Generative Models For Time-lapse Cell Datasets: A Structured Review of GAN and Diffusion Approaches

Abstract

Time-lapse cell microscopy provides detailed information on cell dynamics, morphology, interactions, and development. However, both the techniques and datasets have limitations that hinder cell analysis. Recent advances in generative models, especially generative adversarial networks (GANs) and diffusion models, offer new tools to alleviate data availability limitations by generating samples that resemble the probability distribution of real data. This review summarizes applications of GANs and diffusion models in time-lapse microscopy of cells. For each group, we analyze representative studies with respect to data, architecture design, evaluation metrics, and biological plausibility. This review provides an insightful overview of the theoretical foundation and a concise summary of existing methods for time-lapse cell microscopy.