According to a study published in Human Brain Mapping, multimodal imaging found 4 features (abnormalities of white matter microstructures and gray matter volume) implicated as potential biomarkers for major depressive disorder. Whereas modeling constructs performed poorly overall, these 4 markers contributed to predictive accuracy across all models and analyses.
Researchers in this study sought to identify biomarkers of major depressive disorder using multiple modalities of imaging and relating neurobiological findings to binary, ordinal, or continuous outcomes. To address heterogeneity in major depressive disorder, they also used various factors associated with depression (severity of psychic depression, motivation, anxiety, psychosis, and sleep disturbance) as outcomes.
The study cohort included 147 participants with major depressive disorder and 52 controls who received multisite, multimodal imaging using diffusion MRI and structural MRI techniques. An additional cohort of 83 individuals with major depressive disorder and 25 controls were used to externally validate the study findings. Researchers examined participants for 225 predictive features of major depressive disorder: age at diagnosis, sex, handedness, 77 white matter features (using diffusion MRI), and 145 gray matter features (using structural MRI). After feature evaluation, a select subset of 39 features was used to build predictive models for binary (major depressive disorder vs controls), ordinal (severe depression vs mild depression vs controls), and continuous outcomes (illness severity).
Researchers applied different modeling techniques, including penalized logistic regression, random forest, and support vector machine, to each classification scheme (binary, ordinal, or continuous). For binary classification, all modeling techniques demonstrated similar accuracy with misclassification rates around 26%. Specificity among all classifiers was low; however, in both binary and ordinal outcomes, specificity slightly increased among the different predictive models at the expense of decreasing sensitivity. External validation was only performed on the binary classifier and resulted in a mean 87.95% sensitivity with 32% specificity. Despite an overall low performance by the prediction models, 4 features contributed to accuracy across all models and analyses. These included 2 features identified via diffusion MRI (average fractional anisotropy in the right cuneus and left insula) and 2 features identified via structural MRI (asymmetry in the volume of the pars triangularis and the cerebellum).
A limitation of the study was excluding people with major depressive disorder who had comorbidities or who were taking medication, preventing the findings from representing all major depressive disorder cases. Despite accounting for the heterogeneous nature of depression, there may be multiple biological pathways that manifest similarly, confounding the study results. Finally, the brain atlas used for analysis may have limited the study as other available atlases use finer parcellations of regional effects, yet are far more complex.
Overall, the observed model performances were too low for clinical application. However, the study revealed that across all analyses, 4 clinically relevant features were implicated: average fractional anisotropy in the right cuneus and left insula and asymmetry in the volume of the pars triangularis and the cerebellum. These features should be analyzed in future studies as potential neurobiomarkers of major depressive disorder.
This study was supported by grants from the National Institute of Mental Health, the Stony Brook School of Medicine and the Office of the Vice President for Research, Targeted Research Opportunity Program – FUSION Award, and the National Center for Advancing Translational Sciences.
Yang J, Zhang M, Ahn H, et al. Development and evaluation of a multimodal marker of major depressive disorder [published online August 16, 2018]. Hum Brain Mapp. doi: 10.1002/hbm.24282