In Vivo Raman Skin-Cancer Seed Spectra and Physically Constrained Augmented Cohort (2026)
Description
Open Raman spectral dataset released with the paper "Physically Constrained Data Augmentation and Interpretable Machine Learning for Class-Imbalanced Raman Diagnosis of Skin Cancer" (ISDS-2026, paper 6661; Can Tho University Journal of Innovation for Sustainable Development, CTUJoISD). The release contains (i) the 54 real in vivo skin-cancer seed spectra (18 basal-cell carcinoma, 17 melanoma, 19 benign; 284.9-1801.6 cm-1) curated from the public Samara University dataset, and (ii) the class-balanced 1,200-spectrum cohort (375 BCC, 375 SCC, 225 melanoma, 225 benign; fingerprint region 800-1800 cm-1 on a 2 cm-1 grid) generated from the seeds by the physically constrained augmentation pipeline described in the paper. Every cohort record carries an origin flag: real (54), synthetic-real-anchored (771), or synthetic-band-model (375). Because no public in vivo squamous-cell-carcinoma (SCC) spectra exist, the SCC class is generated entirely from a literature-derived band model; cross-validated metrics on the cohort (overall accuracy 94.8%, balanced accuracy 93.8% with the interpretable PCA-30 + logistic-regression model) are therefore an upper-bound characterisation of the pipeline. The conservative anchor is the train-on-synthetic/test-on-real evaluation on the 54 real spectra: 87.0% overall accuracy. Binary malignant-vs-benign screening on the cohort reaches sensitivity 1.000 (975/975) and specificity 0.978 (220/225).
Acquisition parameters
Acquisition conditions are recorded in full so you can judge whether the dataset fits your purpose. An empty field means the information has not been recorded, not that it does not apply.
| Parameter | Value |
|---|---|
| Instrument | Portable Raman + autofluorescence spectroscopic system, Samara National Research University (Khristoforova et al., J. Biophotonics 12, e201800400, 2019) |
| Laser wavelength (nm) | 785 |
| Power at sample (mW) | — |
| Wavenumber range (cm⁻¹) | 284.9–1801.6 cm⁻¹ (seeds); 800–1800 cm⁻¹ (cohort) |
| Spectral resolution (cm⁻¹) | — |
| Objective / numerical aperture | — |
| Sample preparation | None — in vivo measurement on unaltered skin lesions |
| Wavenumber calibration standard | — |
| Most recent calibration date | — |
| Number of samples | 54 in vivo spectra (18 BCC, 17 melanoma, 19 benign) |
If you need a parameter that is not listed here, contact us — we keep the raw record of every measurement.
How to cite
Please cite this dataset exactly as shown below in any publication that uses it. Full citation lets the community trace data provenance and is a condition of the licence.
Plain-text citation
Hoang, C.D., Tung, N.T., Son, L.H., Le, C.H. & BIRALAB (2026). In vivo Raman skin-cancer seed spectra and physically constrained augmented cohort (version 1.0) [dataset]. BIRALAB Data Hub, International School, Vietnam National University, Hanoi. https://biralab.org/en/data-hub/datasets/raman-skin-cancer-augmentation-2026
BibTeX
@misc{biralab2026ramanskincanceraugmentat,
title = {In Vivo Raman Skin-Cancer Seed Spectra and Physically Constrained Augmented Cohort (2026)},
author = {{BIRALAB}},
year = {2026},
howpublished = {BIRALAB Data Hub},
version = {1.0},
url = {https://biralab.org/en/data-hub/datasets/raman-skin-cancer-augmentation-2026},
note = {dataset}
}DOI
No DOI yet
Download the dataset
Downloads: No downloads yet
By downloading you agree to comply with the dataset licence and to cite it as described above.
Data dictionary
The meaning of every field in the data files, with its type and unit.
| Field name | Type | Unit | Description |
|---|---|---|---|
id | integer | — | Record identifier in the original release |
diagnosis | string | — | Clinical diagnosis as released |
type | string | — | Lesion type (contains class label text) |
location | string | — | Anatomical site |
gender | string | — | Patient gender |
label | string | — | Harmonised class: BCC / Melanoma / Benign |
284.893653497196 … 1801.6 (504 columns) | float | a.u. | Raw Raman intensity at each wavenumber (cm⁻¹), native instrument grid |
label | string | — | Class: BCC / SCC / Melanoma / Benign |
origin | string | — | real (54) / synthetic-real-anchored (771) / synthetic-band-model (375) |
800.0 … 1800.0 (501 columns) | float | SNV units | Preprocessed intensity on the 2 cm⁻¹ fingerprint grid |
Preprocessing applied
This dataset was processed with the BIRALAB preprocessing standard: cosmic ray removal, wavenumber-axis calibration against a reference material, baseline correction, Savitzky–Golay smoothing, intensity normalisation and resampling onto a common wavenumber grid. The exact parameters of each step are recorded in the metadata shipped with the download.
Publications using this dataset
BIRALAB publications that reference this dataset or its DOI.
No publication in our records references this dataset yet. If you have used it in your own research, send us the reference and we will add it to this list.