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

Generalizability remains a significant challenge for artificial intelligence (AI) models in ultrasound imaging, primarily due to the scarcity of large-scale, diverse, and comprehensively annotated multi-organ data. To address this gap, we present FOCUS-7D (FOundational Cross-organ Ultrasound Set for 7 Diseases), a large-scale, multi-organ ultrasound dataset designed to foster the development of foundational AI models. This resource integrates 13,920 images—6,432 prospectively collected from four tertiary hospitals and 7,488 curated from public sources—across seven critical anatomical regions: thyroid, breast, liver, kidney, fetal head, heart, and appendix. The dataset supports both segmentation and classification tasks. All annotations underwent a rigorous multi-tier validation process, beginning with initial segmentation by junior annotators, followed by refinement under senior sonographer supervision, and concluding with final verification by physicians with over 10 years of experience. To demonstrate its utility, we established baseline benchmarks. Models trained on FOCUS-7D achieved high performance, with a notable mean Dice coefficient for segmentation and strong accuracy for classification, confirming the dataset’s quality and the robustness of the established benchmarks. By harmonizing diverse and clinically representative data, FOCUS-7D bridges the gap between AI research and real-world application, enabling the development of generalizable ultrasound AI models for enhanced diagnostic workflows.