Using machine learning to reveal two distinct neuroanatomical subtypes of first-episode, drug-naïve major depressive...

Major depressive disorder (MDD) is a highly heterogeneous condition, complicating biomarker discovery and precision medicine. Identifying biologically distinct subtypes using structural MRI (sMRI) offers a promising approach to address this heterogeneity. This study employed sMRI features to define neuroimaging-based subtypes in first-episode, drug-naïve (FEDN) MDD patients.

In this study, we analyzed T1-weighted anatomical images from 169 first-episode, drug-naïve (FEDN) MDD patients and 169 healthy controls (HCs) obtained from the rest-meta-MDD project. Patient symptom severity was assessed using the 17-item Hamilton Depression Rating Scale (HAMD-17) and its subscales. We employed region-specific gray matter volume (GMV) feature-based heterogeneity through discriminant analysis (HYDRA) to explore neuroanatomical subtypes of FEDN MDD patients and validate their stability. Furthermore, we examined demographic and symptomatic differences between identified subtypes.

We identified two distinct neuroanatomical subtypes of FEDN MDD patients (FEDN MDD 1: n = 85, FEDN MDD 2: n = 84), which exhibited significant differences in GMV alterations. Compared with HCs, FEDN MDD 1 showed widespread GMV increases, while FEDN MDD 2 demonstrated significant GMV reductions. These two subtypes also demonstrated significant differences in HAMD anxiety/somatization subscale scores (t = 2.845, p < 0.01) and age distribution (t = 3.886, p < 0.001). Furthermore, reproducibility analyses confirmed the robustness of these subtypes.

Our findings revealed two clinically significant neuroanatomical subtypes of MDD, providing new insights into the neurobiological heterogeneity of this disorder. These findings may serve as a valuable reference for future precision diagnosis and treatment strategies.