Related Experiment Video
Updated: Sep 8, 2025
09:50
Author Spotlight: Aiding Research in Kidney Biology by Labeling Glomeruli in Cleared Tissues
Published on: February 9, 2024
1.4K
Data augmentation based on spatial deformations for histopathology: An evaluation in the context of glomeruli
Florian Allender1, Rémi Allègre1, Cédric Wemmert1
1ICube, Université de Strasbourg CNRS, France.
Computer Methods and Programs in Biomedicine
|June 14, 2022
Summary
Augmenting deep learning models with random spatial deformations significantly improves the segmentation of kidney structures in histopathology images. Stronger distortions yielded the best results, enhancing model performance for digital pathology applications.
Area of Science:
- Digital pathology
- Medical image analysis
- Deep learning in histopathology
Background:
- Deep learning for digital histopathology is limited by a scarcity of annotated images.
- Accurate segmentation of glomerular structures in renal histopathology slides is crucial.
Purpose of the Study:
- To evaluate the impact of data augmentation using random spatial deformations on U-Net model performance for supervised segmentation.
- To investigate the effect of varying distortion strengths in data augmentation.
Main Methods:
- Utilized a U-Net model for supervised segmentation of glomerular structures.
- Applied random spatial deformations to augment training data.
- Systematically evaluated the impact of different deformation strengths and parameters.
Main Results:
- Data augmentation with spatial deformations improved average Dice scores by up to 0.23.
- Stronger spatial distortions led to greater performance improvements compared to weaker ones.
- Developed a framework to assess the impact of spatial deformations on U-Net segmentation models.
Conclusions:
- This study is the first to assess random spatial deformations for histopathology image segmentation.
- The findings and framework offer practical guidance for optimizing deep learning models in digital pathology.
