The field of biomedical image processing is moving towards the development of more efficient and effective methods for image restoration, registration, and translation. Researchers are focusing on creating lightweight models that can achieve high-quality results while requiring fewer computational resources. This is particularly important in biomedical imaging, where large amounts of data need to be processed quickly and accurately. Another trend is the use of self-supervised learning techniques, which can learn from noisy or unpaired data, making them useful for applications where paired datasets are scarce or difficult to obtain. Noteworthy papers in this area include:

• A paper that presents an ultra-lightweight model for detail-preserving biomedical image restoration, which achieves fast denoising and high-quality image restoration while requiring only a fraction of the computational resources of existing methods.
• A paper that proposes a general framework for self-supervised denoising with patch aggregation, which leverages only noisy OCT images to improve image clarity and diagnostic outcomes in clinical practice.
• A paper that introduces a novel registration framework that dynamically adjusts elastic regularization based on the gradient norm of the deformation field, ensuring robustness and anatomical plausibility in deformable medical image registration.
• A paper that presents a curvilinear structure-preserving unpaired cross-domain medical image translation framework, which explicitly preserves fine curvilinear structures during translation, improving diagnostic reliability and quantitative analysis in medical imaging.