Princut-Auto: An Unsupervised 3D Cell Detection Tool for Embryonic Data

We present PrinCut-Auto, a GUI software designed for unsupervised 3D cell detection and manual detection refinement. PrinCut-Auto is built upon statistic-based multi-scale principal curvature (MSPC), showcasing a capability that matches or potentially surpasses human-level accuracy in detecting inter-cellular gaps in noisy data. We also developed an innovative pipeline that integrates MSPC, resulting in remarkable cell detection accuracy across various scenarios. The performance of the PrinCut-Auto without manual refinement is evaluated on three benchmark datasets, which not only significantly outperforms other unsupervised methods, but also achieves comparable or even superior performance compared to popular supervised pre-trained models. PrinCut-Auto is freely available at https://github.com/yu-lab-vt/PrinCut-Auto.