Capturing the Individual Deviations From Normative Models of Brain Structure for Depression Diagnosis and Treatment

Abstract

Background

The high heterogeneity of depression prevents us from obtaining reproducible and definite anatomical maps of brain structural changes associated with the disorder, which thereafter limits the individualized diagnosis and treatment of patients. In this study, we investigated the clinical issues related to depression according to individual deviations from normative ranges of grey matter volume (GMV).

Methods

We enrolled 1,092 participants totally, including 187 depression patients and 905 healthy controls (HCs). Structural MRI of HCs from the Human Connectome Project (n=510) and REST-meta-MDD Project (n=229) were used to establish normative model across the lifespan in 18-65 years for each brain region. Deviations from normative range for 187 patients and 166 HCs, recruited from two local hospitals, were captured as normative probability maps (NPMs), which was used to identify the disease risk and treatment-related latent factors.

Results

Unlike case-control results, our normative modeling approach revealed highly individualized patterns of anatomic abnormalities in depressed patients (less than 11% extreme deviation overlapping for any regions). Based on our classification framework, models trained with individual NPMs (AUC range, 0.7146-0.7836) showed better performance than those trained with original GMV (AUC range, 0.6800-0.7036), which was verified in an independent external test set. Furthermore, different latent brain structural factors in relation to antidepressant treatment were revealed by a Bayesian model based on NPMs, suggesting distinct treatment response and inclination.

Conclusion

Capturing the personalization deviations from normative range could help understand the heterogeneous neurobiology of depression, thus contribute to guide diagnosis and treatment for depression clinically.