Search Results (11,841)

Acute Lymphoblastic Leukemia (ALL) is a highly prevalent hematological malignancy, especially in children, for whom precise and prompt subtype identification is essential to establish suitable treatment protocols. Current deep learning-based computer-aided diagnosis (CAD) methods for identifying ALL are hindered by numerous drawbacks, such as a dependence on solely spatial feature depictions, elevated feature dimensions, computationally extensive deep learning architectures, inadequate multi-layer feature utilization, and poor interpretability. This paper introduces LUMINA-Net, a custom, lightweight, and interpretable deep learning CAD for the automated identification and subtype diagnosis of ALL using microscopic blood smear pictures. LUMINA-Net makes four principal contributions: first, it integrates a self-attention module within a lightweight custom Convolution Neural Network (CNN) to effectively capture long-range spatial relationships across clinically pertinent cytological patterns while preserving a compact design. Second, it employs a Discrete Wavelet Transform (DWT)-based wavelet pooling layer that decreases feature dimensions by up to 96.875% while enhancing the obtained depictions with spatial-spectral information. Third, it utilizes a multi-layer feature fusion strategy that combines wavelet-pooled features from two deep layers with a third fully connected layer to create a discriminating multi-scale feature vector. Fourth, it incorporates Gradient-weighted Class Activation Mapping as a dedicated explainability process to furnish clinicians with apparent visual explanations for each classification decision. Withoit the need for image enhancement or segmentation preprocessing, LUMINA-Net outperforms the competing state-of-the-art methods on the same dataset, achieving a peak accuracy of 99.51%, specificity of 99.84%, and sensitivity of 99.51% on the publicly available Kaggle ALL dataset. This demonstrates that LUMINA-Net has the potential to be a dependable, effective, and clinically interpretable CAD tool for ALL diagnosis. Full article

Transmission tower climbing is a critical specialized operation in ultra-high-voltage power maintenance and communication infrastructure servicing. However, existing lower-limb exoskeletons used for tower climbing still suffer from insufficient motion intention recognition accuracy under complex operational environments. To address this issue, this study proposes an inertial measurement unit (IMU)-based bidirectional temporal deep learning method for motion intention recognition. First, a one-dimensional convolutional neural network (1D-CNN) is employed to extract local temporal features from multi-channel IMU signals. Subsequently, a bidirectional long short-term memory network (Bi-LSTM) is introduced to model the forward and backward temporal dependencies of motion sequences. Furthermore, a temporal attention mechanism is incorporated to emphasize discriminative features at critical movement phases, enabling the precise recognition of short-duration and transitional motions. Experimental results demonstrate that the proposed method outperforms traditional machine learning approaches and unidirectional temporal models in terms of accuracy, F1-score, and other evaluation metrics. In particular, this method demonstrates significant advantages in identifying the flexion/extension phases and transitional states. This study provides an offline method for analyzing movement intentions in lower-limb exoskeleton control for power transmission tower climbing scenarios and offers a reference for developing assistive control strategies for assisted climbing tasks in this specific context. Full article

In underground coal mining, the gear of a shearer’s rocker arm endures extreme stress and environmental fluctuations. Failures in this vital component can pose serious safety hazards, cause prolonged operational downtime, and result in significant financial losses. Therefore, accurate gear fault diagnosis is crucial. However, conventional diagnostic methods often struggle with limited feature extraction and poor performance when dealing with non-stationary, noisy signals typical of this environment. To address these challenges, a hybrid model consisting of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) network, and Markov Transition Model (MTM) is proposed. In this framework, the CNN is used to extract both global and local features related to gear fault. A time-distributed feature extractor is then integrated with the LSTM to capture the temporal progression of these features, aiding in effective modeling of fault evolution over time. Finally, the MTM further refines classification by incorporating probabilistic state transition between fault conditions, thereby improving diagnostic stability and robustness under noise. Experimental validation was done using vibration data from the Taizhong Coal Machinery rocker arm test platform and gear data from Southeast University and achieved up to 99.79% accuracy. These results show this proposed method outperformed other advanced diagnostic methods, offering dependable fault diagnosis and strong noise resistance even under extreme noise conditions of −5 dB SNR. Full article

