Search Results (4,422)

Facial emotion recognition (FER) can provide supplementary affective information for human–robot interaction, but deployment on resource-constrained assistive robots requires a balance between recognition performance and computational efficiency. This study presents an edge-based FER framework using a compact CNN operating on 48 × 48 grayscale facial images and retaining all seven FER-2013 expression categories. Square-root-smoothed inverse-frequency weighting is employed to mitigate class imbalance without excessively emphasizing rare classes. On the held-out FER-2013 test set, the proposed model achieves 63.78% Accuracy and 59.32% Macro-F1, achieving higher Accuracy and Macro-F1 than the evaluated ImageNet-pretrained MobileNetV2 and MobileNetV3-Small baselines. INT8 post-training quantization reduces model size by 74.13% relative to FP32, with decreases of only 0.91 and 0.50 percentage points in Accuracy and Macro-F1, respectively. On a Raspberry Pi 3 Model B+, INT8 achieves a mean model-only inference latency of 19.30 ms and a model-only throughput of 51.82 FPS. Using an actor-disjoint RAVDESS protocol comprising 416 videos, EMA stabilization reduces prediction switching by 54.86% on the held-out test actors. These results support the feasibility of compact edge-based FER for assistive robotic interaction while emphasizing that the framework provides supplementary affective cues rather than clinical diagnosis or autonomous decision-making. Full article

Wheat (Triticum aestivum L.) is considered one of the most important cereals globally, contributing significantly to the human population’s caloric and protein requirements. Therefore, ensuring a sufficient yield of wheat for global consumption plays a significant role in maintaining food security in different parts of the world. With increasing demand and dwindling production capacity, due to increasingly uncertain growing conditions, projections indicate that there should be an upswing of 60–70% in wheat productivity by 2050 to fulfill the requirement. However, drought represents the most significant and widespread abiotic limitation to global wheat production, currently resulting in approximately 10% yield losses worldwide. Furthermore, each additional 1 °C increase in temperature is anticipated to decrease staple calorie production by 4.4%. The factors contributing to drought in wheat, as well as its impact on the plant’s biochemical, physiological, and morphological structures, include altered rainfall patterns, elevated atmospheric CO₂ levels, increased temperatures, hot and dry winds, and restricted soil water availability. These factors initiate a series of morphological, physiological, and biochemical disruptions that hinder wheat growth and productivity. Drought impact on wheat starts at biochemical levels through reactive oxygen species (ROS) generation and degradation of chlorophylls, and tolerance to stress is influenced by a polygenic system where numerous genes contribute minor effects and interact significantly with environmental factors transitioning to osmoprotectants. At the physiological level, drought alters the water content in the plant body, leading to reduced net photosynthetic rates, stomatal conductance, transpiration rates, and water utilization efficiency. At the morphological level, drought impacts all kinds of structures such as roots, shoots, leaves and reproductive parts. To counter these effects, wheat develops a set of tolerant mechanisms called drought escape, avoidance and tolerance. An increase in trichomes and leaf waxes, alteration of root–shoot ratios, the staying green phenomenon, production of stress proteins like proline, activity of enzymes including superoxide dismutase (SOD), ascorbate peroxidase, catalase, etc., osmotic adjustment, abscisic acid (ABA) accumulation, expression of dehydration proteins called dehydrin, etc., contribute towards drought tolerance. This comprehensive review investigates the intricate interactions between drought and various wheat genotypes, emphasizing their substantial impacts on plant physiology, biochemistry, growth dynamics, and grain yield. Additionally, this review assesses a variety of genetic and biotechnological strategies aimed at enhancing the resilience of wheat genotypes to drought stress. By integrating recent research findings with practical applications, this review provides a detailed framework for improving the adaptive capacity of wheat plants to withstand the escalating threats of drought stress, thereby supporting sustainable wheat production in a changing climate. Addressing drought stress through genetic and biotechnological management practices is crucial for maintaining wheat productivity. Full article

