Xiaoyun Xie

With the advancement of image processing techniques, few-shot learning (FSL) has gradually become a key approach to addressing the problem of data scarcity. However, existing FSL methods often rely on unimodal information under limited sample conditions, making it difficult to capture fine-grained differences between categories. To address this issue, we propose a multimodal few-shot learning method based on category name expansion and image feature enhancement. By integrating the expanded category text with image features, the proposed method enriches the semantic representation of categories and enhances the model’s sensitivity to detailed features. To further improve the quality of cross-modal information transfer, we introduce a cross-modal residual connection strategy that aligns features across layers through progressive fusion. This approach enables the fused representations to maximize mutual information while reducing redundancy, effectively alleviating the information bottleneck caused by uneven entropy distribution between modalities and enhancing the model’s generalization ability. Experimental results demonstrate that our method achieves superior performance on both natural image datasets (CIFAR-FS and FC100) and a medical image dataset.

Due to insufficient samples in rare arrhythmia classes, current research in ECG classification has reported poor performance in rare arrhythmia classes and difficulties in promptly generalizing to novel, unseen classes. This paper introduces a novel multi-level metric fusion framework specifically designed for few-shot electrocardiogram (ECG) classification. The framework aims to extract multi-level deep local descriptors and learn the intricate relationships between samples across different levels, thereby enhancing the few-shot ECG classification capability. In the proposed method, the episodic training mechanism based on N-way K-shot task is introduced to effectively train deep neural networks with limited samples. Then a unique multi-level metric fusion framework is developed for similarity learning, which jointly utilizes heartbeat-level and segment-level features for metric learning. Furthermore, the multi-level metric fusion module integrates weighted fusion of heartbeat-level and segment-level scores along the branch dimension, forming a a global decision similarity.The proposed method was evaluated on a few-shot ECG database, demonstrating remarkable performance with average accuracies of 72.74%, 69.81%, and 49.84% for 5-way 10-shot, 5-shot, and 1-shot scenarios, respectively. These results significantly surpass existing methods, highlighting the effectiveness of the proposed approach. The study presents a straightforward yet highly effective solution for multi-lead ECG classification, particularly suitable for few-shot scenarios. Furthermore, the proposed multi-level metric fusion module effectively leverages multi-level features, demonstrating its potential for improved generalization and accuracy in ECG classification tasks. This is the first study to achieve wide-range class classification in few-shot scenarios without relying on auxiliary techniques such as transfer learning or pre-training, and demonstrating important impact to clinical applications.

Electrocardiogram (ECG) denoising is very important for heart disease diagnosis. The traditional ECG denoising models have problems such as single noise type and poor interpretability of deep neural networks. The innovation of the proposed method is to incorporate the precious achievements of traditional methods into the design of neural networks and to build a bridge between them. Therefore, a novel interpretable deep denoising framework based on sparse representation is proposed in this study, and the half quadratic splitting (HQS) algorithm is applied to decompose the denoising method into sparse representations as an iterative solution process. In addition, a new weight distribution (WD) module is designed to extract adaptive hyperparameters based on ECG correlation instead of empirical values and greatly improves the efficiency of hyperparameter selection. To demonstrate the fairness and effectiveness of the proposed method, four different denoising models with different data preprocessing techniques are used for comparison. The extensive experimental validation and simulation studies demonstrated that the proposed framework has an excellent performance in quantitative and visual evaluation.

The precise identification of arrhythmia is critical in electrocardiogram (ECG) research. Many automatic classification methods have been suggested so far. However, efficient and accurate classification is still a challenge due to the limited feature extraction and model generalization ability. We integrate attention mechanism and residual skip connection into the U-Net (RA-UNET); besides, a skip connection between the RA-UNET and a residual block is executed as a residual attention convolutional neural network (RA-CNN) for accurate classification. The model was evaluated using the MIT-BIH arrhythmia database and achieved an accuracy of 98.5% and scores for the classes and of 82.8% and 91.7%, respectively, which is far superior to other approaches.