Cloud cover significantly reduces the electrical power output of photovoltaic systems, making accurate short-term cloud movement predictions essential for reliable solar energy production planning. This article presents a deep learning framework that directly estimates cloud movement from ground-based all-sky camera images, rather than predicting future production from past power data. The system is based on a three-step process: First, a lightweight Convolutional Neural Network segments cloud regions and produces probabilistic masks that represent the spatial distribution of clouds in a compact and computationally efficient manner. This allows subsequent models to focus on the geometry of clouds rather than irrelevant visual features such as illumination changes. Second, a Vector Quantized Variational Autoencoder compresses these masks into discrete latent token sequences, reducing dimensionality while preserving fundamental cloud structure patterns. Third, a GPT-style autoregressive transformer learns temporal dependencies in this token space and predicts future sequences based on past observations, enabling iterative multi-step predictions, where each prediction serves as the input for subsequent time steps. Our evaluations show an average intersection-over-union ratio of 0.92 and a pixel accuracy of 0.96 for single-step (5 s ahead) predictions, while performance smoothly decreases to an intersection-over-union ratio of 0.65 and an accuracy of 0.80 in 10 min autoregressive propagation. The framework also provides prediction uncertainty estimates through token-level entropy measurement, which shows positive correlation with prediction error and serves as a confidence indicator for downstream decision-making in solar energy forecasting applications. Full article

Osteoarthritis (OA) is a complex degenerative joint disease for which early diagnosis and clear molecular characterization remain limited. DNA methylation has been increasingly recognized as an important regulatory factor in OA pathogenesis. In this study, we proposed an integrative computational framework combining statistical analysis, machine learning, deep learning, and functional genomics to identify and validate OA-associated genes and methylation biomarkers for diagnostic and biological interpretation. Candidate CpG sites were obtained using two complementary strategies: differential methylation analysis and selection of loci located near transcription start sites of previously reported OA-related genes. Key features were further refined using support vector machine recursive feature elimination and random forest algorithms. Based on the selected loci, we developed a feature-fusion diagnostic model that combines Transformer and convolutional neural networks with adaptive weighting to capture both global dependency structures and local methylation patterns. A panel of 220 methylation sites demonstrated stable and reproducible diagnostic performance in an independent cohort. Functional annotation and pathway analysis highlighted several established OA-associated genes, including TGFBR2, SMAD3, PPARG, and MAPK3, and suggested INHBB as a potential novel effector gene, with additional support for AMH and INHBE involvement. Overall, this study presents a robust methylation-based framework for identifying key OA-associated genes and provides new insights into the epigenetic mechanisms underlying OA. Full article

Prolonged sitting with poor posture is associated with musculoskeletal disorders, reduced productivity, and long-term health risks. Many existing posture monitoring systems predominantly rely on single-modality sensing, such as pressure or vision-based approaches, limiting their ability to capture both static alignment and dynamic micro-movements. This study proposes a multimodal smart-skin system integrating pressure, temperature, and vibration sensors for sitting posture recognition. A total of 42 sensors distributed across 14 anatomical locations were deployed, generating 15,037 samples collected over three independent sessions to evaluate cross-session temporal generalization across nine posture classes under controlled experimental conditions. Two deep learning architectures—Temporal Convolutional Networks with Attention (TCN + Attn) and Convolutional Neural Network–Long Short-Term Memory (CNN − LSTM)—were compared under Leave-One-Session-Out (LOSO) cross-validation. TCN + Attn achieved 85.23% LOSO accuracy, outperforming CNN − LSTM by 2.56 percentage points while reducing training time by 36.7% and inference latency by 33.9%. Ablation analysis revealed that temperature sensing was the most discriminative unimodal modality (71.5% accuracy), and full multimodal fusion improved LOSO accuracy by 22.93% compared to pressure-only configurations. These results demonstrate the feasibility of multimodal smart-skin sensing combined with temporal convolutional modeling for cross-session posture recognition and indicate potential for efficient real-time, privacy-preserving ergonomic monitoring. This study should be interpreted as a controlled, single-subject proof-of-concept, and further validation in multi-subject and real-world environments is required to establish broader generalizability. Full article

