Search Results (13,239)

Acidic mine drainage (AMD) poses a severe global environmental threat due to its high acidity and elevated levels of toxic hexavalent chromium (Cr(VI)), for which biogenic iron sulfide (FeS) nanoparticles have emerged as a promising remediation agent; however, their practical application is hindered by aggregation and oxidative deactivation. This research synthesized biogenic FeS nanoparticles via sulfate-reducing bacteria (SRB) and employed humic acid (HA) as a stabilizing agent to enhance Cr(VI) removal performance in simulated AMD conditions. Single-factor experiments combined with response surface methodology identified the optimal biosynthetic conditions for FeS: yeast extract powder dosage of 2.2 g/L, Fe/S molar ratio of 0.8, and NH₄Cl dosage of 3.1 g/L. Under these conditions, the material achieved 84.25% Cr(VI) removal, with the Fe/S molar ratio identified as the most influential parameter governing synthesis and performance. Introducing HA at an optimal dosage of 2 mg/L drove marked improvements in both nanoparticle yield and reactivity: FeS yield increased to 1096.26 mg/L, Cr(VI) removal efficiency reached 99.62%, and residual Cr(VI) dropped from 15.75 mg/L to just 0.38 mg/L. Kinetic and isotherm analyses, paired with SEM/TEM imaging and zeta potential measurements, revealed that HA stabilization improved particle dispersion and reduced lamellar stacking, resulting in a surface-controlled Cr(VI) removal process. FTIR and 2D-COS analyses demonstrated that HA-derived oxygen-containing functional groups, including O–H/N–H, C=O, and C–O moieties, played a central role in interfacial interactions during Cr(VI) sequestration. XRD results confirmed that Cr(VI) was reduced to Cr(III) and primarily immobilized as low-solubility CrOOH and Cr₂S₃, while the formation of Fe–Cr spinel-like phases remains tentative without X-ray Photoelectron Spectroscopy (XPS) validation. Further investigation via surface-sensitive spectroscopy and dynamic leaching tests is needed to fully assess the long-term stability of the reaction products. Full article

Background: Hepatocellular carcinoma (HCC) stands as a prevalent global health issue with increasing incidence and mortality rates. Hepatocellular carcinoma (HCC) exhibits profound molecular and clinical heterogeneity, which limits the effectiveness of current therapeutic strategies. Circadian rhythm disruption has been implicated in metabolic reprogramming, proliferation, and immune modulation in cancer, but its role in shaping HCC heterogeneity remains poorly defined. Methods: Four public HCC transcriptomic cohorts (TCGA-LIHC, CHCC, LIRI, LICA) were integrated using RMA normalization and ComBat for batch correction. Consensus clustering based on 31 core circadian clock genes (CCGs) identified robust molecular subtypes. Multi-omics characterization—including genomic alterations, pathway activity (GSEA/GSVA), immune microenvironment profiling (CIBERSORT, EPIC, MCP-counter, xCell), and drug-sensitivity prediction (pRRophetic/oncoPredict)—was performed to delineate subtype-specific biological properties. A nine-gene CCG-based RiskScore model was constructed using LASSO Cox regression to internally validate subtype robustness and intra-subtype risk stratification. Results: Using consensus clustering of 31 core CCGs in TCGA-LIHC and three independent validation cohorts (CHCC, LIRI, LICA), we identified three reproducible subtypes—Cluster-1 (metabolic–quiescent), Cluster-2 (transition–intermediate), and Cluster-3 (proliferation–inflammatory)—which were recapitulated across cohorts and showed distinct overall survival (Cluster-3 worst; log-rank p values significant across datasets). Multi-omic characterization revealed that Cluster-3 exhibits the highest tumor mutational burden and CNV burden with enrichment of TP53/AXIN1/TERT alterations, strong activation of cell-cycle, E2F, and G2M programs, and an immune-hot yet immunosuppressed microenvironment enriched for TAMs, Tregs and MDSCs. By contrast, Cluster-1 shows relative genomic stability, dominant hepatic metabolic signatures (fatty-acid oxidation, bile-acid and xenobiotic metabolism) and an immune-cold phenotype. Single-cell mapping linked ALAS1 expression to malignant hepatocytes predominating in Cluster-1, whereas NONO and CSNK1D localized to stromal (CAFs/TECs) and both malignant/immune compartments respectively in Cluster-3, providing a cellular mechanism for subtype-specific metabolism, angiogenesis and immune modulation. Finally, a nine-gene CCG-based RiskScore validated prognostic stratification and drug-sensitivity predictions indicated subtype-specific therapeutic vulnerabilities (notably increased predicted TKI sensitivity in Cluster-3). Conclusion: In conclusion, this study proposes a robust circadian rhythm-based molecular classification of hepatocellular carcinoma, revealing three biologically and clinically distinct subtypes characterized by divergent genomic alterations, metabolic programs, immune microenvironment states, and prognostic patterns. By integrating bulk and single-cell transcriptomic data, we identify subtype-specific roles of key circadian regulators—including ALAS1, NONO, and CSNK1D—in shaping tumor metabolism, proliferation, stromal remodeling, and immune suppression. These findings highlight circadian dysregulation as a potential upstream factor associated with HCC heterogeneity and provide a conceptual framework for developing subtype-tailored mechanistic studies and circadian-informed therapeutic strategies. Full article

