Deep Bayesian Active-Learning-to-Rank for Endoscopic Image Data

Signal peptides play essential roles in protein secretion and localization, and their accurate identification is critical for understanding protein synthesis, transport, and functional regulation. However, severe class imbalance in signal peptide data sets leads to substantially lower recognition performance for minor classes compared with major classes. Here, we propose a structure-aware multimodal signal peptide prediction network (SaSPNet), which incorporates structural modality information into conventional sequence modeling and uses a graph convolutional network (GCN)-based structure encoder to learn structural representations of signal peptides for both signal peptide type and cleavage-site prediction. SaSPNet significantly improves the prediction performance for minor signal peptide classes on the USPNet data set, achieving more than a 10% gain over existing methods on key minor-class metrics. Feature visualization and explainability analyses show that the structure encoder learns more discriminative structural patterns for minor signal peptides, revealing the mechanism by which the structural modality enhances model performance. In addition, comparative analyses using three-dimensional structures generated by different structure prediction models demonstrate that SaSPNet is robust to variations in structural data quality. We further construct an independent test set, SP-MinorEval, specifically for minor signal peptides, and evaluations on this data set show that SaSPNet maintains strong performance across domains, providing an effective tool for minor-class signal peptide prediction, protein secretion mechanism studies, and functional protein discovery.

Informative sample generation using class aware generative adversarial networks for classification of chest Xrays

Neural networks trained with class-imbalanced data are known to perform poorly on minor classes of scarce training data. Several recent works attribute this to over-fitting to minor classes. In this paper, we provide a novel explanation of this issue. We found that a neural network tends to first under-fit the minor classes by classifying most of their data into the major classes in early training epochs. To correct these wrong predictions, the neural network then must focus on pushing features of minor class data across the decision boundaries between major and minor classes, leading to much larger gradients for features of minor classes. We argue that such an under-fitting phase over-emphasizes the competition between major and minor classes, hinders the neural network from learning the discriminative knowledge that can be generalized to test data, and eventually results in over-fitting. To address this issue, we propose a novel learning strategy to equalize the training progress across classes. We mix features of the major class data with those of other data in a mini-batch, intentionally weakening their features to prevent a neural network from fitting them first. We show that this strategy can largely balance the training accuracy and feature gradients across classes, effectively mitigating the under-fitting then over-fitting problem for minor class data. On several benchmark datasets, our approach achieves the state-of-the-art accuracy, especially for the challenging step-imbalanced cases.

Automatic sleep staging is a classification process with severe class imbalance and suffers from instability of scoring stage N1. Decreased accuracy in classifying stage N1 significantly impacts the staging of individuals with sleep disorders. We aim to achieve automatic sleep staging with expert-level performance in both N1 stage and overall scoring. A neural network model combines an attention-based convolutional neural network and a classifier with two branches is developed. A transitive training strategy is employed to balance universal feature learning and contextual referencing. Parameter optimization and benchmark comparisons are conducted using a large-scale dataset, followed by evaluation on seven datasets in five cohorts. The proposed model achieves an accuracy of 88.16%, Cohen's kappa of 0.836, and MF1 score of 0.818 on the SHHS1 test set, also with comparable performance to human scorers in scoring stage N1. Incorporating multiple cohort data improves its performance. Notably, the model maintains high performance when applied to unseen datasets and patients with neurological or psychiatric disorders. The proposed algorithm demonstrates strong performance and generalizablility, and its direct transferability is noteworthy among similar studies on automated sleep staging. It is publicly available, which is conducive to expanding access to sleep-related analysis, especially those associated with neurological or psychiatric disorders.

Timely detection of sliding bearing oil-film faults is critical for preventing costly failures and ensuring equipment reliability. Yet, diagnosis accuracy is often hindered by data scarcity and severe class imbalance in industrial environments. To overcome these challenges, this study proposes a physics-informed generative framework that couples sliding bearing mechanics with data-driven modeling. The Reynolds equation is embedded into a physics-informed neural network to generate physically consistent samples for data augmentation, while a self-attention–enhanced convolutional neural network is employed to extract features from three-dimensional oil-film pressure distributions. To the best of our knowledge, this is the first work to integrate physics-informed generative modeling with self-attention mechanisms for sliding bearing fault diagnosis. Comparative experiments against conventional finite difference and other imbalanced approaches demonstrate that the proposed method not only solves the oil-film pressure field with higher efficiency but also improves fault classification accuracy on imbalanced datasets by up to 37.5%, thereby offering a new paradigm for robust diagnosis under limited data conditions.

