Dataset for: Raman-grounded multimodal sensing of CaCO₃ polymorphs during microfluidic biomineralization

This deposit contains paired Raman spectra and brightfield/polarized-light micrographs of calcium carbonate crystals grown across five microfluidic biomineralization experiments. Polymorph labels (calcite, early-stage vaterite, late-stage vaterite) are derived from Raman ground-truth measurements at known crystal locations.

REPRODUCIBILITY

Companion code: https://github.com/d-ericksonlab/polymorphThe notebook reproduces the segmentation and classification pipeline and regenerates the manuscript figures from the data in this deposit.

CONTENTS

- publication_data/raman_data.parquet — 550 Raman-labeled crystal instances across 5 experimental sets, with polymorph class, image-pixel coordinates, and references to the corresponding microscopy images. Embedded raw spectra (wavenumber and intensity arrays) are included for four representative channels spanning all three polymorph classes.- publication_data/images/ — Brightfield (*_ra.png) and polarized-light (*_pol.png) micrographs for the four representative channels.- publication_data/raman_metadata/ — WiTec instrument metadata (.txt) for the included spectra.- models/cellpose_5set — Trained Cellpose instance segmentation model.- models/model5_3class_best.pth — Trained 5-channel EfficientNet-B0 polymorph classifier.

METHODS SUMMARY

Raman labels were extracted from spectra acquired at known stage coordinates on a WiTec confocal Raman microscope and quality-checked. Stage coordinates were aligned to the brightfield image coordinate system. Crystal instance segmentation was performed with Cellpose, retrained on this dataset. Polymorph classification used a 5-channel EfficientNet-B0 (brightfield + binary instance mask + polarized R/G/B) trained with stratified 60/20/20 splits.

LICENSE

Data: CC-BY 4.0. Code (linked GitHub): MIT.

CITATION

If you use this dataset, please cite the accompanying paper and this Zenodo deposit (DOI: 10.5281/zenodo.19868282).