This paper focuses on variable selection and parameter estimation for mixed-effects Poisson count regression models. To simultaneously select important variables in both fixed effects and random effects, we propose a double-penalized Poisson count regression model with the Smoothly Clipped Absolute Deviation (SCAD) penalty imposed on both components. To estimate the unknown parameters, we develop a new iterative algorithm called the Double SCAD–Local Quadratic Approximation (DSCAD-LQA) algorithm. Under regularity conditions, the consistency and Oracle property of the proposed estimator are established. Simulation studies are conducted under two types of penalty parameter selection criteria: the Schwarz Information Criterion (SIC) and the Generalized Approximate Cross-Validation (GACV). We evaluate the performance of the proposed method under different levels of correlation among explanatory variables and different covariance structures of random effects. Comparisons are also carried out with the non-penalized model, the single-penalized model, and the double LASSO-penalized model. The results demonstrate that the proposed double SCAD penalty method performs better than the other three methods in terms of important variable selection and coefficient estimation, and is especially effective for sparse models. Full article

Long-term home-based sleep monitoring requires wearable sensing devices that strictly balance signal precision with power constraints. This study presents the design and implementation of a low-noise, low-power wearable single-channel electroencephalography (EEG) system for automatic sleep staging. The hardware architecture integrates a TI ADS1298 analog front-end with an STM32F4 microcontroller, utilizing differential sampling and hardware-based filtering to effectively suppress power-line interference and baseline drift. System-level testing demonstrates an average power consumption of approximately 150.85 mW, enabling over 24.6 h of continuous operation on a 1000 mAh battery, which meets the requirements for overnight monitoring. To achieve accurate staging without draining the wearable’s battery, we adopted and deployed a lightweight deep learning model, SleePyCo, on the cloud backend. This architecture was specifically optimized for our edge–cloud collaborative execution, which combines contrastive representation learning with temporal dependency modeling. Validation on the ISRUC dataset yielded an overall accuracy of 79.3% ± 3.0%, with a notable F1-score of 88.3% for Deep Sleep (N3). Furthermore, practical field trials involving 10 healthy subjects verified the system’s engineering stability, achieving a valid data rate exceeding 97% and a Bluetooth packet loss rate of only 0.8%. These results confirm that the proposed hardware–software co-designed system provides a robust, energy-efficient IoMT sensing solution for daily sleep health management. Full article

The Survey of Health, Ageing and Retirement in Europe (SHARE) is a cross-national panel study including approximately 160,000 adults aged ≥50 years from 29 countries. While multiple cognitive subtests are available, the SHARE consortium does not currently recommend a standardised approach to cognitive screening. This scoping review and mapping study aimed to (1) assess how cognition is measured in SHARE publications, (2) identify whether any cognitive screening instruments (CSIs) are validated in the SHARE, and (3) explore the potential to replicate additional CSIs using cognitive measures available in recent waves that include an expanded battery of subtests. SHARE-related publications were identified by searching PubMed, and a dedicated online registry of SHARE publications. Methodical details were extracted and quantitative counts calculated. Among 234 SHARE publications, the most common choices were using single subtests (n = 94), CSIs (n = 56), and standardised scores (n = 50). From 22 unique CSIs used in the SHARE, only the SHARE Cognitive Instrument and Langa–Weir Criteria were formally validated. Cognitive impairment was assessed in 36 studies, yet no validated recognised definition of mild cognitive impairment (MCI) was found. Mapping other potential CSIs (n = 81) identified the 10-Point Cognitive Screener, Six-Item Screener and Mini-Cog as other potential CSIs for use across SHARE waves. Further research is needed to validate existing CSIs and to better operationalise MCI in the SHARE. Full article

