A survey of deep learning-based microscopic cell image understanding

Enhancement and recognition of microscopic cell images are crucial for advancing biomedical research, with deep learning methods rapidly developing for various tasks. Deep learning drives the processing of multiple image types, including microscopic cell images. Numerous surveys have focused on clinical images like X-rays or CT scans, or on enhancement and segmentation tasks for typical cells and tissues. This inspired us to conduct a comprehensive survey on deep learning-based image processing of microscopic cell images. After introducing recent imaging techniques in the biomedical field, we classified existing deep learning-based methods into two categories: enhancement and recognition. The enhancement section addresses image denoising and super-resolution of 2D images, while 3D image enhancement methods are further divided into 3D reconstruction and isotropic reconstruction. The recognition section is organized into three main aspects: single cell recognition, organelles and subcellular recognition, and live cell dynamic behavior recognition. Finally, we discuss the unique challenges and potential improvements for deep learning-based methods.