Universal Ultrasound Image & Video Analysis Challenge: Multi-Organ Classification and Segmentation Across B-mode and Contrast-Enhanced Ultrasound

Ultrasound is a cornerstone of modern clinical imaging and is widely used for screening, diagnosis, and treatment guidance across a wide range of organs and diseases. In routine clinical practice, clinicians rely not only on static ultrasound images but also on ultrasound videos and contrast-enhanced ultrasound (CEUS) to assess anatomical
structures, lesion boundaries, and tissue perfusion patterns. These complementary imaging forms provide critical information for disease characterization, staging, and follow-up. However, the interpretation of multi-organ, multi-modality ultrasound data remains highly dependent on operator experience and is subject to considerable
inter-observer variability.

From a technical perspective, building unified ultrasound models that generalize across both classification and segmentation tasks while jointly capturing spatial structures, temporal dynamics, and contrast enhancement patterns remains an open challenge. Existing approaches are often tailored for specific organs or single data modalities, which limits their scalability and robustness in real-world clinical scenarios. This challenge is therefore designed as a comprehensive benchmark for advancing universal ultrasound modeling, to encourage methods that learn consistent and transferable representations across heterogeneous ultrasound data.

Compared with UUSIC 2025, this benchmark additionally incorporates B-mode ultrasound videos and CEUS videos, substantially expanding the data modalities and bringing the benchmark closer to real-world clinical workflows. The benchmark datasets consist of a collection of public datasets together with an additional dataset bundle contributed by partner hospitals. The public dataset includes multi-organ B-mode ultrasound images and cardiac B-mode ultrasound videos. The partner-hospital datasets further provide multi-organ B-mode ultrasound images, CEUS videos covering multiple organs, and cardiac B-mode ultrasound videos. The B-mode ultrasound
image dataset covers 8 organs, including breast, thyroid, liver, kidney, fetal head, heart, appendix, and prostate, and supports both classification and segmentation tasks depending on the organ. The CEUS dataset includes 4 organs, breast, thyroid, liver, and prostate, and is designed for both classification and segmentation tasks. In addition, a cardiac B-mode ultrasound video dataset is provided for the cardiac structural segmentation task. In total, the public datasets contain 6,740 ultrasound images and 500 cardiac video samples, while the partner-hospital datasets include 5,262 ultrasound images, approximately 800 CEUS samples, and more than 200 cardiac video samples. Leveraging these data, we construct around 15 distinct evaluation tasks, covering multi-organ classification and segmentation under both image and video settings. For evaluation, the area under the receiver operating characteristic curve (AUC) is expected to be adopted for classification tasks, while the Dice similarity coefficient (Dice) is expected to be used for segmentation tasks.