Table 2.

Key novelties of 3DCellSeg.

Aspect	Novelty
Network	Based on the characteristics of cell membrane images, a light-weight network, 3DCellSeg, is designed to yield a fast inference speed while achieving an accuracy comparable or superior to the existing cutting-edge approaches
Loss function	A new loss function, 3DCellSeg Loss, is proposed to tackle the clumped cell problem
Post-processing	Inspired by DBSCAN (Density-based Spatial Clustering of Applications with Noise)48, a new clustering method, TASCAN (Touching Area-based Spatial Clustering of Applications with Noise) is proposed for 3D cell instance segmentation; TASCAN operates faster, achieves better performance, and requires only one single manually selected hyperparameter
Model usability	3DCellSeg pipeline is robust, easy to fine-tune, and outperforms existing cutting-edge methods across different experimental datasets