near-Infrared (UV/Vis and NIR) spectroscopy combined with machine learning algorithms. Spectral data from the samples were collected and analyzed in two ranges: 170–870 nm (UV/Vis) and 900–2170 nm (NIR). Four supervised learning algorithms, including Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), the Multilayer Perceptron (MLP) neural network, and Decision Tree, were evaluated for modeling. To quantitatively assess model performance, we employed not only the accuracy metric but also complementary performance indicators including precision, recall, and the F1-score to provide a more comprehensive evaluation of classification effectiveness. The models developed in the 900–2170 nm spectral range demonstrated highly significant performance, with most models achieving 100% accuracy on the independent test set. To reduce data dimensionality and enhance computational efficiency, a hybrid feature selection method combining SVM with five algorithms—League Championship Algorithm (LCA), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Imperialist Competitive Algorithm (ICA)—was employed. Upon evaluation of each method, the SVM-LCA was selected as the optimal feature selection technique. This algorithm successfully extracted the most effective wavelengths with the highest correlation and lowest error, which maintained or improved the accuracy of the classification models. This study confirms the high potential of UV/Vis and NIR spectroscopy as rapid, non-destructive, and precise tools for detecting adulteration in turmeric. The findings can pave the way for the development of intelligent quality control systems in the food and pharmaceutical industries, playing a crucial role in ensuring consumer health and safety. Full article

Background/Objectives: Heart failure (HF) remains a major cause of global mortality and morbidity; it is, therefore, of paramount importance that diagnosis and prognostication are made timely in order to better improve outcomes and reduce healthcare expenditure. This research presents a novel predictive model of heart failure that combines clinical criteria with demographic factors in order to maximize predictive performance and act as a reliable tool for individualized healthcare intervention. Methods: Complex machine learning techniques, including decision trees, random forest, and deep learning, are applied in analyzing a large dataset of subjects with heart failure. We collected a diverse dataset comprising clinical indicators such as echocardiographic data, biomarkers, electrocardiogram (ECG) features, and demographic information. Data preprocessing techniques, such as feature normalization and handling of missing values, were applied to ensure the integrity and reliability of the dataset. Results: The results indicate that integrating both clinical indicators and demographic characteristics significantly improves the predictive power of the model, compared to models based on clinical indicators alone. Specifically, the hybrid model demonstrated a superior ability to predict short- and long-term outcomes in heart failure patients, offering enhanced accuracy in risk stratification and prognosis prediction. Conclusions: This research highlights the potential of artificial intelligence (AI) and machine learning in revolutionizing heart failure care by providing healthcare professionals with more accurate, data-driven decision support tools. The proposed model not only holds promise for clinical applications but also offers insights for future research into personalized medicine. Full article

Background: The precise identification of cancer subtypes through the integration of multi-omics data has emerged as a key research direction in bioinformatics. Among existing multi-omics integration methods, similarity graph-based clustering algorithms have attracted widespread interest owing to their capacity to effectively characterize the association patterns between samples. However, the majority of existing methods primarily focus on first-order relationships among samples while ignoring the prevalent high-order neighborhood relationships, and fail to fully exploit the complementary information from different omics. Methods: To address these limitations, we propose an innovative multi-omics integration framework termed MHSGTR, which integrates multi-omics data by combining Motif high-order similarity graphs and tensor regularization to identify cancer subtypes. Specifically, MHSGTR introduces Motif theory to construct a high-order similarity graph and designs a high-order graph learning term to obtain a hybrid similarity that integrates both first-order and high-order information, thereby capturing the latent high-order structural information among samples. For multi-omics data integration, we employ third-order tensor regularization constraints to explore complementary information across multi-omics data, coupled with an attention module to adaptively learn omics-specific weights for constructing a consensus similarity graph. Final clusters are derived via spectral clustering. Results: Comprehensive experiments on eight TCGA cancer datasets and a case study on adrenocortical carcinoma (ACC) demonstrate that MHSGTR achieves superior clustering performance and identifies cancer subtypes with significant biological differences, showcasing its effectiveness in robust multi-omics integration. Full article

