Heart Attack Outcome Predictions Using FMM Models

As part of the PhysioNet Challenge 2023, our team FM-MGroup.Uva presents an original approach for the prediction of the outcome of coma patients after a heart attack. Our methodology involves the integration of two types of EEG features extracted from 10-second epoch data, analyzed at various time intervals with patient clinical data. The first type is the FMM features created from the parameters of an FMM (Frequency Modulated Möbius) model fitted to the epoch data. The other type is indices taken from the literature that include spectral, entropy, and background measures. The mean feature values for patients are combined with clinical variables into classification models to obtain an outcome score in a given time block. Our best performance in the official challenge phase achieved a score of 0.45 at 72 hours on the test dataset. Additionally, we introduce an alternative proposal in this paper, displaying promising results in our laboratory. Unfortunately, this proposal was not ranked due to difficulties in submitting it on time without errors.