Usefulness of approximate entropy in the diagnosis of schizophrenia

Diagnosis of the psychiatric diseases is a bit challenging at the first interview due to this fact that qualitative criteria are not as accurate as quantitative ones. Here, the objective is to classify schizophrenic patients from the healthy subject using a quantitative index elicited from their electroencephalogram (EEG) signals.

Ten right handed male patients with schizophrenia who had just auditory hallucination and did not have any other psychotic features and ten age-matched right handed normal male control participants participated in this study. The patients used haloperidol to minimize the drug-related affection on their EEG signals. Electrophysiological data were recorded using a Neuroscan 24 Channel Synamps system, with a signal gain equal to 75K (150 xs at the headbox). According to the observable anatomical differences in the brain of schizophrenic patients from controls, several discriminative features including AR coefficients, band power, fractal dimension, and approximation entropy (ApEn) were chosen to extract quantitative values from the EEG signals.

The extracted features were applied to support vector machine (SVM) classifier that produced 88.40% accuracy for distinguishing the two groups. Incidentally, ApEn produces more discriminative information compare to the other features.

This research presents a reliable quantitative approach to distinguish the control subjects from the schizophrenic patients. Moreover, other representative features are implemented but ApEn produces higher performance due to complex and irregular nature of EEG signals.