Multiple myeloma (MM) is a biologically heterogeneous plasma cell malignancy in which prognosis is strongly influenced by cytogenetic abnormalities. Recent updates from the International Myeloma Working Group (IMWG), along with the European Hematology Association (EHA) and European Myeloma Network (EMN), have refined the definition of high-risk (HR) disease by integrating TP53 alterations, chromosome 1 abnormalities, and specific combinations of cytogenetic lesions. However, validation of these criteria in real-world patient populations remains limited. We conducted a retrospective, single-center study including 738 patients diagnosed with MM between 2017 and 2025, of whom 408 had available fluorescence in situ hybridization (FISH) data at diagnosis. Patients were reclassified according to the latest IMWG/IMS high-risk criteria proposed in international literature. Cytogenetic abnormalities, treatment patterns, and clinical outcomes, including overall survival (OS), progression-free survival (PFS), response rates, and relapse, were analyzed. Survival was estimated using the Kaplan–Meier method. A total of 103 patients (25%) were reclassified as high-risk according to IMWG/IMS high-risk criteria. Cytogenetic HR abnormalities were identified in 17.2% of cases, with del(17p) being the most frequent (14.7%). Median OS and PFS in HR patients were 52.4 months and 16 months, respectively, compared with 68.4 months and 28 months in standard-risk patients (log-rank test p values of 0.0197 and 0.0004, respectively). Although overall response rates were high (83% in HR vs. 91% in standard-risk), relapse remained frequent in HR patients. Outcomes varied significantly according to cytogenetic complexity. Isolated del(17p) was associated with improved survival compared with cases harboring additional abnormalities, while double-hit and triple-hit profiles demonstrated inferior outcomes. The presence of chromosome 1 abnormalities, particularly in combination with IGH translocations, further worsened prognosis. Among HR patients, 44% underwent autologous stem cell transplantation (ASCT), including 10 cases of TANDEM ASCT. No survival benefit was observed for TANDEM compared with single ASCT, with median OS of 52.9 vs. 78.3 months, respectively (log-rank test p values of 0.2516). Our real-world analysis supports the prognostic relevance of the updated IMS/IMWG high-risk criteria in MM. Cytogenetic complexity, rather than individual abnormalities alone, is a key determinant of outcome. Despite high response rates achieved with modern therapies, survival remains inferior in HR patients. TANDEM ASCT did not confer additional benefit in this cohort, supporting a more individualized approach to treatment intensification. Full article

The reduction of carbon dioxide related to the energy sector is one of the greatest challenges of this century. To ensure a proper transition towards a sustainable electric power system, innovative solutions are fundamental for the efficient integration of renewable energy sources. Hybrid Renewable Energy Systems (HRES) play a crucial role in this scenario; they can ensure a stable and reliable electricity supply thanks to the combination of different renewable technologies, particularly thanks to the integration of storage systems. However, the optimal sizing process of such systems is a complex challenge due to the multiple uncertainties that can be present, involving demand fluctuations and electricity zonal price variations. The aim of this work was to develop a Mixed-Integer Linear Programming (MILP) optimization approach for the robust sizing of a HRES under multiple sources of uncertainty. The developed hybrid model consists of a wind farm, a photovoltaic (PV) plant, a Battery Energy Storage System (BESS), and an industrial load with the entire infrastructure for connection to the national power grid. Additionally, the model includes the capability to manage the over-generation of renewable resources through curtailment mechanisms. The objective of the sizing tool is to minimize the Net Present Cost (NPC) of the plant, while ensuring the reliability of the system. The developed tool can represent a useful assistant for the evaluation of different possible configurations, helping the decision-making process during the design of a HRES. The results will show the best trade-off between economic and reliability aspects, highlighting the impact that the uncertainty has on the optimal size of the plant. In particular, the best configuration analyzed is able to reduce the NPC of more than 50% compared to a plant with a single renewable source. Full article

Tomato (Solanum lycopersicum) cultivation is highly vulnerable to fungal, bacterial, and viral leaf diseases that can significantly reduce crop yield and fruit quality when not detected at early stages. Although recent deep learning approaches have achieved remarkable performance in plant disease classification, many state-of-the-art architectures remain computationally expensive and therefore difficult to deploy on resource-constrained edge devices commonly used in smart agriculture environments. To address this challenge, this paper introduces EcoTomHybridNet, an adaptive resource-aware CNN–Transformer framework designed for efficient tomato leaf disease classification under edge-computing constraints. The proposed architecture combines a lightweight convolutional backbone with a dual-branch inference mechanism composed of a fast convolutional branch for computationally efficient prediction and a Transformer-enhanced branch with local self-attention for richer contextual feature extraction. Unlike conventional lightweight hybrid models relying on static inference pipelines, EcoTomHybridNet integrates a lightweight policy-guided routing mechanism that dynamically allocates inputs between the fast convolutional branch and the Transformer-enhanced branch according to input complexity. This adaptive inference strategy dynamically reduces unnecessary Transformer computations for simpler samples while preserving strong predictive performance on more challenging inputs through policy-guided branch allocation. To further improve representation capability without significantly increasing computational complexity, the proposed student network is trained using knowledge distillation from a ViT-Tiny teacher model. Experimental results on the PlantVillage tomato dataset demonstrate that EcoTomHybridNet achieves 99.42% test accuracy and 99.0% validation accuracy under the full hybrid inference configuration. Additional validation strategies, including 5-fold cross-validation and robustness evaluation under Gaussian noise and motion blur perturbations, indicate stable performance across different data splits and moderate image degradations, suggesting improved generalization capability beyond simple dataset memorization. Furthermore, adaptive routing experiments using a lightweight threshold-based policy mechanism achieved 99.20% test accuracy while reducing computational complexity from 0.36 GFLOPs to 0.25 GFLOPs per image, corresponding to approximately 30% computational savings. These results demonstrate the effectiveness of policy-guided adaptive inference for balancing predictive performance and computational efficiency in edge-oriented plant disease classification. Overall, EcoTomHybridNet provides an efficient and adaptive framework for intelligent plant disease monitoring in IoT-enabled smart agriculture systems. Full article

