Dataset of Ultrasound Video Frames from Gelatin Phantoms

This dataset contains ultrasound video frames acquired from gelatin-based breast phantoms with embedded hyperechoic and anechoic lesions (4\u201320 mm). The data were collected to support the development and evaluation of artificial intelligence (AI) models for lesion detection, classification, and real-time deployment on portable ultrasound devices. The dataset includes: (1) raw ultrasound videos, (2) extracted video frames, (3) bounding-box labels identifying lesion regions, and (4) pretrained model weights for reference AI architectures. The dataset was designed to facilitate research in lightweight, edge-AI deployment, predictive modeling, and point-of-care ultrasound (POCUS) applications in low-resource settings. By providing both labeled images and video data, this resource enables training, benchmarking, and reproducibility of lesion detection pipelines. All files are organized into separate folders for frames, labels, videos, and weights for ease of use.