Standardized Instance-Level Quantification of CD34-Positive Vessels in Lymph Node Whole-Slide Images Using U-Net

Area of Science:

• Biomedical Image Analysis
• Computational Pathology
• Deep Learning in Histopathology

Background:

• Quantitative analysis of blood vessels in whole-slide images (WSIs) is crucial for disease understanding.
• Existing deep learning methods for vessel counting primarily use semantic segmentation, which may not be optimal for complex structures like lymph node high endothelial venules (HEVs).
• Systematic evaluation of vessel counting in lymph nodes, particularly HEVs, using artificial intelligence (AI) is lacking.

Purpose of the Study:

• To develop and evaluate an instance-level deep learning approach for detecting and counting individual CD34-positive vessels in lymph node WSIs.
• To address the gap in systematic evaluation of AI-based vessel quantification in lymph nodes.
• To provide a standardized method for quantifying lymph node vascular morphology.

Main Methods:

• Developed a convolutional neural network (CNN) based on the U-Net architecture for biomedical image segmentation.
• Trained the CNN on 4.0 million iterations to quantify CD34-positive vessels in lymph node WSIs.
• Employed a fixed post-processing procedure to extract individual vessel instances and applied an instance-level detection approach.

Main Results:

• The trained model, particularly at the 3.0 million iteration checkpoint, demonstrated the highest fidelity in capturing CD34-positive vascular morphology and morphological similarity in the validation set.
• The instance-level detection performance was robust, showing minimal variation across different lymph nodes and training checkpoints.
• The method achieved accurate vessel detection and counting in lymph node WSIs using a straightforward, fixed procedure.

Conclusions:

• The developed instance-level deep learning model provides accurate and robust vessel detection and counting in lymph node WSIs.
• This approach offers a practical foundation for standardized quantification of lymph node vasculature.
• The model facilitates the study of lymph node vascular morphology in normal, inflammatory, and lymphomatous conditions.