The 12-lead electrocardiogram (ECG) is widely used in the clinical diagnosis of cardiovascular disease, and deep learning has become an effective approach to automatic ECG classification. Generally, current research simply regards the 12-lead ECG signal as an ordinary 2-D array and does not specifically consider the intrinsic relationship between different leads when building the neural networks. However, ECG classes, from the biomedical perspective, mainly show specific patterns on one or several leads rather than all 12 leads, which suggests that it would be more efficient to learn class-specific intrinsic features from corresponding leads. To make use of such domain knowledge, in this study, we present a multilabel 12-lead ECG classification method based on the leadwise grouping multibranch network. A simple yet effective leadwise grouping strategy is proposed to incorporate domain knowledge to ECG classification models. Meanwhile, a multibranch network is designed accordingly, and spatial and temporal features are extracted by a BranchNet for each branch, which corresponds to one lead group. Moreover, an extended focal loss is presented to solve the class imbalance problem for multilabel classification. The proposed method was evaluated on two large-scale real-world ECG databases and yielded values of 0.9599, 0.7920, 0.7490, 0.5537, 0.1282, and 1.5354 for the area under receiver operating characteristic curve (AUROC), area under precision–recall curve (AUPRC), F1, accuracy (Acc), one-error (OE), and coverage (Cove), respectively, on the PhysioNet/Computing in Cardiology Challenge 2020 (CinC2020) database and values of 0.9531, 0.8975, 0.8102, 0.7484, 0.1791, and 0.5392 on the Shandong Provincial Hospital (SPH) database. The results are better than the existing work and are reached with fewer parameters and lower computational cost, demonstrating the effectiveness of the proposed method and leadwise grouping strategy.

In ECG applications, the correct recognition of R-peaks is extremely important for detecting abnormalities, such as arrhythmia and ventricular hypertrophy. In this work, a novel ECG enhancement and R-peak detection method based on window variability is presented, and abbreviated as SQRS. Firstly, the ECG signal corrupted by various high or low-frequency noises is denoised by moving-average filtering. Secondly, the window variance transform technique is used to enhance the QRS complex and suppress the other components in the ECG, such as P/T waves and noise. Finally, the signal, converted by window variance transform, is applied to generate the R-peaks candidates, and the decision rules, including amplitude and kurtosis adaptive thresholds, are applied to determine the R-peaks. A special squared window variance transform (SWVT) is proposed to measure the signal variability in a certain time window, and this technique reduces false detection rate caused by the various types of interference presented in ECG signals. For the MIT-BIH arrhythmia database, the sensitivity of R-peak detection can reach 99.6% using the proposed method.

Semi-supervised classification methods result in higher performance for hyperspectral images, because they can utilize the relationship between unlabeled samples and labeled samples to obtain pseudo-labeled samples. However, how generating an effective training sample set is a major challenge for semi-supervised methods, In this paper, we propose a novel semi-supervised classification method based on extended label propagation (ELP) and a rolling guidance filter (RGF) called ELP-RGF, in which ELP is a new two-step process to make full use of unlabeled samples. The first step is to implement the graph-based label propagation algorithm to propagate the label information from labeled samples to the neighboring unlabeled samples. This is then followed by the second step, which uses superpixel propagation to assign the same labels to all pixels within the superpixels that are generated by the image segmentation method, so that some labels wrongly labeled by the above step can be modified. As a result, so obtained pseudo-labeled samples could be used to improve the performance of the classifier. Subsequently, an effective feature extraction method, i.e., RGF is further used to remove the noise and the small texture structures to optimize the features of the initial hyperspectral image. Finally, these produced initial labeled samples and high-confidence pseudo-labeled samples are used as a training set for support vector machine (SVM). The experimental results show that the proposed method can produce better classification performance for three widely-used real hyperspectral datasets, particularly when the number of training samples is relatively small.

In this paper, a novel feature extraction method based on Non-convex Robust Principal Component Analysis (NRPCA) and minimum noise fraction is proposed for hyperspectral image classification. The proposed method consists of following steps. First, the spectral dimension of the hyperspectral image is reduced with averaging-based image fusion. Then, each band of the hyperspectral image which dimension has been reduced are processed with NRPCA technique. Next, minimum noise fraction is performed to reduce low-rank data dimensionality and noise. Finally, SVM as a pixel-wise classifier is used to obtain final classification. Experimental results demonstrate that the proposed method is able to obtain much higher classification accuracy than other methods in hyperspectral image.