Classifying benign and malignant lung nodules from computed tomography (CT) images remains difficult because lung nodules can be hard to classify, and unimodal models cannot capture complementary diagnostic information. Despite the success of deep learning, existing methods rely only on image information and miss semantic information that can be obtained from an expert radiologist’s knowledge. Hence, the authors propose a new multimodal lung nodule classification model in this study, named the Graph-Guided Fusion Network (R-GGFN), that combines three-dimensional (3D) CT image features and structured radiologist annotations. The proposed architecture consists of three models. A 3D ResNet-18 network for image feature extraction, an MLP network for encoding semantic information, and a Graph Attention Network (GAT) for capturing inter-nodule relationships and fusing multimodal information with the graph. We add a tabular skip connection to preserve discriminative semantic features and use focal loss to address imbalance during training. To prevent data leakage, we partitioned the publicly available LIDC-IDRI dataset at the patient level. Experimental results on a held-out patient-level test set, accessed only once after model selection was finalized, show that the proposed R-GGFN achieves an accuracy of 85.21%, an AUROC of 0.9147, a PR-AUC of 0.9213, and an F1-score of 0.8609. Among all unimodal and multimodal baselines internally evaluated, R-GGFN achieved the best value on every reported metric, including accuracy, AUROC, PR-AUC, F1-score, Precision, sensitivity, and specificity. Furthermore, the proposed approach enhances model transparency by combining explainable AI techniques (e.g., 3D Grad-CAM, SHAP, and graph visualization) to explain the model at the image, feature, and graph levels. The results show that the graph-guided multimodal fusion method can fully leverage complementary image and semantic information, improving diagnostic accuracy and interpretability. The framework proposed here is a good and understandable computer-aided diagnosis decision-support system for lung cancer and a step towards future external dataset validation. Full article

Background: Intracranial hemorrhage (ICH) is a time-critical neurological emergency in which delayed diagnosis significantly worsens patient outcomes. This challenge is amplified in resource-limited, high-workload settings where rapid neuroimaging interpretation may be constrained. Many high-performing deep learning approaches for disease detection rely on large-scale models trained on extensive data and evaluated only on internal datasets, limiting their generalization across heterogeneous clinical environments. This study aims to develop and evaluate a resource-efficient framework for automated ICH detection from CT scans, with internal and external validation across heterogeneous clinical settings. Methods: A hybrid deep learning framework was developed, combining an EfficientNetV2-S-based feature extractor with a bidirectional GRU model for scan-level prediction. The model was trained on a stratified subset of 6000 CT scans from the RSNA Intracranial Hemorrhage Detection dataset and evaluated using an internal test set and two external validation cohorts (PhysioNet and CQ500). Results: On an internal held-out test set, the model achieved scan-level AUROC and AUPRC of 0.980 and 0.977, respectively, and slice-level AUROC and AUPRC of 0.981 and 0.923. External validation on the PhysioNet and CQ500 datasets yielded scan-level AUROC/AUPRC values of 0.914/0.932 and 0.905/0.909, respectively, demonstrating consistent performance across datasets differing in institution, geography, patient population, and acquisition protocols. Conclusions: Despite its compact architecture and reduced training subset, the proposed framework achieves performance competitive with substantially larger and more computationally demanding models, completing training in under 26 h on single-GPU hardware. These results support the feasibility of reproducible, resource-efficient ICH detection systems for automated triage in emergency radiology workflows across different clinical settings. Full article

Deep-water gravity-flow depositional systems are characterized by intricate internal architectural configurations and pronounced heterogeneity. Traditional outcrop studies rely heavily on qualitative observations and often fail to capture the sub-seismic 3D spatial heterogeneity required for precise hydrocarbon exploration. This study leverages an integrated multi-geophysical framework to achieve a high-resolution, quantitative characterization of internal structures within gravity-flow fan units in field outcrops. Focusing on the Upper Ordovician Lashizhong Formation at the Beishan outcrop in Wuhai, northwestern Ordos Basin, the research integrates Terrestrial Laser Scanning (LiDAR) to construct a 3D topographic framework and uses Ground-Penetrating Radar (GPR) to probe subsurface architecture. Furthermore, Full-Waveform Inversion (FWI) is employed for the high-resolution reconstruction of relative permittivity, culminating in the generation of a 3D geological model via Sequential Gaussian Simulation (SGS). Six primary lithofacies were identified: (1) massive-to-graded medium-to-fine-grained sandstone, (2) graded fine- to silty sandstone, (3) parallel-laminated fine sandstone, (4) climbing ripple-laminated siltstone, (5) wave ripple-laminated siltstone, and (6) horizontally bedded mudstone. Geometrically, six channel–levee architectural stages and one sheet-lobe stage within a mid-fan setting. Quantitative analysis shows that channel axes maintain a high net-to-gross (N/G) ratio of 0.82 with 92% connectivity, whereas levee and lobe margins exhibit a significantly reduced N/G of 0.28 and 42% connectivity. The overall cross-validation accuracy of the 3D model reached 86.4%. This research provides an objective, quantitative technical framework for 3D spatial characterization and reservoir modeling of complex gravity-flow systems. Full article

