USE-Net

USE-Net segments individual nuclei in histopathology images to enable morphometric feature extraction for disease characterization and treatment-response analysis.

Key Features:

• U-Net variation with squeeze and excitation blocks (USE): Incorporates squeeze and excitation blocks for feature recalibration to emphasize informative features and suppress less useful ones.
• Output of shape markers: Produces segmentation masks and shape markers to facilitate separation of closely clustered nuclei and improve delineation in dense regions with indistinct boundaries.
• Robustness to annotation issues: Addresses subjective annotation and mislabeling to maintain performance when trained on relatively small datasets.
• Evaluation on MoNuSeg: Demonstrates promising results on unseen data from the 2018 MICCAI Multi-Organ-Nuclei-Segmentation (MoNuSeg) challenge dataset.

Scientific Applications:

• Computational pathology: Enables precise nuclei instance segmentation for downstream computational pathology analyses.
• Morphometric feature analysis: Facilitates extraction of morphometric features that inform understanding of disease mechanisms and prediction of treatment responses.
• Annotation-limited studies: Supports studies where manual annotation by pathologists is laborious or limited by dataset size.

Methodology:

Train a modified U-Net architecture on histopathological images with integrated squeeze and excitation blocks to dynamically adjust feature importance; output segmentation masks and shape markers to distinguish individual nuclei within dense clusters, with evaluation on unseen data from the 2018 MICCAI Multi-Organ-Nuclei-Segmentation (MoNuSeg) dataset.