3DCellSeg

3DCellSeg performs three-dimensional cell instance segmentation in densely packed cell membrane images to identify and separate individual cells.

Key Features:

• Two-Stage Segmentation Pipeline: Implements a two-stage segmentation framework that performs voxel-wise mask prediction followed by instance separation.
• 3DCellSegNet Deep Convolutional Neural Network: Uses a lightweight deep convolutional neural network architecture to generate voxel-wise segmentation masks from 3D cell membrane images.
• 3DCellSeg Loss Function: Applies a custom loss function optimized for distinguishing clumped or closely packed cells during training.
• Touching Area-Based Clustering Algorithm (TASCAN): Separates individual cell instances from foreground segmentation masks using a clustering algorithm based on touching-area relationships.
• Multi-Dataset Evaluation: Demonstrates segmentation performance across datasets including ATAS, LRP, Ovules, and HMS cell imaging datasets.

Scientific Applications:

• Histopathological Image Analysis: Enables segmentation of densely packed cells in histopathological images for cellular-level analysis.
• Cellular Morphology Studies: Supports quantitative analysis of cell structures in complex biological imaging datasets.

Methodology:

3DCellSeg generates voxel-wise segmentation masks using the 3DCellSegNet deep convolutional neural network trained with the 3DCellSeg Loss function, and separates individual cells from foreground masks using the touching area-based clustering algorithm TASCAN.