Tumor Detection, Segmentation and Classification Challenge on Automated 3D Breast Ultrasound: The TDSC-ABUS Challenge

Gongning Luo, Mingwang Xu, Hongyu Chen, Xinjie Liang, Xing Tao, Dong Ni, Hyunsu Jeong, Chulhong Kim, Raphael Stock, Michael Baumgartner, Yannick Kirchhoff, Maximilian Rokuss, Klaus Maier-Hein, Zhikai Yang, Tianyu Fan, Nicolas Boutry, Dmitry Tereshchenko, Arthur Moine, Maximilien Charmetant, Jan Saue · 2025

Read Tumor Detection, Segmentation and Classification Challenge on Automated 3D Breast Ultrasound: The TDSC-ABUS Challenge by Gongning Luo, Mingwang Xu, Hongyu Chen, Xinjie Liang, Xing Tao, Dong Ni, Hyunsu Jeong, Chulhong Kim, Raphael Stock, Michael Baumgartner, Yannick Kirchhoff, Maximilian Rokuss, Klaus Maier-Hein, Zhikai Yang, Tianyu Fan, Nicolas Boutry, Dmitry Tereshchenko, Arthur Moine, Maximilien Charmetant, Jan Saue (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-2501-15588.txt.

Introduction

Gongning Luo and a very large team of co-authors run a medical imaging challenge in twenty twenty-five. Breast cancer is a leading cause of death for women worldwide. Early detection cuts those deaths. Automated 3D breast ultrasound is a newer screening route, with gains in safety, speed, and detection over handheld methods.

The challenge, TDSC-ABUS, asks algorithms to do three jobs on such scans. They must detect tumors, segment their boundaries, and classify them. Dozens of teams took part under the challenge banner. Shared data and scoring let methods be compared fairly. Readers get a snapshot of how well machines now handle a full screening pipeline. The work aims to speed tools that catch cancer while it is still curable.

Catalog record

CREATORGongning Luo, Mingwang Xu, Hongyu Chen, Xinjie Liang, Xing Tao, Dong Ni, Hyunsu Jeong, Chulhong Kim, Raphael Stock, Michael Baumgartner, Yannick Kirchhoff, Maximilian Rokuss, Klaus Maier-Hein, Zhikai Yang, Tianyu Fan, Nicolas Boutry, Dmitry Tereshchenko, Arthur Moine, Maximilien Charmetant, Jan Saue
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-15588

