A lightweight, human-in-the-loop microscopy image analysis app.
Mycol is a Streamlit-based application that makes machine-learning-assisted microscopy analysis accessible to non-specialists. It enables fast annotation, automated segmentation and classification, model fine-tuning, and quantitative phenotyping, all on a standard laptop and without coding.
- Upload images and optional masks
- Manual mask drawing and editing
- SAM2-guided segmentation
- Automated Cellpose segmentation (single or batch mode)
- Interactive classification (manual or DenseNet-based)
- Train Cellpose (segmentation) and DenseNet (classification) models directly in the app
- Default training settings for general use
- Diagnostic outputs:
- Loss curves
- IoU scores
- True vs. predicted counts
- Accuracy, precision, F1, confusion matrix
- Download trained models and training summaries
- Automatic computation of cell descriptors (size, shape, elongation, compactness, etc.)
- Visual comparison of phenotypic classes
- Export plots and tabulated descriptors
- Built-in explanations for descriptor interpretation
- Runs locally on standard hardware
- Minimal dependencies
- Designed for small-scale workflows
conda create -n mycol_env
conda activate mycol_env
git clone https://github.com/<your-username>/mycol.git
cd mycol
pip install -r requirements.txtstreamlit run app.py- Rapid cell counting
- Creating curated datasets of annotated images
- Automating image annotation (with human QC)
- Morphology-based phenotypic comparison
MIT