Official implementation of CoFi, our paper on glomerular basement membrane (GBM) segmentation using a lightweight few-shot model and SAM2-based refinement.
🔬 Fast, accurate, and annotation-efficient segmentation for grayscale EM images.
Accurate segmentation of the glomerular basement membrane (GBM) in electron microscopy (EM) images is fundamental for quantifying membrane thickness and supporting the diagnosis of various kidney diseases. While supervised deep learning approaches achieve high segmentation accuracy, their reliance on extensive pixel-level annotation renders them impractical for clinical workflows. Few-shot learning can reduce this annotation burden but often struggles to capture the fine structural details necessary for GBM analysis. In this study, we introduce CoFi, a fast and efficient coarse-to- fine few-shot segmentation pipeline designed for GBM delineation in EM images. CoFi first trains a lightweight neural network using only three annotated images to produce an initial coarse segmentation mask. This mask is then automatically processed to generate high-quality point prompts with morphology-aware pruning, which are subsequently used to guide SAM in refining the segmentation. The proposed method achieved exceptional GBM segmentation performance, with a Dice coefficient of 0.745 and an inference speed of 1.9 FPS. We demonstrate that CoFi not only alleviates the annotation and computational burdens associated with conventional methods, but also achieves accurate and reliable segmentation results. The pipeline’s speed and annotation efficiency make it well-suited for both research and clinical applications in renal pathology
CoFi delivers:
- Dice: 74.54%
- IoU: 61.87%
- Accuracy: 94.26%
- Speed: 1.9 FPS
📄 [Paper on arXiv (coming soon)]
- Fast few-shot training: Trains on just 3 support images.
- Coarse-to-fine segmentation: Uses DeepLabV3 for initial prediction.
- Anatomy-aware prompts: Automatically extracts spatially diverse points based on GBM structure.
- SAM2 refinement: Refines masks with Segment Anything Model 2 for subpixel accuracy.
- Supports grayscale EM images.
git clone https://github.com/ddrrnn123/CoFi.git
cd CoFi
pip install -r requirements.txt