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Tumor Detection, Segmentation and Classification Challenge on Automated 3D Breast Ultrasound: The TDSC-ABUS Challenge Gongning Luo, Mingwang Xu, Hongyu Chen, Xinjie Liang, Xing Tao, Dong Ni, Hyunsu Jeong, Chulhong Kim, Raphael Stock, Michael Baumgartner, Yannick Kirchhoff, Maximilian Rokuss, Klaus Maier-Hein, Zhikai Yang, Tianyu Fan, Nicolas Boutry, Dmitry Tereshchenko, Arthur Moine, Maximilien Charmetant, Jan Sauer, Hao Du, Xiang-Hui Bai, Vipul Pai Raikar, Ricardo Montoya-del-Angel, Robert Martí, Miguel Luna, Dongmin Lee, Abdul Qayyum, Moona Mazher, Qihui Guo, Changyan Wang, Navchetan Awasthi, Qiaochu Zhao, Wei Wang, Kuanquan Wang, Qiucheng Wang, Suyu Dong Abstract Breast cancer is one of the most common causes of death among women worldwide. Early detection helps in reducing the number of deaths. Automated 3D Breast Ultrasound (ABUS) is a newer approach for breast screening, which has many advantages over handheld mammography such as safety, speed, and higher detection rate of breast cancer. Tumor detection, segmentation, and classification are key components in the analysis of medical images, especially challenging in the context of 3D ABUS due to the significant variability in tumor size and shape, unclear tumor boundaries, and a low signal-to-noise ratio. The lack of publicly accessible, well-labeled ABUS datasets further hinders the advancement of systems for breast tumor analysis. Addressing this gap, we have organized the inaugural T umor D etection, S egmentation, and C lassification Challenge on A utomated 3D B reast U ltra s ound 2023 (TDSC-ABUS2023). This initiative aims to spearhead research in this field and create a definitive benchmark for tasks associated with 3D ABUS image analysis. In this paper, we summarize the top-performing algorithms from the challenge and provide critical analysis for ABUS image examination. We offer the TDSC-ABUS challenge as an open-access platform at https://tdsc-abus2023.grand-challenge.org/ to benchmark and inspire future developments in algorithmic research. 1 Introduction Breast cancer, the most commonly diagnosed cancer worldwide, is also the fifth leading cause of death globally [ 38 ] . The past decade has seen significant strides in reducing mortality and improving the 5-year survival rate, largely thanks to advancements in early detection and diagnostics [ 23 ] . However, accurately diagnosing malignant breast tumors at an early stage remains a paramount yet challenging goal.- Central to breast cancer screening and diagnosis is the use of ultrasound technology. Traditional 2D hand-held ultrasound systems (HHUS), despite their widespread use, are hampered by operator dependency, time-consuming processes, and their limitation to two-dimensional imaging [ 5 ] . The Automated Breast Ultrasound System (ABUS) represents a groundbreaking advancement in this field. It not only standardizes the scanning process but also minimizes the influence of the operator on the quality of the images. By providing a detailed 3D representation of breast tissue, ABUS enables multi-angle evaluations and enriches the potential for retrospective analyses, thereby outperforming traditional ultrasound in tumor detection [ 22 , 45 , 41 ] . Nonetheless, the intricate interpretation of ABUS images requires extensive clinical experience, and current techniques have yet to fully harness the wealth of data offered by 3D imaging. Thus, there is a pressing need for the development of more advanced and efficient computer-aided diagnosis (CAD) algorithms for use in clinical settings. In the realm of CAD, tumor detection, segmentation, and classification are three fundamental tasks, each serving as a critical preliminary step for further analysis. Numerous studies have been conducted to address these challenges across various medical contexts [ 17 , 20 , 1 , 2 , 9 , 8 , 11 , 43 ] . Focusing on ABUS images, significant progress has been made. For instance, in tumor detection, [ 32 ] proposed a novel approach using a GoogLeNet-based fully convolutional network, leveraging both axial and reconstructed sagittal slices [ 39 ] . This method, while effective, highlighted the need for a more comprehensive approach to spatial information integration. In response, [ 40 ] developed a 3D U-net-based network that directly processes 3D ABUS images, thus overcoming the limitations of previous methods and enhancing diagnostic accuracy. The segmentation of tumors in ABUS images is particularly challenging due to indistinct lesion boundaries. Addressing this, [ 7 ] innovated a dual-path U-net architecture, achieving promising results. Moreover, [ 46 ] further advanced segmentation techniques by devising a cross-model attention-guided network, incorporating an improved 3D Mask R-CNN head into a V-Net framework. Their methodology showed marked improvements over existing segmentation approaches. Finally, in the field of tumor classification, [ 48 ] introduced a shallowly dilated convolutional branch network (SDCB-Net), specifically designed for classifying breast tumors in ABUS images. Additionally, [ 24 ] proposed an innovative branch network that integrates segmenting mask information into the training regime, noting that the performance varied depending on the positioning of the mask branch network in relation to mass and cancer classifications. Figure 1 : Representative ABUS Image Cases. (a) A small-sized tumor exhibiting a jagged boundary. (b) A comparatively larger tumor characterized by a smooth periphery. Despite the advancements in 3D Automated Breast Ultrasound (ABUS) imaging, developing efficient and accurate CAD systems remains a significant challenge. As shown in Fig. 1 , although ABUS addresses some limitations of 2D ultrasound, it still presents distinct challenges for CAD systems: (1) Unclear Boundaries: The inherent characteristics of ultrasound imaging often result in blurred lesion boundaries, complicating the tasks of annotation and segmentation. (2) High Similarity Between Lesion and Background: Lesions may exhibit features akin to the surrounding tissue, leading to potential misclassification as tumors or, conversely, missed detections. (3) Wide Variation in Lesion Size and Shape: The size of lesions can vary considerably, posing challenges for algorithms to effectively adapt to extreme cases. Additionally, the shape of tumors can range from smooth to rough textures, making it difficult to differentiate based on these characteristics alone. (4) Small Proportion of Lesion in the Total Image: Lesions often occupy a minor fraction of the overall image area, which can make detection challenging due to the predominance of non-lesion tissue. (5) Unpredictable Lesion Location: The location of lesions within the image can be highly variable, adding complexity to the detection process as there is no standard region to focus the analysis. Over recent decades, despite extensive research into Automated Breast Ultrasound (ABUS) imaging, there remains a notable absence of a publicly available ABUS dataset that serves as a standard benchmark for the fair evaluation of algorithms. To bridge this gap, we have initiated the T umor D etection, S egmentation, and C lassification Challenge on A utomated 3D B reast U ltra s ound 2023 (TDSC-ABUS2023). This challenge was held as part of the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) 2023, in Turkey. The competition invited participants to develop algorithms that address detection, segmentation, or classification of tumors in ABUS images. Entrants had the flexibility to tackle any combination of these tasks, with each task being independently evaluated using specific metrics. In order to promote a comprehensive approach, we also introduced an overall leaderboard encompassing all three tasks. This paper presents a comprehensive overview of the TDSC-ABUS 2023 challenge, highlighting the most effective algorithms. Our primary contributions can be encapsulated as follows: 1. We have established the inaugural TDSC-ABUS challenge, the first of its kind to focus on the trifecta of detection, segmentation, and classification tasks in ABUS

Read the complete text