In the node classification task on graphs, mainstream graph neural networks often adopt the same neighborhood aggregation strategy for all nodes. This approach results in biased learning toward features from the majority nodes while devaluing those from the minority nodes under class imbalance. In particular, for marginal nodes, limited neighboring information can severely impact classification performance. To address this challenge, this paper proposes a feature fusion‐based hybrid quantum‐classical graph residual neural network (QGRNN). Leveraging the nonlinear expressive capacity of qubits in modeling complex feature interactions, the model innovatively integrates a structure‐driven node selection mechanism with a quantum feature enhancement module, while also dynamically fusing classical features and Hamiltonian expectation values through a gated residual fusion mechanism to compensate for representational deficiencies of marginal nodes overlooked by classical methods. Experimental results show that QGRNN consistently outperforms baselines across a range of node classification tasks. In binary and ternary classification settings, it exhibits strong discriminative capability and robustness, especially maintaining high accuracy under severe class imbalance. In addition, QGRNN also demonstrates strong generality in other tasks, in recommendation scenarios, achieving an average improvement of 36.3% on NDCG@5, 44.3% on Recall@5, and 21.9% on Recall@10 compared to the baseline.

Climate change poses significant challenges to agriculture, leading to increased crop damage owing to extreme weather conditions. Detecting and analyzing such damage is crucial for mitigating its effects on crop yield. This study proposes a novel autoencoder (AE)-based model, termed “Memory Ganomaly,” designed to detect and analyze weather-induced crop damage under conditions of significant class imbalance. The model integrates memory modules into the Ganomaly architecture, thereby enhancing its ability to identify anomalies by focusing on normal (undamaged) states. The proposed model was evaluated using apple and peach datasets, which included both damaged and undamaged images, and was compared with existing robust Convolutional neural network (CNN) models (ResNet-50, EfficientNet-B3, and ResNeXt-50) and AE models (Ganomaly and MemAE). Although these CNN models are not the latest technologies, they are still highly effective for image classification tasks and are deemed suitable for comparative analyses. The results showed that CNN and Transformer baselines achieved very high overall accuracy (94–98%) but completely failed to identify damaged samples, with precision and recall equal to zero under severe class imbalance. Few-shot learning partially alleviated this issue (up to 75.1% recall in the 20-shot setting for the apple dataset) but still lagged behind AE-based approaches in terms of accuracy and precision. In contrast, the proposed Memory Ganomaly delivered a more balanced performance across accuracy, precision, and recall (Apple: 80.32% accuracy, 79.4% precision, 79.1% recall; Peach: 81.06% accuracy, 83.23% precision, 80.3% recall), outperforming AE baselines in precision and recall while maintaining comparable accuracy. This study concludes that the Memory Ganomaly model offers a robust solution for detecting anomalies in agricultural datasets, where data imbalance is prevalent, and suggests its potential for broader applications in agricultural monitoring and beyond. While both Ganomaly and MemAE have shown promise in anomaly detection, they suffer from limitations—Ganomaly often lacks long-term pattern recall, and MemAE may miss contextual cues. Our proposed Memory Ganomaly integrates the strengths of both, leveraging contextual reconstruction with pattern recall to enhance detection of subtle weather-related anomalies under class imbalance.

Limited experimental data remains a key challenge in applying machine learning to drug discovery, particularly for cancer-related targets. In this study, we present a data-efficient active meta-deep learning framework to predict mitogen-activated protein kinase 1 (MAPK1) inhibitors, which are promising candidates for cancer-related therapies. Our approach integrates active learning (AL) with a meta-model that combines four deep architectures: a convolutional neural network, an attention, a graph convolutional network, and a graph neural network-attention, trained on molecular descriptors and graph-based representations. These models generate four probability-based features that feed into an attention-based meta-learner, improving predictive performance by 5.12% in the area under the precision-recall curve (AUPRC) and 5.48% in the Matthews correlation coefficient (MCC) using only 10% of the training data. Among the AL sampling strategies evaluated, entropy sampling showed competitive performance in selecting informative molecules for model improvement. Overall, our framework achieves an AUPRC of 0.835 ± 0.017 and MCC of 0.817 ± 0.017, on par with a traditional training method despite using only 26.7% of the training data. Compared to a conventional random forest model trained on brute-force, a 100% full training set, our approach shows a 10.6% improvement in AUPRC and modest gains in MCC, confirming the effectiveness of the proposed framework. Under severe class imbalance, balanced accuracy steadily increased across AL iterations, reaching values greater than 0.85 at the final iteration for all uncertainty-driven strategies. Molecular docking confirmed successful prioritization of the top four predicted compounds. Evaluation on an external MAPK1 data set demonstrated generalizability, with our approach achieving an AUPRC of 0.818 and an MCC of 0.403, comparable to the independent test set. These results highlight the potential of combining intelligent data selection with deep learning architectures through the meta-model to accelerate predictive performance in data-scarce drug discovery. Scientific contribution: This study contributes a novel, data-efficient active meta-deep learning framework for predicting MAPK1 inhibitors, addressing the challenge of limited experimental data in a cancer-specific target. By integrating AL with a meta-model composed of four deep architectures, the approach significantly enhances the predictive performance using only a fraction of the training data. The framework achieves superior metrics compared to traditional training methods, highlighting its potential to accelerate drug discovery in data-scarce settings.

