Integrating functional neuroimaging and serum proteins improves the diagnosis of major depressive disorder.

ABSTRACT

BACKGROUND: The lack of effective objective diagnostic biomarkers for major depressive disorder (MDD) leads to high misdiagnosis. Compared with healthy controls (HC), abnormal brain functions and protein levels are often observed in MDD. However, it is unclear whether combining these changed multidimensional indicators could help improve the diagnosis of MDD. METHODS: Sixty-three MDD and eighty-one HC subjects underwent resting-state fMRI scans, among whom 37 MDD and 45 HC provided blood samples. Amplitudes of low-frequency fluctuation (ALFF), regional homogeneity (ReHo), and serum levels of brain-derived neurotrophic factor (BDNF), cortisol, and multiple cytokines were measured and put into the linear discriminant analysis (LDA) to construct corresponding MDD diagnostic models. The area under the receiver operating characteristic curve (AUC) of 5-fold cross-validation was calculated to evaluate each model's performance. RESULTS: Compared with HC, MDD patients' spontaneous brain activity, serum BDNF, cortisol, interleukin (IL)-4, IL-6, and IL-10 levels changed significantly. The combinations of unidimensional multi-indicator had better diagnostic performance than a single one. The model consisted of multidimensional multi-indicator further exhibited conspicuously superior diagnostic efficiency than those constructed with unidimensional multi-indicator, and its AUC, sensitivity, specificity, and accuracy of 5-fold cross-validation were 0.99, 92.0 %, 100.0 %, and 96.3 %, respectively. LIMITATIONS: This cross-sectional study consists of relatively small samples and should be replicated in larger samples with follow-up data to optimize the diagnostic model. CONCLUSIONS: MDD patients' neuroimaging features and serum protein levels significantly changed. The model revealed by LDA could diagnose MDD with high accuracy, which may serve as an ideal diagnostic biomarker for MDD.