Reproducibility assessment of Dextrusion: A deep learning pipeline for detecting epithelial cell extrusion events

Reproducibility is essential for scientific software but can be hindered by technical and maintenance challenges. As part of the Global Bioimage Analyst Society reproducibility initiative, we evaluated Dextrusion, a deep learning pipeline for detecting epithelial cell extrusion events published by Villars et al. The code is openly available and supported by example datasets, ImageJ macros, and Jupyter notebooks. We can independently confirm the repeatability (using the provided example data) and reproducibility (using new semi-synthetic confocal data) to detect cell death and division events using Dextrusion. This mini-review highlights both the strengths of Dextrusion, including transparency and performance, and the wider challenges of long-term sustainability and platform compatibility in publishing scientific software.