In cancer research, examining the effectiveness of individual treatments using cell cultures is increasingly common. In addition to being ethically more advantageous compared to animal experiments, cell experiments can also mimic tumors. Determining the effectiveness of each method is aided by the evaluation of microscopic images of cells, which are still somewhat performed manually today. Nowadays, artificial intelligencebased solutions are used more frequently to automate this lengthy and error-prone process. However, their accuracy does not yet enable widespread and efficient application. To improve the results, the availability of an appropriate amount and quality of training data would be necessary, among other factors. This work aims to present a tool that can be used to generate synthetic data based on microscopic images of cell experiments. With the generated images, neural networks used on real-world data can be trained more effectively. Due to broader customizability, manual labeling and annotation of the training data are not required. To compare the results, a cell-counting neural network was trained on the generated images. Based on the accuracy of cell count estimates for real and generated images, the usability of synthetic data can be determined for training neural networks. Further development may enable data generation according to recordings made with different microscopic techniques, thereby ensuring wider applicability
beyond cancer research