BioHackathon 2026 — Filament Segmentation Pipeline

A pipeline for detecting and analyzing filaments (brighter spots) in microscopy images. Includes segmentation methods (Ilastik, MorphologicalSegmenter), evaluation against Label Studio annotations, and measurement of filament length and persistence.

---

Features

• Segmentation: Ilastik (probability maps), MorphologicalSegmenter (threshold + morphological ops), and classical methods (Otsu, Frangi, Watershed, etc.)
• Evaluation: IoU, Dice, and Boundary Dice vs ground truth from Label Studio polygon annotations
• Measurement: Filament length (ellipse major axis) and persistence time (longest consecutive detection run)
• WebServer: Streamlit app with Single Image, Batch Processing, and Stack Analysis modes

---

Project Structure

BioHackathon2026/
├── workflow/                    # Python package
│   ├── segmentation.py          # Otsu, MorphologicalSegmenter, Frangi, Watershed, etc.
│   ├── data_loader.py           # RawImageLoader
│   ├── segment_tracker.py       # SegmentTracker
│   ├── evaluation.py            # SegmentationEvaluator
│   └── experiments.py           # SampleCollector, visualizations
├── WebServer/                   # Streamlit app + Docker
│   ├── app.py                   # Main app (Ilastik, evaluation, measurement)
│   ├── Dockerfile
│   └── filament_model_v1.ilp    # Ilastik project (required for Docker)
├── filament_segmentation_pipeline.ipynb
├── filament_evaluation.ipynb
├── TestSetRawTIF/               # Test images + Label Studio export JSON
├── requirements.txt
├── pyproject.toml
└── README.md

---

Installation

Option 1: Install workflow package (for notebooks)

cd BioHackathon2026
pip install -e .

Option 2: Install from requirements.txt

pip install -r requirements.txt

Main dependencies:

• numpy, Pillow, matplotlib, scipy, scikit-image, pandas
• imageio[ffmpeg], imagecodecs
• tifffile (for TIFF I/O)
• torch, sam2 (optional, for SAM2-based segmentation in notebooks)
• streamlit (for WebServer)

---

Usage

Jupyter notebooks

1. filament_segmentation_pipeline.ipynb — Load data, run segmentation methods, visualize
2. filament_evaluation.ipynb — Compare Ilastik, MorphologicalSegmenter, and others vs Label Studio ground truth

Ensure workflow/data/ contains your multi-frame TIFFs (or adjust paths in the notebooks).

WebServer (Streamlit)

Locally (no Ilastik):

cd WebServer
pip install streamlit pillow tifffile numpy scikit-image matplotlib pandas
streamlit run app.py

With Docker (includes Ilastik):

cd WebServer
docker build -t filament-detector-webserver .
docker run -p 8501:8501 filament-detector-webserver

Open http://localhost:8501

WebServer modes

Mode	Description
Single Image	Preprocess TIFF → Ilastik → probability heatmap + overlay
Batch Processing	Process multiple TIFFs in parallel. Optional: upload Label Studio JSON for evaluation (IoU, Dice, Boundary Dice) of Ilastik and MorphologicalSegmenter
Stack Analysis	Upload multi-frame TIFF → filament length distribution + persistence (longest consecutive run of detections)

---

Evaluation

Ground truth is provided as a Label Studio export JSON with polygon annotations. Frame matching:

• JSON: URA7_URA8_002-crop6_frame_N.png → frame index N
• Uploaded files: frame_042.tif → frame index 42

Place your Label Studio export JSON (e.g. export_244107_project-244107-at-2026-03-21-11-44-ab41ab26.json) in TestSetRawTIF/ or upload it in the WebServer evaluation panel.

---

License

[Specify your license here, e.g. MIT, Apache 2.0]

---

Acknowledgments

Developed for BioHackathon 2026.