Glomerular Basement Membrane Thickness Estimation and Stratification via Active Semi-Supervised Learning Model.

Area of Science:

• Nephrology
• Medical Imaging
• Computational Pathology

Background:

• Glomerular basement membrane (GBM) thickness is crucial for diagnosing kidney glomerular diseases.
• Manual GBM thickness measurement from transmission electron microscopy (TEM) images is subjective and time-consuming.

Purpose of the Study:

• To develop a fully automated pipeline for GBM segmentation and thickness estimation.
• To improve the accuracy and reproducibility of GBM thickness measurements.

Main Methods:

• Utilized a convolutional neural network with active semi-supervised learning for GBM segmentation.
• Employed computer vision techniques and pixel distance matrices for thickness estimation.
• Trained a machine learning model for automated GBM thickness classification.

Main Results:

• Achieved a high correlation (Pearson's R2 = 0.85) between automated and manual GBM thickness measurements.
• Demonstrated robust GBM segmentation across various image magnifications and complexities.
• Classified GBM thickness into normal, thin, and thick categories with 0.76 accuracy.

Conclusions:

• The automated pipeline achieves state-of-the-art performance in GBM segmentation and thickness estimation.
• This technology can assist clinicians by speeding up routine diagnostic procedures with high accuracy.
• The method offers a robust and reproducible solution for assessing GBM thickness in clinical practice.