Published in Applications of Artificial Intelligence and Neural Systems to Data Science, Smart Innovation, Systems and Technologies, Springer, pages 17-27 (proceedings of WIRN 2022).
The paper proposes three ways to extract topological features from ECG signals and classify heartbeats with a graph isomorphism network, a type of graph neural network. On the PTB Diagnostic dataset the three methods reached 99.38%, 98.76% and 91.93% accuracy.
Authors: Kamyar Zeinalipour, Marco Gori.