Objective: Patients with prostate-specific antigen (PSA) levels in the 4–10 ng/mL gray zone present a diagnostic challenge, as PSA alone cannot reliably distinguish clinically significant prostate cancer (csPCa) from benign conditions or indolent disease, potentially leading to unnecessary biopsies or missed clinically significant disease. This study aimed to develop a risk prediction model for csPCa in this population using clinical indicators and magnetic resonance imaging (MRI) quantitative data, and to evaluate its potential value for risk stratification and as a supplementary tool for biopsy decision-making. Methods: We retrospectively included 210 patients with PSA levels in the 4–10 ng/mL range and confirmed pathological diagnoses, who were admitted to our hospital between January 2018 and June 2025. csPCa was defined as a Gleason score ≥ 7. Patients were stratified by pathological diagnosis (csPCa vs. non-csPCa) and randomly divided into training and internal validation sets in a 7:3 ratio. The following variables were collected for analysis: age, PSA, apparent diffusion coefficient (ADC), and relative T2 value. Univariate and multivariate logistic regression were used to identify independent predictors of csPCa. We constructed a prediction model and nomogram. The discriminating ability, calibration, and stability of the model were assessed using receiver operating characteristic (ROC) curves, calibration curves, and Bootstrap internal validation. Decision curve analysis (DCA) was performed to evaluate the model’s net benefit across clinically relevant threshold probabilities. Risk stratification was performed based on predicted probabilities. Results: Of the 210 patients, 50 had csPCa and 160 had non-csPCa lesions. Univariate regression showed associations between age, ADC value, and relative T2 value. Multivariate logistic regression identified age and relative T2 value as independent predictors in the final model. Internal validation using 1000 Bootstrap resampling with optimism correction yielded a corrected AUC of 0.698 (95% CI: 0.669–0.716), with a mean optimism of only 0.016, indicating minimal overfitting. The area under the curve (AUC) of the model was 0.720 in the training set and 0.711 in the validation set. The calibration curve for the training set demonstrated good agreement between predicted and observed probabilities. DCA showed net benefits across certain threshold probabilities. Risk stratification based on predicted probability showed csPCa detection rates of 14.7%, 28.3%, and 47.4% in low-, intermediate- and high-risk groups, respectively. Conclusions: The logistic binary classification model based on age and relative T2 value may assist in risk assessment for csPCa in PSA gray zone patients. However, its moderate discrimination suggests it should be used as a supplementary tool rather than a standalone diagnostic test. Risk stratification based on predicted probability may help identify different csPCa risk populations, but given the exploratory nature of this single-center retrospective study, multi-center, large-sample, and prospective studies are required for external validation and optimization before clinical implementation. Full article

China’s “Dual Carbon” goals impose rigid constraints on high-emission industries, yet why similarly endowed large mining groups exhibit divergent green and low-carbon performance remains poorly understood. The existing literature predominantly relies on single-factor net-effect analyses, failing to capture the configurational complexity and causal asymmetry underlying transformation outcomes. To address this gap, we extend the Technology–Organization–Environment (TOE) framework by integrating a green practice dimension, constructing a TOE-P configurational model. Using balanced panel data of A-share listed mining companies from 2015 to 2025, we combine Necessary Condition Analysis (NCA) with dynamic fuzzy-set Qualitative Comparative Analysis (fsQCA) to identify the multiple conditions and combinational paths driving high green and low-carbon development. Our findings reveal three key insights. First, no single condition constitutes a necessary condition; high performance is inherently configuration-driven, requiring synergistic coordination among technology, organization, environment, and practice. Second, three equivalent paths, which are innovation breakthrough, regulation–innovation–practice synergy, and innovation-substitution-for-investment, emerge with green invention patents as the sole core condition shared across all paths, underscoring technological innovation as the foundational capability. Third, pronounced causal asymmetry exists between high and non-high outcomes; non-high performance stems from distinct mechanisms such as resource–practice decoupling rather than simply lacking success conditions. These results support differentiated, technology-driven transformation strategies and tailored policy designs for mining enterprises with varying resource constraints. Full article