Multichannel synthetic aperture radar (SAR) achieves high-resolution imaging while significantly enhancing the spatial freedom of the SAR system. As SAR hardware performance continues to improve, observed scenes often transition from simple to complex scenes. The increasingly complex clutter components introduced by complex scenes make clutter suppression increasingly challenging. Traditional multichannel clutter suppression algorithms usually assume that the observed scene, as a whole, satisfies the independent and identical distribution (IID) condition. However, in actual complex scenes, this assumption often proves difficult to uphold. Therefore, how to achieve more effective clutter suppression for complex scenes is a challenge for SAR. In this paper, we propose a knowledge-aided (KA) multichannel SAR clutter suppression algorithm for complex scenes. First, the single-channel image is processed at the superpixel level and a superpixel fusion algorithm is proposed, which adaptively realizes the refined classification of the complex scene. Then, a two-step clutter suppression processing method with multi-strategy clutter suppression preprocessing and sparse Bayesian residual clutter suppression is proposed. This method not only provides effective classification information for complex scenes but also achieves more efficient clutter suppression in complex scenes based on this classification information. Finally, the clutter suppression performance of this algorithm in complex scenes was validated through measured data. Full article

Deepwater drilling operations face critical challenges including narrow pore-fracture pressure windows, wellbore instability, and environmental concerns from drilling discharge. This paper presents a comprehensive systematic review of Riserless Mud Recovery (RMR) technology, tracing its evolution from its conceptual origins to its current applications, critically analyzing its technical limitations, and identifying future research directions. A systematic literature review was conducted covering peer-reviewed journals, SPE/IADC conference proceedings, industry technical reports, and independent academic studies from 1990 to 2025. Databases searched included Web of Science, Scopus, OnePetro, and Google Scholar, supplemented by Derwent Innovation Index for patents. After screening over 100 publications, approximately 60 references were selected following a two-step process excluding vendor-only promotional materials. Key findings reveal the following: (1) RMR technology has evolved through three distinct hardware generations—flexible hose systems, steel-pipe return lines with tandem pumps enabling deepwater breakthrough to 1419 m, and hybrid riser configurations for conceptual designs beyond 3000 m; (2) documented field benefits include 70% drilling fluid reduction, 9 days’ time savings per well, and successful mitigation of shallow geohazards across more than 1000 global well applications; (3) integration with casing-while-drilling and managed pressure cementing has enabled record-breaking performance of 1710 m in a single run; (4) independent academic validation confirms fatigue mechanisms affecting mud return lines; (5) systematic failure mode analysis identifies critical reliability issues in suction hoses, seals, and control systems; (6) quantitative economic analysis shows RMR cost-effectiveness depends on water depth, geological conditions, and environmental regulations. RMR technology has matured into a reliable drilling solution, yet its continued evolution requires addressing hardware limitations, developing dedicated well-control protocols, expanding to ultra-deepwater and emerging applications, and integrating digitalization for real-time optimization. Full article

Accurate time-domain photovoltaic (PV) models are needed to evaluate performance under outdoor variability beyond STC datasheet conditions. This paper presents a traceable modeling workflow based on the standard single-diode formulation, implemented in MATLAB/Simulink (R2023a) as a modular white-box architecture that explicitly resolves photocurrent generation and loss mechanisms (diode recombination, shunt leakage, and series resistance effects) with temperature-consistent propagation through V_T(T) and saturation-current terms. The method couples optical boundary conditions to the electrical model by embedding plane-of-array (POA) excitation via the incidence angle Θ(t) and roof albedo directly into the photocurrent source term, preserving the causal chain from mounting geometry to electrical response. Calibration is separated from prediction by initializing key parameters using the standard Simulink PV block and then freezing them for time-domain evaluation. The workflow is validated on a 395 W rooftop prototype using 1 min resolved POA irradiance (ISO 9060:2018 Class A radiometric chain) and module temperature (IEC 60751 Class A Pt100), synchronized with electrical measurements. Over a multi-week campaign, the model exhibits high fidelity, with a worst-case relative current error of ~1.1% and a consistently low bias and dispersion, quantified by ME, MAE, RMSE, σ_e, and thresholded MAPE. Full article

