Skin OCT Surface Localization Dataset

Description

This dataset contains optical coherence tomography (OCT) B-scans acquired with an in-house robot-mounted OCT/OCE imaging system for skin-surface localization.

The dataset was used to evaluate DINOCT, a robust skin-surface localizer for artifact-prone OCT images affected by structured polarization-maintaining-fiber artifacts.

Data

Each raw OCT B-scan has size 512 × 500 pixels.

Visible-surface scans include a manually annotated skin-surface centerline, represented as one depth coordinate per image column.

The dataset also includes non-visible/background frames where no valid skin surface is visible.

Splits

Train, validation, and test splits are provided at the recording/acquisition level.

Subject identifiers and detailed demographic metadata were not consistently recorded, so the splits should not be interpreted as guaranteed subject-wise splits.

Intended Use

This dataset is intended for research on:

• OCT skin-surface localization
• OCT boundary detection
• artifact robustness
• robot-mounted OCT/OCE image guidance

Not Intended For

This dataset is not intended for:

• clinical diagnosis
• treatment planning
• autonomous medical decision-making
• demographic or subgroup analysis
• subject identification or re-identification

Human Subjects and Privacy

Human-subject imaging was approved by the University of Washington Institutional Review Board, and written informed consent was obtained from all participants.

Released files do not include direct subject identifiers.

Do not attempt to identify participants or link this dataset with external personal data.

Limitations

All data were acquired on a single in-house OCT/OCE platform. Subject metadata, anatomical site metadata, and formal inter-annotator variability measurements are not included. The real-artifact stress set should be interpreted as an artifact-stress evaluation set rather than a population-level out-of-distribution benchmark.

Citation

Please cite the associated paper if you use this dataset:

@article{dinoct_2026,
  title   = {DINOCT: robust skin-surface localization in OCT under structured PM-fiber artifacts},
  author  = {Bawornkitchaikul, Raveeroj and Pelivanov, Ivan and O'Donnell, Matt},
  year    = {2026}
}