Accurate classification of rice plant diseases is essential for early intervention and precision agriculture. However, real-world datasets often suffer from complex backgrounds, high-dimensional features, and severe class imbalances, which compromise classification performance. This study proposes an integrated framework combining image segmentation using U-Net, feature selection via Ant Colony Optimization (ACO), hybrid sampling to handle class imbalance, and final classification using a Convolutional Neural Network (CNN). Segmentation isolates disease-affected areas, ACO optimizes feature subsets, and hybrid sampling balances class distribution using undersampling and SMOTE. The proposed method was tested on four rice leaf disease datasets—Brown Spot, Leaf Blast, Leaf Blight, and Leaf Scald—exhibiting significant class imbalance. Experimental results show that the proposed approach outperforms baseline models (SegNet, PspNet, and E-Net) across multiple metrics: Accuracy, IoU, Precision, and Recall. This indicates the framework’s robustness and potential for real-world deployment in precision agriculture. Future work will focus on model compression and real-time implementation in IoT systems.

Application of a burned area Spatio-Temporal Stratification–Based positive sample augmentation strategy in Fine-Scale wildfire susceptibility mapping

Laser Additive Manufacturing (LAM) has emerged as a transformative technology in the aerospace and biomedical industries due to its ability to fabricate and repair high-value components with high precision. However, the intricate laser-material interactions involved in LAM processes, such as Directed Energy Deposition (DED) and Powder Bed Fusion (PBF-L), can lead to defects like porosity, which compromise the structural integrity of the final parts. Traditional non-destructive evaluation methods, such as high-resolution X-ray Computed Tomography (XCT), are commonly used to identify internal defects. But the process is very time-consuming and requires labor-intensive manual inspection. In contrast, in-situ thermal monitoring of melt pool dynamics presents a promising alternative, offering real-time insights into defect formation. Despite recent advances, accurately interpreting melt pool thermal gradients remains a challenge for conventional analytical methods. This research proposes a domain-informed artificial intelligence (AI) framework that addresses three interconnected challenges in LAM, namely large amounts of unlabeled data with limited labels and severe class imbalance by performing (i) unsupervised porosity segmentation in unlabeled XCT scan datasets using the Segment Anything Model (SAM) guided by prompts created through unsupervised clustering, (ii) semi-supervised classification of limited-labeled melt pool thermal images by combin- ing a self-supervised Masked Autoencoders (MAE) and Vision Transformers (ViT) classifier, and (iii) handling severe class imbalance by generating synthetic minority samples using a domain-aware class- conditioned diffusion model with high fidelity samples with mode coverage. The proposed methodologies are validated on in-house experimental datasets, highlighting real-world applicability for limited labeled or unlabeled datasets. The research was divided into two main parts: in-house experimentation- tion with data acquisition and labeling, and customized modeling to address key challenges and perform downstream tasks. Overall, this work demonstrates the potential of customizing state-of-the-art AI techniques for complex and label-scarce manufacturing datasets, contributing towards robust, scalable, and efficient defect detection frameworks in the LAM domain.

Modern power systems are increasingly complex, and the risk of transient instability is rising accordingly. Data-driven transient stability assessment (TSA) is attractive for its efficiency, yet in practice the number of unstable events is much smaller than that of stable ones, leading to severe class imbalance and degraded accuracy. This paper proposes a SHAP-guided, classifier-controlled diffusion augmentation framework to mitigate imbalance and enhance TSA. First, SHAP analysis identifies critical unstable and near-boundary samples, ensuring that augmentation targets the most informative regions of the state space. Then, a classifier-guided conditional diffusion model—with a Transformer-based denoising network—generates class-faithful synthetic trajectories that capture long-range temporal dependencies and inter-variable couplings. Case studies on the IEEE 10-machine 39-bus system show that the proposed method consistently surpasses traditional over-sampling (e.g., SMOTE/ADASYN) and deep generative baselines (e.g., CGAN/TimeGAN) in terms of accuracy, precision, recall, and F1-score. Moreover, the approach maintains strong performance under small-sample settings and shortened time-series inputs, demonstrating favorable adaptability and robustness. These results indicate that the proposed augmentation framework offers a practical and effective solution for TSA under severe class imbalance.

