ECG-Grounding dataset

ECG-Grounding provides more accurate, holistic, and evidence-driven interpretations with diagnoses grounded in measurable ECG features. Currently, it contains 30,000 instruction pairs annotated with heartbeat-level physiological features. This is the first high-granularity ECG grounding dataset, enabling evidence-based diagnosis and improving the trustworthiness of medical AI. We will continue to release more ECG-Grounding data and associated beat-level features progressively.

Project Page: 📖 Page

Code: 💻 Code

Paper: 📄 Arxiv

Model: 🤗 GEM

Data: 🤗 ECG-Grounding

File descriptions

├── ecg_jsons
    └── ECG_Grounding_30k: ECG_Instruct plus the grounding_train_30k (training data for GEM-7B)
    └── grounding_train_30k: the proposed 30k high-granularity instruction data for ECG grounding
    └── ecg-grounding-test-mimiciv: test data from MIMIC-IV database for Grounded ECG Understanding task
    └── ecg-grounding-test-ptbxl: test data from PTB-XL database for Grounded ECG Understanding task

├── ecg_images
    └── gen_images
        └── mimic_gen: 30k ECG images for grounding_train_30k
        └── ptb-xl-gen-images: generated ECG images for ecg-grounding-test-ptbxl

├── ecg_bench
    └── ECG_Bench files with associated ecg paths