Oculo

Oculo is a publicly available multi-label ocular B-scan ultrasound dataset of 1,630 images from 1,242 patients, annotated for five ophthalmic abnormalities. It accompanies the MICCAI 2026 paper Oculo: A Multilabel Dataset for Identification of Ocular Abnormalities from Ultrasound Images.

Benchmark code: https://github.com/Sri-Kanchi-Kamakoti-Medical-Trust/OculoBenchmark

Contents

• images/ — 1,630 anonymized 900×900 PNGs named {image_id}.png
• data.csv — labels and the fixed train/val/test split (one row per image)

data.csv columns: image_id, split (train/val/test), the eight binary abnormality labels, onh_scan (optic-nerve-head view), and diagnosis (1 = any abnormality).

Label	Abbr.	Positives	Evaluated
vitreous_dot_echo	VDE	508	yes
membranous_echo	ME	339	yes
retinal_detachment	RD	186	yes
mass_lesion	ML	115	yes
abnormal_contour	AC	92	yes
posterior_vitreous_detachment	PVD	17	rare
choroidal_detachment	CD	29	rare
phthisis	PB	19	rare

Machine text overlays (identifiers, timestamps) were removed via inpainting, so the images contain no burned-in patient identifiers.

Citation

@inproceedings{oculo2026,
  title     = {Oculo: A Multilabel Dataset for Identification of Ocular Abnormalities from Ultrasound Images},
  booktitle = {Medical Image Computing and Computer Assisted Intervention (MICCAI)},
  year      = {2026}
}