ClinSegAI: A post-processing framework for superior histopathology segmentation accuracy, radiomics feature...

Area of Science:

• Computational pathology
• Digital pathology
• Medical image analysis

Background:

• Accurate cell segmentation is crucial for radiomics and multi-omic analysis in digital pathology.
• Foundation models for segmentation often require refinement for clinical applications.

Purpose of the Study:

• To introduce ClinSegAI, a post-processing tool to refine cell segmentation outputs from the BiomedParse foundation model.
• To preserve radiomic feature integrity in H&E stained whole slide images.

Main Methods:

• ClinSegAI analyzes whole-slide images, preprocesses them, and uses BiomedParse for initial nucleus segmentation.
• A refinement algorithm corrects segmentation boundaries, merging or splitting regions to maintain morphological fidelity.
• Evaluation involved lung and other cancer pathology slides, comparing ClinSegAI against six alternative segmentation approaches.

Main Results:

• ClinSegAI achieved the highest average Dice Similarity Coefficient (DSC) near 0.80.
• It significantly reduced segmentation errors, demonstrating the lowest 95th-percentile Hausdorff distance (HD95) and average symmetric surface distance (ASSD).
• Refined segmentations better preserved radiomic feature distributions (shape, intensity, texture) compared to ground truth.

Conclusions:

• ClinSegAI enhances segmentation accuracy and radiomics reliability in computational pathology.
• Improved segmentation enables more reliable integration of histopathology with other modalities for enhanced prognostic models and treatment prediction.
• The tool demonstrates the value of targeted post-processing for foundational visual transformer models.