Identifying two distinct neuroanatomical subtypes of first-episode depression using heterogeneity through discriminative analysis

Abstract

Background: Neurobiological heterogeneity in depression remains largely unknown, leading to inconsistent neuroimaging findings.

Methods: Here, we adopted a novel proposed machine learning method ground on gray matter volumes (GMVs) to investigate neuroanatomical subtypes of first-episode treatment-naïve depression. GMVs were obtained from high-resolution T1-weighted images of 195 patients with first-episode, treatment-naïve depression and 78 matched healthy controls (HCs). Then we explored distinct subtypes of depression by employing heterogeneity through discriminative analysis (HYDRA) with regional GMVs as features.

Results: Two prominently divergent subtypes of first-episode depression were identified, exhibiting opposite structural alterations compared with HCs but no different demographic features. Subtype 1 presented widespread increased GMVs mainly located in frontal, parietal, temporal cortex and partially located in limbic system. Subtype 2 presented widespread decreased GMVs mainly located in thalamus, cerebellum, limbic system and partially located in frontal, parietal, temporal cortex. Subtype 2 had smaller TIV and longer illness duration than Subtype 1. And TIV in Subtype 1 was positively correlated with age of onset while not in Subtype 2, probably implying the different potential neuropathological mechanisms.

Limitations: Despite results obtained in this study were validated by employing another brain atlas, the conclusions were acquired from a single dataset.

Conclusions: This study revealed two distinguishing neuroanatomical subtypes of first-episode depression, which provides new insights into underlying biological mechanisms of the heterogeneity in depression and might be helpful for accurate clinical diagnosis and future treatment.

Keywords: Depression; Gray matter volume; Machine learning; Neuroanatomical heterogeneity; Subtypes of depression.