TY - JOUR

T1 - Application of Machine Learning to Diagnostics of Schizophrenia Patients Based on Event-Related Potentials

AU - Шанарова, Надежда Леонидовна

AU - Пронина, Марина Владимировна

AU - Липкович, Михаил

AU - Пономарев, Валерий

AU - Muller, Andreas

AU - Кропотов, Юрий Дмитриевич

PY - 2023/1/30

Y1 - 2023/1/30

N2 - Schizophrenia is a major psychiatric disorder that significantly reduces the quality of life. Early treatment is extremely important in order to mitigate the long-term negative effects. In this paper, a machine learning based diagnostics of schizophrenia was designed. Classification models were applied to the event-related potentials (ERPs) of patients and healthy subjects performing the visual cued Go/NoGo task. The sample consisted of 200 adult individuals ranging in age from 18 to 50 years. In order to apply the machine learning models, various features were extracted from the ERPs. The process of feature extraction was parametrized through a special procedure and the parameters of this procedure were selected through a grid-search technique along with the model hyperparameters. Feature extraction was followed by sequential feature selection transformation in order to prevent overfitting and reduce the computational complexity. Various models were trained on the resulting feature set. The best model was support vector machines with a sensitivity and specificity of 91% and 90.8%, respectively.

AB - Schizophrenia is a major psychiatric disorder that significantly reduces the quality of life. Early treatment is extremely important in order to mitigate the long-term negative effects. In this paper, a machine learning based diagnostics of schizophrenia was designed. Classification models were applied to the event-related potentials (ERPs) of patients and healthy subjects performing the visual cued Go/NoGo task. The sample consisted of 200 adult individuals ranging in age from 18 to 50 years. In order to apply the machine learning models, various features were extracted from the ERPs. The process of feature extraction was parametrized through a special procedure and the parameters of this procedure were selected through a grid-search technique along with the model hyperparameters. Feature extraction was followed by sequential feature selection transformation in order to prevent overfitting and reduce the computational complexity. Various models were trained on the resulting feature set. The best model was support vector machines with a sensitivity and specificity of 91% and 90.8%, respectively.

KW - electroencephalogram; event-related potential; schizophrenia; machine learning models; support vector machine

KW - electroencephalogram

KW - event-related potential

KW - machine learning models

KW - schizophrenia

KW - support vector machine

UR - https://www.mendeley.com/catalogue/6af1acc8-caeb-3bb1-9534-739e5e35c870/

U2 - https://doi.org/10.3390/diagnostics13030509

DO - https://doi.org/10.3390/diagnostics13030509

M3 - Article

C2 - 36766614

VL - 13

JO - Diagnostics

JF - Diagnostics

SN - 2075-4418

IS - 3

ER -