Test-time augmentation for deep learning-based cell segmentation on microscopy images
importance 3/10
ai assisted
Cell Image Analysis Techniques microscopy image analysis stage: research
Problem
Deep learning-based cell segmentation on microscopy images requires massive annotated training sets to reach high accuracy; the paper addresses improving accuracy without more annotated data.
Why it matters
Accurate cell segmentation underpins single-cell analysis in biology and medicine.
What the AI did
Incorporated test-time augmentation into two segmentation approaches: U-Net semantic segmentation and Mask R-CNN instance segmentation, merging predictions from transformed image versions.
Results
not stated in source
Impact
Suggests a way to improve segmentation accuracy without collecting more annotated training data.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Nikita Moshkov, Botond Mathe, Attila Kertész‐Farkas, Réka Hollandi, Péter Horváth |
| Citations | 228 (OpenAlex, at import time) |
| References | 26 |
| Model or method | U-Net, Mask R-CNN |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-020-61808-3 |
| Open access | gold |
| Keywords | test-time augmentation, cell segmentation, U-Net, Mask R-CNN, microscopy |
Importance rating
3
incremental / niche
Improving cell segmentation accuracy without more annotated data aids single-cell analysis, but is an incremental methodological gain; no scale stated.