Glo-In-One: holistic glomerular detection, segmentation, and lesion characterization with large-scale web image...

Area of Science:

• Digital Pathology
• Computational Pathology
• Renal Pathology

Background:

• Quantitative analysis of glomeruli in Whole Slide Imaging (WSI) is crucial for computer-assisted diagnosis and research.
• Traditional methods require significant programming expertise, limiting accessibility for non-technical users.
• There is a need for streamlined tools for glomerular quantification and characterization.

Purpose of the Study:

• To develop the Glo-In-One toolkit for holistic glomerular detection, segmentation, and characterization.
• To provide a user-friendly solution for glomerular analysis, accessible via a single command line.
• To release a large dataset of glomerular images to advance self-supervised deep learning in renal pathology.

Main Methods:

• The Glo-In-One toolkit processes WSIs to output multi-class circle glomerular detection, segmented glomerular image patches, and lesion characterization.
• Self-supervised deep learning, utilizing large-scale web-mined images, is employed for enhanced glomerular quantification.
• Fine-grained characterization of global glomerulosclerosis (GGS) types, including assessed-solidified-GSS, disappearing-GSS, and obsolescent-GSS, is implemented.

Main Results:

• The GGS classification model demonstrated strong performance using only 10% of annotated data compared to supervised methods.
• Glomerular detection achieved an average precision of 0.627 using circle representations.
• Glomerular segmentation yielded a high patch-wise Dice similarity coefficient of 0.955.

Conclusions:

• The open-source Glo-In-One toolkit offers a user-friendly, command-line interface for comprehensive glomerular analysis in digital renal pathology.
• The toolkit facilitates non-technical users in performing complex glomerular detection, segmentation, and lesion characterization.
• The publicly available toolkit and dataset will aid further research and development in automated renal pathology analysis.