Segmentation of nuclei (Wahlby algorithm)

Segmentation of nuclei (Wahlby algorithm) segments nuclei in microscopy images stained with DNA-specific dyes such as DAPI for quantitative analysis in cytology, histopathology, and developmental biology.

Key Features:

• Region-based segmentation: Uses morphological filtering and gradient magnitude analysis to delineate nuclear structures.
• Seed-based watershed segmentation: Generates object and background seeds from morphological operations on the original image and its gradient magnitude, ensuring at least one seed per nucleus.
• Initial over-segmentation handling: Refines over-segmentation by merging neighboring regions based on gradient magnitude along boundaries and removing poorly contrasted objects.
• Cluster separation based on shape: Separates closely packed or overlapping nuclei using shape analysis.
• Support for 2D and 3D images: Applicable to both two-dimensional and three-dimensional microscopy image datasets.
• Minimal parameter requirement: Operates with five input parameters that can be configured on a test image and applied across similarly acquired datasets.
• Validation against manual counts: Performance was verified by comparison with manual counts from identical image fields, achieving approximately 90% correct segmentation accuracy.

Scientific Applications:

• Cytology: Enables quantitative analysis of nuclear number and morphology in cytological images stained with DNA-specific dyes such as DAPI.
• Histopathology: Provides nucleus-level segmentation for tissue-level quantitative and morphometric analyses.
• Developmental biology: Facilitates analysis of nuclear positions and counts in developmental imaging studies.
• 2D and 3D imaging datasets: Supports both planar and volumetric nuclear segmentation for diverse experimental setups.

Methodology:

Computational steps comprise morphological operations on the original image and its gradient magnitude to generate object/background seeds, watershed segmentation, merging of over-segmented regions based on gradient magnitude along boundaries, and shape-based separation of clusters.