TUS-REC2024: A Challenge to Reconstruct 3D Freehand Ultrasound Without External Tracker
Qi Li, Shaheer U. Saeed, Yuliang Huang, Mingyuan Luo, Zhongnuo Yan, Jiongquan Chen, Xin Yang, Dong Ni, Nektarios Winter, Phuc Nguyen, Lucas Steinberger, Caelan Haney, Yuan Zhao, Mingjie Jiang, Bowen Ren, SiYeoul Lee, Seonho Kim, MinKyung Seo, MinWoo Kim, Yimeng Dou, Zhiwei Zhang, Yin Li, Tomy Varghe · 2025
Read TUS-REC2024: A Challenge to Reconstruct 3D Freehand Ultrasound Without External Tracker by Qi Li, Shaheer U. Saeed, Yuliang Huang, Mingyuan Luo, Zhongnuo Yan, Jiongquan Chen, Xin Yang, Dong Ni, Nektarios Winter, Phuc Nguyen, Lucas Steinberger, Caelan Haney, Yuan Zhao, Mingjie Jiang, Bowen Ren, SiYeoul Lee, Seonho Kim, MinKyung Seo, MinWoo Kim, Yimeng Dou, Zhiwei Zhang, Yin Li, Tomy Varghe (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-21765.txt.
Introduction
Qi Li and co-authors report a public contest to rebuild 3D ultrasound volumes from freehand 2D sweeps, with no tracking gear, in a 2025 paper. A tracker-free scan is cheap and portable, a boon for poor clinics, but guessing the probe's long path is hard. The TUS-REC2024 contest set the first shared benchmark for the task, with a large open dataset, a starter model, and strict scoring.
Forty-three teams signed up, and six filed twenty-one working entries. Their methods ranged widely: state-space and recurrent nets, registration polish, attention, and physics-guided designs. The report weighs each on shared metrics and maps where the field stands. Strengths and limits both show: fine for short sweeps, shaky on long, curved paths. All data and code sit open for reuse. The contest ran at a 2024 imaging conference and returns in 2025 as a living benchmark.
Catalog record
| CREATOR | Qi Li, Shaheer U. Saeed, Yuliang Huang, Mingyuan Luo, Zhongnuo Yan, Jiongquan Chen, Xin Yang, Dong Ni, Nektarios Winter, Phuc Nguyen, Lucas Steinberger, Caelan Haney, Yuan Zhao, Mingjie Jiang, Bowen Ren, SiYeoul Lee, Seonho Kim, MinKyung Seo, MinWoo Kim, Yimeng Dou, Zhiwei Zhang, Yin Li, Tomy Varghe |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 115 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-21765
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
TUS-REC2024: A Challenge to Reconstruct 3D Freehand Ultrasound Without External Tracker Qi Li, Shaheer U. Saeed, Yuliang Huang, Mingyuan Luo, Zhongnuo Yan, Jiongquan Chen, Xin Yang, Dong Ni, Nektarios Winter, Phuc Nguyen, Lucas Steinberger, Caelan Haney, Yuan Zhao, Mingjie Jiang, Bowen Ren, SiYeoul Lee, Seonho Kim, MinKyung Seo, MinWoo Kim, Yimeng Dou, Zhiwei Zhang, Yin Li, Tomy Varghese, Dean C. Barratt, Matthew J. Clarkson, Tom Vercauteren, Yipeng Hu Abstract Trackerless freehand ultrasound reconstruction aims to reconstruct 3D volumes from sequences of 2D ultrasound images without relying on external tracking systems, offering a low-cost, portable, and widely deployable alternative for volumetric imaging. However, it presents significant challenges, including accurate inter-frame motion estimation, minimisation of drift accumulation over long sequences, and generalisability across scanning protocols. The TUS-REC2024 Challenge was established to benchmark and accelerate progress in trackerless 3D ultrasound reconstruction by providing a publicly available dataset for the first time, along with a baseline model and evaluation framework. The Challenge attracted over 43 registered teams, of which 6 teams submitted 21 valid dockerized solutions. Submitted methods spanned a wide range of algorithmic approaches, including recurrent models, registration-driven volume refinement, attention, and physics-informed models. This paper presents an overview of the Challenge design, summarises the key characteristics of the dataset, provides a concise literature review, introduces the technical details of the underlying methodology working with tracked freehand ultrasound data, and offers a comparative analysis of submitted methods across multiple evaluation metrics. The results highlight both the progress and current limitations of state-of-the-art approaches in this domain, and inform directions for future research. The data, evaluation code, and baseline are publicly available to facilitate ongoing development and reproducibility. As a live and evolving benchmark, this Challenge is designed to be continuously developed and improved. The Challenge was held at MICCAI 2024 and will be organised again at MICCAI 2025, reflecting its growing impact and the sustained commitment to advancing this field. 1 Introduction Ultrasound imaging remains a cost-effective, non-invasive modality with real-time capabilities, making it a valuable tool across a wide range of clinical applications. Although the underlying data is inherently three-dimensional (3D), standard ultrasound typically captures single two-dimensional (2D) frame without spatial localisation across frames. This poses challenges for applications requiring accurate volumetric information, such as biometric quantification, image registration, and 3D visualisation. While expert clinicians can often infer 3D structure mentally or through standardised acquisition protocols (e.g., standard planes), the absence of inter-frame positional data limits reproducibility and the integration of ultrasound into advanced image analysis workflows. Ongoing work seeks to address this limitation by using 3D ultrasound probes to enable 3D reconstruction. 