EmbedSeg

EmbedSeg performs spatial embedding-based instance segmentation to detect and segment biological objects in 2D and 3D biomedical image data for quantitative image analysis.

Key Features:

• Embedding-Based Segmentation: Uses spatial embeddings to assign pixels/voxels to individual object instances for instance-level segmentation.
• Versatility in Dimensions: Supports analysis of both 2D and 3D biomedical image data.
• Benchmark Performance: Demonstrated comparable or improved instance segmentation performance on four 2D and seven 3D benchmark datasets relative to state-of-the-art methods.
• Creation of Training Data: Provisioned annotated training data for three new 3D datasets where such data were previously unavailable.

Scientific Applications:

• Cellular Biology: Enables automated instance segmentation of cells and subcellular structures for quantitative cellular analyses.
• Developmental Biology: Facilitates segmentation of developing tissues and multicellular structures in timepoint and volumetric imaging.
• Pathology: Supports segmentation of histological features and lesions in biomedical imaging for morphological assessment.
• Neurobiology: Assists in segmenting neurons, glia, and other neural structures in 2D and 3D neuroimaging datasets.

Methodology:

Implements spatial embedding-based instance segmentation and is trained and evaluated on annotated datasets, including four 2D and seven 3D benchmarks, with additional annotated training data generated for three new 3D datasets.