Review of cell image synthesis for image processing
| Authors | |
|---|---|
| Year of publication | 2022 |
| Type | Chapter of a book |
| MU Faculty or unit | |
| Citation | |
| Description | Opposites attract, also in the biomedical field and during the processing of cell microscopy images. In the same spirit, image processing, the indispensable analyst tool, is often supported by image synthesis applications. Image synthesis is a methodology implemented in computer program intended to create artificial cell images similar to images from real microscopy. The generation of artificial images has had a stable tradition in image processing and is currently gaining more attention with the rising popularity of deep learning. This chapter reviews the current state of cell image synthesis, including terminology, broader context, goals, and peculiarities. It offers a brief historical introspection and, most importantly, surveys all contemporary methodology and applications. The light descriptions of procedural methods with explicit parameters and deep learning-based methods with implicit parameters, such as the generative adversarial networks, are also included. Last but not least, this chapter discusses what kind of artificial images and ground-truth data the methods generate, including the subsequent usage of this data for image processing such as cell segmentation or data augmentation for deep learning. Among the covered methods are approaches generating artificial cell microscopy images of fluorescence stained proteins, actin filaments, chromatin stained nuclei, membranes, and even populations of cells or full cells in differential inference contrast microscopy, to name a few. The generated data is often accompanied by ground truth annotation, whose forms are also discussed, including cell detection markers, full cell segmentation, and cell tracking data. |
| Related projects: |