Background: Triple-negative breast cancer lacks established targeted therapies, and only a subset of patients achieves a pathologic complete response to neoadjuvant chemotherapy. We aimed to integrate bulk cohorts with an exploratory single-cell multi-omic dataset from only five patients to identify tumor and immune-related features associated with chemotherapy response. Methods: Bulk analyses were performed in two public breast cancer cohorts (GSE76275 and GSE25065) to compare triple-negative versus non-triple-negative tumors and to relate pretreatment transcriptional and inferred immune infiltration patterns to neoadjuvant chemotherapy response. Separately, in a hypothesis-generating single-cell cohort of five triple-negative breast cancers (n = 5; four responders, one non-responder), we performed single-cell RNA sequencing, T cell and B cell receptor sequencing, single-cell ATAC sequencing, and glycosylation tag profiling. Results: In bulk data, triple-negative tumors showed a loss of luminal estrogen receptor-associated programs, higher proliferation, and CIBERSORT-estimated enrichment of myeloid-associated immune fractions compared with non-triple-negative tumors. Chemotherapy response was associated with modest transcriptional shifts and inferred immune composition differences in triple-negative tumors and more pronounced epithelial, stromal, and inflamed immune changes in non-triple-negative disease. Single-cell data suggested that responder tumors were enriched for T and natural killer cells, antigen-presenting myeloid cells, expanded and diverse T and B cell clonotypes, and immune-associated glycosylation signals, whereas the non-responder sample was dominated by epithelial and fibroblast compartments with secretory, adhesion, and potential immune evasion programs. Checkpoint-related analyses reflected expression patterns and predicted ligand–receptor communication, nominating TIGIT–NECTIN2 as a candidate axis for further investigation. Conclusions: Integrating public bulk cohorts with exploratory single-cell multi-omics supports a model in which chemotherapy sensitivity in triple-negative breast cancer is linked to inflamed, antigen-presenting microenvironments and adaptable antitumor immunity, whereas resistance is associated with stromal and tumor dominance. These candidate biomarkers and pathways require validation in larger independent cohorts, and clinical translation is premature given the exploratory single-cell cohort. Full article

Objectives: Oral leukoplakia (OL) is the most prevalent oral potentially malignant disorder and requires accurate diagnosis, safe excision, and reliable margin evaluation to minimize recurrence and malignant transformation. Diode laser excision is increasingly adopted due to its precision and favorable clinical outcomes; however, laser-induced thermal effects at surgical margins raise concerns regarding tissue integrity and histopathological reliability. This study aimed to evaluate optical coherence tomography (OCT) as a real-time, high-resolution, non-invasive imaging modality for assessing peri-incisional thermal effects during diode laser excision of non-dysplastic OL. The primary objective was to validate OCT for ultrastructural and morphometric tissue analysis while ensuring preservation of diagnostic readability. Methods: A single-center observational case series was conducted at the University of Turin. Thirty patients with clinically and histopathologically confirmed oral leukoplakia without epithelial dysplasia were enrolled and allocated to two groups: 15 lesions excised using a 980 nm diode laser in continuous-wave contact mode (laser group) and 15 lesions removed by conventional scalpel biopsy (control group). Laser excisions were performed with standardized parameters and a circumferential safety margin of 5 mm. Immediately after excision, specimens underwent ex vivo spectral-domain OCT (SD-OCT) imaging to evaluate the epithelial and connective tissue microarchitecture at surgical margins and central lesion areas. OCT acquisition sites were precisely correlated with histological sections. Quantitative OCT measurements of epithelial thickness, lamina propria thickness, and laser-induced thermal alterations were compared with corresponding histological findings. Results: OCT consistently provided high-resolution visualization of oral mucosal microarchitecture in both groups, allowing clear identification of epithelial stratification, basement membrane continuity, and lamina propria organization. In the laser group, OCT detected superficial optical alterations at the surgical margins consistent with laser-induced thermal effects, while deeper tissue layers remained structurally readable. Histological analysis revealed mean epithelial and connective tissue thermal alterations of 288.9 μm and 430.3 μm, respectively. OCT-derived measurements showed high concordance with histology, with an overall agreement of 88.5% and no statistically significant differences between OCT and histological assessments. Importantly, laser-induced thermal effects did not impair definitive histopathological diagnosis in any specimen. Comparison with the control group confirmed preserved tissue architecture in scalpel-excised samples and highlighted OCT sensitivity in detecting laser-related structural remodeling. Conclusions: OCT proved to be a reliable, non-invasive imaging technique for real-time assessment of diode laser-induced thermal effects during OL excision. The technique accurately delineated tissue microstructure and surgical margins without compromising histopathological interpretation. Integration of OCT into the laser-assisted management of oral potentially malignant disorders may enhance surgical precision, optimize margin control, reduce diagnostic uncertainty, and support individualized follow-up strategies. Full article

