The study is dedicated to the development of mathematical models and software for addressing the task of automatically determining the number of viable cells in a studied culture based on the results of morphological analysis. The input data for the algorithms consist of microimages of a cell culture monolayer, obtained through phase-contrast and fluorescence microscopy. To enable visual differentiation of cells, a staining method is employed using specific DNA-binding dyes that interact differently with live and dead cells. These dyes penetrate dead cells, allowing their visualization under a fluorescence microscope, while live cells remain unstained. The L929 cell line, derived from mouse connective tissue, was selected as the research object for the development and testing of the created models and software. The operation of the developed algorithm comprises two stages. In the first stage, the number of non-viable cells is determined from fluorescence images using classical computer vision techniques. In the second stage, the total number of cells within the microscope’s field of view is calculated from phase-contrast images, employing deep learning methods. For model training, a dataset was compiled consisting of 95 high-quality images of L929 cell monolayers, categorized into five subgroups based on morphological characteristics such as cell size, density, and membrane visibility. To enhance training efficiency, transfer learning was applied using the LIVECell dataset, which includes over 5,000 annotated phase-contrast microscopy images of cells visually similar to L929. Testing demonstrated that the developed system outperforms manual methods, achieving an error rate of 4–12% (comparable to or better than human performance) and a processing speed of less than 1 s per image. This tool holds promise for applications in biomedical research, diagnostics, and cytotoxicity studies, enhancing the efficiency and reproducibility of cell viability assessments.