Answers are grounded in this paper and cite their sources.
The rapid advancement of artificial intelligence (AI) has catalyzed a profound transformation within the field of bioimaging. Currently, there are two prevailing strategies in developing AI-based bioimage analysis methods: the model-centric and data-centric strategies (Fig. 1 ). In general, many endeavors align with the model-centric paradigm, where researchers iterate on models and algorithms against fixed benchmark datasets (Fig. 1a ), such as the Cell Segmentation benchmark 1 , the Light My Cell challenge ( https://lightmycells.grand-challenge.org/ ) and the CellMap challenge ( https://cellmapchallenge.janelia.org/ ), to achieve higher evaluation scores over the state-of-the-art. These researches have undeniably played a pivotal role in motivating the development of cutting-edge AI methods, propelling progress from AlexNet to ResNet and the subsequent evolution into Vision Transformers, as well as nnUNet 2 . In contrast, the data-centric paradigm 3 focuses on systematic data engineering, i.e., carefully preparing and refining datasets (e.g., data cleaning, de-biasing, etc.), usually with a fixed model architecture, to achieve the desired performance.
No takes yet. Share an insight, caveat, or question.
Cao et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: