polymorph

Companion code for "Raman-Grounded Multimodal Sensing of CaCO₃ Polymorphs during Microfluidic Biomineralization" (Filanoski & Erickson, ACS Sensors, under review).

What's here

A single Colab notebook (polymorph_pipeline.ipynb) that walks through the full polymorph identification pipeline on four representative channels:

1. Load Raman ground-truth labels
2. Plot raw Raman spectra grouped by polymorph class
3. Overlay Raman point locations on the brightfield image
4. Color the points by Raman-derived polymorph label
5. Cellpose instance segmentation
6. Label propagation: Cellpose masks colored by Raman ground truth
7. EfficientNet-B0 polymorph classifier predictions
8. Classifier predictions vs. Raman ground truth

Running the notebook

Open polymorph_pipeline.ipynb in Google Colab. The notebook expects the companion dataset (Zenodo DOI 10.5281/zenodo.19868283), including the trained Cellpose and EfficientNet-B0 model weights. Edit PUB_DIR, MODEL_PATH, and CLASSIFIER_PATH in the relevant cells to point at your local copy.

GPU recommended; works on CPU with longer runtime.

Data

Available on Zenodo: https://doi.org/10.5281/zenodo.19868283

Includes:

• raman_data.parquet — 550 Raman-labeled training points across 5 sets
• Raw spectra (CSV + TXT) for the 4 example channels
• Brightfield + polarized-light PNGs for the 4 example channels
• Trained Cellpose model (cellpose_5set)
• Trained polymorph classifier (model5_3class_best.pth)

Citation

If you use this code or data, please cite the accompanying paper and the Zenodo deposit.

License

MIT (see LICENSE).