Maximum Credibility Voting (MCV) An Integrative Approach for Accurate Diagnosis of Major Depressive Disorder from Clinically Readily Available Data

Abstract

Diagnosis of Major Depressive Disorder (MDD) is currently a lengthy procedure, due to the low diagnostic accuracy of clinically readily available biomarkers. We integrate predictions from multiple datasets based on a credibility parameter defined on the probabilistic distributions of individual sparse prediction models. We demonstrate by means of structural and resting-state functional Magnetic Resonance Imaging and blood markers obtained from 62 treatment naive MDD patients (age 40.63±9.28, 36 female, HRSD 20.03±4.94) and 66 controls without mental disease history (age 35.52±12.91, 30 female), that our method called Maximum Credibility Voting (MCV) significantly increases diagnostic accuracy from about 65% average classification accuracy of individual biomarker models to 80% (accuracy after integration of the models). Classification results from different combinations of the available datasets validate the method's stability with respect to redundant or contradictory predictions. By definition, MCV is applicable to any desired data and compatible with missing values, ensuring continued improvement of diagnostic accuracy and patient comfort as new data acquisition methods and markers emerge.