Brain tumor detection from magnetic resonance imaging (MRI) is fundamental to computer-aided diagnosis, yet automated models must remain robust to heterogeneous imaging conditions. Despite strong recent progress, many deep learning and transformer-based approaches primarily optimize performance accuracy without explicitly improving feature selectivity and spatial localization, and they typically produce deterministic output without uncertainty estimates, which limits reliability. To overcome these limitations, we introduce UA-EffNet-DA, an uncertainty-aware EfficientNet framework that addresses these limitations through three complementary components. First, EfficientNet-B4 serves as an efficient backbone for discriminative feature extraction. Second, lightweight dual attention modules, comprising channel and spatial attention in parallel, are applied to the model to emphasize what and where discriminative features to focus within MRI slices. Third, Monte Carlo dropout is employed during inference to quantify predictive uncertainty and enable confidence-aware decision. Experiments on two public benchmarks demonstrate strong performance, yielding accuracies of 98.73% on the Figshare dataset and 99.23% on the Kaggle dataset. In addition, explainable AI analysis using Gradient-weighted Class Activation Mapping (Grad-CAM) further indicates that the proposed model concentrates on diagnostically relevant tumor regions rather than background structures, supporting transparent decision-making. Ablation studies confirm the complementary contribution of dual attention refinement and uncertainty-aware inference. Overall, the proposed UA-EffNet-DA framework offers an efficient and interpretable approach for brain tumor detection that supports more reliable clinical decision support through uncertainty-aware predictions.
Facial emotion recognition (FER) plays an important role in healthcare, human–computer interaction, and intelligent security systems. However, despite recent advances, many state-of-the-art FER methods depend on computationally intensive CNN or transformer backbones and large-scale annotated datasets while suffering noticeable performance degradation under cross-dataset evaluation because of domain shift. These limitations hinder practical usage in resource-constrained and real-world environments. To address this issue, we propose Cross-Adaptive Gated Fusion Network (CA-GFNet), a lightweight dual-stream FER framework that explicitly combines shallow structural features with deep semantic representations. The proposed architecture integrates domain-robust gradient-based descriptors with compact deep features extracted from a VGG-based backbone. After face detection and normalization, the structural stream captures fine-grained local appearance cues, whereas the semantic stream encodes high-level facial configurations. The two feature streams are projected into a shared latent space and adaptively fused using a gated fusion mechanism that learns sample-specific weights, allowing the model to prioritize the more reliable feature source under dataset shift. Extensive experiments on KDEF along with zero-shot cross-dataset evaluation on CK+ using a strict train-on-KDEF/test-on-CK+ protocol with subject-independent splits demonstrate the effectiveness of the proposed method. CA-GFNet achieves 99.30% accuracy on KDEF and 98.98% on CK+ while requiring significantly fewer parameters than conventional deep FER models. These results confirm that adaptive gated fusion of shallow and deep features can deliver both high recognition accuracy and strong cross-dataset robustness.
Reliable acquisition of vital signs is essential in prehospital emergency care; however, conventional contact-based sensors may delay assessment, cause patient discomfort, increase cross-contamination risk, and perform poorly during motion. To address these limitations, this study developed a body-worn smart camera system for paramedics incorporating a remote photoplethysmography (rPPG)-based signal-processing algorithm to estimate heart rate (HR), oxygen saturation (SpO2), and blood pressure (BP) in real time from subtle facial blood volume fluctuations. The system was evaluated in a repeated-measures full-factorial experiment with three adults under varying lighting, camera positions, and stabilization conditions. Performance was assessed using mean absolute error (MAE) relative to contact-based reference devices. HR (MAE: 1.20–5.60 bpm) and SpO2 (MAE: 0.93–3.00%) showed relatively small errors relative to the reference measurements across the tested conditions, whereas BP estimation showed larger errors (SYS: 5.00–14.20 mmHg; DIA: 3.87–11.60 mmHg), indicating the need for further algorithm refinement. The head-mounted configuration provided the best performance (HR MAE: 1.96 bpm). These preliminary findings support the technical feasibility of a wearable, contactless rPPG-based vital sign monitoring system under the tested conditions, while identifying BP estimation as the primary target for future improvement. Further validation in larger and more diverse populations is required before clinical or prehospital applicability can be established.
Cracks indicate the deterioration of civil engineering structures, and early detection through regular inspection is crucial for structural safety. However, traditional manual inspection is time- and labor-intensive and depends heavily on the inspector’s expertise. Although deep learning-based crack detection has been actively studied, most prior studies focus on a single environment such as asphalt or concrete, and research integrating asphalt, concrete, and tunnel environments remains scarce. In this study, we propose a Dense U-Net++-based model trained on a unified dataset of drone-captured crack images across all three environments. The model combines dense connections with U-Net++’s nested skip pathways to mitigate the semantic gap in the encoder–decoder structure, enabling effective feature fusion and precise restoration of fine crack boundaries. Focal loss addresses the severe class imbalance between background and crack pixels, and area-based postprocessing suppresses spurious detections. The proposed method achieved a precision of 95.47%, a recall of 92.35%, an F1-score of 93.91%, and an IoU of 88.58%, outperforming both the baseline Dense U-Net++ and Mask R-CNN. Qualitative evaluation confirmed reliable detection across rough asphalt textures, striped concrete patterns, and low illumination in tunnels. These results indicate that the proposed framework achieves promising crack segmentation performance for automated structural inspection.