The CACHET-CADB contains more than 1,000 samples of annotated ECG and contextual data collected from more than 40 patient in a free-living ambulatory setting. 

Over the last decade, advances in machine learning and deep learning have brought tremendous improvement in computer-aided arrhythmia detection capabilities. However, the generalizability and validation of atrial fibrillation (AF) detection models under ambulatory, free-living conditions remain an open challenge. The publicly available ECG datasets are often recording in a clinic under very controlled circumstances and hence do not represent all the ECG morphological changes occurring in ambulatory conditions. The models trained and validated on such "clinical" datasets may result in false positives when applied on ECG data collected in an ambulatory setting. Moreover, the lack of knowledge of the patient's context under free-living conditions adds to the complexity of AF analysis as many confounding noise and motion artifacts mimic the arrhythmias.

To overcome these challenges and help train and validate the models that can work underambulatory conditions, we are building the CACHET Contextualized Arrhythmia Database (CACHET-CADB), which is a multi-site contextualized single-channel ECG database collected under free-living conditions.

Along with the details on data source, recording information, ECG rhythm annotations process, and the patients' clinical baseline parameters (i.e, age, gender, height, weight),CACHET-CADB also contains the users' contextual data such as activity, body position, movement accelerations, as well as patient-reported data on experienced events during the recording period and self-reported stress level, sleep quality, and food intakes during the ECG recording.

The CACHET-CADB is in the process of being annotated by multiple cardiologists and currently contains more than 1,000 10-seconds samples of four rhythms types, namely: "AF", "NSR", "noise", and "other" from more than 40 patients.

The CACHET-CADB will be made publicly available for the researchers and will be periodically updated as the new annotations become available. 


Jakob Eyvind Bardram
Head of Sections, Professor
DTU Health Tech
+45 45 25 53 11