Fetal Head Ultrasound Images with Multi-Class Polygon Segmentation Masks

Nizar, M. Fazri provides 3,832 fetal head ultrasound images with YOLO-format polygon segmentation labels for three anatomical structures: Brain, Cavum Septi Pellucidi, and Lateral Ventricles. The dataset includes an inter-patient stratified split of 70/15/15 for training, validation, and testing, totaling 7,664 files. It was published on the Harvard Dataverse platform in April 2026.

Use Cases

• Train a YOLO-based model to segment the Brain, Cavum Septi Pellucidi (CSP), and Lateral Ventricles (LV) structures from 959×661 px B-mode ultrasound frames.
• Benchmark segmentation performance using the provided inter-patient split to prevent data leakage, as defined in the split_info.json file.
• Develop multi-class instance segmentation pipelines using the normalized polygon vertices (x1 y1 x2 y2 … xN yN) stored in the YOLO label files.
• Reproduce results from the associated research paper using the seeded (seed=42) train/val/test partitions for full reproducibility.
• Validate segmentation model generalizability across distinct patient data, as all frames from a single patient are confined to one partition.

Strengths

• Contains 3,832 ultrasound images with 6,551 total annotated instances across three anatomical classes.
• Labels are physician-verified, with an inter-rater reliability (ICC) reported as ≥ 0.859.
• Provides a reproducible, patient-level data split (2,654 train, 603 val, 575 test images) to prevent leakage.
• Includes a dataset.yaml configuration file and split_info.json for immediate use with YOLO frameworks.

Limitations

• Limited to fetal head anatomy; does not include other fetal structures or maternal health data.
• Image resolution is fixed at 959×661 pixels, which may not be suitable for all analysis scales.
• Temporal and geographic coverage of the original source images is not specified, potentially limiting demographic generalizability.

Provenance

Source

HC18 Grand Challenge (van den Heuvel et al., 2018) and a large-scale fetal head biometry annotation dataset (Alzubaidi et al., 2023).

Collection Method

B-mode ultrasound frames were annotated with polygon masks by a senior attending physician and a radiologic technologist.

Freshness

Last updated April 2026.