Time Labeled Visibility Graph for Privacy-Preserved Physiological Time Series Classification

With the development of artificial intelligence (AI) and the continuous improvement of medical informatization, health assessment and auxiliary diagnosis based on physiological time series has become a hot research topic. However, the direct use of raw time series data is inappropriate due to privacy protection regulations in medial scenarios. Therefore, we designed a privacy-preserved framework based on Visibility Graph (VG) transformation and Graph Neural Network (GNN) for physiological data multi-classification. In particular, we proposed a Time Labeled Visibility Graph (TLVG), which uses the idea of VG to protect privacy while retaining more information that is useful for classification. Experiments are conducted based on the ECG5000 electrocardiogram dataset of the UCR time series classification archive. The comparison with existing classic and transformation-based classifiers shows the effectiveness and stable performance of our proposed method, providing an alternative and reasonable way for diseases diagnosis decision supports. Furthermore, from this research, it is discovered that the time sequence of each node in VG is an important feature in time series classification tasks.