Backgrounds: Clinical studies to date have yet to establish distinct boundaries between depression in bipolar disorder (BD) and unipolar depression (UD), leading to misdiagnoses and even exacerbation of the conditions. This study aimed to explore the distinctions in the local gyrification index (LGI) between BD and UD, and to evaluate its potential diagnostic value as a biomarker.
Methods: LGI values across 68 cortical regions were measured from 42 patients with BD, 45 patients with UD, and 45 healthy controls (HCs) based on the Desikan-Killiany atlas. General linear model was performed to compare LGI values among the three groups. XGBoost classifier was implemented to develop a binary classification model for distinguishing BD from UD. Additionally, the correlation between clinical characteristics and LGI values was investigated separately within the BD and UD groups.
Results: Compared to HCs, individuals with BD and UD exhibited significantly reduced LGI values in various cortical regions. Nine LGI regions in the BD group displayed reduced values compared to the UD group, except for a singular increase in the left frontal pole (ηp2 = 0.173; P = 0.006). No significant association was found between LGI values and clinical characteristics within the patient groups. The XGBoost classifier achieved a distinction accuracy of 73.7 % between BD and UD, with the left frontal pole making the most significant contribution to the model.
Conclusions: The findings suggest that LGI could be a relatively stable neuroimaging biomarker for distinguishing between BD and UD.
Keywords: Bipolar disorder; Local gyrification index; MRI; Machine learning; Major depressive disorder.
Copyright © 2024. Published by Elsevier B.V.