The European Union has set an ambitious goal of achieving net-zero emissions by 2050, and 90% reduction by 2040, through its Green Deal policy. A promising solution to this challenge lies in the adoption of Fifth-Generation District Heating Networks (5GDHNs) that operate at low temperatures, collecting and distributing heat from diverse sources (energy geostructures, asphalt pavement solar thermal collectors, industrial waste heat and waste heat from buildings’ cooling plants). The system’s design allows for heat storage underground, primarily during summer months, with distribution occurring via pipelines during winter. As part of the REHEAT project, a study has been conducted focusing on a simulation model developed using TRNSYS software. This model incorporates solar thermal collectors installed beneath the parking area asphalt pavements as a thermal energy source, coupled with a borehole thermal energy storage system. The setup is designed to meet the heating and cooling demands of a multiplex cinema situated in Northern Italy. The study presents the optimization of the system, reporting the monthly and annual data on energy balances and system efficiency. The findings demonstrate significant energy savings when compared to traditional heating and cooling systems (50.6% non-renewable primary energy reduction) and even greater CO₂ emission reduction (63.2%). Also, the economic analysis reveals positive results both from the point of view of operating costs and taking into account investment costs, highlighting the potential of 5GDHN as a sustainable solution for urban energy needs for a real case as the main novelty of this study. Full article

Background/Objectives: Electroencephalography (EEG), with its high temporal resolution, non-invasiveness, and cost-effectiveness, provides a suitable modality for investigating task-related signal patterns associated with Visual Sustained Attention (VSA) and Visual Internally Directed Cognition (VIDC), but cross-subject generalization remains challenging in reduced-channel settings. This study evaluated PDC-Net under an offline GPU setting. Methods: PDC-Net uses a Temporal Representation Adaptation Block (TRAB) for local temporal transformation and feature-channel recalibration and a Cross-Branch Gated Mamba Interaction (CGMI) module for input-dependent bilateral information exchange and long-range sequence modeling. Results: Under the strict LOSO cross-validation setting, PDC-Net achieved an average decoding accuracy of 76.76%, representing the highest mean accuracy among the 11 evaluated models. The model also maintained a favorable balance between cross-subject decoding performance and computational cost under the evaluated offline GPU setting. Conclusions: These findings support the feasibility of offline, subject-independent VSA/VIDC decoding from dual-channel Fp1/Fp2 EEG under the evaluated controlled conditions and provide a basis for subsequent external, cross-session, online, and hardware-specific evaluation. Full article

Objective: Although echocardiography is the gold standard for diagnosing left ventricularoutflow tract obstruction (LVOTO) in hypertrophic cardiomyopathy (HCM), relying solely on imaging is insufficient for precise risk stratification, particularly in borderline or atypical patients, and fails to capture systemic pathophysiological alterations. In recent years, novel inflammatory biomarkers derived from routine blood tests have shown sensitivity in capturing micro-inflammatory states; however, their specific roles in the obstructive phenotype of HCM remain unclear. This study aims to screen hematological and cardiac parameters associated with HCM obstruction and to construct an individualized predictive model. Therefore, this study aims to screen hematological and cardiac parameters associated with HCM obstruction and to develop and validate a nomogram for individualized prediction of current LVOTO risk in HCM patients. Methods: A total of 230 HCM patients hospitalized at The Central Hospital of Wuhan from January 2019 to December 2025 were retrospectively enrolled. Based on the left ventricular outflow tract pressure gradient (LVOTPG), they were divided into a non-obstruction group (n = 177) and an obstruction group (n = 53). Least absolute shrinkage and selection operator (LASSO) regression was used to screen feature variables, and multivariate logistic regression analysis was employed to identify independent predictors and construct a nomogram prediction model. The discrimination, calibration, and clinical utility of the model were evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Results: Multivariate logistic regression analysis revealed that a history of alcohol consumption (OR = 3.68, 95% CI: 1.49–9.09), peak LVOTPG (OR = 3.96, 95% CI: 1.42–11.05), maximum ventricular wall thickness (OR = 1.02, 95% CI: 1.00–1.04), peak mitral valve E-wave velocity (OR = 1.02, 95% CI: 1.00–1.04), and neutrophil count (OR = 2.42, 95% CI: 1.13–5.21) were independent predictors of LVOTO in HCM patients. The nomogram model constructed based on these factors achieved area under the curve (AUC) values of 0.810 and 0.850 in the training and validation sets, respectively. The calibration curve demonstrated good agreement between predicted and actual probabilities, and DCA indicated that the model provided clinical net benefit within a threshold probability range of 10% to 30%.Conclusions: A nomogram integrating alcohol consumption history, peak LVOTPG, maximum ventricular wall thickness, peak mitral E-wave velocity, and neutrophil count accurately predicts LVOTO risk in HCM patients (AUC: 0.810–0.850), providing a low-cost tool for early risk stratification. External prospective validation is warranted. Full article

