Artificial Intelligence-enabled Electrocardiogram Analyses for the Development of Novel Clinical Screening Tools

The non-invasive biomarker electrocardiogram (ECG) is a direct representation of the electrophysiological activity of the heart, which can be affected by numerous cardiovascular diseases. While in many cases these changes can be interpreted by a medical professional, the early disease can lead to more subtle alterations of the physiological ECG, being little or not recognizable to physicians. Yet, especially the early identification of patients at risk permits an optimized personal health care and a better utilization of health care resources. Training Convolutional Neural Networks (CNN) enables the identification of subtle differences in ECGs by analyzing large amounts of data, potentiating the utility of this long-standing diagnostic modality. ECGs are standardized, cheap, easy to obtain from patients, and are already abundantly available in many databases, making them especially suitable for CNNs. Using cutting-edge CNN methods and combining data from the large population-based cohort studies Hamburg City Health Study (HCHS), Study of Health in Pomerania (SHIP), and Gutenberg Health Study (GHS), we aim to develop novel screening tools for several cardiovascular risk factors and diseases. Ultimately, this could lead to the development of instruments which can be used by a physician in medical practice. Only through a collaboration in expertise and data, a goal like this is achievable.