The emission of hydrogen cyanide (HCN) from formaldehyde (CH₂O) during ammonia-assisted selective catalytic reduction (NH₃-SCR) remains a critical challenge for aftertreatment of bio-hybrid fuel combustion exhaust. The mechanistic details of HCN formation are still poorly understood, especially on widely deployed commercial catalysts like Cu-SSZ-13. In this work, we employed density functional theory calculations in combination with the Energetic Span Model to elucidate HCN formation pathways from CH₂O in the presence of NO₂ and H₂O over Cu-SSZ-13. The results revealed the HCN formation pathway with intermediate methylene imine as the dominant one under typical reaction conditions. These findings resonate very well with reports of hexamethylenetetramine (HMT) formation during NH₃-SCR with CH₂O, for which methylene imine is a critical intermediate. Turnover frequency (TOF) estimations highlighted the strong influence of NO₂ and H₂O: higher NO₂ concentrations promoted CO selectivity and suppressed HCN by oxidizing CH₂O to HCOOH, while lower H₂O enhanced HCN formation. These findings establish a detailed mechanistic framework for HCN emission on Cu-SSZ-13 and suggest that controlling NO₂/NO_x ratios and water content can mitigate HCN formation during NH₃-SCR. Full article

Tobacco use remains a major public health challenge in Jordan, where cigarette smoking and waterpipe use are both common and dual use is increasingly prevalent. Community pharmacies are highly accessible healthcare settings, yet structured smoking-cessation services remain underutilized. This study evaluated the clinical effectiveness and implementation of Dual-Quit Digital, a pharmacist-delivered cessation counseling program tailored to the type of tobacco used, paired with a 6-month automated WhatsApp^® (Menlo Park, CA, USA) follow-up system. We conducted a pragmatic, two-arm, parallel-group, Hybrid Type 2 cluster randomized controlled trial in 16 community pharmacies in Jordan, randomized 1:1 to intervention or usual care. A total of 320 adult tobacco users were enrolled (160 per arm). The intervention combined a structured in-pharmacy pharmacist consultation, tailored behavioral support, phenotype-stratified pharmacotherapy support, and 6 months of semi-automated WhatsApp^® follow-up with telepharmacy escalation for predefined red-flag responses. The control arm received usual care, consisting of opportunistic brief advice and standard over-the-counter sales without proactive follow-up. The primary outcome was biochemically verified continuous abstinence at 6 months, defined as exhaled carbon monoxide (CO) < 10 ppm and analyzed using intention-to-treat principles. Secondary outcomes included 7-day point prevalence abstinence (PPA) at 3 and 6 months, 30-day PPA at 6 months, both-product abstinence among baseline dual users, pharmacotherapy uptake and adherence, and implementation-relevant outcomes, including service reach, feasibility of recruitment, and digital engagement metrics. All 16 pharmacies were retained, and all 320 randomized participants were included in the intention-to-treat analysis. At 6 months, CO-verified continuous abstinence was achieved by 26.3% of participants in the intervention arm compared with 11.3% in the control arm (adjusted odds ratio [aOR] 2.84, 95% CI 1.55–5.18; p < 0.001). The intervention also improved 7-day PPA at 3 months (33.1% vs. 15.6%; aOR 2.68, 95% CI 1.56–4.60; p < 0.001), 7-day PPA at 6 months (30.6% vs. 14.4%; aOR 2.62, 95% CI 1.48–4.62; p = 0.001), and 30-day PPA at 6 months (28.1% vs. 11.9%; aOR 2.89, 95% CI 1.59–5.24; p < 0.001). Among baseline dual users, both-product abstinence was higher in the intervention arm (21.9% vs. 7.8%; aOR 3.30, 95% CI 1.12–9.75; p = 0.026). Pharmacotherapy initiation was more frequent in the intervention arm (72.5% vs. 28.1%; p < 0.001), as was self-reported adherence for at least 8 weeks among initiators (56.0% vs. 26.7%; p = 0.002). In the intervention arm, active patient response rates to scheduled WhatsApp^® messages remained substantial, with 88.1% responding at Week 1, 73.8% at Week 4, 67.5% at Month 3, and 61.3% at Month 6; 145 red-flag triggers were captured from 62 participants, and 84.1% of escalations resulted in successful pharmacist follow-up within 48 h. The Dual-Quit Digital model significantly improved smoking-cessation outcomes compared with usual care and proved operationally feasible. These findings support integrating phenotype-stratified pharmacist counselling, pharmacotherapy support, and low-burden digital follow-up as a pragmatic cessation model for Jordan and similar settings. Full article

