Conception of an AI-based recommender system for labeling ballistocardiographic data

Abstract

This paper proposes a tool for assisting in the labeling ballistocardiographic data, as such data is rare. There are not many databases and those that we found are barely labeled or not at all. As such, we are developing an ai-based system that can reliably recognize anomalous areas and display them for evaluation. The system is supposed to detect multiple types of anomalies (point, collective and contextual), which can then be rejected, modified and labeled by medical professionals or researchers. For initial testing, the AER-network [1] was used on the database recorded by Carlson et al. [2] and preliminary results show that it is able to detect anomalous areas. While our system is still in its initial stage, further investigations lead us to conclude that such a system would be of great value to the medical community both in practice and in research.