CHAOS (CHAOS - Combined (CT-MR) Healthy Abdominal Organ Segmentation)

CHAOS 挑战旨在从 CT 和 MRI 数据中分割腹部器官（肝脏、肾脏和脾脏）。 CHAOS 现场部分于 2019 年 4 月 11 日在意大利威尼斯举行的 IEEE 国际生物医学成像研讨会 (ISBI) 上举行。仍然欢迎在线提交！ \textbf{挑战说明} 了解复杂医疗程序的先决条件对于手术的成功起着重要作用。为了丰富理解水平，医生使用 3D 可视化和打印等高级工具，这需要从 DICOM 图像中提取感兴趣的对象。因此，腹部器官（即肝脏、肾脏和脾脏）的精确分割对于包括但不限于活体供体移植手术的肝脏预评估或腹部器官的详细分析等临床程序至关重要。在腹主动脉手术前确定从它们产生和进入的血管，以便正确定位移植物。这激发了正在进行的研究以实现更好的分割结果并克服源自腹部高度灵活的解剖特性和反映到图像特征的模态限制的无数挑战。在这种情况下，提出的挑战有两个独立但相关的目标：1) 从计算机断层扫描 (CT) 数据集分割肝脏，这些数据集是在注射造影剂后在门静脉期获得的，用于对活体肝移植供体进行预评估。 2) 从用两个不同序列(T1-DUAL 和T2-SPIR) 采集的磁共振成像(MRI) 数据集分割四个腹部器官(即肝脏、脾脏、右肾和左肾)。 CHAOS 任务包含这些器官分割的组合。 \textbf{Tasks} 参赛队伍可以参加五个比赛类别并提交他们的结果： 1) 肝脏分割（CT 和 MRI）：这也称为“交叉模式”[1]，它只是基于使用单个系统，该系统可以从 CT 和 MRI 中分割肝脏。例如，机器学习方法的训练和测试集将包含来自两种模式的图像，而无需明确地为模型提供相应的信息。关于此的独特研究是下面的参考，该任务是挑战中最有趣的任务之一。请记住，不同模式的单个系统的融合（即两个模型，一个在 CT 上工作，另一个在 MRI 上工作）对这一类别无效。它们可以在任务 2 和 3 中作为单独的系统进行评估。另一方面，在此任务中，允许在 MR 序列之间融合单独的系统（即两个模型，一个在 T1-DUAL 上工作，另一个在 T2-SPIR 上工作）是允许的. 2）肝脏分割（仅限CT）：这主要是CT肝脏分割的常规任务，（例如SLIVER07）。这项任务比 SLIVER07 更容易，因为它只包含沿相同方向和患者位置排列的健康肝脏。然而，具有挑战性的部分是由于注射造影剂而增强的血管结构（门相）。 3）肝脏分割（仅MRI）：类似于“任务2”，这也是MRI肝脏分割的常规任务。它包括两种不同的脉冲序列：T1-DUAL 和 T2-SPIR。此外，T1-DUAL 有两种形式（进相和出相）。开发的系统应该在这两个序列上工作。在此任务中，允许在 MR 序列之间融合单个系统（即两个模型，一个在 T1-DUAL 上工作，另一个在 T2-SPIR 上工作）。 4) 腹部器官的分割（CT & MRI）：这项任务是任务 1 在 MRI 数据中对肾脏和脾脏的扩展。在这个任务中，有趣的部分是 CT 数据集只有肝脏，但 MRI 数据集有四个带注释的腹部器官（肝脏、肾脏、脾脏）。请记住，针对不同模式（即两个模型，一个在 CT 上工作，另一个在 MRI 上工作）的单个系统的融合对于此类别无效。另一方面，在此任务中，允许在 MR 序列之间融合单个系统（即两个模型，一个在 T1-DUAL 上工作，另一个在 T2-SPIR 上工作）。 5）腹部器官的分割（仅MRI）：与“任务3”中给出的相同任务，但扩展到四个腹部器官；肝、肾、脾。在此任务中，允许在 MR 序列之间集成或融合单个系统（即两个模型，一个在 T1-DUAL 上工作，另一个在 T2-SPIR 上工作）。 [1] Valindria, V. 等人。 （2018 年 3 月）。来自未配对图像的多模态学习：在 CT 和 MRI 中的多器官分割中的应用。 2018 年 IEEE 计算机视觉应用冬季会议 (WACV)（第 547-556 页）。 IEEE。 https://doi.ieeecomputersociety.org/10.1109/WACV.2018.00066

extbf{CHAOS Challenge} aims to segment abdominal organs (liver, kidneys and spleen) from CT and MRI data. The in-person session of CHAOS was held on April 11, 2019 at the IEEE International Symposium on Biomedical Imaging (ISBI) in Venice, Italy. Online submissions are still welcome!

