The paper introduces u-Segment3D, a novel software toolbox designed for creating universal consensus three-dimensional (3D) segmentations of cells from two-dimensional (2D) segmented stacks derived from microscopy images. This method addresses the significant challenge of 3D cell segmentation, particularly the difficulty of acquiring sufficient, densely annotated 3D training data, by unifying and enhancing existing 2D-to-3D segmentation techniques. The framework, which is compatible with any 2D segmentation method, converts 2D instance segmentations from orthogonal views (x-y, x-z, and y-z slices) into a cohesive 3D structure using gradient descent and distance transforms without requiring new 3D training data. Validation across numerous real-life datasets demonstrates that u-Segment3D is highly effective for various cell morphologies, often achieving performance comparable to, or exceeding, native 3D segmentation models, and includes parallel computing capabilities for tissue-scale analysis.
References:
- Zhou F Y, Marin Z, Yapp C, et al. Universal consensus 3D segmentation of cells from 2D segmented stacks[J]. Nature Methods, 2025: 1-14.