Laser wire directed energy deposition (LWDED) has garnered significant attention for the fabrication of large metallic components. However, the complex coupling effects among its process parameters pose challenges for porosity control. Optimizing parameter combinations to effectively minimize porosity is therefore critical to the broader adoption of this technology. In this study, systematic experiments and modeling were conducted to optimize the LWDED process parameters and predict porosity. First, single-factor and orthogonal experiments were performed to evaluate the individual effects of laser power, scanning speed, wire feeding speed, and air pressure on porosity. Subsequently, range analysis and analysis of variance were employed to determine the influence of each parameter and the significance of their interactions. Four machine learning models—SVR, RF, GPR, and XGBoost—were then trained and compared. Among them, the SVR model exhibited the best predictive performance, achieving an R² of 0.8960, an RMSE of 0.19, and an MAE of 0.15, outperforming the other three models. Based on this, the SVR model was further utilized to establish the mapping between process parameters and porosity. Contour maps and three-dimensional surface plots were generated to visualize porosity variation patterns under interacting parameters. Validation experiments showed that the maximum relative error between model predictions and experimental measurements was 0.514%, with an average error of 0.251%. This study provides a reliable reference for selecting low-porosity parameter combinations in the LWDED fabrication of 5356 aluminum alloy components. Full article

Background: Esophagectomy remains a technically demanding oncologic procedure with substantial morbidity, despite ongoing advances in minimally invasive and robotic techniques. Limitations in intraoperative visualization and anatomical recognition contribute to complications such as nerve injury and bleeding. Artificial intelligence (AI)-based intraoperative video analysis has emerged as a potential adjunct to enhance surgical perception and safety, but its application in esophagectomy has not been comprehensively reviewed. Methods: A systematic review was conducted in accordance with PRISMA guidelines. PubMed, Scopus, and Web of Science were searched without a lower date limit to identify eligible studies published up to January 2026, capturing early and contemporary applications of intraoperative AI in esophagectomy. Human studies involving any surgical approach were included. Data on the AI task, methodology, validation strategy, performance metrics, and reported clinical outcomes was extracted. Risk of bias was assessed using the ROBINS-I tool. Results: Six studies met the inclusion criteria, predominantly evaluating AI-driven analysis of intraoperative video during minimally invasive or robotic esophagectomy. Reported applications included real-time anatomical structure recognition, recurrent laryngeal nerve segmentation, detection of excessive nerve traction, instrument and event recognition, and surgical phase identification. Across studies, AI systems demonstrated performance comparable to expert surgeons for selected tasks and achieved real-time or near–real-time inference. One study reported earlier detection of excessive recurrent laryngeal nerve traction compared to conventional nerve integrity monitoring. However, most studies were retrospective, single-center, and feasibility-focused, with limited external validation and minimal assessment of patient-centered clinical outcomes. Conclusions: Artificial intelligence-based intraoperative analysis in esophagectomy is increasingly achievable and may enhance anatomical recognition, intraoperative risk detection, and procedural awareness. Nevertheless, current evidence remains preliminary, heterogeneous, and largely exploratory. Prospective, multicenter studies with standardized reporting and clinically meaningful outcome evaluation are required before routine implementation. Until such data is available, AI should be regarded as a complementary intraoperative tool rather than a standalone clinical decision-making system. Full article