Cross-condition fault identification for crane slewing bearings is difficult because low rotational speed, heavy loading, and operating-condition variation jointly weaken fault-induced impulses and alter the distribution of monitoring data. In addition, vibration and acoustic emission signals describe different aspects of bearing degradation, making fixed multi-sensor fusion insufficient when sensor sensitivity changes across fault states. This study proposes a Lightweight Domain-Adaptive Graph Convolutional Network (LDAGCN) for cross-condition diagnosis. The model uses six vibration channels and one acoustic emission channel as synchronized heterogeneous inputs. Two compact one-dimensional encoders first learn modality-specific temporal representations, after which a feature-wise gate determines the relative contribution of each sensing modality. The fused embeddings in every mini-batch are regarded as graph nodes, and a sparse sample graph is reconstructed from Top-k cosine similarities. Multi-receptive-field graph convolution then aggregates one-hop and higher-order neighborhood information, while a residual connection limits excessive modification of the original fused features. To reduce the discrepancy between operating conditions, the training objective combines source-domain classification, adversarial domain discrimination, maximum mean discrepancy, and supervision from a small labeled target-domain adaptation subset. Experiments were carried out on a dedicated crane slewing-bearing test rig containing normal, inner-race fault, outer-race fault, and B1 localized-fault states. On the target-condition test set, LDAGCN achieved an accuracy of 97.22%, a Macro-F1 score of 97.21%, a Macro-Precision of 97.37%, and a Macro-Recall of 97.22%. The proposed model also outperformed 1D-CNN, ResNet1D, CNN-LSTM, DANN, MMD-DAN, and DeepCORAL under the same data partition. The confusion matrix and t-SNE projection indicate that the learned representation reduces the source–target distribution gap while retaining fault-class separation. The ablation results further indicate that the acoustic emission information, gated fusion, graph-based association learning, and domain adaptation complement each other in terms of the final diagnostic performance, while maintaining a lightweight architecture that is suitable for practical monitoring. Full article

Individual-tree-level information on crown distribution and morphology underpins forest inventory, biomass estimation, and carbon accounting. High-resolution Unmanned Aerial Vehicle (UAV) imagery resolves single-tree detail, but automated crown extraction remains difficult: fully supervised segmentation depends on costly pixel-level annotation that generalizes poorly, while the Segment Anything Model (SAM), though training-free, cannot locate trees on its own and existing SAM-based methods restore this ability only by adding task-specific training. We present an end-to-end, annotation-free toolkit for tree crown extraction and ecological analysis. A RetinaNet-based DeepForest detector produces coarse boxes; an adaptive module then removes duplicate boxes and non-vegetation false positives using an intersection-over-union rule and a global greenness index, converting noisy boxes into clean prompts; these prompts drive SAM to decode irregular crown masks without task-specific training; and geometric and texture features are extracted and grouped by principal component analysis and K-means clustering to map ecological patterns. We evaluated the toolkit on multi-biome imagery from the public OAM-TCD dataset. Because pixel-exact metrics are unstable at 10 cm resolution, where wind sway, shadow shift, and small labeling offsets are strongly amplified, we assessed accuracy under an absolute physical-distance tolerance. At a 2.0 m tolerance, consistent with the effective radius of a mature crown, the toolkit reached a precision of 91.25%, a recall of 86.40%, and an F1-score of 88.76%; bootstrap and Monte Carlo resampling confirmed these values are stable. Without manual annotation, it characterized more than 4700 individual crowns and recovered distinct vegetation patterns across geographic settings, offering a highly adaptable, low-cost baseline tool that demonstrates robust performance across the diverse multi-biome scenes within the OAM-TCD dataset. Full article

