Dense Multi-Object 3D Glomerular Reconstruction and Quantification on 2D Serial Section Whole Slide Images.

Area of Science:

• Renal Pathology
• Medical Imaging
• Computational Biology

Background:

• Precise glomerular quantification is crucial in renal pathology research and clinical practice.
• Manual analysis of 3D glomeruli from 2D serial sections is labor-intensive, time-consuming, and limited in accuracy.
• Current methods lack 3D visualization and quantitative analysis capabilities for glomeruli.

Purpose of the Study:

• To develop an automated, end-to-end deep learning pipeline for 3D glomeruli reconstruction and quantification from whole slide images (WSIs).
• To enable large-scale, 3D-contextual assessment of glomeruli, including volume estimation and 3D phenotype analysis.
• To overcome the limitations of manual analysis in accuracy and efficiency.

Main Methods:

• An end-to-end deep learning method was developed for automatic glomeruli detection, segmentation, and multi-object tracking (MOT) in 3D.
• The pipeline processes high-resolution WSIs as input to generate 3D glomerular reconstructions.
• The method was evaluated using MOT metrics (IDF1, MOTA) and Spearman correlation for volume estimation accuracy.

Main Results:

• The MAP3D+ pipeline achieved MOT performance scores of 81.8 in IDF1 and 69.1 in MOTA.
• The proposed volume estimation method demonstrated a strong correlation (0.84 Spearman) with manual annotations.
• The system successfully performs large-scale glomerular-registered assessment in a 3D context.

Conclusions:

• The end-to-end MAP3D+ pipeline offers a novel approach for automated 3D glomeruli reconstruction from 2D serial section WSIs.
• This method significantly enhances the efficiency and accuracy of glomerular volume quantification and 3D analysis.
• MAP3D+ facilitates extensive 3D glomerular assessment, advancing renal pathology research and clinical applications.