Background/Objectives: Peptide receptor radionuclide therapy (PRRT) is an established treatment for metastatic neuroendocrine tumors (NETs), yet long-term disease control occurs only in a subset of patients. Predicting progression-free survival (PFS) could support individualized treatment planning. This study evaluates laboratory, imaging, and multimodal deep learning models for PFS prediction in PRRT-treated patients. Methods: In this retrospective, single-center study 116 patients with metastatic NETs undergoing [177Lu]Lu-DOTATOC were included. Clinical characteristics, laboratory values, and pretherapeutic somatostatin receptor positron emission tomography/computed tomographies (SR-PET/CTs) were collected. Seven models were trained to classify low- vs. high-PFS groups, including unimodal (laboratory, SR-PET, or CT) and multimodal fusion approaches. Performance was assessed via repeated 3-fold cross-validation with area under the receiver operating characteristic curve (AUROC) and area under the precision–recall curve (AUPRC). Explainability was evaluated by feature importance analysis and gradient based saliency maps. Results: Forty-two patients (36%) displayed short PFS (≤1 year) and 74 patients displayed long PFS (>1 year). Groups were similar in most characteristics, except for higher baseline chromogranin A (p = 0.003), elevated $\gamma$-GT (p = 0.002), and fewer PRRT cycles (p < 0.001) in short-PFS patients. The Random Forest model trained only on laboratory biomarkers reached an AUROC of 0.59 ± 0.02. Unimodal three-dimensional convolutional neural networks using SR-PET or CT performed worse (AUROC 0.42 ± 0.03 and 0.54 ± 0.01, respectively). A multimodal fusion model integrating laboratory values, SR-PET, and CT—augmented with a pretrained CT branch—achieved the best results (AUROC 0.72 ± 0.01, AUPRC 0.80 ± 0.01). Explainability analyses provided insights into model predictions, with explainability patterns in the fusion model appearing physiologically plausible and predominantly tumor-focused. Conclusions: Multimodal deep learning combining SR-PET, CT, and laboratory biomarkers outperformed unimodal approaches for PFS prediction after PRRT. Upon external validation, such models may support risk-adapted follow-up strategies. Full article

Photocatalytic anti-corrosion materials are an emerging intelligent protective material that has been widely used in marine and offshore engineering in recent years. However, its failure mechanism under multi-factor coupling is complex, and it is difficult for traditional methods to achieve accurate life prediction and mechanism analysis. This article takes submarine pipelines as the research object and designs an innovative multi-source environmental data-driven method combined with deep learning (DL), aiming to establish an intelligent prediction model for the failure of the material. This article first systematically collects the multi-source heterogeneous data of materials during service; on this basis, this article constructs a hybrid DL model. Firstly, a multi-scale multimodal image feature fusion network (MMFCT) based on the combination of convolutional neural network (CNN) and Transformer is adopted to automatically extract corrosion features from microscopic images and capture the dynamic correlation between environmental temporal data and performance degradation; then, the Sparrow Search Algorithm (SSA) was constructed to optimize the BP neural network (BPNN) model for predicting the ultimate bearing capacity of submarine corroded pipelines. Simulation experiments show that the proposed method achieves accurate prediction of material remaining life and key performance degradation paths. The corrosion recognition precision reaches 94.7%, and the bearing capacity prediction error remains below 3.1%. Full article

To address the urgent need for enhancing landslide risk monitoring in reservoir embankment slopes, a core component of water conservancy projects, this paper proposes the LRES-YOLO algorithm for real-time landslide detection on reservoir embankments. In LRES-YOLO, we first integrate coordinate attention into basic feature extraction convolutional blocks to form the CACBS attention module, which enhances the model’s ability to identify and locate landslide targets in complex reservoir terrain, overcoming positional information insensitivity in deep networks. Second, we add novel downsampling DP modules and ELAN-W modules to the backbone network, improving feature recognition efficiency for embankment slopes with diverse hydrological and topographical interference. Third, we optimize the feature fusion network with targeted concatenation and pooling operations, balancing semantic information enhancement with computational load reduction to mitigate overfitting in variable reservoir environments. Finally, we adopt Focal Loss and EIoU Loss to accelerate training convergence and strengthen target feature representation for small or obscured landslides on embankments. Experimental results show that LRES-YOLO outperforms traditional algorithms in detecting landslides across diverse reservoir embankment scenarios: it achieves an average improvement of 8.4 percentage points in mean mAP over the best-performing baseline across five independent trials, a detection speed of 8.2 ms per image, and memory usage of 139 MB. This lightweight design makes it suitable for edge computing devices, providing robust technical support for intelligent monitoring systems in water conservancy projects. Full article

