FOCUS-7D: A Large-Scale, Multi-Organ Ultrasound Dataset for Multi-Task AI Model Development

The previously non-public component of FOCUS-7D contains 6,432 de-identified ultrasound images and their corresponding annotations and is divided into two subsets, Private and Private_challenge, according to their different data sources. The Private subset comprises 2,192 cardiac, fetal-head, and kidney ultrasound images collected from participating medical institutions and is primarily intended for segmentation tasks. The Private_challenge subset comprises 4,240 images from separate challenge-project data sources, covering the appendix, breast, heart, fetal head, kidney, liver, and thyroid, and supports both segmentation and classification tasks. All newly generated annotations underwent multi-tier review and quality control. This dataset can be used for multi-organ ultrasound image segmentation, classification, multi-task learning, and the development and validation of general-purpose ultrasound AI models.