ACDC(Automated Cardiac Diagnosis Challenge)

自动心脏诊断挑战 (ACDC) 挑战的目标是：比较自动方法在将左心室心内膜和心外膜分割为舒张末期和收缩末期实例的右心室心内膜方面的性能；比较自动方法对五类检查（正常病例、心力衰竭伴梗死、扩张型心肌病、肥厚型心肌病、右心室异常）的分类性能。整个 ACDC 数据集是根据在第戎大学医院获得的真实临床检查创建的。获得的数据完全匿名，并根据第戎医院（法国）当地伦理委员会制定的规定进行处理。我们的数据集涵盖了几个定义明确的病理学，并有足够的案例来 (1) 正确训练机器学习方法和 (2) 清楚地评估从电影 MRI 获得的主要生理参数的变化（特别是舒张期容积和射血分数）。该数据集由 150 个检查（全部来自不同的患者）组成，分为 5 个均匀分布的亚组（4 个病理组和 1 个健康受试者组），如下所述。此外，每位患者都附带以下附加信息：体重、身高以及舒张期和收缩期瞬间。该数据库在个人注册后通过专用在线评估网站的两个数据集提供给参与者：i) 100 名患者的培训数据集以及基于一位临床专家分析的相应手册参考； ii) 由 50 名新患者组成的测试数据集，没有手动注释，但有上面给出的患者信息。原始输入图像通过 Nifti 格式提供。

The goal of the Automated Cardiac Diagnosis Challenge (ACDC) is two-fold: firstly, to compare the performance of automated methods in segmenting the left ventricular endocardium and epicardium, as well as the right ventricular endocardium, for end-diastolic and end-systolic instances; secondly, to compare the classification performance of automated methods across five categories of examinations: normal cases, heart failure with infarction, dilated cardiomyopathy, hypertrophic cardiomyopathy, and right ventricular abnormalities. The entire ACDC dataset was constructed from real clinical examinations performed at the University Hospital of Dijon. The acquired data was fully anonymized and processed in accordance with the regulations formulated by the local ethics committee of Dijon Hospital (France). Our dataset covers several well-defined pathologies, with sufficient cases to (1) properly train machine learning methods and (2) clearly evaluate the variations in major physiological parameters derived from cine MRI, specifically end-diastolic volume and ejection fraction. The dataset comprises 150 examinations, all from distinct patients, divided into five evenly distributed subgroups (four pathological groups and one healthy subject group), as detailed below. Furthermore, each patient is accompanied by the following additional information: body weight, height, and the end-diastolic and end-systolic time points. This database is made available to participants via a dedicated online evaluation website after personal registration, with two datasets provided: i) a training dataset of 100 patients with corresponding manual references based on the analysis of a single clinical expert; ii) a test dataset consisting of 50 new patients, without manual annotations but with the patient information provided above. The original input images are provided in Nifti format.

搜集汇总

背景与挑战

背景概述

ACDC数据集是一个用于自动心脏诊断挑战的生物医学数据集，专注于心脏MRI图像分割和疾病分类。它包含150个真实临床患者检查，分为训练集和测试集，支持左心室、右心室分割及五类心脏病理分类，旨在评估机器学习方法在心脏生理参数计算中的性能。数据集由多所大学和医院于2018年发布，提供Nifti格式图像和患者临床信息。