The dataset was collected using the US-7a SPACE FAN phantom (Kyoto Kagaku, Japan) in conjunction with a Clarius C3 HD3 point-of-care ultrasound (POCUS) device (Clarius, Canada). This phantom simulates a pregnancy at 23 weeks of gestation. Videos were acquired with the POCUS device, capturing at a maximum depth of 16 cm and at 24 frames per second. To gather all the collected videos, a total of 45 volunteers were asked to perform four different sweeps: vertical, horizontal, and two diagonal trajectories. In total, 90 videos were acquired, resulting in 19407 frames, which will be detailed in the tables below. Videos Sweep # of videos Position # of videos Vertical 21 Occiput Posterior (OP) 34 Horizontal 27 Sacrum Posterior (SP) 26 Diagonal 1 22 Occiput Anterior (OA) 15 Diagonal 2 20 Sacrum Anterior (SA) 15 Frames Plane # of planes Plane # of planes Biparietal Plane (Value: 0) 42 Abdominal Plane (Value: 1) 63 Heart Plane (Value: 2) 61 Spine Plane (Value: 3) 134 Femur Plane (Value: 4) 46 No Plane (Value: 5) 19061 Image annotation was performed by a radiologist, an obstetrician, and pre-trained medical students using the Labelbox platform (Labelbox, USA). Labels were subject to a review process for annotations performed by medical students, where the radiologist validated them.

In low-income countries, particularly in remote communities with a shortage of trained sonographers and high maternal mortality rates, developing AI tools to assist non-experts in accurately identifying relevant fetal planes and potential anomalies during ultrasound exams is crucial. This dataset includes 19407 ultrasound frames collected from 90 videos of a 23-week gestational age fetal ultrasound phantom, recorded through free-hand sweeps by non-experts. These data are valuable for advancing research on second-trimester pregnancies. The frames were extracted from videos captured using a point-of-care ultrasound (POCUS) device in obstetric mode, set to a maximum depth of 16 cm. In total, 45 volunteers with no prior ultrasound experience recorded the videos while following four predefined scanning paths: vertical, horizontal, and two diagonal trajectories, with four different fetal poses. This approach creates a dataset that reflects real-world variability in non-expert settings, simulating ultrasound exams conducted by untrained personnel.

For more information: NatalIA-PBF-US1 This project was funded by CLIAS (Centro de Inteligencia Artificial y Salud para América Latina y el Caribe), an initiative of CIIPS (Centro de Implementación e Innovación de Políticas de Salud) at IECS (Instituto de Efectividad Clínica y Sanitaria), with support from IDRC (International Development Research Centre).

NatalIA PBF-US1 is dataset designed to support the development of AI-based tools for detecting relevant fetal planes in ultrasound videos captured by non-trained personnel, such as midwives or nurses.