Abstract
Modern biomedical research requires highly accurate and scalable image annotation, especially when working with cryomicroscopic data. One of the key stages in analyzing such images is the annotation of cellular structures, which is essential for training machine learning models. This paper explores an approach to optimizing the process of cell annotation by combining manual and automatic labeling within the CVAT (Computer Vision Annotation Tool) environment. The proposed method integrates the precision of manual annotation with the efficiency of automated segmentation algorithms, significantly reducing labor costs and improving reproducibility. During the study, the capabilities of CVAT were analyzed, built-in automated tools were tested, and scenarios for their combined use with manual corrections were described. The advantages of the hybrid approach, common errors, and ways to minimize them are discussed. The results demonstrate that combined annotation enables faster preparation of training datasets without significant loss of quality, especially when working with large volumes of images. This approach may be beneficial for automated biomedical data analysis systems where rapid scaling is required without compromising accuracy.