The current conflicting neuroimaging findings of insomnia disorder (ID) may be attributed to heterogeneity in ID. The present study aims to clarify the high heterogeneity in ID and examine the objective neurobiological subtypes of ID by using a novel machine learning method based on gray matter volumes (GMVs). We recruited 56 patients with ID and 73 healthy controls (HCs). The T1-weighted anatomical images were obtained for each participant. We investigated whether the ID has higher interindividual heterogeneity in GMVs. Then, we used a heterogeneous machine learning algorithm by discriminative analysis (HYDRA) to identify subtypes of ID with features of brain regional GMVs. We found that patients with ID have higher interindividual variability than HCs. HYDRA identified two distinct and reliable neuroanatomical subtypes of ID. Two subtypes showed significantly different aberrance in GMVs compared with HCs. Specifically, subtype 1 exhibited widespread decreased GMVs in some brain regions, including the right inferior temporal gyrus, left superior temporal gyrus, left precuneus, right middle cingulate, and right supplementary motor area. Subtype 2 only demonstrated increased GMVs in the right superior temporal gyrus. Additionally, the GMVs of altered brain regions in subtype 1 were significantly correlated with daytime functioning, but in subtype 2, they were significantly correlated with sleep disturbance. These results explain conflicting neuroimaging findings and propose a potential objective neurobiological classification contributing to ID's precise clinical diagnosis and treatment.
Keywords: Gray matter volume; Heterogeneity; Insomnia; Machine learning; Subtypes.
Copyright © 2023 Elsevier B.V. All rights reserved.