Research Purpose: This study investigates the role of digital investor behavior, measured through Google Trends, alongside technical indicators such as RSI and Bollinger Bands, in forecasting volatility in the Dubai Financial Market. The aim is to develop a hybrid analytical framework that integrates behavioral and technical dimensions to enhance predictive accuracy in emerging markets. Study Methodology: Daily data from 2020 to 2025 were collected, covering both crisis and post-crisis periods. Digital attention was quantified using Google Trends search indices, while technical indicators included RSI and Bollinger Bands calculated over a 7-day horizon. Volatility was modeled using ARCH, GARCH, and EGARCH frameworks, with Max Drawdown employed as a complementary risk metric to capture extreme market movements. Findings: Digital investor attention shows a predictive association with volatility, particularly when combined with technical indicators. Models incorporating both behavioral and technical variables demonstrated superior predictive performance. The EGARCH model successfully captured the asymmetric impact of negative shocks (γ < 0, p < 0.05), while Max Drawdown provided additional insights into risk exposure during periods of heightened market stress, achieving an R² of 95.36%. Scientific value: This study positions digital attention as a complementary variable that improves forecasting, moving beyond conventional price-based models in volatility modeling; by integrating Google Trends with technical analysis, the research introduces a hybrid forecasting framework that can be adapted to other emerging markets. Practical Implications: The findings offer practical value for policymakers and investors. Regulators can use digital attention measures as early warning signals to anticipate volatility, while investors can integrate behavioral and technical indicators to improve risk management and trading strategies. From a foresight perspective, the study contributes to building more resilient financial systems by embedding behavioral data into predictive tools. Full article

Background. Preoperative risk stratification of pancreatic neuroendocrine tumors (PNETs) is constrained by the unavailability of histologic grade before resection. We hypothesized that a panel of biologically informed CT-radiomic signatures, combined with patient-level Δ-radiomics referenced to the contralateral pancreas, would support preoperative discrimination of progression and grade in a two-center pilot cohort. Methods. Forty-four patients with histologically confirmed PNET who underwent contrast-enhanced preoperative CT and surgical resection at two academic centers were analyzed. Lesion and contralateral non-tumor-bearing pancreatic parenchyma regions of interest were revised in 3D Slicer by a board-certified pancreatic surgeon and verified intraoperatively against surgical pathology. PyRadiomics v3.0 features were extracted with IBSI-concordant settings. Parametric ComBat batch correction was applied across the two centers (biological-covariate balance verified beforehand), and Δ-radiomic features (lesion combat–pancreas combat) were computed for the 106 intensity/texture primitives. We constructed a panel of biology-informed hybrid signatures partitioned into a preoperative lesion-only family (Family A; seven signatures) and a preoperative Δ-radiomic family (Family B; three signatures). Candidate features were filtered through correlation clustering, baseline-adjusted likelihood-ratio testing with Benjamini–Hochberg FDR control, and 100-bootstrap stability selection. Three predictor blocks were compared per target with three classifiers each (Logistic Regression, Random Forest, Gradient Boosting): M0 (five-variable clinical baseline), MA (M0 + Family A), and MB (M0 + Family B). Discrimination was reported as AUC with bootstrap 95% CI; calibration was assessed using the Brier score and TRIPOD-recommended calibration intercept and slope; and cross-center generalization was evaluated with leave-one-center-out (LOCO) cross-validation. Univariable Cox regression with bootstrap and permutation inference was used for progression-free survival (PFS). Results. The cohort had 16 progression events and eight deaths (median follow-up was 38 months, IQR 14–59). Prespecified clinical–radiomic and Δ-radiomic signatures were associated with progression-free survival, including B2 = ΔBusyness × Ki-67 (HR 0.38, 95% CI 0.19–0.76, p = 0.006). For progression prediction, the Δ-radiomic model achieved the strongest discrimination, with a nested cross-validation AUC of 0.85 and leave-one-center-out AUC of 0.87. For higher-grade disease, radiomic models also demonstrated high discrimination, with AUCs up to 0.93. Conclusions. Radiomics-derived shape and texture features, especially when combined with clinical markers, may noninvasively identify aggressive PNET phenotypes and support preoperative risk stratification. Prospective validation in larger multicenter cohorts is warranted. Full article