CPSS: Fusing consistency regularization and pseudo-labeling techniques for semi-supervised deep cardiovascular disease detection using all unlabeled electrocardiograms.

Deep learning usually achieves good performance in the supervised way, which requires a large amount of labeled data. However, manual labeling of electrocardiograms (ECGs) is laborious that requires much medical knowledge. Semi-supervised learning (SSL) provides an effective way of leveraging unlabeled data to improve model performance, providing insight for solving this problem. The objective of this study is to improve the performance of cardiovascular disease (CVD) detection by fully utilizing unlabeled ECG. A novel SSL algorithm fusing consistency regularization and pseudo-labeling techniques (CPSS) is proposed. CPSS consists of supervised learning and unsupervised learning. For supervised learning, the labeled ECGs are mapped into prediction vectors by the classifier. The cross-entropy loss function is used to optimize the classifier. For unsupervised learning, the unlabeled ECGs are weakly and strongly augmented, and a consistency loss is used to minimize the difference between the classifier's predictions for the two augmentations. Pseudo-labeling techniques include positive pseudo-labeling (PL) and ranking-based negative pseudo-labeling (RNL). PL introduces pseudo-labels for data with high prediction confidence. RNL assigns negative pseudo-labels to the lower-ranked categories in the prediction vectors to leverage data with low prediction confidence. In this study, VGGNet and ResNet are used as classifiers, which are jointly optimized by labeled and unlabeled ECGs. CPSS has been validated on several databases. With the same number of labeled ECGs (10%), it improves the accuracies over pure supervised learning by 13.59%, 4.60%, and 5.38% in the CPSC2018, PTB-XL, and Chapman databases, respectively. CPSS achieves comparable results to the fully supervised method with only 10% of labeled ECGs, which reduces the labeling workload by 90%. In addition, to verify the practicality of CPSS, a cardiovascular disease monitoring system is designed by heterogeneously deploying the trained classifiers on an SoC (system-on-a-chip), which can detect CVD in real time. The results of this study indicate that the proposed CPSS can significantly improve the performance of CVD detection using unlabeled ECG, which reduces the burden of ECG labeling in deep learning. In addition, the designed monitoring system makes the proposed CPSS promising for real-world applications.

Shi J ，Liu W ，Zhang H ，Chang S ，Wang H ，He J ，Huang Q ... -  《-》

Semantic contrast with uncertainty-aware pseudo label for lumbar semi-supervised classification.

Lumbar disc herniation (LDH) is a prevalent spinal disease that can result in severe pain, with Magnetic resonance imaging (MRI) serving as a commonly diagnostic tool. However, annotating numerous MRI images, necessary for deep learning based LDH diagnosis, can be challenging and labor-intensive. Semi-supervised learning, mainly utilizing pseudo labeling and consistency regularization, can leverage limited labeled images and abundant unlabeled images. However, consistency regularization solely focuses on maintaining the semantic consistency of transformed unlabeled data but fails to utilize the semantic information from labeled data to guide the unlabeled data, and additionally, pseudo labeling is prone to confirmation bias. We propose SeCoFixMatch, an innovative approach that seamlessly integrates semantic contrast and uncertainty-aware pseudo labeling into semi-supervised learning. Semantic contrast constraints the semantic consistency between labeled and unlabeled images. Pseudo labels are generated by combining predictive confidence and uncertainty, with uncertainty computing by optimizing the Kullback-Leibler (KL) loss between predictive and target Dirichlet distribution. Comparison with other semi-supervised models and ablation experiment with varying labeled data demonstrate the effectiveness and generalization of proposed model. Notably, SeCoFixMatch, trained with just 40 labels, outperforms the baseline model trained with 200 labels, reducing the annotation effort by a remarkable 80%. Proposed pseudo labeling algorithm generates more precise pseudo labels for semantic contrastive learning and semantic contrastive learning facilitates better feature representation, thereby further improving the prediction accuracy of pseudo label. The mutual reinforcement of pseudo labeling and semantic contrast constraints boosts the performance of semi-supervised algorithm.

Hai J ，Chen J ，Qiao K ，Liang N ，Su Z ，Lv H ，Yan B ... -  《-》

Deep virtual adversarial self-training with consistency regularization for semi-supervised medical image classification.

Convolutional neural networks have achieved prominent success on a variety of medical imaging tasks when a large amount of labeled training data is available. However, the acquisition of expert annotations for medical data is usually expensive and time-consuming, which poses a great challenge for supervised learning approaches. In this work, we proposed a novel semi-supervised deep learning method, i.e., deep virtual adversarial self-training with consistency regularization, for large-scale medical image classification. To effectively exploit useful information from unlabeled data, we leverage self-training and consistency regularization to harness the underlying knowledge, which helps improve the discrimination capability of training models. More concretely, the model first uses its prediction for pseudo-labeling on the weakly-augmented input image. A pseudo-label is kept only if the corresponding class probability is of high confidence. Then the model prediction is encouraged to be consistent with the strongly-augmented version of the same input image. To improve the robustness of the network against virtual adversarial perturbed input, we incorporate virtual adversarial training (VAT) on both labeled and unlabeled data into the course of training. Hence, the network is trained by minimizing a combination of three types of losses, including a standard supervised loss on labeled data, a consistency regularization loss on unlabeled data, and a VAT loss on both labeled and labeled data. We extensively evaluate the proposed semi-supervised deep learning methods on two challenging medical image classification tasks: breast cancer screening from ultrasound images and multi-class ophthalmic disease classification from optical coherence tomography B-scan images. Experimental results demonstrate that the proposed method outperforms both supervised baseline and other state-of-the-art methods by a large margin on all tasks.

