An integrated region-, boundary-, shape-based active contour for multiple object overlap resolution in histological...

Area of Science:

• Medical image analysis
• Computer vision
• Computational pathology

Background:

• Active contours and active shape models (ASM) are common for image segmentation but struggle with intersecting objects and occlusion.
• Existing methods often segment overlapping objects as one and face challenges with landmark identification and single-object segmentation.

Purpose of the Study:

• To present a novel synergistic active contour model for improved image segmentation.
• To address limitations of existing methods in resolving overlapping and occluded object boundaries.
• To apply the model for segmenting nuclear and glandular structures in histopathology images.

Main Methods:

• Developed a synergistic boundary and region-based active contour model using a level set formulation.
• Incorporated shape priors and automated initialization via watershed.
• Utilized multiple level sets for simultaneous segmentation of multiple objects.
• Model energy functional includes prior shape, boundary detection, and region statistics terms.

Main Results:

• The model successfully resolves object overlap and separates occluded boundaries of multiple objects simultaneously.
• Evaluated on 100 prostate and 14 breast cancer histology images for nuclei and lymphocyte segmentation.
• Outperformed geodesic active contour and Rousson shape-based models.
• Achieved an average resolution of 91% for overlapping/occluded structures.

Conclusions:

• The proposed synergistic active contour model offers superior performance in segmenting complex structures with overlaps and occlusions.
• Demonstrates significant potential for applications in digital pathology and cancer diagnosis.
• Provides a robust solution for overcoming limitations of traditional active contour and ASM methods.