Single-field-of-view (FoV) histopathological image classification is vital for colorectal cancer (CRC) diagnosis in mid- to low-tier hospitals lacking whole-slide imaging (WSI) scanners and storage, yet suffers from severe class imbalance and degraded performance. To address this, we propose a dual-level imbalance mitigation (DIM) framework integrating data-level and algorithm-level approaches. Specifically: (1) A global context generative adversarial network (GCGAN) generates realistic minority-class images for augmentation to balance the dataset. (2) A frequency-aware adaptive focal loss (FAFL) applies a frequency-aware offset and adaptive modulation to better separate overlapping classes. (3) A lightweight receptive field-based convolutional neural network (LRF-CNN) is trained under DIM to leverage both augmentation and loss modulation for improved classification. Extensive experiments on the single-FoV colorectal histopathology dataset demonstrate that DIM-equipped LRF-CNN outperforms five state-of-the-art models (SOTA) across multiple metrics. Furthermore, each DIM component enhances performance when applied individually to those SOTA models, and additional validation on six single-FoV histopathological datasets confirms the generalizability and effectiveness of the proposed DIM framework. Our code is available at https://github.com/Lingling-Yuan/DIM.

Purpose: Endoscopic digital images contain rich information on texture and color. There is increasing interest in differentiating tissue pathology based on the patterns of color and texture features of endoscopic images extracted by digital image analysis (DIA). Objective: In this study, we applied techniques of DIA to endoscopic images of diminutive polyps in order to identify the color and texture features of endoscopic images that may differentiate adenomatous polyps from hyperplastic polyps. Methods: Methods: Our endoscopic image database was searched to retrieve and store digital images of diminutive polyps with diagnosis confirmed by histopathology. Texture analysis was performed on multiple regions of interest (ROIs) digitally selected in these images using the MatLab image processing toolbox; the following texture parameters were extracted: histogram (first order statistics) run-length and co-occurrence matrix (measuring features of runs of pixels and distribution of pairs of pixels), gradient analysis (spatial distribution of pixels), auto-regressive analysis (measurement of local interaction amongst pixels of different grey level values) and also wavelet analysis (which measures parameters of spatial frequency). Principal component analysis (PCA) was used for data reduction, and an artificial neural network (NN) based predictive model was built, trained and validated using the extracted texture features for classification of adenomatous polyps from hyperplastic polyps. Results: Results: A total of 110 and 109 ROIs were selected from endoscopic images of 30 and 28 patients with diminutive adenomatous and hyperplastic polyps, respectively. The sensitivity and specificity of the multilayered perceptron neural network classification model were both 82% with an area under the receiver operating characteristic (ROC) curve of 0.87. Three types of features of pixel distribution had high discriminatory power: Kurtosis (measure of the steepness of the pixel distribution), angular second momentum and horizontal and vertical grey level non-uniformity (parameters estimating the second order joint conditional probability density functions of spatial distribution of two neighboring pixels) Conclusion: Conclusion: Results of this exploratory study suggest that the technique DIA may be a clinically useful adjunct for non-invasive automated diagnosis of diminutive colon polyps, particularly if available as a real-time application.

Visual inspection, lesion detection, and differentiation between malignant and benign features are key aspects of an endoscopist's role. The use of machine learning for the recognition and differentiation of images has been increasingly adopted in clinical practice. This study aimed to establish convolutional neural network (CNN) models to automatically classify gastric neoplasms based on endoscopic images. Endoscopic white-light images of pathologically confirmed gastric lesions were collected and classified into five categories: advanced gastric cancer, early gastric cancer, high grade dysplasia, low grade dysplasia, and non-neoplasm. Three pretrained CNN models were fine-tuned using a training dataset. The classifying performance of the models was evaluated using a test dataset and a prospective validation dataset. A total of 5017 images were collected from 1269 patients, among which 812 images from 212 patients were used as the test dataset. An additional 200 images from 200 patients were collected and used for prospective validation. For the five-category classification, the weighted average accuracy of the Inception-Resnet-v2 model reached 84.6 %. The mean area under the curve (AUC) of the model for differentiating gastric cancer and neoplasm was 0.877 and 0.927, respectively. In prospective validation, the Inception-Resnet-v2 model showed lower performance compared with the endoscopist with the best performance (five-category accuracy 76.4 % vs. 87.6 %; cancer 76.0 % vs. 97.5 %; neoplasm 73.5 % vs. 96.5 %; P < 0.001). However, there was no statistical difference between the Inception-Resnet-v2 model and the endoscopist with the worst performance in the differentiation of gastric cancer (accuracy 76.0 % vs. 82.0 %) and neoplasm (AUC 0.776 vs. 0.865). The evaluated deep-learning models have the potential for clinical application in classifying gastric cancer or neoplasm on endoscopic white-light images.