The field of medical imaging analysis is rapidly advancing, with a focus on developing innovative methods for image classification, segmentation, and denoising. Recent research has explored the use of multiple instance learning, transformer models, and reinforcement learning to improve the accuracy and efficiency of medical image analysis. Notably, the integration of semantic and structural cues has shown promise in improving robustness against noisy annotations. Additionally, the development of hybrid attention networks and collaborative learning frameworks has enhanced image quality and diagnostic accuracy.

Some noteworthy papers in this area include: SemaMIL, which achieves state-of-the-art accuracy in whole slide image classification with fewer FLOPs and parameters. Double-Constraint Diffusion Model, which outperforms state-of-the-art methods in ultra-low-dose PET reconstruction and generalizes well to unknown dose reduction factors. GSD-Net, which improves robustness against noisy annotations in medical image segmentation. PPORLD-EDNetLDCT, which achieves superior denoising performance in low-dose CT images. SAC-MIL, which performs spatial-aware correlated multiple instance learning for histopathology whole slide image classification. Dual Interaction Network, which effectively exploits mutual complementary information from original and enhanced images for medical image segmentation. Hybrid Swin Attention Networks, which achieves superior denoising performance in low-dose PET and CT images. Co-Seg, which collaboratively segments tissue regions and nuclei instances for tumor microenvironment and cellular morphology analysis.