PULSE-ECG/ECGInstruct

ECGInstruct是一个全面且大规模的指令调优数据集，专为心电图图像解读设计。该数据集中的心电图图像是从原始信号记录生成的，并包含模拟现实世界打印心电图图像的各种失真。数据集经过精心策划，来源于临床医生定义的心电图任务、原始诊断、临床报告和多种任务类型。为确保高质量，还应用了额外的检查来过滤掉得分较低的指令。ECGInstruct is a comprehensive and large-scale instruction-tuning dataset designed for ECG image interpretation. The ECG images in this dataset are generated from raw signal recordings and include a range of distortions that simulate real-world printed ECG images. ECGInstruct is carefully curated, drawing from clinician-defined ECG tasks, original diagnoses, clinical reports, and a variety of task types. Additional quality checks are applied to filter out lower-scored instructions.