In this project, we are specifically developing early detection and prognosis of sporadic atrial fibrillation and ventricular tachycardia based on long-term ECG data (Holter). This can not only save lives, but also prevent serious health implications and simplify therapies and improve their outcomes. In addition, we are developing algorithms for measuring vital body signals such as blood pressure and respiration rate, as well as detecting sleep apnea directly from ECG measurements without additional sensors. These serve cardiologists for a more comprehensive assessment. No additional sensors are needed for the patient.
For early detection and prognosis of sporadic atrial fibrillation and ventricular tachycardia, we are investigating and developing algorithms based on deep neural networks, which have shown to detect traces in the ECG as predictive markers of arrhythmia during normal sinus rhythm.
We try to capture the person-specific and especially gender-specific differences/variations in the ECG signal, i.e. signs of myocardial infarction in the ECG, with the AI model and thereby enable more accurate early detections and prognoses. In concrete terms, this means that we train the AI with data from a wide variety of individuals (women/men) so that the AI can cover person-specific variations.
For the measurement of the vital signals blood pressure and respiration rate, we are investigating approaches based on feature extraction and classical machine learning. All algorithms are developed for long-term ECG measurements (1..3 leads) with a duration of 1..10 days. Arrhythmia detection is optimized taking into account the patient's activities detected by accelerometers in the ECG device.