Abstract

Schizophrenia is a chronic and debilitating illness that includes a wide range of emotional, social, and cognitive disorders associated with impaired performance. It affects more than 21 million people in the world, with a 2 to 3 times higher mortality rate and a 10 to 20 years shorter life expectancy than a healthy person. There is evidence that an early diagnosis substantially improves clinical outcomes, but this a complex task, since there is no single symptom that is exclusive to this severe mental illness. Currently, the accurate diagnosis of this type of disorder could take a 6-month minimum and is based mainly on interviews and the existence of some observable and repetitive behavior or symptom as an indicator. In this study, we propose a computer-assisted diagnosis for the detection of schizophrenia in 3 min, which is obtained after the processing of brain micro-states, the brain signal to process is acquired by an electroencephalogram (EEG). The methodology relies heavily on the processing of brain micro-states through a trajectory of successive random steps in time (that is, treated through a random walk built on the basis of previously analyzed studies). Also, we used a Convolutional Neural Network (CNN) as a deep learning method that allowed the exploration and automatic extraction of the main characteristics of the random walk. The EEG data was taken from an open-source repository made up of 28 recordings from 14 healthy subjects and 14 patients with schizophrenia. The results provide an automated method and a substantial improvement for the classification of people with schizophrenia, obtaining a 93% success rate from 3 min of signal capture. Finally, three possible future challenges are open related to improve the classifier, the optimization of electrodes, and the development of a device for the schizophrenia diagnosis.