Multi-scale Multi-site Renal Microvascular Structures Segmentation for Whole Slide Imaging in Renal Pathology

Area of Science:

• Renal pathology
• Digital pathology
• Medical image analysis

Background:

• Manual segmentation of kidney microvasculature in whole slide images (WSI) is laborious and impractical for large datasets.
• Existing deep learning methods for automatic segmentation are often limited to single-site and single-scale training data.
• Accurate segmentation of microvascular structures (arterioles, venules, capillaries) is crucial for renal pathology.

Purpose of the Study:

• To develop a novel deep learning method, Omni-Seg, for accurate and efficient segmentation of renal microvascular structures.
• To address the limitations of single-site, single-scale training data in current segmentation models.
• To utilize partially labeled images for training a robust segmentation network.

Main Methods:

• Developed Omni-Seg, a single dynamic network trained on multi-site, multi-scale data.
• Employed partially labeled images, with only one tissue type labeled per image, for training.
• Utilized datasets from HuBMAP and NEPTUNE across various magnifications (40×, 20×, 10×, 5×).

Main Results:

• Omni-Seg demonstrated superior performance in segmenting microvascular structures compared to existing methods.
• Achieved high accuracy, validated by Dice Similarity Coefficient (DSC) and Intersection over Union (IoU) metrics.
• Successfully trained a single deep network on diverse, multi-site, and multi-scale data.

Conclusions:

• Omni-Seg offers a powerful computational tool for quantitative analysis of renal microvascular structures.
• The method overcomes limitations of traditional and current deep learning segmentation approaches.
• Enables large-scale digital pathology analysis in renal research and clinical practice.