Wang X ，Chen H ，Xiang H ，Lin H ，Lin X ，Heng PA ... -  《-》

FaxMatch: Multi-Curriculum Pseudo-Labeling for semi-supervised medical image classification.

Semi-supervised learning (SSL) can effectively use information from unlabeled data to improve model performance, which has great significance in medical imaging tasks. Pseudo-labeling is a classical SSL method that uses a model to predict unlabeled samples and selects the prediction with the highest confidence level as the pseudo-labels and then uses the generated pseudo-labels to train the model. Most of the current pseudo-label-based SSL algorithms use predefined fixed thresholds for all classes to select unlabeled data. However, data imbalance is a common problem in medical image tasks, where the use of fixed threshold to generate pseudo-labels ignores different classes of learning status and learning difficulties. The aim of this study is to develop an algorithm to solve this problem. In this work, we propose Multi-Curriculum Pseudo-Labeling (MCPL), which evaluates the learning status of the model for each class at each epoch and automatically adjusts the thresholds for each class. We apply MCPL to FixMatch and propose a new SSL framework for medical image classification, which we call the improved algorithm FaxMatch. To mitigate the impact of incorrect pseudo-labels on the model, we use label smoothing (LS) strategy to generate soft labels (SL) for pseudo-labels. We have conducted extensive experiments to evaluate our method on two public benchmark medical image classification datasets: the ISIC 2018 skin lesion analysis and COVID-CT datasets. Experimental results show that our method outperforms fully supervised baseline, which uses only labeled data to train the model. Moreover, our method also outperforms other state-of-the-art methods. We propose MCPL and construct a semi-supervised medical image classification framework to reduce the reliance of the model on a large number of labeled images and reduce the manual workload of labeling medical image data.

Peng Z ，Zhang D ，Tian S ，Wu W ，Yu L ，Zhou S ，Huang S ... -  《-》

PolypMixNet: Enhancing semi-supervised polyp segmentation with polyp-aware augmentation.

AI-assisted polyp segmentation in colonoscopy plays a crucial role in enabling prompt diagnosis and treatment of colorectal cancer. However, the lack of sufficient annotated data poses a significant challenge for supervised learning approaches. Existing semi-supervised learning methods also suffer from performance degradation, mainly due to task-specific characteristics, such as class imbalance in polyp segmentation. The purpose of this work is to develop an effective semi-supervised learning framework for accurate polyp segmentation in colonoscopy, addressing limited annotated data and class imbalance challenges. We proposed PolypMixNet, a semi-supervised framework, for colorectal polyp segmentation, utilizing novel augmentation techniques and a Mean Teacher architecture to improve model performance. PolypMixNet introduces the polyp-aware mixup (PolypMix) algorithm and incorporates dual-level consistency regularization. PolypMix addresses the class imbalance in colonoscopy datasets and enhances the diversity of training data. By performing a polyp-aware mixup on unlabeled samples, it generates mixed images with polyp context along with their artificial labels. A polyp-directed soft pseudo-labeling (PDSPL) mechanism was proposed to generate high-quality pseudo labels and eliminate the dilution of lesion features caused by mixup operations. To ensure consistency in the training phase, we introduce the PolypMix prediction consistency (PMPC) loss and PolypMix attention consistency (PMAC) loss, enforcing consistency at both image and feature levels. Code is available at https://github.com/YChienHung/PolypMix. PolypMixNet was evaluated on four public colonoscopy datasets, achieving 88.97% Dice and 88.85% mIoU on the benchmark dataset of Kvasir-SEG. In scenarios where the labeled training data is limited to 15%, PolypMixNet outperforms the state-of-the-art semi-supervised approaches with a 2.88-point improvement in Dice. It also shows the ability to reach performance comparable to the fully supervised counterpart. Additionally, we conducted extensive ablation studies to validate the effectiveness of each module and highlight the superiority of our proposed approach. PolypMixNet effectively addresses the challenges posed by limited annotated data and unbalanced class distributions in polyp segmentation. By leveraging unlabeled data and incorporating novel augmentation and consistency regularization techniques, our method achieves state-of-the-art performance. We believe that the insights and contributions presented in this work will pave the way for further advancements in semi-supervised polyp segmentation and inspire future research in the medical imaging domain.

Jia X ，Shen Y ，Yang J ，Song R ，Zhang W ，Meng MQ ，Liao JC ，Xing L ... -  《-》