Four-Class Classification of Neuropsychiatric Disorders by Use of Functional Near-Infrared Spectroscopy Derived...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Current neuropsychiatric disorder diagnosis relies heavily on subjective measures, impacting clinical decision reliability.
• Objective diagnostic biomarkers are needed to improve accuracy and consistency in identifying neurological and psychiatric conditions.

Purpose of the Study:

• To develop and evaluate a machine learning-based classification method for the objective diagnosis of three neuropsychiatric disorders.
• To utilize functional near-infrared spectroscopy (fNIRS) derived hemodynamic and cognitive features for automated classification.

Main Methods:

• Functional near-infrared spectroscopy (fNIRS) monitored prefrontal cortex hemodynamics in 13 healthy adolescents and 67 patients (migraine, OCD, schizophrenia) during a Stroop task.
• Hemodynamic and cognitive features were extracted and used to train three supervised learning algorithms: naïve Bayes (NB), linear discriminant analysis (LDA), and support vector machines (SVM).
• Algorithm performance was assessed using a ten-fold cross-validation procedure for four-class classification.

Main Results:

• All tested algorithms achieved classification accuracies above 81% and specificities above 94%.
• Support vector machines (SVM) demonstrated the highest performance with an accuracy of 85.1 ± 1.77%, sensitivity of 84 ± 1.7%, specificity of 95 ± 0.5%, precision of 86 ± 1.6%, and F1-score of 85 ± 1.7%.
• fNIRS-derived features, when used in automated classification, are free from subjective report bias.

Conclusions:

• The proposed machine learning approach using fNIRS shows significant potential for the objective diagnosis of neuropsychiatric disorders.
• This methodology can assist in diagnosing conditions associated with frontal lobe dysfunction, improving clinical decision-making reliability.