Anthropogenic marine debris (AMD) in shallow coastal benthic areas poses serious threats to ecosystems, human health, and the economy. Addressing this issue is hindered by limited data on AMD distribution and classification. We explored the use of semantic segmentation, specifically Pyramid Scene Parsing Network (PSPNet) and Deep Convolutional Neural Network for Semantic Image Segmentation, Version 3, (DeepLabV3) models, for automated AMD detection and classification. The performance was evaluated using mean intersection over union (mIoU), pixel accuracy, and frames per second (FPS). PSPNet achieved a higher mIoU (77.03%) than DeepLabV3 (75.98%), indicating better object identification. However, DeepLabV3 outperformed PSPNet in pixel accuracy (92.24% vs. 92.01%) and FPS (8.83 vs. 6.92), making it more appropriate for real-time applications. To enable real-time identification and classification of AMD, the models are deployed in a minicomputer with adequate processing power, significantly enhancing the models’ frame rate during real-time image processing. While both models are effective, DeepLabV3 is recommended for real-time AMD segmentation. The study contributes to improving AMD monitoring and management in coastal environments through AI-driven solutions. Full article

Accurate stress measurement is essential to evaluating structural integrity and plays a pivotal role in the health monitoring and predicting the service life of steel infrastructures. This study proposes a deep learning approach for stress prediction based on longitudinal critically refracted (LCR) ultrasonic waves. The model integrates gated recurrent units (GRU), attention mechanisms, and one-dimensional convolutional neural networks (1D-CNN), enabling direct stress prediction from raw ultrasonic signals without the need for manual feature extraction or explicit physical modeling. To validate the approach, LCR signals were acquired using a custom-built piezoelectric ultrasonic system from 20# steel specimens subjected to uniaxial stresses ranging from 0 to 200 MPa. A dataset comprising 4200 samples was augmented to enhance training efficiency. The proposed model achieved a mean absolute error of 1.94 MPa. Generalization tests demonstrated high accuracy across diverse stress levels, with average errors below 3 MPa, highlighting the model’s robustness. This research presents an accurate, intelligent, and calibration-free ultrasonic method for stress evaluation, providing practical support for stress evaluation in steel structures under actual operating conditions. Full article

The reliability of high-frequency (HF) frequency selection technology relies on the prediction accuracy of the Maximum Usable Frequency of the ionospheric F2 layer (MUF-F2). To improve its short-term prediction performance, a novel hybrid deep learning prediction model is proposed, which achieves accurate modeling of the complex spatiotemporal variation patterns of MUF-F2 by integrating a feature enhancement mechanism, a dual-branch feature extraction structure, and a bidirectional temporal dependency capture network. The hybrid prediction model integrates the Channel Attention mechanism (CA), Dual-Branch Convolutional Neural Network (DCNN), and Bidirectional Long Short-Term Memory network (BiLSTM). The model is trained and validated using MUF-F2 data from 5 communication links over China during geomagnetically quiet periods and 4 during geomagnetic storm periods, with the difference in the number of links attributed to experimental constraints and the disruptive effects of geomagnetic storms. Its performance is evaluated via multiple metrics, and a comparative analysis is conducted with commonly used prediction models such as the Long Short-Term Memory (LSTM) network. Experimental results show that during geomagnetically quiet periods, the proposed model achieves lower prediction errors (Root Mean Square Error (RMSE) < 1.1 MHz, Mean Absolute Percentage Error (MAPE) < 3.8%) and a higher goodness of fit (coefficient of determination (R²) > 0.94), with the average error reduction across all links ranging 8 from 6.2% to 46.9% compared with the baseline model. Under geomagnetic storm disturbance conditions, the model still maintains robust prediction performance, with R² > 0.89 for all communication links, as well as RMSE < 0.6 MHz, Mean Absolute Error (MAE) < 0.4 MHz, and MAPE < 3.3%. The study demonstrates that the proposed CA-DCNN-BiLSTM model exhibits excellent prediction accuracy and anti-interference capability under different geomagnetic activity conditions, which can effectively improve the short-term prediction accuracy of MUF-F2 and provide more reliable technical support for HF communication frequency decision-making. Full article

