Abstract:The application of medical image segmentation technology facilitates precise tissue localization, aiding doctors in accurate diagnoses. To address limitations of current methods, we propose a Semi‐supervised Contextual Cognitive Augmentation‐based Cross‐teaching Network. This network incorporates a Contextual Cognitive Enhancement Module, employing data augmentation techniques and information extraction mechanisms to improve segmentation accuracy. By utilizing a cross‐teaching strategy and hybrid loss function, our approach encourages knowledge sharing between networks, leading to substantial improvements in multiclass medical image segmentation over existing single‐framework networks, as demonstrated in experimental results. The application of medical image segmentation technology enables accurate localization of human tissues, providing doctors with a reliable foundation for diagnosis. While deep learning methods have proven effective in this task, most current approaches rely on a single prediction framework, which overlooks Edge semantic features and results in flawed texture features. Moreover, existing supervised methods face challenges due to limited availability of high‐quality annotations in the field of medical imaging. In this article, a Semi‐supervised Contextual Cognitive Augmentation‐based Cross‐teaching Network is proposed. A Contextual Cognitive Enhancement Module is introduced consisting of two components: data augmentation and information extraction. The data augmentation component provides multi‐level data distribution by incorporating diverse perturbation strategies such as Discrete Cosine Transform and Gaussian noise. The information extraction component employs the Comprehensive Information Extraction module, which consists of Global Perception Information Extraction module and Multi‐channel Information Extraction module to extract perceptual information from images and enhance interaction between image channels, respectively. Additionally, a cross‐teaching strategy is adopted and a hybrid loss function is utilized to encourage knowledge sharing among the networks, leveraging the advantages of dual networks for improved performance. Experimental results demonstrate significant enhancements in multiclass medical image segmentation compared to several state‐of‐the‐art single‐framework networks.

Online image augmentation via regional cross-attention

Efficient Classification Using Salient Regions

Cross-modality Online Distillation for Multi-View Action Recognition

Fine-Grained Image Classification Based on Cross-Attention Network

Boosting Few-Shot Segmentation via Instance-Aware Data Augmentation and Local Consensus Guided Cross Attention

Region Mixup

Image Enhancement Using Adaptive Region- Guided Multi-Step Exposure Fusion Based on Reinforcement Learning

Saliency information and mosaic based data augmentation method for densely occluded object recognition

Semi‐supervised contextual cognitive augmentation‐based cross‐teaching network for multiclass medical image segmentation

Self-Calibrated Cross Attention Network for Few-Shot Segmentation

LGCOAMix: Local and Global Context-and-Object-Part-Aware Superpixel-Based Data Augmentation for Deep Visual Recognition

Multi-Channel Attention Selection GAN With Cascaded Semantic Guidance for Cross-View Image Translation

Enhancing Learned Image Compression via Cross Window-based Attention

Gradsalmix: gradient saliency-based mix for image data augmentation

Coarse-to-fine multi-scale attention-guided network for multi-exposure image fusion

Decoupled Data Augmentation for Improving Image Classification

Cross-View Image Translation Based on Local and Global Information Guidance

PatchMix: patch-level mixup for data augmentation in convolutional neural networks

RegionCL: Can Simple Region Swapping Contribute to Contrastive Learning?

See Better Before Looking Closer: Weakly Supervised Data Augmentation Network for Fine-Grained Visual Classification