Digital-agriculture development increasingly requires multi-criteria decision analysis (MCDA) approaches to evaluate complex and interrelated influencing factors. However, existing methods often face difficulties in representing uncertain expert judgments, capturing inter-factor influence structures, and integrating complementary weighting information. To address these challenges, this study develops an integrated Pythagorean Neutrosophic Sets (PNS)–Weighted Influence Non-linear Gauge System (WINGS)–Analytic Hierarchy Process (AHP)–Entropy Weight Method (EWM) framework for digital-agriculture factor evaluation. Specifically, PNS is employed to represent uncertain linguistic assessments through truth, indeterminacy, and falsity information; WINGS is applied to identify the direction and intensity of expert-perceived influence relationships among factors; and a coordinated weighting mechanism integrates AHP-based preference information with entropy-based cross-expert dispersion information. The proposed framework is applied to a digital-agriculture case to identify key influencing factors and determine development priorities. The results indicate that Economic Development Level (W1) ranks first in the final composite priority assessment, followed by Digital Technology Level (W6) and Digital Infrastructure (W5). Furthermore, the WINGS analysis reveals the perceived influence structure among factors and identifies Digital Infrastructure (W5) as the strongest net-influencing factor within the network. Comparative analyses suggest that the proposed framework provides a systematic representation of uncertainty, influence relationships, and complementary weighting information for digital-agriculture factor evaluation. This study highlights the application potential of integrating established MCDM techniques to support complex agricultural decision-making problems. Full article

Cardiovascular deterioration is a clinically important event in intensive care units, but early prediction remains challenging because risk may be reflected across structured clinical variables, physiological time series, and clinical text. In this work, we present MultiCardioNet, a multimodal deep learning framework for early prediction of a composite ICU deterioration endpoint using MIMIC-IV data. The prediction target was defined as vasopressor initiation and/or early death during the 24–72 h outcome window, making the task broader than mortality prediction alone but also related to treatment escalation and circulatory support. The model combines structured clinical variables, 24-h vital sign time series, and timestamp-filtered radiology reports. Structured information was represented using enriched first-24-h clinical features, while physiological dynamics were modeled using a transformer-based time series encoder and radiology reports were represented using CXR-BERT-specialized embeddings. On the held-out test set, MultiCardioNet achieved an AUROC of 0.9014, AUPRC of 0.8481, and F1-score of 0.7723. These findings suggest that the three modalities provide complementary information for this composite deterioration endpoint in the internal test setting. Full article

Wearable glucose monitoring demands low-power local processing, but conventional neural networks rely on energy-intensive multiply–accumulate (MAC) operations that limit battery life. This study shows that a Spiking Neural Network (SNN), built on a regression-adapted Leaky Integrate-and-Fire (LIF) neuron, can estimate blood glucose from multi-frequency bioimpedance and auxiliary biosignals with clinically auditable accuracy at low computational and memory cost. Using data from 98 patients (717 measurements, eGluco3 device, Azambuja Hospital, Brusque, Brazil) evaluated by 5-fold walk-forward cross-validation under ISO 15197:2013, three main findings emerge. First, a new calibration method—the Patient Fingerprint, built from each patient’s first K sensor readings—outperforms conventional one-hot patient encoding (14.2 ± 2.6 mg/dL vs. 15.4 ± 3.3 mg/dL mean absolute error) and, unlike one-hot, requires only these K readings rather than the patient’s presence in the training set; a leave-patients-out analysis confirms that the fingerprint captures individual physiology and that unseen-patient accuracy improves with calibration depth but remains clinically insufficient (MAE $94.8\to 65.9$ mg/dL from $K=3$ to $K=5$), positioning clinical-grade cross-patient generalization on a larger cohort as the primary scaling axis. Second, the direct-injection fingerprint model reaches 100% of the samples within Consensus Error Grid Zones A+B across all validation folds (the rate-coding variant reaches 98.8%, just below the 99% Criterion B threshold), without requiring any demographic or clinical metadata; sensor history alone renders such records redundant; and Criterion A, however, stays below the 95% normative threshold, so the results support clinical safety rather than formal certification. Third, replacing the analog input encoding with a multiplication-free rate-coding scheme removes all first-layer MAC operations at a cost of 2.7 mg/dL additional error; because the additional microticks raise the total operation count, this defines a design lever whose energy payoff is specific to neuromorphic hardware rather than a net saving on conventional microcontrollers. Together, these results demonstrate that SNNs offer a clinically auditable, self-calibrating, and memory-efficient path to continuous glucose estimation on embedded wearable devices. Full article