Deep learning has emerged as an effective tool for automatic waste classification, supporting cleaner cities and more sustainable recycling systems. Because environmental protection is central to the United Nations Sustainable Development Goals (SDGs), improving the sorting and processing of everyday waste is a practical step toward this broader objective. In many real-world settings, however, waste is still sorted manually, which is slow, labor-intensive, and prone to human error. Although convolutional neural networks (CNNs) can automate this task with high accuracy, many state-of-the-art models remain too large and computationally demanding for low-cost edge devices intended for deployment in homes, schools, and small recycling facilities. In this work, we investigate lightweight waste-classification models suitable for TinyML deployment while preserving competitive accuracy. We first benchmark multiple CNN architectures to establish a strong baseline, then apply complementary compression strategies including quantization, pruning, singular value decomposition (SVD) low-rank approximation, and knowledge distillation. In addition, we evaluate an RL-guided multi-teacher selection benchmark that adaptively chooses one teacher per minibatch during distillation to improve student training stability, achieving up to 85% accuracy with only 0.496 M parameters (FP32 ≈ 1.89 MB; INT8 ≈ 0.47 MB). Across all experiments, the best accuracy–size trade-off is obtained by combining knowledge distillation with post-training quantization, reducing the model footprint from approximately 16 MB to 281 KB while maintaining 82% accuracy. The resulting model is feasible for deployment on mobile applications and resource-constrained embedded devices based on model size and TensorFlow Lite Micro compatibility. Full article

Magnetic resonance imaging (MRI) is a critical clinical tool that requires precise and reliable interpretation for effective brain tumor diagnosis and timely treatment planning. Deep learning methods have advanced automated tumor classification greatly in the last few years, but many of the current methods are still challenged by a lack of interpretability, a lack of testing on region-focused data, and a lack of model robustness testing. Such limitations reduce clinical trust and limit the practice of automated diagnostic systems. To address these challenges, this study proposes an interpretable deep learning model for classifying brain tumors using the PMRAM dataset, which is a Bangladeshi brain MRI collection containing four categories: glioma, meningioma, pituitary tumor, and normal brain.. The proposed pipeline combines image preprocessing and feature enhancement methods, and then it trains a series of squeeze-and-excitation (SE)-enhanced convolutional neural networks such as VGG19, DenseNet201, MobileNetV3-Large, InceptionV3, and EfficientNetB3. The SE-enhanced EfficientNetB3 performed best, with 98.70% accuracy, 98.77% precision, 98.70% recall, and 98.70% F1-score. Cross-validation also demonstrated stable performance, with a mean accuracy of 96.89%. The model also exhibited efficient inference with low GPU memory consumption, enabling predictions in about 2–4 s per MRI image. Grad-CAM++ and saliency maps were used to improve the transparency of the results, and it was found that the network was concentrated on the clinically significant parts of the tumor, which affected the model predictions. Further robustness analysis and cross-dataset testing are additional evidence of the generalization possibility of the model. An online application was also implemented to allow real-time prediction and visual explanation of brain tumors. Overall, the proposed framework offers a precise, interpretable, and promising solution to automated brain tumor classification using MRI images. Full article

Modern automotive paints have a thin undercoat and color coat layer protected by a thick clear coat layer. All too often, only the clear coat layer of the automotive paint is recovered at the crime scene of a vehicle-related fatality. Searches for motor vehicle paint databases of clear coats using commercial software typically generate large hitlists that are difficult for a forensic paint examiner to work through unless additional information is provided for the search. To address this problem, deep learning has been applied to the infrared spectra of automotive clear coats to identify patterns in their spectra indicative of the motor vehicle manufacturer. An in-house automotive paint library of 2796 clear coat infrared spectra from six automotive manufacturers and 100 assembly plants was partitioned into training, validation, and prediction sets. Each spectrum has 1880 measurements over the spectral range of 4000 cm⁻¹ to 376 cm⁻¹. Several multilayer perceptron neural network models, each with three hidden layers, were developed that achieved high classification success rates for the training, validation, and prediction sets. The addition of convolutional layers to the deep learning neural network models did not improve the performance of these models. Full article