This paper evaluates the current maturity of automatic code-generation workflows for deploying modern CNN-based object detectors on embedded GPU platforms. We compare a native pipeline against a code generation pipeline through a Model-Based Engineering (MBE) approach, using YOLOv8/YOLOv9 inference on NVIDIA Jetson Orin Nano and Jetson AGX Orin as representative edge-GPU workloads. We report detection-quality metrics (mAP, PR curves) and system-level metrics (latency distribution and initialization overhead) under a controlled single-class scenario based on a CARLA-generated sequence with frame-level annotations. Absolute accuracy and latency values are scenario-dependent and may vary under different camera optics, illumination, motion blur, sensor noise, occlusion patterns, and multi-class scene. Results quantify the performance gap between code generation and native pipelines and show that, for the evaluated workloads, the automated pipeline remains less competitive in both latency and accuracy. We discuss the implications of this gap for deployment workflows in safety-oriented domains, and we outline bottlenecks that should be addressed. The study is intended as a controlled traffic-light detection micro-benchmark and does not aim to validate full ADAS perception stacks. Full article

Background and Objectives: Burnout is highly prevalent among emergency and acute care healthcare workers (HCWs), yet biological correlates remain debated because candidate biomarkers are strongly shaped by circadian timing, shift work, sleep loss, and overlapping affective symptoms. We mapped post-2018 evidence of biological biomarkers assessed alongside validated burnout measures in emergency department (ED), emergency medical services (EMS), and related acute care settings. Specifically, we asked whether reproducible biological correlates of burnout can be identified in emergency and acute-care healthcare workers when biomarker endpoint class and sampling context are systematically considered. Materials and Methods: We conducted a systematic scoping review with evidence mapping (PRISMA-ScR). PubMed/MEDLINE and the MDPI platform were searched for English-language studies published from 2018 onward (through January 2026). Eligible quantitative studies enrolled ED/EMS or acute care HCWs, assessed burnout using validated instruments, and reported at least one biological biomarker. Evidence was charted by biomarker domain and endpoint class (basal measures, stress reactivity paradigms, and chronic indices such as hair-based markers). Results: Overall, 19 studies were included in mapping/synthesis. Biomarker selection clustered around the hypothalamic–pituitary–adrenal axis (cortisol; n = 10/19), with fewer studies focused on autonomic function (heart rate variability; n = 2/19) and immune–inflammatory markers (n = 2/19), and single-study coverage for oxidative stress (n = 1/19), cardiometabolic candidates (n = 1/19), cellular aging (n = 1/19), neuroglial/multi-system candidates (n = 1/19), and feasibility-oriented multi-marker designs (n = 1/19). Reported associations with burnout were heterogeneous in direction and magnitude, but were more interpretable when endpoint class, timing anchors, and shift/sleep-related covariates were explicitly reported. Rates of confounder adjustment were low across studies (e.g., only 3/19 reported multivariable adjustment, and none systematically measured sleep or circadian factors), substantially limiting interpretability. Conclusions: The 2018+ literature does not support a single reproducible biomarker for burnout in emergency and acute care workforces. Evidence instead suggests multi-system dysregulation that is highly sensitive to endpoint class, sampling timing, and contextual confounding. Future studies should prioritize timing-anchored repeated-measures protocols across shift and recovery windows, jointly model sleep/circadian factors and depressive symptoms, and evaluate multi-marker panels and intervention responsiveness. Full article

Manual microscopy for malaria diagnosis is labor-intensive and prone to inter-observer variability. This study presents an automated binary classification approach for detecting malaria ring-form infections in thin blood smear single-cell images using a parallel neural network framework. Utilizing a balanced Kaggle dataset of 27,558 erythrocyte crops, images were standardized to 128 × 128 pixels and subjected to on-the-fly augmentation. The proposed architecture employs a dual-branch fusion strategy, integrating a convolutional neural network for local morphological feature extraction with a multi-head self-attention branch to capture global spatial relationships. Performance was rigorously evaluated using 10-fold stratified cross-validation and an independent 10% hold-out test set. Results demonstrated high-level discrimination, with all models achieving an ROC–AUC of approximately 0.99. The primary model (Model#1) attained a peak mean accuracy of 0.9567 during cross-validation and 0.97 accuracy (macro F1-score: 0.97) on the independent test set. In contrast, increasing architectural complexity in Model#3 led to a performance decline (0.95 accuracy) due to higher false-positive rates. These findings suggest that moderate-capacity feature fusion, combining convolutional descriptors with attention-based aggregation, provides a robust and generalizable solution for automated malaria screening without the risks associated with over-parameterization. Despite a strong performance, immediate clinical use remains limited because the model was developed on pre-segmented single-cell images, and external validation is still required before routine implementation. Full article