MEETI

MEETI（MIMIC-IV-Ext ECG-Text-Image）是一个大规模多模态心电图数据集，旨在促进心血管人工智能研究。该数据集基于MIMIC-IV-ECG数据库，包含约800,000条10秒12导联心电图记录，覆盖160,597名患者。MEETI独特之处在于同步整合了四种数据模态：(1) 原始ECG信号（500Hz采样率），(2) 高分辨率绘制的心电图图像（100dpi，标准临床布局），(3) 通过FeatureDB工具包提取的每搏参数（如P波、QRS波群、T波的时程和振幅），(4) 由GPT-4生成的结构化文本解读。数据集采用统一标识符关联各模态数据，支持跨模态对齐分析。MEETI为开发可解释的多模态Transformer模型提供了完整框架，适用于心律失常检测、心肌缺血诊断等心血管AI任务，同时支持心电图教学和临床决策支持系统的开发。数据以标准化的MAT和PNG格式存储，按患者ID分层组织，并附带详细的元数据文件。MEETI (MIMIC-IV-Ext ECG-Text-Image) is a large-scale multimodal electrocardiogram (ECG) dataset designed to advance cardiovascular artificial intelligence (AI) research. Built on the MIMIC-IV-ECG database, it contains approximately 800,000 10-second 12-lead ECG records spanning 160,597 unique patients. What makes MEETI unique is its synchronous integration of four data modalities: (1) Raw ECG signals sampled at 500 Hz, (2) High-resolution ECG images (100 dpi, standard clinical layout), (3) Beat-by-beat parameters extracted via the FeatureDB toolkit (e.g., duration and amplitude of P waves, QRS complexes, and T waves), and (4) Structured textual interpretations generated by GPT-4. All modalities in the dataset are linked via uniform identifiers, enabling cross-modal alignment analysis. MEETI provides a comprehensive framework for developing interpretable multimodal Transformer models, applicable to cardiovascular AI tasks such as arrhythmia detection and myocardial ischemia diagnosis. It also supports the development of ECG teaching materials and clinical decision support systems. The dataset is stored in standardized MAT and PNG formats, stratified by patient ID, and accompanied by detailed metadata files.

搜集汇总

背景与挑战

背景概述

MEETI是一个基于MIMIC-IV-ECG的大规模多模态心电图数据集，包含约80万条12导联心电图记录，覆盖超过16万名患者。其核心特点是同步整合了原始ECG信号、高分辨率图像、每搏参数和GPT-4生成的结构化文本解读四种数据模态，通过统一标识符关联支持跨模态分析，旨在促进心血管人工智能研究，适用于心律失常检测、心肌缺血诊断等任务以及教学和临床决策支持。