3D ultrasound probes is capable of acquiring volumetric data directly, using dedicated mechanical probe or 2D array transducers. While these probes provide valuable 3D imaging capabilities and offer flexible scanning trajectories, their higher cost and limited availability may restrict their use in some clinical settings, such as low-resource environments, point-of-care scenarios, or mobile and emergency units where portability and affordability are critical. A key advantage of freehand 2D ultrasound is its widespread availability and long-standing integration into clinical workflows. It has been used for decades across a broad range of applications, and clinicians are highly familiar with both its handling and interpretation. Building on this established foundation, tracker-based freehand ultrasound reconstruction techniques have been introduced to enable the generation of 3D anatomical representations. These methods aim to enhance conventional 2D ultrasound by incorporating spatial information from external tracking systems, such as optical or electromagnetic (EM) trackers. This enables conventional 2D ultrasound probes to be used for 3D imaging, providing a more flexible and accessible solution in clinical and research applications where dedicated and bulky 3D ultrasound systems may be impractical. However, optical and EM tracking systems present additional challenges in clinical environments. Optical tracking requires a continuous, unobstructed line of sight between the tracker and the camera, although some solutions, such as using multiple cameras, have been proposed to mitigate this limitation. EM tracking remains sensitive to nearby metal objects and electromagnetic interference, which can affect accuracy. Trackerless freehand ultrasound reconstruction refers to a class of techniques aimed at generating 3D volumetric representations from sequential 2D ultrasound frames, without the use of external tracking systems. Instead, these methods compute the relative spatial transformations among frames using internal data sources, such as image contents and signals from internal sensors. Common approaches include image-based registration, speckle decorrelation, inertial measurement unit (IMU) integration, or learning-based motion estimation such as convolutional neural networks or recurrent models. Additionally, trackerless freehand ultrasound reconstruction may further enhance existing 3D ultrasound systems, rather than serving solely as alternatives. However, trackless reconstruction remains challenging due to: 1) the difficulty of maintaining accuracy over long sequences of ultrasound frames, where small frame-to-frame errors can accumulate significantly; and 2) the high variability across different tasks and datasets, which complicates the validation and fair comparison of methods. While benchmarking is essential to address this variability, progress has been limited by the scarcity of publicly available datasets, which are critical for both performance evaluation and the development of learning-based approaches. This application involves practical challenges such as handling both 2D and 3D imaging data, incorporating tracking information, and managing multiple spatial coordinate systems, all of which contribute to a significant barrier for newcomers to the field and may impede broader progress and adoption. Furthermore, comparisons of methods in the existing literature are often conducted on relatively small, private datasets, using a variety of evaluation metrics to assess performance. This variability complicates the comparison of strengths and weaknesses across methods and may lead to biased conclusions, due to dataset characteristics, evaluation metric choices, and inherent differences in the methods’ underlying assumptions. For example, learning-based approaches may assume that training and testing data come from similar distributions, while classical methods may rely on consistently available speckle patterns. To address these gaps, we present a significant study of trackerless freehand ultrasound reconstruction, aiming to provide researchers with clear technical insights and consistent terminology. This effort is formalised in the form of the TUS-REC2024 Challenge, which is designed to foster both algorithmic innovation and practical clinical applicability by promoting reproducibility, benchmarking, and methodological transparency. TUS-REC2024 Challenge provides a large-scale in vivo ultrasound dataset, consisting of scans from both the left and right forearms of 85 volunteers (2,040 scans, 1,025,448 frames in total), acquired using a time-synchronised optical tracking system. We aim to conduct a comprehensive comparison among methods, evaluating their strengths and weaknesses on a common, large-scale dataset, using a consistent set of carefully-defined performance metrics. This approach will ensure a more objective and transparent assessment of the methods’ relative efficacy, and more importantly, to drive the development of new techniques for freehand ultrasound reconstruction. The significance of this Challenge lies in three key contributions. First, it establishes a rigorous and standardised benchmarking framework for trackerless freehand ultrasound reconstruction, advancing the development of novel algorithms and promoting objective performance evaluation through withheld test data and unified assessment metrics. Second, it provides the necessary infrastructure to support this benchmarking effort, including the public release of a comprehensive freehand ultrasound dataset, the largest publicly available dataset to date in the field, alongside detailed preliminary materials and accompanying code that describe the end-to-end pipeline for trackerless reconstruction. Third, beyond