OpenUS: A Fully Open-Source Foundation Model for Ultrasound Image Analysis via Self-Adaptive Masked Contrastive Learning
Xiaoyu Zheng, Xu Chen, Awais Rauf, Qifan Fu, Benedetta Monosi, Felice Rivellese, Myles J. Lewis, Shaogang Gong, Gregory Slabaugh
2025-11-18
Summary
This paper introduces OpenUS, a new, openly available artificial intelligence model designed to interpret ultrasound images. It's built to be a strong starting point for other researchers and doctors who want to use AI to analyze ultrasound scans.
What's the problem?
Ultrasound is a really useful medical imaging tool because it's cheap, safe, and shows images in real-time, but reading those images is tricky and depends a lot on the skill of the person doing it. Plus, ultrasound images can be grainy and hard to standardize, making it difficult to create AI that works well across different situations and with limited labeled data. Existing AI models struggle to generalize because of these variations and lack of good training data.
What's the solution?
The researchers created OpenUS by training an AI model on a huge collection of over 308,000 ultrasound images from many different sources. They used a special type of AI architecture called a 'vision Mamba' which is good at understanding both small details and big picture patterns in images. They also developed a clever training method that focuses the AI's attention on the most important parts of the image, and gradually increases the difficulty of the training process. This helps the AI learn more effectively.
Why it matters?
OpenUS is important because it provides a freely available, powerful AI model that can be adapted to many different ultrasound-related tasks. Because it's a 'foundation model,' it doesn't need a ton of new labeled data to be useful for specific problems, which saves time and resources. This could lead to more accurate and consistent ultrasound diagnoses, ultimately improving patient care.
Abstract
Ultrasound (US) is one of the most widely used medical imaging modalities, thanks to its low cost, portability, real-time feedback, and absence of ionizing radiation. However, US image interpretation remains highly operator-dependent and varies significantly across anatomical regions, acquisition protocols, and device types. These variations, along with unique challenges such as speckle, low contrast, and limited standardized annotations, hinder the development of generalizable, label-efficient ultrasound AI models. In this paper, we propose OpenUS, the first reproducible, open-source ultrasound foundation model built on a large collection of public data. OpenUS employs a vision Mamba backbone, capturing both local and global long-range dependencies across the image. To extract rich features during pre-training, we introduce a novel self-adaptive masking framework that combines contrastive learning with masked image modeling. This strategy integrates the teacher's attention map with student reconstruction loss, adaptively refining clinically-relevant masking to enhance pre-training effectiveness. OpenUS also applies a dynamic learning schedule to progressively adjust the difficulty of the pre-training process. To develop the foundation model, we compile the largest to-date public ultrasound dataset comprising over 308K images from 42 publicly available datasets, covering diverse anatomical regions, institutions, imaging devices, and disease types. Our pre-trained OpenUS model can be easily adapted to specific downstream tasks by serving as a backbone for label-efficient fine-tuning. Code is available at https://github.com/XZheng0427/OpenUS.