extbf{Challenge Description}
Understanding the prerequisites of complex medical procedures plays a critical role in surgical success. To enhance the level of comprehension, clinicians utilize advanced tools such as 3D visualization and printing, which require extracting regions of interest from DICOM images. Therefore, accurate segmentation of abdominal organs (i.e., liver, kidneys and spleen) is essential for clinical workflows including but not limited to pre-operative evaluation of living donor liver transplantation or detailed analysis of abdominal organs. Identifying the blood vessels originating from and entering abdominal organs prior to abdominal aortic surgery enables proper graft placement. This has spurred ongoing research to achieve superior segmentation results and overcome numerous challenges arising from the highly flexible anatomical characteristics of the abdomen and modality-specific limitations inherent in image features.

Against this background, the proposed challenge has two independent but related core objectives: 1) Segment the liver from computed tomography (CT) datasets acquired in the portal venous phase following contrast agent administration, for pre-operative evaluation of living liver transplant donors. 2) Segment four abdominal organs (i.e., liver, spleen, right kidney and left kidney) from magnetic resonance imaging (MRI) datasets acquired using two distinct sequences: T1-DUAL and T2-SPIR. The CHAOS tasks encompass a combination of these organ segmentation tasks.

extbf{Tasks}
Participating teams may select from five competition categories and submit their results:
1. Liver Segmentation (CT and MRI): Also known as "cross-modality" [1], this task requires a single unified system capable of segmenting the liver from both CT and MRI scans. For instance, the training and test datasets for machine learning models will include images from both modalities without explicitly providing modality information to the model. The unique research referenced below exemplifies work in this area, making this one of the most intriguing tasks of the challenge. Please note that fusing separate single-modality systems (i.e., two models, one optimized for CT and the other for MRI) is not permitted for this category. Such systems may instead be evaluated individually in Tasks 2 and 3. Conversely, fusing single systems across MR sequences (i.e., two models, one for T1-DUAL and the other for T2-SPIR) is allowed in this task.
2. Liver Segmentation (CT Only): This is a conventional CT liver segmentation task (e.g., the SLIVER07 challenge). This task is easier than SLIVER07, as it only includes healthy livers aligned to the same orientation and patient positioning. However, the challenge lies in the enhanced vascular structures (portal phase) resulting from contrast agent injection.
3. Liver Segmentation (MRI Only): Similar to Task 2, this is a conventional MRI liver segmentation task. It involves two different pulse sequences: T1-DUAL and T2-SPIR. Additionally, T1-DUAL has two variants: in-phase and out-of-phase. The developed system must perform well on both sequences. Fusing single systems across MR sequences (i.e., two models, one for T1-DUAL and the other for T2-SPIR) is permitted for this task.
4. Abdominal Organ Segmentation (CT & MRI): This task extends Task 1 to include kidney and spleen segmentation on MRI data. A key nuance of this task is that CT datasets only contain liver annotations, while MRI datasets include four annotated abdominal organs (liver, spleen, kidneys). Please note that fusing separate single-modality systems (i.e., two models, one optimized for CT and the other for MRI) is not permitted for this category. Conversely, fusing single systems across MR sequences (i.e., two models, one for T1-DUAL and the other for T2-SPIR) is allowed in this task.
5. Abdominal Organ Segmentation (MRI Only): This task extends the scope of Task 3 to include four abdominal organs: liver, spleen, and both kidneys. Integrating or fusing single systems across MR sequences (i.e., two models, one for T1-DUAL and the other for T2-SPIR) is permitted for this task.

[1] Valindria, V. et al. (March 2018). Multimodal learning from unpaired images: Application to multi-organ segmentation in CT and MRI. *2018 IEEE Winter Conference on Applications of Computer Vision (WACV)*, pp. 547-556. IEEE. https://doi.ieeecomputersociety.org/10.1109/WACV.2018.00066

AI搜集汇总

背景与挑战

背景概述

CHAOS数据集是一个用于健康腹部器官分割的医学图像数据集，结合了CT和MRI两种模态，专注于肝脏、肾脏和脾脏的分割任务。该数据集旨在支持研究和挑战赛，特别是用于临床程序如活体供体移植手术的预评估，具有多模态融合和单模态分割的多种任务设置。