Campbellsville University

Abstract Accurate white blood cell classification is an important component of hematological image analysis, yet robust automation remains challenging because microscopic smear images often exhibit substantial morphological similarity across subtypes, staining variation, acquisition heterogeneity, and sensitivity to surrounding background content. Existing single-backbone approaches may not fully capture the complementary local texture, cellular morphology, and global contextual information required for reliable recognition across heterogeneous white blood cell images, while the effects of preprocessing choices such as cropping and image standardization are often insufficiently examined. To address these issues, this study proposes a three-branch hybrid fusion architecture for five-class white blood cell classification on a combined PBC-Normal and Raabin-WBC dataset comprising 27,606 images, including 3337 basophils, 6158 eosinophils, 4685 lymphocytes, 4,108 monocytes, and 9318 neutrophils. The framework integrates MobileNetV3, DenseNet121, and pretrained DINOv2 feature branches to capture lightweight convolutional cues, densely reused local representations, and transformer-based global semantic features, respectively. Branch-wise 512-dimensional embeddings are combined through feature stacking, layer normalization, multi-head attention, and a residual-style classification head for final five-class prediction. Performance was examined under four image-preparation settings, namely not cropped and unprocessed, not cropped and processed, cropped and unprocessed, and cropped and processed, together with dataset-specific analyses on PBC-Normal and Raabin-WBC. Across these comparative experiments, the proposed model achieved the strongest overall performance, with particularly high results on the PBC-Normal dataset and consistently favorable behavior across the combined-dataset evaluation settings, while processed inputs generally improved convergence and classification quality relative to unprocessed counterparts. These findings indicate that attention-guided CNN-transformer fusion, combined with controlled preprocessing evaluation, provides an effective and robust framework for automated five-class white blood cell classification across heterogeneous microscopic images.

Abstract The rapid expansion of Industrial Internet of Things (IIoT) infrastructures has increased the need for intrusion detection systems that remain reliable under severe class imbalance, heterogeneous protocol behavior, and rare but high-impact attacks. Existing IIoT detectors often achieve high aggregate accuracy but remain sensitive to manual hyperparameter tuning and underperform on minority attack classes. This study proposes an Attention-Guided Quantum-Inspired Extreme Learning Machine with Grey Wolf Optimization for rare-attack detection in IIoT networks. The framework represents each Edge-IIoT flow as a fixed-order 48-feature vector and uses a multi-head self-attention encoder with learned feature-index embedding to model cross-feature structural dependencies. A Quantum-Inspired Extreme Learning Machine provides closed-form classification through quantum-rotation-based hidden-layer initialization, while Grey Wolf Optimization selects seven key hyperparameters. Focal Loss with inverse-frequency class weighting is used to improve learning from underrepresented classes. The model was evaluated on the Edge-IIoT dataset containing 1,909,671 records, 15 traffic classes, and 48 features under binary and multi-class settings. In binary Normal-versus-Attack classification, the proposed model achieved 99.54% accuracy, 0.9942 F1-score, and 0.9997 AUC. In the 15-class setting, it achieved 98.94% accuracy, 0.9871 macro-precision, 0.9848 macro-recall, 0.9859 macro-F1, and 0.0079 FPR. Rare-class gains were most evident for MITM, where F1 increased by 6.57% points over the strongest baseline, and for Ransomware, where F1 increased by 1.34% points. Across five matched runs, the proposed model obtained the highest mean macro-F1 with the lowest variability. Friedman testing confirmed significant rank differences among models (χ² = 50.00, df = 10, p = 2.67 × 10⁻⁷), and Holm-corrected paired tests showed statistically reliable improvements over all baselines. The results indicate that combining attention-based feature modeling, closed-form QELM classification, metaheuristic tuning, and imbalance-aware learning improves rare-attack detection while maintaining efficient inference.

The convergence of quantum computing and medical image analysis has emerged as a promising frontier for advancing diagnostic intelligence beyond the limits of classical machine learning. Recent developments in quantum hardware and quantum machine learning (QML) algorithms have stimulated growing interest in their potential to enhance complex medical imaging tasks, including disease classification, image segmentation, and clinical decision support (Wei et al., 2023; Ullah and Garcia-Zapirain, 2024). This study presents a comprehensive review and synthesis of state-of-the-art QML methodologies applied to medical imaging, systematically examining both theoretical foundations and empirical implementations. We analyze quantum-based and hybrid quantum–classical models employed across modalities such as MRI, CT, X-ray, and histopathological imaging, with particular attention to classification accuracy, computational efficiency, and diagnostic robustness (Maheshwari et al., 2022; ElBedoui et al., 2025). Our findings reveal clear trends indicating that hybrid quantum–classical approaches currently dominate practical applications, consistently outperforming purely classical baselines in select low-dimensional and domain-specific diagnostic tasks. Performance gains are most evident in feature-rich yet data-constrained scenarios, where quantum kernels and variational circuits demonstrate enhanced representational capacity (Ajlouni et al., 2023; Landman et al., 2022). However, these advantages remain context-dependent and sensitive to noise, circuit depth, and encoding strategies. Overall, QML holds significant promise for improving medical image-based diagnosis, but its widespread clinical adoption is constrained by hardware scalability, limited qubit coherence, and unresolved data encoding challenges. Addressing these limitations will be critical for translating QML from experimental feasibility to routine medical practice (Marengo and Santamato, 2025; Rawas and AlSaeed, 2026).

Large language models can generate fluent clinical summaries, yet their reliability is constrained by hallucinations, limited provenance, and safety risks when used in conjunction with automated image-based predictions. This study presents an end-to-end screening-oriented decision support pipeline that integrates a dual-path vision model with a domain-constrained retrieval-augmented generation framework to produce evidence-grounded explanations. The vision component, Attention-Guided Dual-Path FusionNet, combines a ConvNeXtV2 backbone for global semantic representation with a Swin Transformer branch for hierarchical local feature modeling, followed by attention-guided feature fusion for four-class CT image classification. The language component, ClinicGraphRAG, performs allowlisted retrieval over curated clinical sources with metadata-preserving provenance and generates constrained outputs using a grounded-only input interface. The dataset comprises 1,535 CT images expanded to 3,600 samples through class-specific, training-only augmentation under a leakage-safe stratified 5-fold cross-validation protocol, which may introduce bias due to reliance on synthetic data and limited sample diversity. The proposed model achieves a mean cross-validation accuracy of 98.74 ± 0.19, outperforming selected baseline architectures; however, these results are derived from a single curated dataset and may not generalize to heterogeneous clinical settings. The system is designed strictly for screening-oriented decision support and education, with explicit constraints to prevent diagnostic or treatment use, and requires external validation and broader cohort evaluation before real-world deployment.

Electron paramagnetic resonance (EPR) in connection with site-directed spin labeling is a structural biology tool that can be employed to obtain structural and conformational properties of various biological systems. Recent advances in methodological and technical improvements have made EPR spectroscopy a rapidly growing tool for gleaning important structural and conformational dynamics of membrane proteins. In this review, we discuss advancements in the popular site-directed spin labeling EPR approaches in brief and their applications to study the structure and conformations of biologically important membrane proteins. Recent examples of electron spin echo envelope modulation (ESEEM), double electron-electron resonance (DEER), and In-cell EPR studies for addressing structural and conformational-related questions of membrane proteins will be highlighted.