Maternal-Fetal Ultrasouno Video Dataset for End-to-end Intrapartum Biometry and Multi-task Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Obstetrics

Background:

• Intrapartum biometry is crucial for monitoring labor progress.
• Existing ultrasound datasets lack video data for AI-based labor assessment.
• A public, multi-category labeled intrapartum ultrasound video dataset is needed.

Purpose of the Study:

• To release the first multi-center, multi-device, multi-category labeled intrapartum ultrasound video dataset.
• To facilitate research on AI-based end-to-end intrapartum biometry and labor progress assessment.
• To bridge the gap in public video datasets for fine-grained classification in obstetrics.

Main Methods:

• Collected 774 intrapartum ultrasound videos from multiple centers and devices.
• Annotated videos with standard plane classification, segmentation of pubic symphysis and fetal head, and labor progress parameters.
• Created a comprehensive dataset for multi-task learning and automated obstetric process development.

Main Results:

• Publicly released a novel dataset of 774 videos (68,106 images).
• Dataset includes multi-category labels for classification, segmentation, and labor parameters.
• First dedicated video benchmark for intrapartum biometry and labor progress assessment.

Conclusions:

• The released dataset supports research in multi-task learning for automated obstetric processes.
• Enables development of end-to-end AI solutions for intrapartum biometry and labor monitoring.
• Facilitates auxiliary decision-making in obstetrics through advanced AI.