Improving the safety of atrial fibrillation monitoring systems through human verification

In this paper we propose a hybrid decision-making process for medical diagnosis. The hypothesis tested is that a deep learning system can provide real-time monitoring of Atrial Fibrillation (AF), a prevalent heart arrhythmia, and a human cardiologist will then verify the results and reach a diagnosis. The verification step adds the necessary checks and balances to increase the safety of the computer-based diagnostic process. In order to test hybrid-decision making, we created a prototype AF monitoring service. The service is based on Heart Rate (HR) sensors for signal acquisition as well as Internet of Things (IoT) technology for data communication and storage. These technologies enable transfer of HR data from patient to central cloud server. A deep learning system is used to analyze the data, which is then presented to a cardiologist when a dangerous condition is detected. This human specialist then works to verify the deep learning results based on the HR data and additional knowledge obtained through patient records or by personal interaction with the patient. A prerequisite for safety in any computer expert system is the clarity of purpose for the decision-making process. Health-care providers are considered customers who register patients with the AF monitoring service. The service delivers real-time diagnostic support by providing timely alarm messages and HR analysis. The safety critical decision then lies with the human practitioner.