IKEM dataset v1.0.0

<strong>IKEM Dataset:</strong> <strong>Number of ECGs (Electrocardiograms):</strong> 98,130 <strong>Number of Patients:</strong> 30,290 <strong>Size:</strong> 6.3 GB The IKEM dataset consists of 12-lead ECG recordings collected from patients who were examined by cardiology or diabetology sections at the Institute for Clinical and Experimental Medicine (IKEM) in Prague, Czech Republic. These recordings were sampled at a rate of 500 Hz for a duration of 10 seconds each. Each entry in the dataset includes raw ECG recordings paired with an anonymized unique patient ID, which groups ECG recordings belonging to the same patient. The dataset is stored in HDF5 files using 16-bit integers with a granularity of 4.88 microvolts. Redundant augmented leads and lead III have been removed to save storage space, resulting in a size decrease of approximately 60%. When inputting the signal to the model, the 8 reduced leads are expanded back to the original 12, producing 12 voltage values for each time instance. <strong>Structure:</strong> HDF5 files each "row" in each file corresponds to exactly 1 ECG recording each file contains 4 columns: exam_id = a unique ID for the ECG recording hashes = SHA1 hash of the ECG recording real_lengths = original length of the ECG recording tracings = a matrix (4096, 8) containing the reduced form of the ECG recording exams.csv metadata for all recordings contained in the HDF5 files 9 columns, 7 contain auxiliary metadata, the remaining 2 important are: exam_id = a unique ID for the ECG recording (corresponds to exam_id in HDF5 files) patient_id = unique patient ID (label for patient identification) We advise you to use our <strong>HDF5 multi-file loader </strong>when working with this dataset. For further information about the context of this dataset, please read our original publication and cite it if you find this dataset or our work useful, thank you very much.