This research proposes a hierarchical probabilistic graphical model to improve gland segmentation in histopathology images, addressing challenges like scale heterogeneity, dense clustering, appearance variability, and architectural complexity.
Key findings
Proposes a Multi-Scale Tissue Architecture Constraint framework to model hierarchical relationships for improved gland segmentation.
Integrates cell-level probability estimation, gland boundary constraints, and tissue-scale architecture priors.
Formulates gland segmentation as a structured prediction problem using a hierarchical Conditional Random Field.
Achieves state-of-the-art performance with interpretable uncertainty estimates and anatomically consistent segmentations.
Limitations & open questions
Further validation required for broader generalization across different tissue types and pathological grades.